Chapter 11: Weather and Climate Extreme Events in a Changing Climate

CoordinatingLead Authors:

Sonia I. Seneviratne (Switzerland), Xuebin Zhang (Canada)

Lead Authors:

Muhammad Adnan (Pakistan), Wafae Badi (Morocco), Claudine Dereczynski (Brazil), Alejandro Di Luca (Australia/Canada/Argentina), Subimal Ghosh (India), Iskhaq Iskandar (Indonesia), James Kossin (United States of America), Sophie Lewis (Australia), Friederike Otto (United Kingdom/Germany), Izidine Pinto (South Africa/Mozambique), Masaki Satoh (Japan), Sergio M. Vicente-Serrano (Spain), Michael Wehner (United States of America), Botao Zhou (China)

Contributing Authors:

Mathias Hauser (Switzerland), Megan Kirchmeier-Young (Canada/United States of America), Lisa V. Alexander (Australia), Richard P. Allan (United Kingdom), Mansour Almazroui (Saudi Arabia), Lincoln M. Alvez (Brazil), Margot Bador (France, Australia/France), Rondrotiana Barimalala (South Africa/Madagascar), Richard A. Betts (United Kingdom), Suzana J. Camargo (United States of America/Brazil, United States of America), Pep G. Canadell (Australia), Erika Coppola (Italy), Markus G. Donat (Spain/Germany, Australia), Hervé Douville (France), Robert J. H. Dunn (United Kingdom/Germany,United Kingdom), Erich Fischer (Switzerland), Hayley J. Fowler (United Kingdom), Nathan P. Gillett (Canada), Peter Greve (Austria/Germany), Michael Grose (Australia), Lukas Gudmundsson (Switzerland/Germany, Iceland), José Manuel Guttiérez (Spain), Lofti Halimi (Algeria), Zhenyu Han (China), Kevin Hennessy (Australia), Richard G. Jones (United Kingdom), Yeon-Hee Kim (Republic of Korea), Thomas Knutson (United States of America), June-Yi Lee (Republic of Korea), Chao Li (China), Georges-Noel T. Longandjo (South Africa/Democratic Republic of the Congo), Kathleen L. McInnes (Australia), Tim R. McVicar (Australia), Malte Meinshausen (Australia/Germany), Seung-Ki Min (Republic of Korea), Ryan S. Padron Flasher (Switzerland/Ecuador, United States of America), Christina M. Patricola (United States of America), Roshanka Ranasinghe (The Netherlands/Sri Lanka, Australia), Johan Reyns (The Netherlands/Belgium), Joeri Rogelj (United Kingdom/Belgium), Alex C. Ruane (United States of America), Daniel Ruiz Carrascal (United States of America/Colombia), Bjørn H. Samset (Norway), Jonathan Spinoni (Italy), Qiaohong Sun (Canada/China), Ying Sun (China), Mouhamadou Bamba Sylla (Rwanda/Senegal), Claudia Tebaldi (United States of America), Laurent Terray (France), Wim Thiery (Belgium), Jessica Tierney (United States of America), Maarten K. van Aalst (The Netherlands), Bart van den Hurk (The Netherlands), Robert Vautard (France), Wen Wang (China), Seth Westra (Australia), Jakob Zscheischler (Germany)

Review Editors:

Johnny Chan (China), Asgeir Sorteberg (Norway), Carolina Vera (Argentina)

Chapter Scientists:

Mathias Hauser (Switzerland), Megan Kirchmeier-Young (Canada/United States of America), Hui Wan (Canada)

Figure 11.1

Figure 11.2

Figure 11.3

Figure 11.4

Figure 11.5

Figure 11.6

Figure 11.7

Figure 11.8

Figure 11.9

Figure 11.10

Figure 11.11

Figure 11.12

Figure 11.13

Figure 11.14

Figure 11.15

Figure 11.16

Figure 11.17

Figure 11.18

Figure 11.19

Figure 11.20

Box 11.1 Figure 1

Box 11.4 Figure 1

Box 11.4 Figure 2

Cross-Chapter Box 11.1, Figure 1

Cross-Chapter Box 11.1, Figure 2

Cross-Chapter Box 11.1, Figure 3

FAQ 11.1 Figure 1

FAQ 11.2, Figure 1

FAQ 11.3, Figure 1

This Chapter should be cited as:

Seneviratne, S.I., X. Zhang, M. Adnan, W. Badi, C. Dereczynski, A. Di Luca, S. Ghosh, I. Iskandar, J. Kossin, S. Lewis, F. Otto, I. Pinto, M. Satoh, S.M. Vicente-Serrano, M. Wehner, and B. Zhou, 2021: Weather and Climate Extreme Events in a Changing Climate. In Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change[Masson-Delmotte, V., P. Zhai, A. Pirani, S.L. Connors, C. Péan, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M.I. Gomis, M. Huang, K. Leitzell, E. Lonnoy, J.B.R. Matthews, T.K. Maycock, T. Waterfield, O. Yelekçi, R. Yu, and B. Zhou (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 1513–1766, doi: 10.1017/9781009157896.013.

Executive Summary

This chapter assesses changes in weather and climate extremes on regional and global scales, including observed changes and their attribution, as well as projected changes. The extremes considered include temperature extremes, heavy precipitation and pluvial floods, river floods, droughts, storms (including tropical cyclones), as well as compound events (multivariate and concurrent extremes). The assessment focuses on land regions excluding Antarctica. Changes in marine extremes are addressed in Chapter 9 and Cross-Chapter Box 9.1. Assessments of past changes and their drivers are from 1950 onward, unless indicated otherwise. Projections for changes in extremes are presented for different levels of global warming, supplemented with information for the conversion to emissions scenario-based projections (Cross-Chapter Box 11.1 and Table 4.2). Since the IPCC Fifth Assessment Report (AR5), there have been important new developments and knowledge advances on changes in weather and climate extremes, in particular regarding human influence on individual extreme events, on changes in droughts, tropical cyclones, and compound events, and on projections at different global warming levels (1.5°C–4°C). These, together with new evidence at regional scales, provide a stronger basis and more regional information for the AR6 assessment on weather and climate extremes.

It is an established fact that human-induced greenhouse gas emissions have led to an increased frequency and/or intensity of some weather and climate extremes since pre-industrial time, in particular for temperature extremes. Evidence of observed changes in extremes and their attribution to human influence (including greenhouse gas and aerosol emissions and land-use changes) has strengthened since AR5, in particular for extreme precipitation, droughts, tropical cyclones and compound extremes (including dry/hot events and fire weather). Some recent hot extreme events would have been extremely unlikely to occur without human influence on the climate system. {11.2, 11.3, 11.4, 11.6, 11.7, 11.8}

Regional changes in the intensity and frequency of climate extremes generally scale with global warming. New evidence strengthens the conclusion from the IPCC Special Report on Global Warming of 1.5°C (SR1.5) that even relatively small incremental increases in global warming (+0.5°C) cause statistically significant changes in extremes on the global scale and for large regions (high confidence). In particular, this is the case for temperature extremes (very likely ), the intensification of heavy precipitation (high confidence) including that associated with tropical cyclones (medium confidence), and the worsening of droughts in some regions (high confidence). The occurrence of extreme events unprecedented in the observed record will rise with increasing global warming, even at 1.5°C of global warming. Projected percentage changes in frequency are higher for the rarer extreme events (high confidence). {11.1, 11.2, 11.3, 11.4, 11.6, 11.9, Cross-Chapter Box 11.1}

Methods and Data for Extremes

Since AR5, the confidence about past and future changes in weather and climate extremes has increased due to better physical understanding of processes, an increasing proportion of the scientific literature combining different lines of evidence, and improved accessibility to different types of climate models (high confidence). There have been improvements in some observation-based datasets, including reanalysis data (high confidence). Climate models can reproduce the sign (direction) of changes in temperature extremes observed globally and in most regions, although the magnitude of the trends may differ (high confidence). Models are able to capture the large-scale spatial distribution of precipitation extremes over land (high confidence). The intensity and frequency of extreme precipitation simulated by Coupled Model Intercomparison Project Phase 6 (CMIP6) models are similar to those simulated by CMIP5 models (high confidence). Higher horizontal model resolution improves the spatial representation of some extreme events (e.g., heavy precipitation events), in particular in regions with highly varying topography (high confidence). {11.2, 11.3, 11.4}

Temperature Extremes

The frequency and intensity of hot extremes (including heatwaves) have increased, and those of cold extremes have decreased on the global scale since 1950 (virtually certain). This also applies at regional scale, with more than 80% of AR6 regions1showing similar changes assessed to be at leastlikely . In a few regions, limited evidence (data or literature) prevents the reliable estimation of trends. {11.3, 11.9}

Human-induced greenhouse gas forcing is the main driver of the observed changes in hot and cold extremes on the global scale (virtually certain ) and on most continents (very likely ). The effect of enhanced greenhouse gas concentrations on extreme temperatures is moderated or amplified at the regional scale by regional processes such as soil moisture or snow/ice-albedo feedbacks, by regional forcing from land-use and land-cover changes, or aerosol concentrations, and decadal and multi-decadal natural variability. Changes in anthropogenic aerosol concentrations have likely affected trends in hot extremes in some regions. Irrigation and crop expansion have attenuated increases in summer hot extremes in some regions, such as the Midwestern USA (medium confidence). Urbanization has likely exacerbated changes in temperature extremes in cities, in particular for nighttime extremes. {11.1, 11.2, 11.3}

The frequency and intensity of hot extremes will continue to increase and those of cold extremes will continue to decrease, at global and continental scales and in nearly all inhabited regions1 with increasing global warming levels. This will be the case even if global warming is stabilized at 1.5°C. Relative to present-day conditions, changes in the intensity of extremes would be at least double at 2°C, and quadruple at 3°C of global warming, compared to changes at 1.5°C of global warming. The number of hot days and hot nights and the length, frequency, and/or intensity of warm spells or heatwaves will increase over most land areas (virtually certain). In most regions, future changes in the intensity of temperature extremes will very likely be proportional to changes in global warming, and up to two to three times larger (high confidence). The highest increase of temperature of hottest days is projected in some mid-latitude and semi-arid regions and in the South American Monsoon region, at about 1.5 times to twice the rate of global warming (high confidence). The highest increase of temperature of coldest days is projected in Arctic regions, at about three times the rate of global warming (high confidence). The frequency of hot temperature extreme events will very likely increase nonlinearly with increasing global warming, with larger percentage increases for rarer events. {11.2, 11.3, 11.9; Table 11.1; Figure 11.3}

Heavy Precipitation andPluvial Floods

The frequency and intensity of heavy precipitation events have likely increased at the global scale over a majority of land regions with good observational coverage. Heavy precipitation has likely increased on the continental scale over three continents: North America, Europe, and Asia. Regional increases in the frequency and/or intensity of heavy precipitation have been observed with at least medium confidence for nearly half of AR6 regions, including WSAF, ESAF, WSB, SAS, ESB, RFE, WCA, ECA, TIB, EAS, SEA, NAU, NEU, EEU, GIC, WCE, SES, CNA, and ENA. {11.4, 11.9}

Human influence, in particular greenhouse gas emissions, is likely the main driver of the observed global-scale intensification of heavy precipitation over land regions. It is likely that human-induced climate change has contributed to the observed intensification of heavy precipitation at the continental scale in North America, Europe and Asia. Evidence of a human influence on heavy precipitation has emerged in some regions (high confidence). {11.4, 11.9, Table 11.1}

Heavy precipitation will generally become more frequent and more intense with additional global warming. At a global warming level of 4°C relative to the pre-industrial level, very rare (e.g., one in 10 or more years) heavy precipitation events would become more frequent and more intense than in the recent past, on the global scale (virtually certain ) and in all continents and AR6 regions. The increase in frequency and intensity is extremely likely for most continents and very likely for most AR6 regions. At the global scale, the intensification of heavy precipitation will follow the rate of increase in the maximum amount of moisture that the atmosphere can hold as it warms (high confidence), of about 7% per 1°C of global warming. The increase in the frequency of heavy precipitation events will be non-linear with more warming and will be higher for rarer events (high confidence), with a likely doubling and tripling in the frequency of 10-year and 50-year events, respectively, compared to the recent past at 4°C of global warming. Increases in the intensity of extreme precipitation at regional scales will vary, depending on the amount of regional warming, changes in atmospheric circulation and storm dynamics (high confidence). {11.4, Box 11.1}

The projected increase in the intensity of extreme precipitation translates to an increase in the frequency and magnitude of pluvial floods – surface water and flash floods – (high confidence), as pluvial flooding results from precipitation intensity exceeding the capacity of natural and artificial drainagesystems. {11.4}

River Floods

Significant trends in peak streamflow have been observed in some regions over the past decades (high confidence). The seasonality of river floods has changed in cold regions where snow-melt is involved, with an earlier occurrence of peak streamflow (high confidence). {11.5}

Global hydrological models project a larger fraction of land areas to be affected by an increase in river floods than by a decrease in river floods (medium confidence). Regional changes in river floods are more uncertain than changes in pluvial floods because complex hydrological processes and forcings, including land cover change and human water management, are involved. {11.5}

Droughts

Different drought types exist, and they are associated with different impacts and respond differently to increasing greenhouse gas concentrations. Precipitation deficits and changes in evapotranspiration govern net water availability. A lack of sufficient soil moisture, sometimes amplified by increased atmospheric evaporative demand, results in agricultural and ecological drought. Lack of runoff and surface water result in hydrological drought. {11.6}

Human-induced climate change has contributed to increases in agricultural andecological droughts in some regions due to evapotranspiration increases (medium confidence). Increases in evapotranspiration have been driven by increases in atmospheric evaporative demand induced by increased temperature, decreased relative humidity and increased net radiation (high confidence). Trends in precipitation are not a main driver in affecting global-scale trends in drought (medium confidence), but have induced increases in meteorological droughts in a few AR6 regions (NES: high confidence; WAF, CAF, ESAF, SAM, SWS, SSA, SAS: medium confidence). Increasing trends in agricultural and ecological droughts have been observed on all continents (WAF, CAF, WSAF, ESAF, WCA, ECA, EAS, SAU, MED, WCE, WNA, NES: medium confidence), but decreases only in one AR6 region (NAU: medium confidence). Increasing trends in hydrological droughts have been observed in a few AR6 regions (MED: high confidence; WAF, EAS, SAU: medium confidence). Regional-scale attribution shows that human-induced climate change has contributed to increased agricultural and ecological droughts (MED, WNA), and increased hydrological drought (MED) in some regions (medium confidence). {11.6, 11.9}

More regions are affected by increases in agricultural and ecological droughts with increasing global warming (high confidence). Several regions will be affected by more severe agricultural and ecological droughts even if global warming is stabilised at 2°C, including MED, WSAF, SAM and SSA (high confidence), and ESAF, MDG, EAU, SAU, SCA, CAR, NSA, NES, SWS, WCE, NCA, WNA and CNA (medium confidence). Some regions are also projected to be affected by more severe agricultural and ecological droughts at 1.5°C (MED, WSAF, ESAF, SAU, NSA, SAM, SSA, CNA, medium confidence). At 4°C of global warming, about 50% of all inhabited AR6 regions would be affected by increases in agricultural and ecological droughts (WCE, MED, CAU, EAU, SAU, WCA, EAS, SCA, CAR, NSA, NES, SAM, SWS, SSA, NCA, CNA, ENA, WNA, WSAF, ESAF, MDG: medium confidence or higher), and only two regions (NEAF, SAS) would experience decreases in agricultural and ecological drought (medium confidence). There is high confidence that the projected increases in agricultural and ecological droughts are strongly affected by evapotranspiration increases associated with enhanced atmospheric evaporative demand. Several regions are projected to be more strongly affected by hydrological droughts with increasing global warming (at 4°C of global warming: NEU, WCE, EEU, MED, SAU, WCA, SCA, NSA, SAM, SWS, SSA, WNA, WSAF, ESAF, MDG: medium confidence or higher). There is low confidence that effects of enhanced atmospheric carbon dioxide (CO2) concentrations on plant water-use efficiency alleviate extreme agricultural and ecological droughts in conditions characterized by limited soil moisture and enhanced atmospheric evaporative demand. There is also low confidence that these effects will substantially reduce global plant transpiration and the severity of hydrological droughts. There is high confidence that the land carbon sink will become less efficient due to soil moisture limitations and associated drought conditions in some regions in higher-emissions scenarios, in particular under global warming levels above 4°C. {11.6, 11.9, Cross-Chapter Box 5.1}

Extreme Storms, Including Tropical Cyclones

The average and maximum rain rates associated with tropical cyclones (TCs), extratropical cyclones and atmospheric rivers across the globe, and severe convective storms in some regions, increase in a warming world (high confidence) . Available event attribution studies of observed strong TCs provide medium confidence for a human contribution to extreme TC rainfall. Peak TC rain rates increase with local warming at least at the rate of mean water vapour increase over oceans (about 7% per 1°C of warming) and in some cases exceeding this rate due to increased low-level moisture convergence caused by increases in TC wind intensity (medium confidence). {11.7, 11.4, Box 11.1}

It is likely that the global proportion of Category 3–5 tropical cyclone instances2has increased over the past four decades. The average location where TCs reach their peak wind intensity has very likely migrated poleward in the western North Pacific Ocean since the 1940s, and TC translation speed has likely slowed over the conterminous USA since 1900. Evidence of similar trends in other regions is not robust. The global frequency of TC rapid intensification events has likely increased over the past four decades. None of these changes can be explained by natural variability alone (medium confidence).

The proportion of intense TCs, average peak TC wind speeds, and peak wind speeds of the most intense TCs will increase on the global scale with increasing global warming (high confidence). The total global frequency of TC formation will decrease or remain unchanged with increasing global warming (medium confidence). {11.7.1}

There is low confidence in past changes of maximum wind speeds and other measures of dynamical intensity of extratropical cyclones. Future wind speed changes are expected to be small, although poleward shifts in the storm tracks could lead to substantial changes in extreme wind speeds in some regions (medium confidence). There is low confidence in past trends in characteristics of severe convective storms, such as hail and severe winds, beyond an increase in precipitation rates. The frequency of spring severe convective storms is projected to increase in the USA, leading to a lengthening of the severe convective storm season (medium confidence); evidence in other regions is limited. {11.7.2, 11.7.3}.

Compound Events, Including Dry/Hot Events, Fire Weather, Compound Flooding, and Concurrent Extremes

The probability of compound events has likely increased in the past due to human-induced climate change and will likely continue to increase with further global warming. Concurrent heatwaves and droughts have become more frequent, and this trend will continue with higher global warming (high confidence). Fire weather conditions (compound hot, dry and windy events) have become more probable in some regions (medium confidence) and there is high confidence that they will become more frequent in some regions at higher levels of global warming. The probability of compound flooding (storm surge, extreme rainfall and/or river flow) has increased in some locations (medium confidence), and will continue to increase due to sea level rise and increases in heavy precipitation, including changes in precipitation intensity associated with tropical cyclones (high confidence). The land area affected by concurrent extremes has increased (high confidence). Concurrent extreme events at different locations, but possibly affecting similar sectors (e.g., critical crop-producing areas for global food supply) in different regions, will become more frequent with increasing global warming, in particular above 2°C of global warming (high confidence). {11.8, Box 11.2, Box 11.4}.

Low-likelihood, High-impact Events Associated With Climate Extremes

The future occurrence of low-likelihood, high-impact events linked to climate extremes is generally associated with low confidence, but cannot be excluded, especially at global warming levels above 4°C. Compound events, including concurrent extremes, are a factor increasing the probability of low-likelihood, high-impact events (high confidence). With increasing global warming, some compound events with low likelihood in past and current climates will become more frequent, and there is a higher chance of occurrence of historically unprecedented events and surprises (high confidence). However, even extreme events that do not have a particularly low probability in the present climate (at more than 1°C of global warming) can be perceived as surprises because of the pace of global warming (high confidence). {Box 11.2}

11.1 Introduction

11.1.1 Scope of the Chapter

This chapter provides assessments of changes in weather and climate extremes (collectively referred to as extremes) framed in terms of the relevance to the Working Group II (WGII) assessment. It assesses observed changes in extremes, their attribution to causes, and future projections, at three global warming levels: 1.5°C, 2°C, and 4°C. This chapter is also one of the four ‘regional chapters’ of the WGI Report (along with Chapters 10 and 12 and the Atlas). Consequently, while it encompasses assessments of changes in extremes at global and continental scales to provide a large-scale context, it also addresses changes in extremes at regional scales.

Extremes are climatic impact-drivers (Annex VII: Glossary, see Chapter 12 for a comprehensive assessment). The IPCC risk framework (Chapter 1) articulates clearly that the exposure and vulnerability to climatic impact-drivers, such as extremes, modulate the risk of adverse impacts of these drivers, and that adaptation which reduces exposure and vulnerability will increase resilience, resulting in a reduction in impacts. Nonetheless, changes in extremes lead to changes in impacts as a direct consequence of changes in their magnitude and frequency, and also through their influence on exposure and resilience.

The Special Report on Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation (referred as the SREX report, IPCC, 2012) provided a comprehensive assessment on changes in extremes and how exposure and vulnerability to extremes determine the impacts and likelihood of disasters. Chapter 3 of that report (Seneviratne et al., 2012, hereafter also referred to as SREX Chapter 3) assessed physical aspects of extremes, and laid a foundation for the follow-up IPCC assessments. Several chapters of the IPCC Fifth Assessment Report (AR5) (IPCC, 2013) addressed climate extremes with respect to observed changes (Hartmann et al., 2013), model evaluation (Flato et al., 2013), attribution (Bindoff et al., 2013), and projected long-term changes (Collins et al., 2013). Assessments were also provided in the IPCC Special Report on Global Warming of 1.5°C (SR1.5) (IPCC, 2018; Hoegh-Guldberg et al., 2018), on climate change and land (SRCCL; (IPCC, 2019a), and on oceans and the cryosphere (SROCC; IPCC, 2019b). These assessments are the starting point for the present assessment.

This chapter is structured as follows (Figure 11.1): This section (11.1) provides the general framing and introduction to the chapter, highlighting key aspects that underlie the confidence and uncertainty in the assessment of changes in extremes, and introducing some main elements of the chapter. To provide readers with a quick overview of past and future changes in extremes, a synthesis of global-scale assessments for different types of extremes is included at the end of this section (Tables 11.1 and 11.2). Section 11.2 introduces methodological aspects of research on climate extremes. Sections 11.3 to 11.7 assess past changes and their attribution to causes, and projected future changes in extremes, for different types of extremes, including temperature extremes, heavy precipitation and pluvial floods, river floods, droughts, and storms, in separate sections. Section 11.8 addresses compound events. Section 11.9 summarizes regional assessments of changes in temperature extremes, in precipitation extremes and in droughts by continents in tables. The chapter also includes several boxes and FAQs on more specific topics.

Figure 11.1 | Visual guide to Chapter 11.

11.1.2 What Are Extreme Events and How are Their Changes Studied?

Building on the SREX report and AR5, this Report defines an extreme weather event as ‘an event that is rare at a particular place and time of year’, and an extreme climate event as ‘a pattern of extreme weather that persists for some time, such as a season’ (see Glossary). The definitions of ‘rare’ are wide ranging, depending on applications. Some studies consider an event as an extreme if it is unprecedented; other studies consider events that occur several times a year as moderate extreme events. Rarity of an event with a fixed magnitude also changes under human-induced climate change, making events that are unprecedented so far rather probable under present conditions, but unique in the observational record – and thus often considered as ‘surprises’ (see Box 11.2).

Various approaches are used to define extremes. These are generally based on the determination of relative (e.g., 90th percentile) or absolute (e.g., 35°C for a hot day) thresholds above which conditions are considered extremes. Changes in extremes can be examined from two perspectives, either focusing on changes in frequency of given extremes, or on changes in their intensity. These considerations in the definition of extremes are further addressed in Section 11.2.1.

11.1.3 Types of Extremes Assessed in this Chapter

The types of extremes assessed in this chapter include temperature extremes, heavy precipitation and pluvial floods, river floods, droughts, and storms. The drought assessment addresses meteorological droughts, agricultural and ecological droughts, and hydrological droughts (see Glossary). The storms assessment addresses tropical cyclones, extratropical cyclones, and severe convective storms. This chapter also assesses changes in compound events – that is, multivariate or concurrent extreme events – because of their relevance to impacts as well as the emergence of new literature on the subject. Most of the considered extremes were also assessed in SREX and AR5. Compound events were not assessed in depth in past IPCC reports (SREX Chapter 3; Section 11.8 of this Report). Marine-related extremes such as marine heatwaves and extreme sea level, are assessed in Section 9.6.4 and Box 9.2 of this Report.

Extremes and related phenomena are of various spatial and temporal scales. Tornadoes have a spatial scale as small as less than 100 metres and a temporal scale as short as a few minutes. In contrast, a drought can last for multiple years, affecting vast regions. The level of complexity of the involved processes differs from one type of extreme to another, affecting our capability to detect, attribute and project changes in weather and climate extremes. Temperature and precipitation extremes studied in the literature are often based on extremes derived from daily values. Studies of events on longer time scales for temperature or precipitation, or on sub-daily extremes, are scarcer, which generally limits the assessment for such events. Nevertheless, extremes on time scales different from daily are assessed for temperature extremes and heavy precipitation, when possible (Sections 11.3 and 11.4). Droughts and tropical cyclones are treated as phenomena in general in the assessment, not limited by their extreme forms, because these phenomena are relevant to impacts (Sections 11.6 and 11.7). Both precipitation and wind extremes associated with storms are considered.

Multiple concomitant extremes can lead to stronger impacts than those resulting from the same extremes had they happened in isolation. For this reason, the occurrence of multiple extremes that are multivariate and/or concurrent and/or happening in succession, also called ‘compound events’ (SREX Chapter 3), are assessed in this chapter based on emerging literature on this topic (Section 11.8). Box 11.2 also provides an assessment on low-likelihood, high-impact scenarios associated with extremes.

The assessment of projected future changes in extremes is presented as function of different global warming levels (Section 11.2.4 and Cross-Chapter Box 11.1). This provides traceability and comparison to the SR1.5 assessment (Hoegh-Guldberg et al., 2018, hereafter referred to as SR1.5 Chapter 3). Also, this is useful for decision makers as actionable information, as much of the mitigation policy discussion and adaptation planning can be tied to the level of global warming. For example, regional changes in extremes, and thus their impacts, can be linked to global mitigation efforts. There is also the advantage of separating uncertainty in future projections due to regional responses as a function of global warming levels from other factors such as differences in global climate sensitivity and emissions scenarios (Cross-Chapter Box 11.1). Information is also provided on the translation between information provided at global warming levels and for single emissions scenarios (Cross-Chapter Box 11.1). This facilitates easier comparison with the AR5 assessment and with some analyses provided in other chapters as function of emissions scenarios.

A global-scale synthesis of this chapter’s assessments is provided in Section 11.1.7. In particular, Tables 11.1 and 11.2 provide a synthesis for observed and attributed changes, and projected changes in extremes, respectively, at different global warming levels (1.5°C, 2°C, and 4°C). Tables on regional-scale assessments for changes in temperature extremes, heavy precipitation and droughts, are provided in Section 11.9.

11.1.4 Effects of Greenhouse Gas and Other External Forcings on Extremes

The SREX, AR5, and SR1.5 assessed that there is evidence from observations that some extremes have changed since the mid-20th century, that some of the changes are a result of anthropogenic influences, and that some observed changes are projected to continue into the future. Additionally, other changes are projected to emerge from natural climate variability under enhanced global warming (SREX Chapter 3; AR5 Chapter 10).

At the global scale, and also at the regional scale to some extent, many of the changes in extremes are a direct consequence of enhanced radiative forcing, and the associated global warming and/or resultant increase in the water-holding capacity of the atmosphere, as well as changes in vertical stability and meridional temperature gradients that affect climate dynamics (see Box 11.1). Widespread observed and projected increases in the intensity and frequency of hot extremes, together with decreases in the intensity and frequency of cold extremes, are consistent with global and regional warming (Section 11.3 and Figure 11.2). Extreme temperatures on land tend to increase more than the global mean temperature (Figure 11.2), due in large part to the land–sea warming contrast, and additionally to regional feedbacks in some regions (Section 11.1.6). Increases in the intensity of temperature extremes scale robustly, and in general linearly, with global warming across different geographical regions in projections up to 2100, with minimal dependence on emissions scenarios (Section 11.2.4, Figure 11.3,and Cross-Chapter Box 11.1; Seneviratne et al., 2016; Wartenburger et al., 2017; Kharin et al., 2018). The frequency of hot temperature extremes (see Figure 11.6), the number of heatwave days and the length of heatwave seasons in various regions also scale well, but nonlinearly (because of threshold effects, Section 11.2.1), with global mean temperatures (Wartenburger et al., 2017; Y. Sun et al., 2018a).

Figure 11.2 | Time series of observed temperature anomalies for global average annual mean temperature (black), land average annual mean temperature (green), land average annual hottest daily maximum temperature (TXx, purple), and land average annual coldest daily minimum temperature (TNn, blue). Global and land mean temperature anomalies are relative to their 1850–1900 means and are based on the multi-product mean annual time series assessed in Section 2.3.1.1.3 (see text for references). TXx and TNn anomalies are relative to their respective 1961–1990 means and are based on the HadEX3 dataset (Dunn et al., 2020) using values for grid boxes with at least 90% temporal completeness over 1961–2018. Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

Changes in annual maximum one-day precipitation (Rx1day) are proportional to mean global surface temperature changes, at about 7% increase per 1°C of warming, that is, following the Clausius–Clapeyron relation (Box 11.1), both in observations (Westra et al., 2013) and in future projections (Kharin et al., 2013) at the global scale. Extreme short-duration precipitation in North America also scales with global surface temperature (Prein et al., 2016b; C. Li et al., 2019a). At the local and regional scales, changes in extremes are also strongly modulated and controlled by regional forcings and feedback mechanisms (Section 11.1.6), whereby some regional forcings, for example, associated with changes in land cover and land use or aerosol emissions, can have non-local or some (non-homogeneous) global-scale effects. In general, there is high confidence in changes in extremes due to global-scale thermodynamic processes (i.e., global warming, mean moistening of the air) as the processes are well understood, while the confidence in those related to dynamic processes or regional and local forcing, including regional and local thermodynamic processes, is much lower due to multiple factors (see the following subsection and Box 11.1).

Figure 11.3 | Regional mean changes in annual hottest daily maximum temperature (TXx) for AR6 land regions and the global land area (except Antarctica), against changes in global mean surface air temperature (GSAT) as simulated by Coupled Model Intercomparison Project Phase 6 (CMIP6) models under different Shared Socio-economic Pathway (SSP) forcing scenarios, SSP1-1.9, SSP1-2.6, SSP2-4.5, SSP3-7. 0, and SSP5-8.5. Changes in TXx and GSAT are relative to the 1850–1900 baseline, and changes in GSAT are expressed as global warminglevel. (a) Individual models from the CMIP6 ensemble (grey), the multi-model median under three selected SSPs (colours), and the multi-modelmedian (black); (b) to (l) Multi-model median for the pooled data for individual AR6 regions. Numbers in parentheses indicate the linear scaling between regional TXx and GSAT. The black line indicates the 1:1 reference scaling between TXx and GSAT. See Atlas.1.3.2 for the definition of regions. Changes in TXx are also displayed in the Interactive Atlas. For details on the methods, see Supplementary Material 11.SM.2.

Since AR5, the attribution of extreme weather events, or the investigation of changes in the frequency and/or magnitude of individual and local- and regional-scale extreme weather events due to various drivers (Section 11.2.3 and Cross-Working Group Box 1.1) has provided evidence that greenhouse gases and other external forcings have affected individual extreme weather events. The events that have been studied are geographically uneven. For example, extreme rainfall events in the UK (Schaller et al., 2016; Vautard et al., 2016; Otto et al., 2018b) or heatwaves in Australia (King et al., 2014; Perkins-Kirkpatrick et al., 2016; Lewis et al., 2017b) have spurred more studies than other events. Many highly impactful extreme weather events have not been studied in the event attribution framework. Studies in the developing world are also generally lacking. This is due to various reasons (Section 11.2) including lack of observational data, lack of reliable climate models and other problems (Otto et al., 2020). While the events that have been studied are not representative of all extreme events that occurred, and results from these studies may also be subject to selection bias, the large number of event attribution studies provide evidence that changes in the properties of these local and individual events are in line with expected consequences of human influence on the climate and can be attributed to external drivers (Section 11.9). Figure 11.4 summarizes assessments of observed changes in temperature extremes, in heavy precipitation and in droughts, and their attribution in a map form.

Figure 11.4 | Overview of observed changes for cold, hot, and wet extremes and their potential human contribution. Shown are the direction of change and the confidence in: 1) the observed changes in cold and hot as well as wet extremes across the world; and 2) whether human-induced climate change contributed to causing these changes (attribution). In each region changes in extremes are indicated by colour (orange – increase in the type of extreme; blue – decrease; both colours – changes of opposing direction within the region, with the signal depending on the exact event definition; grey – there are no changes observed; and no fill – the data/evidence is too sparse to make an assessment). The squares and dots next to the symbol indicate the level of confidence for observing the trend and the human contribution, respectively. The more black dots/squares, the higher the level of confidence. The information on this figure is based on regional assessment of the literature on observed trends, detection and attribution and event attribution in Section 11.9.

Box 11.1 | Thermodynamic and Dynamic Changes in Extremes Across Scales

Changes in weather and climate extremes are determined by local exchanges in heat, moisture, and other related quantities (thermodynamic changes) and those associated with atmospheric and oceanic motions (dynamic changes). While thermodynamic and dynamic processes are interconnected, considering them separately helps to disentangle the roles of different processes contributing to changes in climate extremes (e.g., Shepherd, 2014).

Temperature extremes

An increase in the concentration of greenhouse gases in the atmosphere leads to the warming of tropospheric air and the Earth’s surface. This direct thermodynamic effect leads to warmer temperatures everywhere, with an increase in the frequency and intensity of warm extremes, and a decrease in the frequency and intensity of cold extremes. The initial increase in temperature leads to other thermodynamic responses and feedbacks affecting the atmosphere and the surface. These include an increase in the water vapour content of the atmosphere (water vapour feedback, see Section 7.4.2.2) and a change in the vertical profile of temperature (lapse rate feedback, see Section 7.4.2.2). While the water vapour feedback always amplifies the initial temperature increases (positive feedback), the lapse rate feedback amplifies near-surface temperature increases (positive feedback) in mid- and high latitudes but reduces temperature increases (negative feedback) in tropical regions (Pithan and Mauritsen, 2014).

Thermodynamic responses and feedbacks also occur through surface processes. For instance, observations and model simulations show that temperature increases, including extreme temperatures, are amplified in areas where seasonal snow cover is reduced due to decreases in surface albedo (see Section 11.3.1). In some mid-latitude areas, temperature increases are amplified by the higher atmospheric evaporative demand (Fu and Feng, 2014; Vicente-Serrano et al., 2020a) that results in a drying of soils in some regions (Section 11.6), leading to increased sensible heat fluxes (soil-moisturetemperature feedback, see Sections 11.1.6 and 11.3.1 for more background). Other thermodynamic feedback processes include changes in the water-use efficiency of plants under enhanced atmospheric carbon dioxide (CO2) concentrations that can reduce the overall transpiration, and thus also enhance temperature in projections (Sections 8.2.3.3, 11.1.6, 11.3 and 11.6).

Changes in the spatial distribution of temperatures can also affect temperature extremes by modifying the characteristics of weather patterns (e.g., Suarez-Gutierrez et al., 2020a). For example, a robust thermodynamic effect of polar amplification is a weakened north-south temperature gradient, which amplifies the warming of cold extremes in the Northern Hemisphere mid- and high latitudes because of the reduction of cold air advection (Holmes et al., 2015; Schneider et al., 2015; Gross et al., 2020). Much less robust is the dynamic effect of polar amplification (Section 7.4.4.1) and the reduced low-altitude meridional temperature gradient that has been linked to an increase in the persistence of weather patterns (e.g., heatwaves) and subsequent increases in temperature extremes (Cross-Chapter Box 10.1; Francis and Vavrus, 2012; Coumou et al. , 2015, 2018; Mann et al., 2017).

Precipitation extremes

Changes in temperature also control changes in water vapour through increases in evaporation and in the water-holding capacity of the atmosphere (Section 8.2.1). At the global scale, column-integrated water vapour content increases roughly following the Clausius–Clapeyron (C-C) relation, with an increase of approximately 7% per 1°C of global-mean surface warming (Section 8.2.1). Nonetheless, at regional scales, water vapour increases differ from this C-C rate due to several reasons (Section 8.2.2), including a change in weather regimes and limitations in moisture transport from the ocean, which warms more slowly than land (Byrne and O’Gorman, 2018). Observational studies (Fischer and Knutti, 2016; Sun et al., 2021) have shown that the observed rate of increased precipitation extremes is similar to the C-C rate at the global scale. Climate model projections show that the increase in water vapour leads to robust increases in precipitation extremes everywhere, with a magnitude that varies between 4% and 8% per 1°C of surface warming (thermodynamic contribution, Box 11.1, Figure 1b). At regional scales, climate models show that the dynamic contribution (Box 11.1, Figure 1c) can be substantial and strongly modify the projected rate of change of extreme precipitation (Box 11.1, Figure 1a) with large regions in the subtropics showing robust reductions and other areas (e.g., equatorial Pacific) showing robust amplifications (Box 11.1, Figure 1c). However, the dynamic contributions show large differences across models and are more uncertain than thermodynamic contributions (Box 11.1, Figure 1c; Shepherd, 2014; Trenberth et al., 2015; Pfahl et al., 2017).

Box 11.1, Figure 1: Multi-model Coupled Model Intercomparison Project Phase 5 (CMIP5) mean fractional changes (in % per degree of warming). (a) changes in annual maximum precipitation (Rx1day); (b) changes in Rx1day due to the thermodynamic contribution; and (c) changes in Rx1day due to the dynamic contribution estimated as the difference between the total changes and the thermodynamic contribution. Changes were derived from a linear regression for the period 1950–2100. Uncertainty is represented using the simple approach: no overlay indicates regions with high model agreement, where ≥80% of models (n=22) agree on the sign of change; diagonal lines indicate regions with low model agreement, where <80% of models agree on the sign of change. For more information on the simple approach, please refer to the Cross-Chapter Box (Atlas 1. A detailed description of the estimation of dynamic and thermodynamic contributions is given in Pfahl et al. (2017). Figure adapted from Pfahl et al. (2017), originally published inNature Climate Change/Springer Nature. Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

Dynamic contributions can occur in response to changes in the vertical and horizontal distribution of temperature (thermodynamics) and can affect the frequency and intensity of synoptic and subsynoptic phenomena, including tropical cyclones, extratropical cyclones, fronts, mesoscale-convective systems and thunderstorms. For example, the poleward shift and strengthening of the Southern Hemisphere mid-latitude storm tracks (Section 4.5.1) can modify the frequency or intensity of extreme precipitation. However, the precise way in which dynamic changes will affect precipitation extremes is unclear due to several competing effects (Shaw et al., 2016; Allan et al., 2020).

Box 11.1

Extreme precipitation can also be enhanced by dynamic responses and feedbacks occurring within storms that result from the extra latent heat released from the thermodynamic increases in moisture(Lackmann, 2013; Willisonet al. , 2013; Marcianoet al. , 2015; Nieet al. , 2018; Mizuta and Endo, 2020). The extra latent heat released within storms has been shown to increase precipitation extremes by strengthening convective updrafts and the intensity of the cyclonic circulation (e.g., Molnar et al., 2015; Nie et al., 2018), although weakening effects have also been found in mid-latitude cyclones (e.g., Kirshbaum et al., 2017). Additionally, the increase in latent heat can also suppress convection at larger scales due to atmospheric stabilization (Nie et al., 2018; Tandon et al., 2018; Kendon et al., 2019). As these dynamic effects result from feedback processes within storms where convective processes are crucial, their proper representation might require improving the horizontal/vertical resolution, the formulation of parametrizations, or both, in current climate models (i.e., Kendon et al. , 2014; Westra et al. , 2014; Ban et al. , 2015; Meredith et al. , 2015; Prein et al. , 2015; Nie et al., 2018).

Droughts

Droughts are also affected by thermodynamic and dynamic processes (Sections 8.2.3.3 and 11.6). Thermodynamic processes affect droughts by increasing atmospheric evaporative demand (Martin, 2018; Gebremeskel Haile et al., 2020; Vicente-Serrano et al., 2020a) through changes in air temperature, radiation, wind speed, and relative humidity. Dynamic processes affect droughts through changes in the occurrence, duration and intensity of weather anomalies, which are related to precipitation and the amount of sunlight (Section 11.6). While atmospheric evaporative demand increases with warming, regional changes in aridity are affected by increasing land–ocean warming contrast, vegetation feedbacks and responses to rising CO2 concentrations, and dynamic shifts in the location of the wet and dry parts of the atmospheric circulation in response to climate change, as well as internal variability (Byrne and O’Gorman, 2015; Kumar et al., 2015; Allan et al., 2020).

In summary, both thermodynamic and dynamic processes are involved in the changes of extremes in response to warming. Anthropogenic forcing (e.g., increases in greenhouse gas concentrations) directly affects thermodynamic variables, including overall increases in high temperatures and atmospheric evaporative demand, and regional changes in atmospheric moisture, which intensify heatwaves, droughts and heavy precipitation events when they occur (high confidence). Dynamic processes are often indirect responses to thermodynamic changes, are strongly affected by internal climate variability, and are also less well understood. As such, there is low confidence in how dynamic changes affect the location and magnitude of extreme events in a warming climate.

11.1.5 Effects of Large-scale Circulation on Changes in Extremes

Atmospheric large-scale circulation patterns and associated atmospheric dynamics are important determinants of the regional climate (Chapter 10). As a result, they are also important to the magnitude, frequency, and duration of extremes (Box 11.4). Aspects of changes in large-scale circulation patterns are assessed in Chapters 2, 3, 4 and 8, and representative atmospheric and oceanic modes are described in Annex IV. This subsection provides some general concepts, through a couple of examples, on why the uncertainty in the response of large-scale circulation patterns to external forcing can cascade to uncertainty in the response of extremes to external forcings. Details for specific types of extremes are covered in the relevant subsections. For example, the occurrence of the El Niño–Southern Oscillation (ENSO) influences precipitation regimes in many areas, favouring droughts in some regions and heavy rains in others (Box 11.4). The extent and strength of the Hadley circulation influences regions where tropical and extratropical cyclones occur, with important consequences for the characteristics of extreme precipitation, drought, and winds (Section 11.7). Changes in circulation patterns associated with land–ocean heat contrast, which affect the monsoon circulations (Section 8.4.2.4), lead to heavy precipitation along the coastal regions in East Asia (Freychet et al., 2015). As a result, changes in the spatial and/or temporal variability of the atmospheric circulation in response to warming affect characteristics of weather systems such as tropical cyclones (Sharmila and Walsh, 2018), storm tracks (Shaw et al., 2016), and atmospheric rivers (Section 11.7; Waliser and Guan, 2017). Changes in weather systems come with changes in the frequency and intensity of extreme winds, extreme temperatures, and extreme precipitation, on the backdrop of thermodynamic responses of extremes to warming (Box 11.1). Floods are also affected by large-scale circulation modes, including ENSO, the North Atlantic Oscillation (NAO), the Atlantic Multi-decadal Variability (AMV), and the Pacific Decadal Variability (PDV) (Kundzewicz et al., 2018; Annex IV). Aerosol forcing, through changes in patterns of sea surface temperatures (SSTs), also affects circulation patterns and tropical cyclone activities (Takahashi et al., 2017).

In general, changes in atmospheric large-scale circulation due to external forcing are uncertain, but there are some robust changes (Sections 2.3.1.4 and 8.2.2.2). Among them, there has been avery likely widening of the Hadley circulation since the 1980s and the extratropical jets and cyclone tracks have likely been shifting poleward since the 1980s (Section 2.3.1.4). The poleward expansion affects drought occurrence in some regions (Section 11.6), and results in poleward shifts of tropical cyclones and storm tracks (Sections 11.7.1 and 11.7.2). Although it is very likely that the amplitude of ENSO variability will not robustly change over the 21st century (Section 4.3.3.2), the frequency of extreme ENSO events (Box 11.4), defined by precipitation threshold, is projected to increase with global warming (Section 6.5 of SROCC). This would have implications for projected changes in extreme events affected by ENSO, including droughts over wide areas (Section 11.6; Box 11.4) and tropical cyclones (Section 11.7.1). A case study is provided for extreme ENSO events in 2015–2016 in Box 11.4 to highlight the influence of ENSO on extremes.

In summary, large-scale atmospheric circulation patterns are important drivers for local and regional extremes. There is overall low confidence about future changes in the magnitude, frequency, and spatial distribution of these patterns, which contributes to uncertainty in projected responses of extremes, especially in the near term.

11.1.6 Effects of Regional-scale Processes and Forcings and Feedbacks on Changes in Extremes

At the local and regional scales, changes in extremes are strongly modulated by local and regional feedbacks (SRCCL, Jia et al., 2019; Seneviratne et al., 2013; Miralles et al., 2014a; Lorenz et al., 2016; Vogel et al., 2017), changes in large-scale circulation patterns (Section 11.1.5), and regional forcings such as changes in land use or aerosol concentrations (Chapters 3 and 7; Findell et al., 2017; Hirsch et al., 2017, 2018; Thiery et al., 2017; Z. Wang et al., 2017b). In some cases, such responses may also include non-local effects (e.g., de Vrese et al., 2016; Persad and Caldeira, 2018; Miralles et al., 2019; Schumacher et al., 2019). Regional-scale forcing and feedbacks often affect temperature distributions asymmetrically, with generally higher effects for the hottest percentiles (Section 11.3).

Land use can affect regional extremes, in particular hot extremes, in several ways (high confidence). This includes effects of land management (e.g., cropland intensification, irrigation, double cropping) as well as of land cover changes (deforestation; Sections 11.3.2 and 11.6). Some of these processes are not well represented (e.g., effects of forest cover on diurnal temperature cycle) or not integrated (e.g., irrigation) in climate models (Sections 11.3.2 and 11.3.3). Overall, the effects of land-use forcing may be particularly relevant in the context of low-emissions scenarios, which include large land-use modifications, for instance those associated with the expansion of biofuels, bioenergy with carbon capture and storage, or re-/afforestation to ensure negative emissions, as well as with the expansion of food production (e.g., SR1.5, Chapter 3; Cross-Chapter Box 5.1 in this Report; van Vuuren et al., 2011; Hirsch et al., 2018). There are also effects on the water cycle through freshwater use (Section 11.6 and Cross-Chapter Box 5.1).

Aerosol forcing also has a strong regional footprint associated with regional emissions, which affects temperature and precipitation extremes (high confidence) (Sections 11.3 and 11.4). From around the 1950s to 1980s, enhanced aerosol loadings led to regional cooling due to decreased global solar radiation (‘global dimming’) which was followed by a phase of ‘global brightening’ due to a reduction in aerosol loadings (Chapters 3 and 7; Wild et al., 2005). King et al. (2016b) show that aerosol-induced cooling delayed the timing of a significant human contribution to record-breaking heat extremes in some regions. However, the decreased aerosol loading since the 1990s has led to an accelerated warming of hot extremes in some regions. Based on Earth system model (ESM) simulations, Dong et al. (2017) suggest that a substantial fraction of the warming of the annual hottest days in Western Europe since the mid-1990s has been due to decreases in aerosol concentrations in the region. Dong et al. (2016b) also identify non-local effects of decreases in aerosol concentrations in Western Europe, which they estimate played a dominant role in the warming of the hottest daytime temperatures in north-east Asia since the mid-1990s, via induced coupled atmosphere–land surface and cloud feedbacks, rather than a direct impact of anthropogenic aerosol changes on cloud condensation nuclei.

In addition to regional forcings, regional feedback mechanisms can also substantially affect extremes (high confidence) (Sections 11.3, 11.4 and 11.6). In particular, soil moisture feedbacks play an important role for extremes in several mid-latitude regions, leading to a marked additional warming of hot extremes compared to mean global warming (Seneviratne et al., 2016; Bathiany et al., 2018; Miralles et al., 2019), which is superimposed on the known land–sea contrast in mean warming (Vogel et al., 2017). Soil moisture–atmosphere feedbacks also affect drought development (Section 11.6). Additionally, effects of land surface conditions on circulation patterns have also been reported (Koster et al., 2016; Sato and Nakamura, 2019). These regional feedbacks are also associated with substantial spread in models (Section 11.3), and contribute to the identified higher spread of regional projections of temperature extremes as a function of global warming, compared with the spread resulting from the differences in projected global warming (global transient climate responses) in climate models (Seneviratne and Hauser, 2020). In addition, there are also feedbacks between soil moisture content and precipitation occurrence, generally characterized by negative spatial feedbacks and positive local feedbacks (Taylor et al., 2012; Guillod et al., 2015). Climate model projections suggest that these feedbacks are relevant for projected changes in heavy precipitation (Seneviratne et al., 2013). However, there is evidence that climate models do not capture the correct sign of the soil moisture–precipitation feedbacks in several regions, in particular spatially, and/or in some cases also temporally (Taylor et al., 2012; Moon et al., 2019). In the Northern Hemisphere high latitudes, the snow- and ice-albedo feedback, along with other factors, is projected to largely amplify temperature increases (e.g., Pithan and Mauritsen, 2014), although the effect on temperature extremes is still unclear. It also remains unclear whether snow-albedo feedbacks in mountainous regions might have an effect on temperature and precipitation extremes (e.g., Gobiet et al., 2014). However, these feedbacks play an important role in projected changes in high-latitude warming (Hall and Qu, 2006), and, in particular, in changes in cold extremes in these regions (Section 11.3).

Finally, extreme events may also regionally amplify one another. For example, this is the case for heatwaves and droughts, with high temperatures and stronger radiative forcing leading to drying tendencies on land due to increased evapotranspiration (Section 11.6), and drier soils then inducing decreased evapotranspiration and higher sensible heat flux and hot temperatures (Box 11.1, Section 11.8; Seneviratne et al., 2013; Miralles et al., 2014a; Vogel et al., 2017; Zscheischler and Seneviratne, 2017; S. Zhou et al., 2019; Kong et al., 2020).

In summary, regional forcings and feedbacks – in particular those associated with land use and aerosol forcings – and soil-moisture–temperature, soil moisture–precipitation, and snow/ice–albedo–temperature feedbacks, play an important role in modulating regional changes in extremes. These can also lead to a higher warming of extreme temperatures compared to mean temperature (high confidence), and possibly cooling in some regions (medium confidence). However, there is only medium confidence in the representation of the associated processes in state-of-the-art ESMs.

11.1.7 Global-scale Synthesis

Tables 11.1 and 11.2 provide a synthesis for observed and attributed changes in extremes, and projected changes in extremes, respectively, at different levels of global warming. This synthesis assessment focuses on the assessed range of observed and projected changes. In this chapter, the assessed likely range in a projection typically corresponds to the 90% range of the multi-model ensemble spread to take into account other sources of uncertainty, unless stated otherwise. Some low-likelihood, high-impact scenarios that can be of high relevance are addressed in Box 11.2.

Building on the assessments from Tables 11.1 and 11.2, Figure 11.5 provides a synthesis on the level of confidence in the attribution and projection of changes in extremes. In the case where the signal in the observations is still relatively weak but the physical processes underlying the changes in extremes in response to human forcing are well understood, confidence in the projections would be higher than in the attribution because of strengthening in the signal with warming. But, when the observed signal is already strong and when observational evidence is consistent with model simulated responses, confidence in the projection may be lower than that in attribution if certain physical processes could be expected to behave differently in a much warmer world and under much higher greenhouse gas forcing, and in particular if such a behaviour is poorly understood.

Figure 11.5 | confidence and likelihood of past changes and projected future changes at 2°C of global warming on the global scale. The information in this figure is based on Tables 11.1 and 11.2.

Further synthesis for regional assessments are provided in Figure 11.4 (event attribution), Figure 11.6 (projected change in hot temperature extremes) and Figure 11.7 (projected changes in precipitation extremes). A synthesis on regional assessments for observed, attributed and projected changes in extremes is provided in Section 11.9 for all AR6 reference regions (see Section 1.4.5 and Figures 1.18 and Atlas.2 for definitions of AR6 regions).

Figure 11.6 | Projected changes in the frequency of extreme temperature events under 1°C, 1.5°C, 2°C, 3°C, and 4°C global warming levels relative to the 1850–1900 baseline. Extreme temperatures are defined as the maximum daily temperatures that were exceeded on average once during a 10-year period (10-year event, blue) and once during a 50-year period (50-year event, orange) during the 1850–1900 base period. Results are shown for the global land area and the AR6 regions. For each box plot, the horizontal line and the box represent the median and central 66% uncertainty range, respectively, of the frequency changes across the multi-model ensemble, and the ‘whiskers’ extend to the 90% uncertainty range. The dotted line indicates no change in frequency. The results are based on the multi-model ensemble from simulations of global climate models contributing to the Coupled Model Intercomparison Project Phase 6 (CMIP6) under different Shared Socio-economic Pathway forcing scenarios. Adapted from Li et al. (2021). Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).
Figure 11.7 | Projected changes in the frequency of extreme precipitation events under 1°C, 1.5°C, 2°C, 3°C, and 4°C global warming levels relative to the 1850–1900 baseline. Extreme precipitation is defined as the annual maximum daily precipitation (Rx1day) that was exceeded on average once during a 10-year period (10-year event, blue) and once during a 50-year period (50-year event, orange) during the 1850–1900 base period. Results are shown for the global land area and the AR6 regions. For each box plot, the horizontal line and the box represent the median and central 66% uncertainty range, respectively, of the frequency changes across the multi-model ensemble, and the ‘whiskers’ extend to the 90% uncertainty range. The dotted line indicates no change in frequency. The results are based on the multi-model ensemble from simulations of global climate models contributing to the Coupled Model Intercomparison Project Phase 6 (CMIP6) under different Shared Socio-economic Pathway forcing scenarios. Adapted from Li et al. (2021). Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

Table 11.1 | Synthesis table on observed changes in extremes and contribution by human influence. Note that observed changes in marine extremes are assessed in Cross-Chapter Box 9.1.

Phenomenon and Direction of Trend

Observed/Detected Trends Since 1950 (for +0.5°C global warming or higher)

Human Contribution to the Observed Trends Since 1950 (for +0.5°C global warming or higher)

Warmer and/or more frequent hot days and nights over most land areas

Warmer and/or fewer cold days and nights over most land areas

Warm spells/heatwaves: increases in frequency or intensity over most land areas

Cold spells/cold waves: decreases in frequency or intensity over most land areas

Virtually certain on global scale {11.3}

Continental-scale evidence:

Asia, Australasia, Europe, North America: Very likely

Central and South America: High confidence

Africa: Medium confidence

{11.3, 11.9}

Extremely likely main contributor on global scale {11.3}

Continental-scale evidence:

North America, Europe, Australasia, Asia: Very likely

Central and South America: High confidence

Africa: Medium confidence

{11.3, 11.9}

Heavy precipitation events: increase in the frequency, intensity, and/or amount of heavy precipitation

Likely on global scale, over majority of land regions with good observational coverage {11.3}

Continental-scale evidence:

Asia, Europe, North America: Likely

Africa, Australasia, Central and South America: Low confidence

{11.3, 11.9}

Likely main contributor to the observed intensification of heavy precipitation in land regions on global scale.

{11.3}

Continental-scale evidence:

Asia, Europe, North America: Likely

Africa, Australasia, Central and South America: Low confidence

{11.3, 11.9}

Increases in agricultural and ecological drought events

Medium confidence some regions {11.6, 11.9}

Increasing trends in agricultural and ecological droughts have been observed in AR6 regions on all continents (medium confidence) {11.6, 11.9}

Medium confidence some regions

{11.6, 11.9}

Increase in precipitation associated with tropical cyclones (TCs)

Medium confidence

{11.7}

High confidence

{11.7}

Increase in likelihood that a TC will be at major TC intensity (Cat. 3–5)

Likely

{11.7}

Medium confidence

{11.7}

Changes in frequency of rapidly intensifying tropical cyclones

Likely

{11.7}

Medium confidence

{11.7}

Poleward migration of tropical cyclones in the western Pacific

Medium confidence

{11.7}

Medium confidence

{11.7}

Decrease in TC forward motion over the USA

It is likely that TC translation speed has slowed over the USA since 1900.

{11.7}

It is more likely than not that the slowdown of TC translation speed over the USA has contributions from anthropogenic forcing.

{11.7}

Severe convective storms (tornadoes, hail, rainfall, wind, lightning)

Low confidence in past trends in hail and winds and tornado activity due to short length of high-quality data records. {11.7}

Low confidence

{11.7}

Increase in compound events

Likely increase in the probability of compound events.

High confidence that concurrent heatwaves and droughts are becoming more frequent under enhanced greenhouse gas forcing at global scale.

Medium confidence that fire weather, i.e. compound hot, dry and windy events, have become more frequent in some regions.

Medium confidence that compound flooding risk has increased in some locations.

{11.8}

Likely that human-induced climate change has increased the probability of compound events.

High confidence that human influence has increased the frequency of concurrent heatwaves and droughts.

Medium confidence that human influence has increased fire weather occurrence in some regions.

Low confidence that human influence has contributed to changes in compound events leading to flooding.

{11.8}

Table 11.2 | Synthesis table on projected changes in extremes. Note that projected changes in marine extremes are assessed in Chapter 9 and Cross-Chapter Box 9.1 (marine heatwaves). Assessments are provided compared to pre-industrial conditions.

Phenomenon and Direction of Trend

Projected Changes at +1.5ºC Global Warming

Projected Changes at +2°C Global Warming

Projected Changes at +4°C Global Warming

Warmer and/or more frequent hot days and nights over most land areas

Warmer and/or fewer cold days and nights over most land areas

Warm spells/heatwaves; increases in frequency or intensity over most land areas

Cold spells/cold waves: decreases in frequency or intensity over most land areas

Virtually certain on global scale

Extremely likely on all continents

Highest increase of temperature of hottest days is projected in some mid-latitude and semi-arid regions, and the South American Monsoon region, at about 1.5 times to twice the rate of global warming (high confidence){11.3, Figure 11.3}

Highest increase of temperature of coldest days is projected in Arctic regions, at about three times the rate of global warming (high confidence){11.3}

Continental-scale projections:

Extremely likely : Africa, Asia, Australasia, Central and South America, Europe, North America

{11.3, 11.9}

Virtually certain on global scale

Virtually certain on all continents

Highest increase of temperature of hottest days is projected in some mid-latitude and semi-arid regions, and the South American Monsoon region, at about 1.5 times to twice the rate of global warming (high confidence){11.3, Figure 11.3}

Highest increase of temperature of coldest days is projected in Arctic regions, at about three times the rate of global warming (high confidence){11.3}

Continental-scale projections:

Virtually certain: Africa, Asia, Australasia, Central and South America, Europe, North America

{11.3, 11.9}

Virtually certain on global scale

Virtually certain on all continents

Highest increase of temperature of hottest days is projected in some mid-latitude and semi-arid regions, and the South American Monsoon region, at about 1.5 times to twice the rate of global warming (high confidence){11.3, Figure 11.3}

Highest increase of temperature of coldest days is projected in Arctic regions, at about three times the rate of global warming (high confidence){11.3}

Continental-scale projections:

Virtually certain: Africa, Asia, Australasia, Central and South America, Europe, North America

{11.3, 11.9}

Heavy precipitation events: increase in the frequency, intensity, and/or amount of heavy precipitation

High confidence that increases take place in most land regions{11.4}

Very likely : Asia, North America

Likely: Africa, Europe

High confidence: Central and South America

Medium confidence: Australasia

{11.4, 11.9}

Likely that increases take place in most land regions{11.4}

Extremely likely : Asia, North America

Very likely : Africa, Europe

Likely: Australasia, Central and South America

{11.4, 11.9}

Very likely that increases take place in most land regions{11.4}

Virtually certain: Africa, Asia, North America

Extremely likely : Central and South America, Europe

Very likely Australasia

{11.4, 11.9}

Agricultural and ecological droughts: increases in intensity and/or duration of drought events

More regions affected by increases in agricultural and ecological droughts compared to observed changes (high confidence). {11.6, 11.9}

Decreased precipitation is going to increase the severity of drought in some regions; atmospheric evaporative demand will continue to increase compared to pre-industrial conditions and lead to further increases in agricultural and ecological droughts due to increased evapotranspiration in some regions. (high confidence)

{11.6, 11.9}

More regions affected by increases in agricultural and ecological droughts than at 1.5°C of global warming (high confidence). {11.6, 11.9}

Decreased precipitation is going to increase the severity of drought in some regions; atmospheric evaporative demand will continue to increase compared to pre-industrial conditions and lead to further increases in agricultural and ecological droughts due to increased evapotranspiration in some regions. (high confidence)

{11.6, 11.9}

More regions affected by increases in agricultural and ecological droughts than at 2°C of global warming (very likely) . {11.6, 11.9}

Decreased precipitation is going to increase the severity of drought in several regions; atmospheric evaporative demand will continue to increase compared to pre-industrial conditions and lead to further increases in agricultural and ecological droughts due to increased evapotranspiration in several regions. (high confidence)

{11.6, 11.9}

Increase in precipitation associated with tropical cyclones (TCs)

High confidence in a projected increase of TC rain rates at the global scale with amedian projected increase due to human emissions of about 11%. {11.7}

Medium confidence that rain rates will increase in every basin. {11.7}

High confidence in a projected increase of TC rain rates at the global scale with a median projected increase due to human emissions of about 14%. {11.7}

Medium confidence that rain rates will increase in every basin. {11.7}

High confidence in a projected increase of TC rain rates at the global scale with a median projected increase due to human emissions of about 28%. {11.7}

Medium confidence that rain rates will increase in every basin. {11.7}

Increase in mean TC lifetime-maximum wind speed (intensity)

Medium confidence

{11.7}

High confidence

{11.7}

High confidence

{11.7}

Increase in likelihood that a TC will reach major TC intensity (Category 4–5)

High confidence for an increase in the proportion of TCs that reach the strongest (Category 4–5) levels. The median projected increase in this proportion is about 10%.

{11.7}

High confidence for an increase in the proportion of TCs that reach the strongest (Category 4–5) levels. The median projected increase in this proportion is about 13%.

{11.7}

High confidence for an increase in the proportion of TCs that reach the strongest (Category 4–5) levels. The median projected increase in this proportion is about 20%.

{11.7}

Severe convective storms

High confidence that the average and maximum rain rates associated with severe convective storms increase in some regions, including the USA. High confidence that convective available potential energy (CAPE) increases in response to global warming in the tropics and subtropics, suggesting more favourable environments for severe convective storms. Medium confidence that the frequency of spring severe convective storms is projected to increase in the USA, leading to a lengthening of the severe convective storm season. {11.7}

Increase in compound events (frequency, intensity)

Likely that probability of compound events will continue to increase with global warming.

High confidence that concurrent heatwaves and droughts will continue to increase under higher levels of global warming, with higher frequency/intensity with every additional 0.5°C of global warming.

High confidence that fire weather, (i.e. compound hot, dry and windy events), will become more frequent in some regions at higher levels of global warming.

High confidence that compound flooding at the coastal zone will increase under higher levels of global warming. {11.8}

Box 11.2 | Changes in Low-likelihood, High-impact Extremes

The SREX (Chapter 3) assigned low confidence to changes in low-likelihood, high-impact (LLHI) events (termed ‘low-probability high-impact scenarios‘). Such events are often not anticipated and thus sometimes referred to as ‘surprises’. There are several types of LLHI events. Abrupt changes in mean climate are addressed in Chapter 4. Unanticipated LLHI events can either result from tipping points in the climate system (Section 1.4.4.3), such as the shutdown of the Atlantic thermohaline circulation (SROCC Chapter 6; Collins et al., 2019) or the drydown of the Amazonian rainforest (SR1.5 Chapter 3, Hoegh-Guldberg et al., 2018; Drijfhout et al., 2015), or from uncertainties in climate processes, including climate feedbacks, that may enhance or damp extremes either related to global or regional climate responses (Seneviratne et al., 2018a; Sutton, 2018). The low confidence does not by itself exclude the possibility of such events occuring, rather it indicates a poor state of knowledge. Such outcomes, while improbable, could be associated with very high impacts, and are thus highly relevant from a risk perspective (see Section 1.4.3 and Box 11.4; Sutton, 2018, 2019). Alternatively, high impacts can occur when different extremes occur at the same time, or in short succession at the same location, or in several regions with shared vulnerability (e.g., food-basket regions Gaupp et al., 2019). These ‘compound events’ are assessed in Section 11.8, and Box 11.4 provides a case study example.

Difficulties persist in determining the likelihood of occurrence and time frame of potential tipping points and LLHI events. However, new literature has emerged on unanticipated and LLHI events. There are some events that are sufficiently rare that they have not been observed in meteorological records, but whose occurrence is nonetheless plausible within the current state of the climate system – see examples below and in McCollum et al. (2020). The rare nature of such events and the limited availability of relevant data makes it difficult to estimate their occurrence probability and thus gives little evidence on whether to include such hypothetical events in planning decisions and risk assessments. The estimation of such potential surprises is often limited to events that have historical analogues (including before the instrumental records began, Wetter et al., 2014), albeit the magnitude of the event may differ. Additionally, there is also a limitation of available resources to exhaust all plausible trajectories of the climate system. As a result, there will still be events that cannot be anticipated. These events can be surprises to many in that the events have not been experienced, although their occurrence could be inferred by statistical means or physical modelling approaches (Chen et al., 2017; van Oldenborgh et al., 2017; Harrington and Otto, 2018a). Another approach focusing on the estimation of low-probability events and of events whose likelihood of occurrence is unknown consists in using physical climate models to create a physically self-consistent storyline of plausible extreme events and assessing their impacts and driving factors in past (Section 11.2.3) or future conditions (Section 11.2.4) (Hazeleger et al. , 2015; Shepherd, 2016; Zappa and Shepherd, 2017; Cheng et al. , 2018; Shepherd et al. , 2018; Sutton, 2018; Schaller et al. , 2020; Wehrli et al., 2020).

In many parts of the world, observational data are limited to 50–60 years. This means that the chance to observe an extreme event at a particular location that occurs once in several hundred or more years is small. Thus, when a very extreme event occurs, it becomes a surprise to many (Bao et al., 2017; McCollum et al., 2020), and very rare events are often associated with high impacts(van Oldenborghet al. , 2017; Philipet al. , 2018b; Tozeret al. , 2020). Attributing and projecting very rare events in a particular location by assessing their likelihood of occurrence within the same larger region and climate thus provides another way to make quantitative assessments regarding events that are extremely rare locally. Some examples of such events include:

  • Hurricane Harvey, that made landfall in Houston, TX in August 2017 (Section 11.7.1.4.)
  • The 2010–2011 extreme floods in Queensland, Australia (Christidis et al., 2013a)
  • The 2018 concurrent heatwaves across the Northern Hemisphere (Box 11.4)
  • Tropical Cyclone Idai in Mozambique (Cross-Chapter Box: Disaster in WGII AR6 Chapter 4)
  • The California fires in 2018 and 2019
  • The 2019–2020 Australia fires (Cross-Chapter Box: Disaster in WGII AR6 Chapter 4)

One factor making such events hard to anticipate is the fact that we now live in a non-stationary climate, and that the framework of reference for adaptation is continuously moving. As an example, the concurrent heatwaves that occurred across the Northern Hemisphere in the summer of 2018 were considered very unusual and were unprecedented given the total area that was concurrently affected(Drouardet al. , 2019; Kornhuberet al. , 2019; Toretiet al. , 2019; Vogelet al. , 2019); however, the probability of this event under 1°C global warming was found to be about 16% (Vogel et al., 2019), which is not particularly low. Similarly, the 2013 summer temperature over eastern China was the hottest on record at the time, but it had an estimated recurrence interval of about four years in the climate of 2013 (Sun et al., 2014). Furthermore, when other aspects of the risk, vulnerability, and exposure are historically high or have recently increased (see WGII, Chapter 16, Section 16.4), relatively moderate extremes can have very high impacts (Otto et al., 2015b; Philip et al., 2018b). As warming continues, the climate moves further away from its historical state we are familiar with, resulting in an increased likelihood of unprecedented events and surprises. This is particularly the case under high global warming levels – for example, the climate of the late 21st century under high-emissions scenarios, above 4°C of global warming (Cross-Chapter Box 11.1).

Another factor highlighted in Section 11.8 and Box 11.4 making events high-impact and difficult to anticipate is that several locations under moderate warming levels could be affected simultaneously, or very repeatedly by different types of extremes (Mora et al., 2018; Gaupp et al., 2019; Vogel et al., 2019). Box 11.4 shows that concurrent events at different locations, which can lead to major impacts across the world, can also result from the combination of anomalous circulation or natural variability (e.g., El Niño–Southern Oscillation) patterns with amplification of resulting responses to human-induced global warming. Also multivariate extremes at single locations pose specific challenges to anticipation (Section 11.8), with low likelihoods in the current climate but the probability of occurrence of such compound events strongly increasing with increasing global warming levels (Vogel et al., 2020a). Therefore, in order to estimate whether, and at what level of global warming, very high impacts arising from extremes would occur, the spatial extent of extremes and the potential of compounding extremes need to be assessed. Sections 11.3, 11.4, 11.7 and 11.8 highlight increasing evidence that temperature extremes, higher intensity precipitation accompanying tropical cyclones, and compound events such as dry/hot conditions conducive to wildfire or storm surges resulting from sea level rise and heavy precipitation events, pose widespread threats to societies already at relatively low warming levels. Studies have already shown that the probability for some recent extreme events is so small in the undisturbed world that these events were extremely unlikely to occur without human influence (Section 11.2.4). Box 11.2, Table 1, provides examples of projected changes in LLHI extremes (single extremes, compound events) of potential relevance for impact and adaptation assessments showing that today’s very rare events can become commonplace in a warmer future.

Box 11.2, Table 1 | Examples of changes in low-likelihood, high-impact extreme conditions (single extremes, compound events) at different global warming levels.

+1°C (Present-day)

+1.5°C

+2°C

+3°C and Higher

Risk ratio for annual hottest daytime temperature (TXx) with 1% of probability under present-day warming (+1°C) (Kharin et al., 2018): Global land

1

3.3 (i.e., 230% higher probability)

8.2 (i.e., 720% higher probability)

Not assessed

Risk ratio for heavy precipitation events (Rx1day) with 1% of probability under present-day warming (+1°C) (Kharin et al., 2018): Global land

1

1.2 (i.e., 20% higher probability)

1.5 (i.e., 50% higher probability)

Not assessed

Number of 1–5 day duration extreme floods with 1% of probability under present-day warming (+1°C) (H. Ali et al., 2019) Indian subcontinent

Up to 3 in individual locations

Up to 5 in individual locations

2–6 in most locations

Up to 12 in individual locations (4°C)

Probability of ‘extreme extremes’ hot days with 1/1000 probability at the end of the 20th century (Vogel et al., 2020a): Global land

About 20 days over 20 years in most locations

About 50 days in 20 years in most locations

About 150 days in 20 years in most locations

About 500 days in 20 years in most locations (3°C)

Probability of co-occurrence in the same week of hot days with 1/1000 probability and dry days with 1/1000 probability at the end of the 20th century (Vogel et al., 2020a): Amazon

0% probability

About one week in 20 years

About 4 to 5 weeks in 20 years

More than 9 weeks in 20 years (3°C)

Projected soil moisture drought duration per year (Samaniego et al., 2018): Mediterranean region

41 days (+46% compared to the late 20th century)

58 days (+107% compared to the late 20th century)

71 days (+154% compared to the late 20th century)

125 days (+346% compared to the late 20th century) (3°C)

Increase in days exposed to dangerous extreme heat – measured in Health Heat Index (HHI) (Q. Sun et al., 2019) global land

Not assessed, baseline is 1981–2000

1.6 times higher risk of experiencing heat >40.6

2.3 times higher risk of experiencing heat >40.6

Around 80% of land area exposed to dangerous heat, tropical regions 1/3 of the year (4°C)

Increase in regional mean fire season length (Q. Sun et al., 2019; Xu et al., 2020) global land

Not assessed, baseline is 1981–2000

6.2 days

9.5 days

About 50 days (4°C)

In summary, the future occurrence of LLHI events linked to climate extremes is generally associated with low confidence, but cannot be excluded, especially at global warming levels above 4°C. Compound events, including concurrent extremes, are a factor increasing the probability of LLHI events (high confidence). With increasing global warming, some compound events with low likelihood in past and current climate will become more frequent, and there is a higher chance of historically unprecedented events and surprises (high confidence). However, even extreme events that do not have a particularly low probability in the present climate (at more than 1°C of global warming) can be perceived as surprises because of the pace of global warming (high confidence).

11.2 Data and Methods

This section provides an assessment of observational data and methods used in the analysis and attribution of climate change specific to weather and climate extremes. It also introduces some concepts used in presenting future projections of extremes. Later sections (Sections 11.3–11.8) also provide additional assessments on relevant observational datasets and model validation specific to the type of extremes to be assessed. General background on climate modelling is provided in Chapters 4 and 10.

11.2.1 Definition of Extremes

In the literature, an event is generally considered extreme if the value of a variable exceeds (or lies below) a threshold. The thresholds have been defined in different ways, leading to differences in the meaning of extremes that may share the same name. For example, two sets of metrics for the frequency of hot/warm days have been used in the literature. One set counts the number of days when maximum daily temperature is above a relative threshold defined as the 90th or higher percentile of maximum daily temperature for the calendar day over a base period. An event based on such a definition can occur at any time of the year, and the impact of such an event would differ depending on the season. The other set counts the number of days in which maximum daily temperature is above an absolute threshold such as 35°C, because exceeding this temperature can sometimes cause health impacts (however, these impacts may depend on location and whether ecosystems and the population are adapted to such temperatures). While both types of hot extreme indices have been used to analyse changes in the frequency of hot/warm events, they represent different events that occur at different times of the year, possibly affected by different types of processes and mechanisms, and possibly also associated with different impacts.

Changes in extremes have also been examined from two perspectives: changes in the frequency for a given magnitude of extremes; or changes in the magnitude for a particular return period (frequency). Changes in the probability of extremes (e.g., temperature extremes) depend on the rarity of the extreme event that is assessed, with a larger change in probability associated with a rarer event (e.g., Kharin et al., 2018). However, changes in the magnitude represented by the return levels of the extreme events may not be as sensitive to the rarity of the event. While the answers to the two different questions are related, their relevance may differ for distinct audiences. Conclusions regarding the respective contribution of greenhouse gas forcing to changes in magnitude versus frequency of extremes may also differ (Otto et al., 2012). Correspondingly, the sensitivity of changes in extremes to increasing global warming is also dependent on the definition of the considered extremes. In the case of temperature extremes, changes in magnitude have been shown to often depend linearly on global surface temperature (Seneviratne et al., 2016; Wartenburger et al., 2017), while changes in frequency tend to be nonlinear and can, for example, be exponential for increasing global warming levels (Fischer and Knutti, 2015; Kharin et al., 2018). When similar damage occurs once a fixed threshold is exceeded, it is more important to ask a question regarding changes in the frequency. But when the exceedance of this fixed threshold becomes a normal occurrence in the future, this can lead to a saturation in the change of probability (Harrington and Otto, 2018a). Also, if the impact of an event increases with the intensity of the event, it would be more relevant to examine changes in the magnitude. Finally, adaptation to climate change might change the relevant thresholds over time, although such aspects are still rarely integrated in the assessment of projected changes in extremes. Framing is considered when forming the assessments of this Chapter, including how extremes are defined and how the questions are asked in the literature .

11.2.2 Data

Studies of past and future changes in weather and climate extremes, and in the mean state of the climate, use the same original sources of weather and climate observations, including in situ observations, remotely sensed data, and derived data products such as reanalyses. Sections 2.3 and 10.2 assess various aspects of these data sources and data products from the perspective of their general use, and in the analysis of changes in the mean state of the climate in particular. Building on these previous chapters, this subsection highlights particular aspects that are related to extremes and are most relevant to the assessment of this Chapter. The SREX (Chapter 3, Seneviratne et al., 2012) and AR5 (Chapter 2, Hartmann et al., 2013) addressed critical issues regarding the quality and availability of observed data and their relevance for the assessment of changes in extremes.

Extreme weather and climate events occur on time scales of hours (e.g., convective storms that produce heavy precipitation) to days (e.g., tropical cyclones, heatwaves), to seasons and years (e.g., droughts). A robust determination of long-term changes in these events can have different requirements for the spatial and temporal scales and sample size of the data. In general, it is more difficult to determine long-term changes for events of fairly large temporal duration, such as ‘megadroughts’ that last several years or longer (e.g., Ault et al., 2014), because of the limitations of the observational sample size. Literature that studies changes in extreme precipitation and temperature often uses indices representing specifics of extremes that are derived from daily precipitation and temperature values. Station-based indices would have the same issues as those for the mean climate regarding the quality, availability, and homogeneity of the data. For the purpose of constructing regional information and/or for comparison with model outputs, such as model evaluation, and detection and attribution, these station-based indices are often interpolated onto regular grids. Two different approaches, involving two different orders of operation, have been used in producing such gridded datasets.

In some cases, such as for the HadEX3 dataset (Dunn et al., 2020), indices of extremes are computed using time series directly derived from stations first, and are then gridded over the space. As the indices are computed at the station level, the gridded data products represent point estimates of the indices averaged over the spatial scale of the grid box. In other instances, daily values of station observations are first gridded (e.g., Contractor et al., 2020a), and the interpolated values can then be used to compute various indices. Depending on the station density, values for extremes computed from data gridded this way represent extremes of spatial scales anywhere from the size of the grid box to a point. In regions with high station density (e.g., North America, Europe), the gridded values are closer to extremes of area means and are thus more appropriate for comparisons with extremes estimated from climate model output, which is often considered to represent areal means (Chen and Knutson, 2008; Gervais et al., 2014; Avila et al., 2015; Di Luca et al., 2020b). In regions with very limited station density (e.g., Africa), the gridded values are closer to point estimates of extremes. The difference in spatial scales among observational data products and model simulations needs to be carefully accounted for when interpreting the comparison among different data products. For example, the average annual maximum daily maximum temperature (TXx) over land computed from the original ERA-Interim reanalysis (at 0.75° resolution) is about 0.4°C warmer than that computed when the ERA-Interim dataset is upscaled to the resolution of 2.5° × 3.75° (Di Luca et al., 2020).

Extreme indices computed from various reanalysis data products have been used in some studies, but reanalysis extreme statistics have not been rigorously compared to observations (Donat et al., 2016a).

In general, changes in temperature extremes from various reanalyses were most consistent with gridded observations after about 1980, but larger differences were found during the pre-satellite era (Donat et al., 2014b). Overall, lower agreement across reanalysis datasets was found for extreme precipitation changes, although temporal and spatial correlations against observations were found to be still significant. In regions with sparse observations (e.g., Africa and parts of South America), there is generally less agreement for extreme precipitation between different reanalysis products, indicating a consequence of the lack of an observational constraint in these regions (Donat et al., 2014b, 2016a). More recent reanalyses, such as ERA5 (Hersbach et al., 2020), seem to have improved over previous products, at least over some regions (e.g., Mahto and Mishra, 2019; Gleixner et al., 2020; Sheridan et al., 2020). Caution is needed when reanalysis data products are used to provide additional information about past changes in these extremes in regions where observations are generally lacking.

Satellite remote sensing data have been used to provide information about precipitation extremes because several products provide data at sub-daily resolution for precipitation, for example, Tropical Rainfall Measuring Mission (TRMM; Maggioni et al., 2016) and clouds, for example, Himawari (Bessho et al., 2016; Chen et al., 2019). However, satellites do not observe the primary atmospheric state variables directly and polar orbiting satellites do not observe any given place at all times. Hence, their utility as a substitute for high-frequency (i.e., daily) ground-based observations is limited. For instance, Timmermans et al. (2019) found little relationship between the timing of extreme daily and five-day precipitation in satellite and gridded station data products over the USA.

Box 11.3 | Extremes in Paleoclimate Archives Compared to Instrumental Records

Examining extremes in pre-instrumental information can help to put events occurring in the instrumental record (referred to as ‘observed’) in a longer-term context. This box focuses on extremes in the Common Era (CE, the last 2000 years), because there is generally higher confidence in pre-instrumental information gathered from the more recent archives from the Common Era than from earlier evidence. It addresses evidence of extreme events in paleoreconstructions, documentary evidence (such as grape harvest data, religious documents, newspapers, and logbooks) and model-based analyses, and whether observed extremes have or have not been exceeded in the Common Era. This box provides overviews of: (i) AR5 assessments; (ii) types of evidence assessed here; evidence of: (iii) droughts; (iv) temperature extremes; (v) paleofloods; and (vi) paleotempests; and (vii) a summary.

(Chapter 5 of AR5 (Masson-Delmotte et al., 2013) concluded with high confidence that droughts of greater magnitude and of longer duration than those observed in the instrumental period occurred in many regions during the preceding millennium. There was high confidence in evidence that floods during the past five centuries in northern and Central Europe, the western Mediterranean region, and eastern Asia were of a greater magnitude than those observed instrumentally, and medium confidence in evidence that floods in the Near East, India and Central North America were comparable to modern observed floods. While AR5 assessed 20th century summer temperatures compared to those reconstructed in the Common Era, it did not assess shorter duration temperature extremes.

Many factors affect confidence in information on pre-instrumental extremes. First, the geographical coverage of paleoclimate reconstructions of extremes is not spatially uniform (Smerdon and Pollack, 2016) and depends on both the availability of archives and records, which are environmentally dependent, and also the differing attention and focus from the scientific community. In Australia, for example, the paleoclimate network is sparser than for other regions, such as Asia, Europe and North America, and synthesized products rely on remote proxies and assumptions about the spatial coherence of precipitation between remote climates (Cook et al., 2016c; Freund et al., 2017). Second, pre-instrumental evidence of extremes may be focused on understanding archetypal extreme events, such as the climatic consequences of the 1815 eruption of Mount Tambora, Indonesia (Veale and Endfield, 2016). These studies provide narrow evidence of extremes in response to specific forcings (M. Li et al., 2017) for specific epochs. Third, natural archives may provide information about extremes in one season only and may not represent all extremes of the same types.

Evidence of shorter duration extreme event types, such as floods and tropical storms, is further restricted by the comparatively low chronological controls and temporal resolution (e.g., monthly, seasonal, yearly, multiple years) of most archives compared to the events (e.g., minutes to days). Natural archives may be sensitive only to intense environmental disturbances, and so only sporadically record short-duration or small spatial-scale extremes. Interpreting sedimentary records as evidence of past short-duration extremes is also complex and requires a clear understanding of natural processes (Wilhelm et al. , 2019). For example, paleoflood reconstructions of flood recurrence and intensity produced from geological evidence (e.g., river and lake sediments), speleothems (Denniston and Luetscher, 2017), botanical evidence (e.g., flood damage to trees, or tree ring reconstructions), and floral and faunal evidence (e.g., diatom fossil assemblages) require understanding of sediment sources and flood mechanisms. Pre-instrumental records of tropical storm intensity and frequency (also called paleotempest records) derived from overwash deposits of coastal lake and marsh sediments are difficult to interpret. Many factors have an impact on whether disturbances are deposited in archives (Muller et al., 2017) and deposits may provide sporadic and incomplete preservation histories (e.g., Tamura et al., 2018).

Overall, the most complete pre-instrumental evidence of extremes occurs for long-duration, large spatial-scale extremes, such as for multi-year meteorological droughts or seasonal- and regional-scale temperature extremes. Additionally, more precise insights into recent extremes emerge where multiple studies have been undertaken, compared to the confidence in extremes reported at single sites or in single studies, which may not necessarily be representative of large-scale changes, or for reconstructions that synthesize multiple proxies over large areas (e.g., drought atlases). Multiproxy synthesis products combine paleoclimate temperature reconstructions and cover sub-continental- to hemispheric-scale regions to provide continuous records of the Common Era(e.g., Ahmedet al. , 2013; Neukomet al. , 2014 for temperature).

There is high confidence in the occurrence of long-duration and severe drought events during the Common Era for many locations, although their severity compared to recent drought events differs between locations and the lengths of reconstruction provided. Recent observed drought extremes in some regions – such as the eastern Mediterranean Levant (Cook et al., 2016a), California in the USA(Cook et al., 2014b; Griffin and Anchukaitis, 2014), and in the Andes (Garreaud et al. , 2017; Domínguez-Castro et al. , 2018) – do not have precedents within the multi-century periods reconstructed in these studies, in terms of duration and/or severity. In some regions (in south-western North America(Asmerom et al., 2013; Cook et al., 2015), the Great Plains region (Cook et al., 2004), the Middle East (Kaniewski et al., 2012), and China (Gou et al., 2015)), recent drought extremes may have been exceeded in the Common Era. In further locations, there is conflicting evidence for the severity of pre-instrumental droughts compared to observed extremes, depending on the length of the reconstruction and the seasonal perspective provided(see Cooket al. , 2016c; Freundet al. , 2017 for Australia). There can also be differing conclusions for the severity, or even the occurrence, of specific individual pre-instrumental droughts when different evidence is compared (e.g., Wetter et al., 2014; Büntgen et al., 2015).

There is medium confidence that the magnitudes of large-scale, seasonal-scale extreme high temperatures in observed records exceed those reconstructed over the Common Era in some locations, such as Central Europe. In one example, multiple studies have examined the unusualness of present-day European summer temperature records in a long-term context, particularly in comparison to the exceptionally warm year of 1540 CE in Central Europe. Several studies indicate that recent extreme summers (2003 and 2010) in Europe have been unusually warm in the context of the last 500 years (Barriopedro et al., 2011; Wetter and Pfister, 2013; Wetter et al., 2014; Orth et al., 2016b), or longer (Luterbacher et al., 2016). Others studies show that summer temperatures in Central Europe in 1540 were warmer than the present-day (1966–2015) mean, but note that it is difficult to assess whether or not the 1540 summer was warmer than observed record extreme temperatures (Orth et al., 2016b).

There is high confidence that the magnitude of floods over the Common Era exceeded observed records in some locations, including Central Europe and eastern Asia. Recent literature supports the AR5 assessments of floods (Masson-Delmotte et al., 2013). For example, high temporally resolved records provide evidence of Common Era floods exceeding the probable maximum flood levels in the Upper Colorado River, USA (Greenbaum et al., 2014) and peak discharges that are double gauge levels along the middle Yellow River, China (Liu et al., 2014). Further studies demonstrate pre-instrumental or early instrumental differences in flood frequency compared to the instrumental period, including reconstructions of high and low flood frequency in the European Alps (e.g., Swierczynski et al., 2013; Amann et al., 2015) and Himalayas (Ballesteros Cánovas et al., 2017). The combination of extreme historical flood episodes determined from documentary evidence also increases confidence in the determination of flood frequency and magnitude, compared to using geomorphological archives alone (Kjeldsen et al., 2014). In regions, such as Europe and China, that have rich historical flood documents, there is strong evidence of high-magnitude flood events over pre-instrumental periods (Kjeldsen et al., 2014; Benito et al., 2015; Macdonald and Sangster, 2017). A key feature of paleoflood records is variability in flood recurrence at centennial timescales (Wilhelm et al., 2019), although constraining climate-flood relationships remains challenging. Pre-instrumental floods often occurred in considerably different contexts in terms of land use, irrigation, and infrastructure, and may not provide direct insight into modern river systems, which further prevents long-term assessments of flood changes being made based on these sources.

There is medium confidence that periods of both more and less tropical cyclone activity (frequency or intensity) than observed occurred over the Common Era in many regions. Paleotempest studies cover a limited number of locations that are predominantly coastal, and hence provide information on specific locations that cannot be extrapolated basin-wide (see Muller et al., 2017). In some locations, such as the Gulf of Mexico and the New England, USA, coast, similarly intense storms to those observed recently have occurred multiple times over centennial timescales (Donnelly et al., 2001; Bregy et al., 2018). Further research focused on the frequency of tropical storm activity. Extreme storms occurred considerably more frequently in particular periods of the Common Era, compared to the instrumental period in north-east Queensland, Australia (Nott et al., 2009; Haig et al., 2014), and the Gulf Coast (e.g., Brandon et al., 2013; Lin et al., 2014).

The probability of finding an unprecedented extreme event increases with a longer length of past record-keeping, in the absence of longer-term trends. Thus, as a record is extended to the past based on paleoreconstruction, there is a higher chance of very rare extreme events having occurred at some time prior to instrumental records. Such an occurrence is not, in itself, evidence of a change, or lack of a change, in the magnitude or the likelihood of extremes in the past or in the instrumental period at regional and local scales. Yet, the systematic collection of paleoclimate records over wide areas may provide evidence of changes in extremes. In one study, extended evidence of the last millennium from observational data and paleoclimate reconstructions using tree rings indicates that human activities affected the worldwide occurrence of droughts as early as the beginning of the 20th century (Marvel et al., 2019).

In summary, there is low confidence in overall changes in extremes derived from paleo-archives. There is high confidence that long-duration and severe drought events occurred at many locations during the last 2000 years. There is also high confidence that high-magnitude flood events occurred at some locations during the last 2000 years, but overall changes in infrastructure and human water management make the comparison with present-day records difficult. But these isolated paleo-drought and paleo-flood events are not evidence of a change, or lack of a change, in the magnitude or the likelihood of relevant extremes.

11.2.3 Attribution of Extremes

Attribution science concerns the identification of causes for changes in characteristics of the climate system (e.g., trends, single extreme events). A general overview and summary of methods of attribution science is provided in the Cross-Working Group Box 1.1. Trend detection using optimal fingerprinting methods is a well-established field, and has been assessed in AR5 (Chapter 10, Bindoff et al., 2013), and Section 3.2.1 of this Report. There are specific challenges when applying optimal fingerprinting to the detection and attribution of trends in extremes and on regional scales where the lower signal-to-noise ratio is a challenge. In particular, the method generally requires the data to follow a Normal (Gaussian) distribution, which is often not the case for extremes. However, recent studies showed that extremes can be transformed to a Gaussian distribution, for example, by averaging over space, so that optimal fingerprinting techniques can still be used (Wen et al., 2013; Zhang et al., 2013; Wan et al., 2019). Non-stationary extreme value distributions, which allow for the detailed detection and attribution of regional trends in temperature extremes, have also been used (Z. Wang et al., 2017a).

Apart from the detection and attribution of trends in extremes, new approaches have been developed to answer the question of whether, and to what extent, external drivers have altered the probability and intensity of an individual extreme event (NASEM, 2016). In AR5, there was an emerging consensus that the role of external drivers of climate change in specific extreme weather events could be estimated and quantified in principle, but related assessments were still confined to particular case studies, often using a single model, and typically focusing on high-impact events with a clear attributable signal.

However, since AR5, the attribution of extreme weather events has emerged as a growing field of climate research with an increasing body of literature (see series of supplements to the annual State of the Climate report (Peterson et al., 2012, 2013a; Herring et al., 2014, 2015, 2016, 2018), including the number of approaches to examining extreme events(described in Easterling et al., 2016; Otto, 2017; Stott et al., 2016)). A commonly used approach – often called the risk-based approach in the literature, and referred to here as the ‘probability-based approach’ – produces statements such as ‘anthropogenic climate change made this event type twice as likely ’ or ‘anthropogenic climate change made this event 15% more intense’. This is done by estimating probability distributions of the index characterizing the event in today’s climate, as well as in a counterfactual climate, and either comparing intensities for a given occurrence probability (e.g., 1-in-100-year event) or probabilities for a given magnitude (see FAQ 11.3). There are a number of different analytical methods encompassed in the probability-based approach, building on observations and statistical analyses (e.g., van Oldenborgh et al., 2012), optimal fingerprint methods (Sun et al., 2014), regional climate and weather forecast models (e.g., Schaller et al., 2016), global climate models (GCMs) (e.g., Lewis and Karoly, 2013), and large ensembles of atmosphere-only GCMs (e.g., Lott et al., 2013). A key component in any event attribution analysis is the level of conditioning on the state of the climate system. In the least conditional approach, the combined effect of the overall warming and changes in the large-scale atmospheric circulation are considered and often utilize fully coupled climate models (Sun et al., 2014). Other more conditional approaches involve prescribing certain aspects of the climate system. These range from prescribing the pattern of the surface ocean change at the time of the event (e.g., Hoerling et al., 2013, 2014), often using Atmospheric Model Intercomparison Project (AMIP) style global models, where the choice of sea surface temperature and ice patterns influences the attribution results (Sparrow et al., 2018), to prescribing the large-scale circulation of the atmosphere and using weather forecasting models or methods (e.g., Pall et al., 2017; Patricola and Wehner, 2018; Wehner et al., 2018a). These highly conditional approaches have also been called ‘storylines’ (Cross-Working Group Box 1.1; Shepherd, 2016) and can be useful when applied to extreme events that are too rare to otherwise analyse, or where the specific atmospheric conditions were central to the impact. These methods are also used to enable the use of very high-resolution simulations in cases were lower-resolution models do not simulate the regional atmospheric dynamics well (Shepherd, 2016; Shepherd et al., 2018). However, the imposed conditions limit an overall assessment of the anthropogenic influence on an event, as the fixed aspects of the analysis may also have been affected by climate change. For instance, the specified initial conditions in the highly conditional hindcast attribution approach often applied to tropical cyclones (e.g., Takayabu et al., 2015; Patricola and Wehner, 2018) permit only a conditional statement about the magnitude of the storm if similar large-scale meteorological patterns could have occurred in a world without climate change, thus precluding any attribution statement about the change in frequency if used in isolation. Combining conditional assessments of changes in the intensity with a multi-model approach does allow for the latter as well (Shepherd, 2016).

The outcome of event attribution is dependent on the definition of the event (Leach et al., 2020), as well as the framing (Otto et al., 2016; Christidis et al., 2018; Jézéquel et al., 2018) and uncertainties in observations and modelling. Observational uncertainties arise in estimating the magnitude of an event as well as its rarity (Angélil et al., 2017). Results of attribution studies can also be very sensitive to the choice of climate variables (Sippel and Otto, 2014; Wehner et al., 2016). Attribution statements are also dependent on the spatial (Uhe et al., 2016; Cattiaux and Ribes, 2018; Kirchmeier‐Young et al., 2019) and temporal (Harrington, 2017; Leach et al., 2020) extent of event definitions, as events of different scales involve different processes (W. Zhang et al., 2020) and large-scale averages generally yield higher attributable changes in magnitude or probability due to the smoothing out of noise. In general, confidence in attribution statements for large-scale heat and lengthy extreme precipitation events have higher confidence than shorter and more localized events, such as extreme storms, an aspect also relevant for determining the emergence of signals in extremes or the confidence in projections (see also Cross-Chapter Box Atlas.1).

The reliability of the representation of the event in question in the climate models used in a study is essential (Angélil et al., 2016; Herger et al., 2018). Extreme events characterized by atmospheric dynamics that stretch the capabilities of current-generation models (Section 10.3.3.4; Shepherd, 2014; Woollings et al., 2018) limit the applicability of the probability-based approach of event attribution. The lack of model evaluation, in particular in early event attribution studies, has led to criticism of the emerging field of attribution science as a whole (Trenberth et al., 2015) and of individual studies (Angélil et al., 2017). In this regard, the storyline approach (Shepherd, 2016) provides an alternative option that does not depend on the model’s ability to represent the circulation reliably. In addition, several ways of quantifying statistical uncertainty (Paciorek et al., 2018) and model evaluation (Lott and Stott, 2016; Philip et al., 2018b, 2020) have been employed to evaluate the robustness of event attribution results. For the unconditional probability-based approach, multi-model and multi-approach (e.g., combining observational analyses and model experiments) methods have been used to improve the robustness of event attribution (Hauser et al., 2017; Otto et al., 2018a; Philip et al., 2018b, 2019, 2020; van Oldenborgh et al., 2018; Kew et al., 2019).

In the regional tables provided in Section 11.9, the different lines of evidence from event attribution studies and trend attributions are assessed alongside one another to provide an assessment of the human contribution to observed changes in extremes in all AR6 regions.

11.2.4 Projecting Changes in Extremes as a Function of Global Warming Levels

The most important quantity used to characterize past and future climate change is global warming relative to its pre-industrial level. Changes in global warming are linked quasi-linearly to global cumulative carbon dioxide (CO2) emissions (IPCC, 2013), and for their part, changes in regional climate, including many types of extremes, scale quasi-linearly with changes in global warming, often independently of the underlying emissions scenarios (SR1.5 Chapter 3; Seneviratne et al., 2016; Matthews et al., 2017; Wartenburger et al., 2017; Kharin et al., 2018; Y. Sun et al., 2018a; Tebaldi and Knutti, 2018; Beusch et al., 2020; Li et al., 2021). In addition, the use of global warming levels in the context of global policy documents – in particular the 2015 Paris Agreement (UNFCCC, 2016) implies that information on changes in the climate system, and specifically extremes, as a function of global warming are of particular policy relevance. Cross-Chapter Box 11.1 provides an overview on the translation between information at global warming levels (GWLs) and scenarios.

The assessment of projections of future changes in extremes as function of GWL has an advantage in separating uncertainty associated with the global warming response (see Chapter 4) from the uncertainty resulting from the regional climate response as a function of GWLs (Seneviratne and Hauser, 2020). If the interest is in the projection of regional changes at certain GWLs, such as those defined by the Paris Agreement, projections based on time periods and emissions scenarios have unnecessarily larger uncertainty due to differences in model global transient climate responses. To take advantage of this feature and to provide easy comparison with SR1.5, assessments of projected changes in this chapter are largely provided in relation to future GWLs, with a focus on changes at +1.5°C, +2°C, and +4°C of global warming above pre-industrial levels (e.g., Tables 11.1 and 11.2 and regional tables in Section 11.9). These encompass a scenario compatible with the lowest limit of the Paris Agreement (+1.5°C), a scenario slightly overshooting the aims of the Paris Agreement (+2°C), and a ‘worst-case’ scenario with no mitigation (+4°C). Cross-Chapter Box 11.1 provides a background on the GWL sampling approach used in AR6, for the computation of GWL projections from climate models contributing to Phase 6 of the Coupled Model Intercomparison Project (CMIP6) as well as for the mapping of existing scenario-based literature for CMIP6 and the CMIP Phase 5 (CMIP5) to assessments as function of GWLs (see also Section 11.9. and Table 11.3 for an example).

While regional changes in many types of extremes do scale robustly with global surface temperature, generally irrespective of emissions scenarios (Section 11.1.4, Figures 11.3, 11.6 and 11.7 and Cross-Chapter Box 11.1), effects of local forcing can distort this relation. For example, emissions scenarios with the same radiative forcing can have different regional extreme precipitation responses resulting from different aerosol forcing (Z. Wang et al., 2017b). Another example is related to forcing from land-use and land cover changes (Section 11.1.6). Climate models often either overestimate or underestimate observed changes in annual maximum daily maximum temperature, depending on the region and considered models (Donat et al., 2017; Vautard et al., 2020). Part of the discrepancies may be due to the lack of representation of some land forcings, in particular crop intensification and irrigation (N.D. Mueller et al., 2016; Findell et al., 2017; Thiery et al., 2017, 2020). Since these local forcings are not represented, and their future changes are difficult to project, these can be important caveats when using GWL scaling to project future changes for these regions. However, these caveats also apply to the use of scenario-based projections.

The SR1.5 (Chapter 3) assessed different climate responses at +1.5°C of global warming, including transient climate responses, short-term stabilization responses, and long-term equilibrium stabilization responses, and their implications for future projections of different extremes. Indeed, the temporal dimension – that is, when the given GWL occurs – also matters for projections, in particular beyond the 21st century, and for some climate variables related to components of the climate system associated with large inertia (e.g., sea level rise and associated extremes). Nonetheless, for assessments focused on conditions within the next decades, and for the main extremes considered in this chapter, derived projections are relatively insensitive to details of climate scenarios and can be well-estimated based on transient simulations (Cross-Chapter Box 11.1; see also SR1.5).

An important question is the identification of the GWL at which a given change in a climate extreme can begin to emerge from climate noise. Figure 11.8 displays analyses of the GWLs at which emergence in hot extremes – annual maximum daily temperature represented by TXx and heavy precipitation represented by Rx1day is identified in AR6 regions for the whole CMIP5 and CMIP6 ensembles. Overall, signals for extremes emerge very early for TXx, already below 0.2°C in many regions (Figure 11.8a,b), and at around 0.5°C in most regions. This is consistent with conclusions from the SR1.5 (Chapter 3 for less-rare temperature extremes (TXx on the yearly time scale), which shows that a difference as small as 0.5°C of global warming – for example, between +1.5°C and +2°C of global warming – leads to detectable differences in temperature extremes in TXx in most WGI AR6 regions in CMIP5 projections (e.g., Wartenburger et al., 2017; Seneviratne et al., 2018b). The GWL emergence for Rx1day is also largely consistent with analyses for less-extreme heavy precipitation events (Rx5day on the yearly time scale) in SR1.5 (see Chapter 3).

Figure 11.8 | Global and regional-scale emergence of changes in temperature (a) and precipitation (b) extremes for the globe (glob.), global oceans (oc.), global lands (land), and the AR6 regions. Colours indicate the multi-model mean global warming level at which the difference in 20-year means of the annual maximum daily maximum temperature (TXx) and the annual maximum daily precipitation (Rx1day) become significantly different from their respective mean values during the 1850–1900 base period. Results are based on simulations from the Coupled Model Intercomparison Project Phases 5 and 6 (CMIP5 and CMIP6) multi-model ensembles. See Atlas.1.3.2 for the definition of regions. Adapted from Seneviratne and Hauser (2020) under the terms of the Creative Commons Attribution licence.

To some extent, analyses as functions of GWLs replace the time axis with a global surface temperature axis. Nonetheless, information on the timing of given changes in extremes is obviously also relevant. (For information on the time frame at which given GWLs are reached, see Cross-Chapter Box 11.1 and Section 4.6). Figure 11.5 provides a synthesis of attributed and projected changes in extremes as function of GWLs (see also Figures. 11.3, 11.6 and 11.7 for regional analyses).

Cross-Chapter Box 11.1 | Translating Between Regional Information at Global Warming Levels Versus Scenarios for End Users

Contributors: Erich Fischer (Switzerland), Mathias Hauser (Switzerland), Sonia I. Seneviratne (Switzerland), Richard Betts (United Kingdom), José M. Gutiérrez (Spain), Richard G. Jones (United Kingdom), June-Yi Lee (Republic of Korea), Malte Meinshausen (Australia/Germany), Friederike Otto (United Kingdom/Germany), Izidine Pinto (Mozambique), Roshanka Ranasinghe (The Netherlands/Sri Lanka/Australia), Joeri Rogelj (Germany/Belgium), Bjørn Samset (Norway), Claudia Tebaldi (United States of America), Laurent Terray (France)

Background

Traditionally, projections of climate variables are summarized and communicated as function of time and emissions scenarios. Recently, quantifying global and regional climate at specific global warming levels (GWLs) has become widespread, motivated by the inclusion of explicit GWLs in the long-term temperature goal of the Paris Agreement (Section 1.6.2). GWLs, expressed as changes in global surface temperature relative to the 1850–1900 period (see Cross-Chapter Box 2.3), are used in SR1.5 and in the assessment of Reasons for Concerns in the WGII reports (see also Cross-Chapter Box 12.1). Cross-Chapter Box 11.1, Figure 1 illustrates how the assessment of the climate response at GWLs relates to the uncertainty in scenarios regarding the timing of the respective GWLs, as well as to the uncertainty in the associated regional climate responses, including extremes and other climatic impact-drivers (CIDs). For many (but not all) climate variables and CIDs, the response pattern for a given GWL is consistent across different scenarios (Chapters 1, 4, 9, 11 and Atlas). GWLs are defined as long-term means (e.g., 20-year averages) compared to the pre-industrial period, are commonly used in the literature, and were also underlying main assessments of SR1.5 (Chapter 3).

Cross-Chapter Box 11.1, Figure 1 | Schematic representation of relationship between emissions scenarios, global warming levels (GWLs), regional climate responses, and impacts. The illustration shows the implied uncertainty problem associated with differentiating between 1.5°C, 2°C, and other GWLs. Focusing on GWLs raises questions associated with emissions pathways to get to these temperatures (scenarios), as well as regional climate responses and the associated impacts at the corresponding GWL (the impacts question). Adapted from James et al. (2017) and Rogelj (2013) under the terms of the Creative Commons Attribution licence.

Numerous studies have compared the regional response to anthropogenic forcing at GWLs in annual and seasonal mean values and extremes of different climate and impact variables across different multi-model ensembles and/or different scenarios (e.g., Frieleret al. , 2012; Scheweet al. , 2014; Hergeret al. , 2015; Schleussneret al. , 2016; Seneviratneet al. , 2016; Wartenburgeret al. , 2017; Bettset al. , 2018; Dosio and Fischer, 2018; Samset et al. , 2019; Tebaldiet al. , 2020; see Sections 4.6.1, 8.5.3, 9.3.1, 9.5, 9.6.3, 10.4.3 and 11.2.4 for further details). The regional response patterns at given GWLs have been found to be consistent across different scenarios for many climate variables (Cross-Chapter Box 11.1 Figure 2; Pendergrasset al. , 2015; Seneviratneet al. , 2016; Wartenburgeret al. , 2017; Seneviratne and Hauser, 2020). The consistency tends to be higher for temperature-related variables than for variables in the hydrological cycle or variables characterizing atmospheric dynamics, and for intermediate to high-emissions scenarios than for low-emissions scenarios (e.g., for mean precipitation in the Representative Concentration Pathway (RCP) 2.6 scenario: Pendergrass et al., 2015; Wartenburger et al., 2017). Nonetheless, Cross-Chapter Box 11.1 Figure 2 illustrates that, even for mean precipitation, which is known to be forcing dependent (Sections 4.6.1 and 8.5.3), scenario differences in the response pattern at a given GWL are smaller than model uncertainty and internal variability in many regions (Herger et al., 2015). The response pattern is further found to be broadly consistent between models that reach a GWL relatively early, and those that reach it later under a given Shared Socio-economic Pathway (SSP; see Cross-Chapter Box 11.1, Figure 2g,h).

Cross-Chapter Box 11.1, Figure 2 | (a–c) Coupled Model Intercomparison Project Phase 6 (CMIP6) multi-model mean precipitation change at 2°C global warming level (GWL) (20-year mean) in three different Shared Socio-economic Pathway (SSP) scenarios relative to 1850–1900. All models reaching the corresponding GWL in the corresponding scenario are averaged. The number of models averaged across is shown at the top right of the panel. The maps for the other two SSP scenarios SSP1-1.9 (five models only) and SSP3-7.0 (not shown) are consistent. (d–f) Same as (a–c) but for annual mean temperature. (g) Annual mean temperature change at 2°C in CMIP6 models with high warming rate reaching the GWL in the corresponding scenario before the earliest year of the assessed very likely range (Section 4.3.4). (h) Climate response at 2°C GWL across all SSP1-1.9, SSP2-2.6, SSP2-4.5. SSP3-7.0 and SSP5-8.5 in all other models not shown in (g). The close agreement of (g) and (h) demonstrates that the mean temperature response at 2°C is not sensitive to the rate of warming, and thereby the global mean surface air temperature (GSAT) warming of the respective models in 2081–2100. Uncertainty is represented using the advanced approach: No overlay indicates regions with robust signal, where ≥66% of models show change greater than the variability threshold and ≥80% of all models agree on the sign of change; diagonal lines indicate regions with no change or no robust signal, where <66% of models show a change greater than the variability threshold; crossed lines indicate regions with conflicting signal, where ≥66% of models show change greater than the variability threshold and <80% of all models agree on the sign of change. For more information on the advanced approach, please refer to the Cross-Chapter Box Atlas.1.

In contrast to linear pattern scaling (Mitchell, 2003; Collins et al., 2013), the use of GWLs as a dimension of integration does not require linearity in the response of a climate variable. It is therefore useful even for metrics that do not show a linear response, such as the frequency of heat extremes over land and oceans (Fischer and Knutti, 2015; Frölicher et al., 2018; Kharin et al., 2018; Perkins-Kirkpatrick and Gibson, 2017) if the relationship of the variable of interest to the GWL is scenario independent. The latter means that the response is independent of the pathway and relative contribution of various radiative forcings. For some more complex indices like warm-spell duration, or for regions with strong aerosol changes, discrepancies can be larger (Z. Wang et al., 2017b; King et al., 2018; Tebaldi et al., 2020). (See also the subsection below on GWLs vs scenarios for further caveats.)

The limited scenario dependence of the GWL-based response for many variables implies that the regional response to emissions scenarios can be split in almost independent contributions of: (i) the transient global warming response to scenarios (see Chapter 4); and (ii) the regional response as function of a given GWL, which has also been referred to as ‘regional climate sensitivity’ (Seneviratne and Hauser, 2020). This property has also been used to develop regionally resolved emulators for global climate models, using global surface temperature as input (Beusch et al., 2020; Tebaldi et al., 2020). Analyses of the CMIP6 and CMIP5 multi-model ensembles shows that the GWL-based responses are very similar for temperature and precipitation extremes across the ensembles (Seneviratne and Hauser, 2020; Wehner, 2020; Li et al. , 2021). This is despite their difference in global warming response (Chapter 4), confirming a substantial decoupling between the two responses (global warming vs GWL-based regional response) for these variables. Thus, the GWL approach isolates the uncertainty in the regional climate response from the global warming uncertainty induced by scenario, global mean model response and internal variability (Cross-Chapter Box, Figure 1).

Mapping between GWL- and scenario-based responses in model analyses

To map scenario-based climate projections into changes at specific GWLs, first, all individual Earth system model (ESM) simulations that reach a certain GWL are identified. Second, the climate response patterns at the respective GWL are calculated using an approach termed here ‘GWL-sampling’ – sometimes also referred to as epoch analysis, time shift, or time sampling approach – taking into account all models and scenarios (Cross-Chapter Box, Figure 3). Note that the range of years when a given GWL is reached in the CMIP6 ensemble is different from the AR6 assessed range of projected global surface temperature (Section 4.3.4; Table 4.5). The latter further takes into account different lines of evidence, including the assessed observed warming between pre-industrial and present day, information from observational constraints on CMIP6, and emulators using the assessed transient climate response (TCR) and equilibrium climate sensitivity (ECS) ranges (Section 4.3.4). Hence the Chapter 4 assessed range (Table 4.5) is the reference to determine when a given GWL is likely reached under given scenarios, while the mapping between scenarios/time frames and GWLs is used to assess the respective regional responses happening at these time frames (which also allows accounting for the global surface temperature assessment, rather than using scenarios analyses directly from CMIP6 output).

In the model-based asssessment of Chapters 4, 8, 10, 11, 12 and the Atlas, the estimation of changes at GWLs are generally defined as the 20-year time period in which the mean global surface air temperature (GSAT; Cross-Chapter Box 2.3) first exceeds a certain anomaly relative to 1850–1900 – for simulations that start after 1850, relative to all years up to 1900 (Cross-Chapter Box Figure 3). The years when each individual model reaches a given GWL for CMIP6 and CMIP5 can be found in Hauser et al. (2021). The changes at given GWLs are identified for each ensemble member (for all scenarios) individually. Thereby, a given GWL is potentially reached a few years earlier or later in different realizations of the same model due to internal variability, but the temperature averaged across the 20-year period analysed in any simulation is consistent with the GWL. Instead of blending the information from the different scenarios, the Interactive Atlas can be used to compare the GWL spatial patterns and timings across the different scenarios (see Section (Atlas 1.3.1).

Cross-Chapter Box11.1, Figure 3 | Illustration of the AR6 global warming level (GWL) sampling approach to derive the timing and the response at a given GWL for the case of Coupled Model Intercomparison Project Phase 6 (CMIP6) data. For the mapping of scenarios/time slices into GWLs for CMIP6, please refer to Table 4.2. Respective numbers for the CMIP5 multi-model experiment are provided in Chapter 11 Supplementary Material (11.SM.1). Note that the time frames used to derive the GWL time slices can also include a different number of years (e.g., 30 years for some analyses).

Mapping between GWL- and scenario-based responsesfor literature

A large fraction of the literature considers scenario-based analyses for given time slices. When GWL-based information is required instead, an approximated mapping of the multi-model mean can be derived based on the known GWL in the given experiments for a particular time period. As a rough approximation, CMIP6 multi-model mean projections for the near-term (2021–2040) correspond to changes at about 1.5°C, and projections for the high-end scenario (SSP5-8.5) for the long-term (2081–2100) correspond to about 4°C–5°C of global warming(see Table 4.2 for changes in the CMIP6 ensemble and the Chapter 11 Supplementary Material (11.SM.1) and Hauser (2021) for details on other time periods and CMIP5). These approximated changes are used for some of the GWL-based assessments provided in the Chapter 11 regional tables (Section 11.9 and Table 11.3) when literature based on scenario projections is used to assess estimated changes at given GWLs.

GWLs versus scenarios

The use of scenarios remains a key element to inform mitigation decisions (Cross-Chapter Box 1.4), to assess which emissions pathways are consistent with a certain GWL (Cross-Chapter Box 1.4, Figure 1), to estimate when certain GWLs are reached (Section 4.3.4), and to assess for which variables it is meaningful to use GWLs as a dimension of integration. The use of scenarios is also essential for variables whose climate response strongly depends on the contribution of radiative forcing (e.g., aerosols) or land-use and land management changes, are time and warming rate dependent (e.g., sea level rise), or differ between transient and quasi-equilibrium states. Furthermore, the use of concentration or emission-driven scenario simulations is required if regional climate assessments need to account for the uncertainty in GSAT changes or climate-carbon feedbacks.

Forcing dependence of the GWL response is found for global mean precipitation (Section 8.4.3), but less for regional patterns of mean precipitation changes (Cross-Chapter Box 11.1, Figure 2). Limited dependence is found for extremes, as highlighted above. In the cryosphere, elements that are quick to respond to warming like sea ice area, permafrost and snow, show little scenario dependence (Sections 9.3.1.1, 9.5.2.3 and 9.5.3.3), whereas slow-responding variables such as ice volumes of glaciers and ice sheets respond with a substantial delay and, due to their inertia, the response depends on when a certain GWL is reached. This also applies to some extent for sea level rise where, for example, the contributions of melting glaciers and ice sheets depend on the pathway followed to reach a given GWL (Section 9.6.3.4).

In addition to the lagged effect, the climate response at a given GWL may differ before and after a period of overshoot, for example in the Atlantic Meridional Overturning Circulation (e.g., Palter et al. 2018). Finally, as assessed in IPCC SR1.5, there is a difference in the response even for temperature-related variables if a GWL is reached in a rapidly warming transient state or in an equilibrium state when the land–sea warming contrast is less pronounced (e.g., King et al. 2020). However, in this Report, GWLs are used in the context of projections for the 21st century when the climate response is mostly not in equilibrium and where projections for many variables are less dependent on the pathway than for projections beyond 2100 (Section 9.6.3.4).

Key conclusions on assessment s based on GWLs

GWL-based projections can inform society and policymakers on how climate would change under GWLs consistent with the aims of the Paris Agreement (stabilization at 1.5°C/well below 2°C), as well as on the consequences of missing these aims and reaching GWLs of 3°C or 4°C by the end of the century. The AR6 assessment shows that every bit of global warming matters and that changes in global warming of 0.5°C lead to statistically significant changes in mean climate and climate extremes on global scale and for large regions (Sections 4.6.2, 11.2.4, 11.3, 11.4, 11.6 and 11.9, Figures 11.8 and 11.9, Atlas and Interactive Atlas), as also assessed in IPCC SR1.5.

11.3 Temperature Extremes

This section assesses changes in temperature extremes at global, continental and regional scales. The main focus is on the changes in the magnitude and frequency of moderate extreme temperatures (those that occur several times a year) to very extreme temperatures (those that occur once in 10 or more years) of time scales from a day to a season, though there is a strong emphasis on the daily scale where literature is most concentrated.

11.3.1 Mechanisms and Drivers

The SREX (IPCC, 2012) and AR5 (IPCC, 2014) concluded that greenhouse gas forcing is the dominant factor for the increases in the intensity, frequency, and duration of warm extremes and the decrease in those of cold extremes. This general global-scale warming is modulated by large-scale atmospheric circulation patterns, as well as by feedbacks such as soil moisture-evapotranspiration–temperature and snow/ice-albedo–temperature feedbacks, and local forcings such as land-use change or changes in aerosol concentrations at the regional and local scales (Sections 11.1.5 and 11.1.6, and Box 11.1). Therefore, changes in temperature extremes at regional and local scales can have heterogeneous spatial distributions. Changes in the magnitudes (or intensities) of extreme temperatures are often larger than changes in global surface temperature, because of larger warming on land than on the ocean surface (Section 2.3.1.1), and because of feedbacks, though they are of similar magnitude to changes in the local mean temperature (Figure 11.2).

Extreme temperature events are associated with large-scale meteorological patterns (Grotjahn et al., 2016). Quasi-stationary anticyclonic circulation anomalies or atmospheric blocking events are linked to temperature extremes in many regions, such as in Australia (Parker et al., 2014; Perkins-Kirkpatrick et al., 2016), Europe (Brunner et al., 2017, 2018; Schaller et al., 2018), Eurasia (Yao et al., 2017), Asia (Chen et al., 2016; Ratnam et al., 2016; Rohini et al., 2016), and North America (Yu et al., 2018, 2019; Zhang and Luo, 2019). Mid-latitude planetary wave modulations affect short-duration temperature extremes such as heatwaves (Perkins, 2015; Kornhuber et al., 2020). The large-scale modes of variability (Annex IV) affect the strength, frequency and persistence of these meteorological patterns and, hence, temperature extremes. For example, cold and warm extremes in the mid-latitudes are associated with atmospheric circulation patterns such as the Pacific-North American (PNA) pattern, as well as atmosphere–ocean coupled modes such as Pacific Decadal Variability (PDV), the North Atlantic Oscillation (NAO), and Atlantic Multi-decadal Variability (AMV) (Section 11.1.5; Kamae et al., 2014; Johnson et al., 2018; Ruprich-Robert et al., 2018; Yu et al., 2018, 2020; Müller et al., 2020; Qasmi et al., 2021). Changes in the modes of variability in response to warming would therefore affect temperature extremes (Clark and Brown, 2013; Horton et al., 2015). The level of confidence in those changes varies, both in the observations and in future projections, affecting the level of confidence in changes in temperature extremes in different regions. As highlighted in Chapters 2 to 4 of this Report, it is likely that there have been observational changes in the extratropical jets and mid-latitude jet meandering (Section 2.3.1.4.3 and Cross-Chapter Box 10.1). There is low confidence in possible effects of Arctic warming on mid-latitude temperature extremes (Cross-Chapter Box 10.1). A large portion of the multi-decadal changes in extreme temperature remains after the removal of the effect of these modes of variability, and can be attributed to human influence (Kamae et al., 2017b; Wan et al., 2019). Thus, global warming dominates changes in temperature extremes at the regional scale and it is very unlikely that dynamic responses to greenhouse-gas induced warming would alter the direction of these changes.

Land–atmosphere feedbacks strongly modulate regional- and local-scale changes in temperature extremes (high confidence) (Section 11.1.6; Seneviratne et al., 2013; Lemordant et al., 2016; Donat et al., 2017; Sillmann et al., 2017b; Hirsch et al., 2019). This effect is particularly notable in mid-latitude regions where the drying of soil moisture amplifies high temperatures, especially through increases in sensible heat flux (Whan et al., 2015; Douville et al., 2016; Vogel et al., 2017). Land–atmosphere feedbacks amplifying temperature extremes also include boundary-layer feedbacks and effects on atmospheric circulation (Miralles et al., 2014a; Schumacher et al., 2019). Soil-moisture–temperature feedbacks affect past and present-day heatwaves in observations and model simulations, both locally (Miralles et al., 2014a; Cowan et al., 2016, 2020; Hauser et al., 2016; Meehl et al., 2016; Wehrli et al., 2019) and beyond the regions of feedback occurrence through changes in regional circulation patterns (Stéfanon et al., 2014; Koster et al., 2016; Sato and Nakamura, 2019). The uncertainty due to the representation of land–atmosphere feedbacks in ESMs is a cause of discrepancy between observations and simulations (Clark et al., 2006; Mueller and Seneviratne, 2014; Meehl et al., 2016). The decrease of plant transpiration or the increase of stomata resistance under enhanced CO2 concentrations is a direct CO2 forcing of land temperatures (warming due to reduced evaporative cooling), which contributes to higher warming on land (Lemordant et al., 2016; Vicente-Serrano et al., 2020b). The snow/ice-albedo feedback plays an important role in amplifying temperature variability in the high latitudes (Diro et al., 2018) and can be the largest contributor to the rapid warming of cold extremes in the mid- and high latitudes of the Northern Hemisphere (Gross et al., 2020).

Regional external forcings, including land-use changes and emissions of anthropogenic aerosols, play an important role in the changes of temperature extremes in some regions (high confidence) (Section 11.1.6). Deforestation may have contributed to about one third of the warming of hot extremes in some mid-latitude regions since the pre-industrial time (Lejeune et al., 2018). Aspects of agricultural practice, including no-till farming, irrigation, and overall cropland intensification, may cool hot temperature extremes (Davin et al., 2014; N.D. Mueller et al., 2016). For instance, cropland intensification has been suggested to be responsible for a cooling of the highest temperature percentiles in Midwest USA (N.D. Mueller et al., 2016). Irrigation has been shown to be responsible for a cooling of hot temperature extremes of up to 1°C–2°C in many mid-latitude regions in the present climate (Thieryet al., 2017, 2020), a process not represented in most of state-of-the-art ESMs (CMIP5, CMIP6). Double cropping may have led to increased hot extremes in the inter-cropping season in part of China (Jeong et al., 2014). Rapid increases in summer warming in western Europe and north-east Asia since the 1980s are linked to a reduction in anthropogenic aerosol precursor emissions over Europe (Nabat et al., 2014; Dong et al., 2016b, 2017), in addition to the effect of increased greenhouse gas forcing (see also Section 10.1.3.1). This effect of aerosols on temperature-related extremes is also noted for declines in short-lived anthropogenic aerosol emissions over North America (Mascioli et al., 2016). On the local scale, the urban heat island (UHI) effect results in higher temperatures in urban areas than in their surrounding regions, and contributes to warming in regions of rapid urbanization, in particular for nighttime temperature extremes (Box 10.3; Phelan et al., 2015; Chapman et al., 2017; Y. Sun et al., 2019). But these local and regional forcings are generally not or not well represented in the CMIP5 and CMIP6 simulations (see also Section 11.3.3), contributing to uncertainty in model simulated changes.

In summary, greenhouse gas forcing is the dominant driver leading to the warming of temperature extremes. At regional scales, changes in temperature extremes are modulated by changes in large-scale patterns and modes of variability, feedbacks including soil-moisture–evapotranspiration–temperature or snow/ice–albedo–temperature feedbacks, and local and regional forcings such as land-use and land-cover changes, or aerosol concentrations, and decadal and multi-decadal natural variability. This leads to heterogeneity in regional changes and their associated uncertainties (high confidence). Changes in anthropogenic aerosol concentrations have likely affected trends in hot extremes in some regions. Irrigation and crop expansion have attenuated increases in summer hot extremes in some regions, such as the Midwestern USA (medium confidence). Urbanization has likely exacerbated the effects of global warming in cities, in particular for nighttime temperature extremes.

11.3.2 Observed Trends

The SREX (IPCC, 2012) reported avery likely decrease in the number of cold days and nights and increase in the number of warm days and nights at the global scale. Confidence in trends was assessed as regionally variable (low to medium confidence) due to either a lack of observations or varying signals in sub-regions.

Since SREX (IPCC, 2012) and AR5 (IPCC, 2014), many regional-scale studies have examined trends in temperature extremes using different metrics that are based on daily temperatures, such as the Commission for Climatology/World Climate Research Program/Commission for Oceanography and Marine Meteorology joint Expert Team on Climate Change Detection and Indices (ETCCDI) indices (Dunn et al., 2020). The additional observational records, along with a stronger warming signal, show very clearly that changes observed at the time of AR5 (IPCC, 2014) continued, providing strengthened evidence of an increase in the intensity and frequency of hot extremes and decrease in the intensity and frequency of cold extremes. While the magnitude of the observed trends in temperature-related extremes varies depending on the region, spatial and temporal scales, and metric assessed, evidence of a warming effect is overwhelming, robust, and consistent. In particular, an increase in the intensity and frequency of hot extremes is almost always associated with an increase in the hottest temperatures and in the number of heatwave days. It is also the case for changes (decreases) in cold extremes. For this reason, and to simplify the presentation, the phrase ‘increase in the intensity and frequency of hot extremes’ is used to represent, collectively, an increase in the magnitude of extreme day and/or night temperatures, in the number of warm days and/or nights, and in the number of heatwave days. Changes in cold extremes are assessed similarly.

On the global scale, evidence of an increase in the number of warm days and nights and a decrease in the number of cold days and nights, and an increase in the coldest and hottest extreme temperatures is very robust and consistent among all variables. Figure 11.2 displays time series of globally averaged TXx and TNn on land. Warming of land mean TXx is similar to the mean temperature warming on land, which is about 45% higher than global warming (Section 2.3.1). Warming of land mean TNn is even higher, with about 3°C of warming since 1960 (Figure 11.2). Figure 11.9 shows maps of linear trends over 1960–2018 in TXx, TNn, and frequency of warm days (TX90p). The maps for TXx and TNn show trends consistent with overall warming in most regions, with a particularly high warming of TXx in Europe and north-western South America, and a particularly high warming of TNn in the Arctic. Consistent with the observed warming in global surface temperature (Section 2.3.1.2) and the observed trends in TXx and TNn, the frequency of TX90p has increased, while that of cold nights (TN10p) has decreased since the 1950s: Nearly all land regions showed statistically significant decreases in TN10p (Alexander, 2016; Dunn et al., 2020), though trends in TX90p are variable with some decreases in Southern South America, mainly during austral summer (Rusticucci et al., 2017). A decrease in the number of cold spell days is also observed over nearly all land surface areas (Easterling et al., 2016) and in the northern mid-latitudes in particular (van Oldenborgh et al., 2019). These observed changes are also consistent when a new global land surface daily air temperature dataset is analysed (P. Zhang et al., 2019). Warming trends in temperature extremes globally, and in most land areas, over the path century are also found to be consistent in a range of observation-based datasets (Fischer and Knutti, 2014; Donat et al., 2016a; Dunn et al., 2020), with the extremes related to daily minimum temperatures changing faster than those related to daily maximum temperatures (Dunn et al., 2020; see Figure 11.2). Seasonal variations in trends in temperature-related extremes have been demonstrated. A warming in warm-season temperature extremes is detected, even during the ‘slower surface global warming’ period from the late 1990s to early 2010s (Cross-Chapter Box 3.1; Kamae et al., 2014; Seneviratne et al., 2014; Imada et al., 2017). Many studies of past changes in temperature extremes for particular regions or countries show trends consistent with this global picture, as summarized below and in Tables 11.4, 11.7, 11.10, 11.13, 11.16 and 11.19.

Figure 11.9 | Linear trends over 1960–2018 for three temperature extreme indices: (a) the annual maximum daily maximum temperature (TXx), (b) the annual minimum daily minimum temperature (TNn), and (c) the annual number of days when daily maximum temperature exceeds its 90th percentile from a base period of 1961–1990 (TX90p); based on the HadEX3 dataset (Dunn et al. , 2020). Linear trends are calculated only for grid points with at least 66% of the annual values over the period and which extend to at least 2009. Areas without sufficient data are shown in grey. No overlay indicates regions where the trends are significant at the p = 0.1 level. Crosses indicate regions where trends are not significant. Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

In Africa (Table 11.4), while it is difficult to assess changes in temperature extremes in parts of the continent because of a lack of data, evidence of an increase in the intensity and frequency of hot extremes and decrease in the intensity and frequency of cold extremes is clear and robust in regions where data are available. These include an increase in the frequency of warm days and nights and a decrease in the frequency of cold days and nights with high confidence (Donat et al., 2013a, 2014b; Kruger and Sekele, 2013; Chaney et al., 2014; Filahi et al., 2016; Moron et al., 2016; Ringard et al., 2016; Barry et al., 2018; Gebrechorkos et al., 2018) and an increase in heatwaves (Russo et al., 2016; Ceccherini et al., 2017). The increase in TNn is more notable than in TXx (Figure 11.9). Cold spells occasionally strike subtropical areas, but are likely to have decreased in frequency (Barry et al., 2018). The frequency of cold events has likely decreased in South Africa (Song et al., 2014; Kruger and Nxumalo, 2017), North Africa (Filahi et al., 2016; Driouech et al., 2021), and the Sahara (Donat et al., 2016a). Over the whole continent, there is medium confidence in an increase in the intensity and frequency of hot extremes and decrease in the intensity and frequency of cold extremes; it is likely thatsimilar changes have also occurred in areas with poor data coverage, as warming is widespread and as projected future changes are similar over all regions (Section 11.3.5).

In Asia (Table 11.7), there is veryrobust evidence for avery likely increase in the intensity and frequency of hot extremes and decrease in the intensity and frequency of cold extremes in recent decades. This is clear in global studies (e.g., Alexander, 2016; Dunn et al., 2020), as well as in numerous regional studies (Table 11.7). The area fraction with extreme warmth in Asia increased during 1951–2016 (Imada et al., 2018). The frequency of warm extremes increased and the frequency of cold extremes decreased in East Asia (B. Zhou et al., 2016; Chen and Zhai, 2017; Yin et al., 2017; W. Lee et al., 2018; Qian et al., 2019) and west Asia (Acar Deniz and Gönençgil, 2015; Erlat and Türkeş, 2016; Rahimi and Hejabi, 2018; Rahimi et al., 2018) with high confidence. The duration of heat extremes has also lengthened in some regions, for example, in southern China (Luo and Lau, 2016), but there is medium confidence of heat extremes increasing in frequency in South Asia (AlSarmi and Washington, 2014; Sheikh et al., 2015; Mazdiyasni et al., 2017; Zahid et al., 2017; Nasim et al., 2018; Khan et al., 2019; Sen Roy, 2019). Warming trends in daily temperature extremes indices have also been observed in central Asia (Hu et al., 2016; Feng et al., 2018), the Hindu Kush Himalaya (Sun et al., 2017), and South East Asia (Supari et al., 2017; Cheong et al., 2018). The intensity and frequency of cold spells in all Asian regions have been decreasing since the beginning of the 20th century (high confidence) (Sheikh et al., 2015; Donat et al., 2016a; Dong et al., 2018; van Oldenborgh et al., 2019).

In Australasia (Table 11.10), there is veryrobust evidence forvery likely increases in the number of warm days and warm nights and decreases in the number of cold days and cold nights since 1950 (Lewis and King, 2015; Jakob and Walland, 2016; Alexander and Arblaster, 2017). The increase in extreme minimum temperatures occurs in all seasons over most of Australia and typically exceeds the increase in extreme maximum temperatures (X.L. Wang et al., 2013b; Jakob and Walland, 2016). However, some parts of Southern Australia have shown stable or increased numbers of frost days since the 1980s (Dittus et al., 2014) (see also Section 11.3.4). Similar positive trends in extreme minimum and maximum temperatures have been observed in New Zealand, in particular in the autumn and winter seasons, although they generally show higher spatial variability (Caloiero, 2017). In the tropical Western Pacific region, spatially coherent warming trends in maximum and minimum temperature extremes have been reported for the period 1951–2011 (Whan et al., 2014; McGree et al., 2019).

In Central and South America (Table 11.13), there is high confidence that observed hot extremes (TN90p, TX90p) have increased, and cold extremes (TN10p, TX10p) have decreased over recent decades, though trends vary among different extremes types, datasets, and regions (Skansi et al., 2013; Dittus et al., 2016; Rusticucci et al., 2017; Meseguer-Ruiz et al., 2018; Salvador and de Brito, 2018; Dereczynski et al., 2020; Dunn et al., 2020; Olmo et al., 2020). An increase in the intensity and frequency of heatwave events was also observed between 1961 and 2014 in an area covering most of South America (Ceccherini et al., 2016; Geirinhas et al., 2018). However, there is medium confidence that warm extremes (TXx and TX90p) have decreased in the last decades over the central region of South-Eastern South America (SES) during austral summer (Tencer and Rusticucci, 2012; Skansi et al., 2013; Rusticucci et al., 2017; Wu and Polvani, 2017). There is medium confidence that TNn extremes are warming faster than TXx extremes, with the largest warming rates observed over North-East Brazil (NEB) and Northern South America (NSA) for cold nights (Skansi et al., 2013).

In Europe (Table 11.16), there is veryrobust evidence for avery likely increase in maximum temperatures and the frequency of heatwaves. The increase in the magnitude and frequency of high maximum temperatures has been observed consistently across regions, including in central Europe (Twardosz and Kossowska-Cezak, 2013; Christidis et al., 2015; Lorenz et al., 2019) and southern Europe (Croitoru and Piticar, 2013; El Kenawy et al., 2013; Christidis et al., 2015; Nastos and Kapsomenakis, 2015; Fioravanti et al., 2016; Ruml et al., 2017). In Northern Europe, a strong increase in extreme winter warming events has been observed (Matthes et al., 2015; Vikhamar-Schuler et al., 2016). Temperature observations for winter cold spells show a long-term decreasing frequency in Europe (Brunner et al., 2018; van Oldenborgh et al., 2019), and typical cold spells, such as that observed during the 2009–2010 winter, had an occurrence probability two times smaller currently than if climate change had not occurred (Christiansen et al., 2018).

In North America (Table 11.19), there is veryrobust evidence for avery likely increase in the intensity and frequency of hot extremes and decrease in the intensity and frequency of cold extremes for the whole continent, though there are substantial spatial and seasonal variations in the trends. Minimum temperatures display warming consistently across the continent, while there are more contrasting trends in the annual maximum daily temperatures in parts of the USA (Figure 11.9; Lee et al., 2014; van Oldenborgh et al., 2019; Dunn et al., 2020). In Canada, there is a clear increase in the intensity and frequency of hot extremes and decrease in the intensity and frequency of cold extremes (Vincent et al., 2018). In Mexico, a clear warming trend in TNn was found, particularly in the northern arid region (Montero-Martínez et al., 2018). The number of warm days has increased and the number of cold days has decreased (García-Cueto et al., 2019). Cold spells have undergone a reduction in magnitude and intensity in all regions of North America (Bennett and Walsh, 2015; Donat et al., 2016a; Grotjahn et al., 2016; Vose et al., 2017; García-Cueto et al., 2019; van Oldenborgh et al., 2019).

Extreme heat events have increased around the Arctic since 1979, particularly over Arctic North America and Greenland (Matthes et al., 2015; Dobricic et al., 2020), which is consistent with summer melt (Section 9.4.1). Observations north of 60˚N show increases in winter warm days and nights over 1979–2015, while cold days and nights declined (Sui et al., 2017). Extreme heat days are particularly strong in winter, with observations showing the warmest mid-winter temperatures at the North Pole rising at twice the rate of mean temperature (Moore, 2016), as well as increases in Arctic winter warm days (Vikhamar-Schuler et al., 2016; Graham et al., 2017). Arctic annual minimum temperatures have increased at about three times the rate of global surface temperature since the 1960s (Figures 11.2 and 11.9), consistent with the observed mean cold season (October–May) warming of 3.1°C in the region (Atlas 11.2).

Trends in some measures of heatwaves are also observed at the global scale. Globally averaged heatwave intensity, heatwave duration, and the number of heatwave days have significantly increased from 1950–2011 (Perkins, 2015). There are some regional differences in trends in characteristics of heatwaves, with significant increases reported in Europe (Russo et al., 2015; Forzieri et al., 2016;Sánchez-Benítez et al., 2020) and Australia (CSIRO and BOM, 2016; Alexander and Arblaster, 2017). In Africa, there is medium confidence that heatwaves, regardless of the definition, have been becoming more frequent, longer-lasting, and hotter over more than three decades (Fontaine et al., 2013; Mouhamed et al., 2013; Ceccherini et al., 2016, 2017; Forzieri et al., 2016; Moron et al., 2016; Russo et al., 2016). The majority of heatwave characteristics examined in China between 1961 and 2014 show increases in heatwave days, consistent with warming (You et al., 2017; Xie et al., 2020). Increases in the frequency and duration of heatwaves are also observed in Mongolia (Erdenebat and Sato, 2016) and India (Ratnam et al., 2016; Rohini et al., 2016). In the UK, the lengths of short heatwaves have increased since the 1970s, while the lengths of long heatwaves (more than 10 days) have decreased over some stations in the south-east of England (M. Sanderson et al., 2017). In Central and South America, there are increases in the frequency of heatwaves (Barros et al., 2015; Bitencourt et al., 2016; Ceccherini et al., 2016; Piticar, 2018), although decreases in Excess Heat Factor (EHF), which is a metric for heatwave intensity, are observed in South America in data derived from HadGHCND (Cavanaugh and Shen, 2015).

In summary, it is virtually certain that there has been an increase in the number of warm days and nights and a decrease in the number of cold days and nights on the global scale since 1950. Both the coldest extremes and hottest extremes display increasing temperatures. It is very likely that these changes have also occurred at the regional scale in Europe, Australasia, Asia, and North America. It is virtually certain that there has been increases in the intensity and duration of heatwaves and in the number of heatwave days at the global scale. These trends likely occur in Europe, Asia, and Australia. There is medium confidence in similar changes in temperature extremes in Africa and high confidence in South America; the lower confidence is due to reduced data availability and fewer studies. Annual minimum temperatures on land have increased about three times more than global surface temperature since the 1960s, with particularly strong warming in the Arctic (high confidence).

11.3.3 Model Evaluation

The AR5 assessed that CMIP3 and CMIP5 models generally captured the observed spatial distributions of the mean state and that the inter-model range of simulated temperature extremes was similar to the spread estimated from different observational datasets; the models generally captured trends in the second half of the 20th century for indices of extreme temperature, although they tended to overestimate trends in hot extremes and underestimate trends in cold extremes (Flato et al., 2013). Post-AR5 studies on the CMIP5 models’ performance in simulating mean and changes in temperature extremes continue to support the AR5 assessment (Fischer and Knutti, 2014; Sillmann et al., 2014; Ringard et al., 2016; Borodina et al., 2017b; Donat et al., 2017; Di Luca et al., 2020b). Over Africa, the observed warming in temperature extremes is captured by CMIP5 models, although it is underestimated in Western and Central Africa (Sherwood et al., 2014; Diedhiou et al., 2018). Over East Asia, the CMIP5 ensemble performs well in reproducing the observed trend in temperature extremes averaged over China (Dong et al., 2015). Over Australia, the multi-model mean performs better than individual models in capturing observed trends in gridded station-based ETCCDI temperature indices (Alexander and Arblaster, 2017).

Initial analyses of CMIP6 simulations (H. Chen et al., 2020; Di Luca et al., 2020a; Kim et al., 2020; Thorarinsdottir et al., 2020; Wehner et al., 2020; Li et al., 2021) indicate that the CMIP6 models perform similarly to the CMIP5 models regarding biases in hot and cold extremes. In general, CMIP5 and CMIP6 historical simulations are similar in their performance in simulating the observed climatology of extreme temperatures (high confidence). The general warm bias in hot extremes and cold bias in cold extremes reported for CMIP5 models (Kharin et al., 2013; Sillmann et al., 2013a) remain in CMIP6 models (Di Luca et al., 2020a). However, there is some evidence that CMIP6 models better represent some of the underlying processes leading to extreme temperatures, such as seasonal and diurnal variability and synoptic-scale variability (Di Luca et al., 2020a). Whether these improvements are sufficient to enhance our understanding of past changes, or to reduce uncertainties in future projections, remains unclear. The relative error estimates in the simulation of various indices of temperature extremes in the available CMIP6 models show that no single model performs the best on all indices, and the multi-model ensemble seems to outperform any individual model due to its reduction in systematic bias (Kim et al., 2020). Figure 11.10 show errors in the 1979–2014 average annual TXx and annual TNn simulated by available CMIP6 models in comparison with HadEX3 and ERA5 (Kim et al., 2020; Wehner et al., 2020; Li et al., 2021). While the magnitude of the model error depends on the reference dataset, the model evaluations drawn from different reference datasets are quite similar. In general, models reproduce the spatial patterns and magnitudes of both cold and hot temperature extremes quite well. There are also systematic biases. Hot extremes tend to be too cool in mountainous and high-latitude regions, but too warm in the eastern USA and South America. For cold extremes, CMIP6 models are too cool, except in north-eastern Eurasia and the southern mid-latitudes. Errors in seasonal mean temperatures are uncorrelated with errors in extreme temperatures and are often of opposite sign (Wehner et al., 2020).

Figure 11.10 | Multi-model mean bias in temperature extremes (°C) for the period 1979–2014, calculated as the difference between the Coupled Model Intercomparison Project Phase 6 (CMIP6) multi-model mean and the average of observations from the values availablein HadEX3. (a) The annual hottest temperature(TXx); and (b) the annual coldest temperature (TNn). Areas without sufficient data are shown in grey. Adapted from Wehner et al. (2020) under the terms of the Creative Commons Attribution licence. Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

Atmospheric Model Intercomparison Project (AMIP) simulations are often used in event attribution studies to assess the influence of global warming on observed temperature-related extremes. These simulations typically capture the observed trends in temperature extremes, though some regional features, such as the lack of warming in daytime warm temperature extremes over South America and parts of North America, are not reproduced in the model simulations (Dittus et al., 2018), possibly due to internal variability, deficiencies in local surface processes, or forcings that are not represented in the sea surface temperatures (SSTs). Additionally, the AMIP models assessed tend to produce overly persistent heatwave events. This bias in the duration of the events does not impact on the reliability of the models’ positive trends (Freychet et al., 2018).

Several regional climate models (RCMs) have also been evaluated in terms of their performance in simulating the climatology of extremes in various regions of the Coordinated Regional Downscaling Experiment (CORDEX) (Giorgi et al., 2009), especially in East Asia (Ji and Kang, 2015; Yu et al., 2015; Park et al., 2016; Bucchignani et al., 2017; Gao et al., 2017a; Niu et al., 2018; Y. Sun et al., 2018b; Wang et al., 2019), Europe (Vautard et al., 2013, 2021; Smiatek et al., 2016; Gaertner et al., 2018; Cardoso et al., 2019; Lorenz et al., 2019; Jacob et al., 2020; Kim et al., 2020), and Africa (J. Kim et al., 2014; Diallo et al., 2015; Dosio, 2017; Samouly et al., 2018; Mostafa et al., 2019). Compared to GCMs, RCM simulations show an added value in simulating temperature-related extremes, though this depends on topographical complexity and the parameters employed (see Section 10.3.3). The improvement with resolution is noted in East Asia (Park et al., 2016; W. Zhou et al., 2016; Shi et al., 2017; Hui et al., 2018). However, in the European CORDEX ensemble, different aerosol climatologies with various degrees of complexity were used in projections (Bartók et al., 2017; Lorenz et al., 2019) and the land surface models used in the RCMs do not account for physiological CO2 effects on photosynthesis leading to enhanced water-use efficiency and decreased evapotranspiration (Schwingshackl et al., 2019), which could lead to biases in the representation of temperature extremes in these projections (Boé et al., 2020). In addition, there are key cold biases in temperature extremes over areas with complex topography (Niu et al., 2018). Over North America, 12 RCMs were evaluated over the ARCTIC-CORDEX region (Diaconescu et al., 2018). Models performed well at simulating climate indices related to mean air temperature and hot extremes over most of the Canadian Arctic, with the exception of the Yukon region where models displayed the largest biases related to topographic effects. Two RCMs were evaluated against observed extremes indices over North America over the period 1989–2009, with a cool bias in minimum temperature extremes shown in both RCMs (Whan and Zwiers, 2016). The most significant biases are found in TXx and TNn, with fewer differences in the simulation of annual minimum daily maximum temperature (TXn) and annual maximum daily minimum temperature (TNx) in Central and Western North America. Over Central and South America, maximum temperatures from the Eta RCM are generally underestimated, although hot days, warm nights, and heatwaves are increasing in the period 1961–1990, in agreement with observations (Chou et al., 2014b; Tencer et al., 2016; Bozkurt et al., 2019).

Some land forcings are not well represented in climate models. As highlighted in the Special Report on Climate Change and Land (SRCCL) Chapter 2, there is high agreement that temperate deforestation leads to summer warming and winter cooling (Anderson et al., 2011; Gálos et al., 2011, 2013; Anderson-Teixeira et al., 2012; Chen et al., 2012; Wickham et al., 2013; Zhao and Jackson, 2014; Ahlswede and Thomas, 2017; Bright et al., 2017; Strandberg and Kjellström, 2019), which has substantially contributed to the warming of hot extremes in the northern mid-latitudes over the course of the 20th century (Lejeune et al., 2018) and in recent years (Strandberg and Kjellström, 2019). However, observed forest effects on the seasonal and diurnal cycle of temperature are not well-captured in several ESMs: while observations show a cooling effect of forest cover compared to non-forest vegetation during daytime (Li et al., 2015), in particular in arid, temperate, and tropical regions (Alkama and Cescatti, 2016), several ESMs simulate a warming of daytime temperatures for regions with forest versus non-forest cover (Lejeune et al., 2017). Also irrigation effects, which can lead to regional cooling of temperature extremes, are generally not integrated in current generations of ESMs (Section 11.3.1).

In summary, there is high confidence that climate models can reproduce the mean state and overall warming of temperature extremes observed globally and in most regions, although the magnitude of the trends may differ. The ability of models to capture observed trends in temperature-related extremes depends on the metric evaluated, the way indices are calculated, and the time periods and spatial scales considered. Regional climate models add value in simulating temperature-related extremes over GCMs in some regions. Some land forcings on temperature extremes are not well-captured (effects of deforestation) or generally not representated (irrigation) in ESMs.

11.3.4 Detection and Attribution, Event Attribution

The SREX (IPCC, 2012) assessed that it is likely anthropogenic influences have led to the warming of extreme daily minimum and maximum temperatures at the global scale. The AR5 concluded that human influence has very likely contributed to the observed changes in the intensity and frequency of daily temperature extremes on the global scale in the second half of the 20th century (IPCC, 2014). With regard to individual, or regionally or locally specific events, AR5 concluded that it is likely human influence has substantially increased the probability of occurrence of heatwaves in some locations.

Studies since AR5 continue to attribute the observed increase in the frequency or intensity of hot extremes and the observed decrease in the frequency or intensity of cold extremes to human influence, dominated by anthropogenic greenhouse gas emissions, on global and continental scales, and for many AR6 regions. These include attribution of changes in the magnitude of annual TXx, TNx, TXn, and TNn, based on different observational datasets including, HadEX2 and HadEX3, CMIP5 and CMIP6 simulations, and different statistical methods (Kim et al., 2016; Z. Wang et al., 2017a; Seong et al., 2021). As is the case for an increase in mean temperature (Section 3.3.1), an increase in extreme temperature is mostly due to greenhouse gas forcing, offset by aerosol forcing. The aerosols’ cooling effect is clearly detectable over Europe and Asia (Seong et al., 2021). As much as 75% of the moderate daily hot extremes (above 99.9th percentile) over land are due to anthropogenic warming (Fischer and Knutti, 2015). New results are found to be more robust due to the extended period that improves the signal-to-noise ratio. The effect of anthropogenic forcing is clearly detectable and attributable in the observed changes in these indicators of temperature extremes, even at country and sub-country scales, such as in Canada (Wan et al., 2019). Changes in the number of warm nights, warm days, cold nights, and cold days, and other indicators such as the Warm Spell Duration Index (WSDI), are also attributed to anthropogenic influence (Christidis and Stott, 2016; Hu et al., 2020).

Regional studies, including for Asia (Dong et al., 2018; Lu et al., 2018), Australia (Alexander and Arblaster, 2017), and Europe (Christidis and Stott, 2016), found similar results. A clear anthropogenic signal is also found in the trends in the Combined Extreme Index (CEI) for North America, Asia, Australia, and Europe (Dittus et al., 2016). While various studies have described increasing trends in several heatwave metrics (heatwave duration, the number of heatwave days, etc.) in different regions (e.g., Cowan et al., 2014; Bandyopadhyay et al., 2016; M. Sanderson et al., 2017), few recent studies have explicitly attributed these changes to causes; most of them stated that observed trends are consistent with anthropogenic warming. The detected anthropogenic signals are clearly separable from the response to natural forcing, and the results are generally insensitive to the use of different model samples, as well as different data availability, indicating robust attribution. Studies of monthly, seasonal, and annual records in various regions (Kendon, 2014; Lewis and King, 2015; Bador et al., 2016; Meehl et al., 2016; C. Zhou et al., 2019) and globally (King, 2017) show an increase in the breaking of hot records and a decrease in the breaking of cold records (King, 2017). Changes in anthropogenically attributablerecord-breaking rates are noted to be largest over the Northern Hemisphere land areas (Shiogama et al., 2016). Yin and Sun (2018) found clear evidence of an anthropogenic signal in the changes in the number of frost and ice days, when multiple model simulations were used. In some key wheat-producing regions of Southern Australia, increases in frost days or frost season length have been reported (Dittus et al., 2014; Crimp et al., 2016); these changes are linked to decreases in rainfall, cloud-cover, and subtropical ridge strength, despite an overall increase in regional mean temperatures (Dittus et al., 2014; Pepler et al., 2018).

A significant advance since AR5 has been a large number of studies focusing on extreme temperature events at monthly and seasonal scales, using various extreme event attribution methods. Diffenbaugh et al. (2017) found that anthropogenic warming has increased the severity and probability of the hottest month by more than 80% of the available observational area on the global scale. Christidis and Stott (2014) provide clear evidence that warm events have become more probable because of anthropogenic forcings. Sun et al. (2014) found that human influence has caused a more than 60-fold increase in the probability of the extreme warm 2013 summer in eastern China since the 1950s. Human influence is found to have increased the probability of the historically hottest summers in many regions of the world, both in terms of mean temperature (B. Mueller et al., 2016) and wet bulb globe temperature (WBGT; C. Li et al., 2017). In most regions of the Northern Hemisphere, changes in the probability of extreme summer average WBGT were found to be about an order of magnitude larger than changes in the probability of extreme hot summers estimated by surface air temperature (C. Li et al., 2017). In addition to these generalized, global-scale approaches, extreme event studies have found an attributable increase in the probability of hot annual and seasonal temperatures in many locations, including Australia (Knutson et al., 2014b; Lewis and Karoly, 2014), China (Sun et al., 2014; Sparrow et al., 2018; Zhou et al., 2020), Korea (Y.-H. Kim et al., 2018) and Europe (King et al., 2015b).

There have also been many extreme event attribution studies that examined short-duration temperature extremes, including daily temperatures, temperature indices, and heatwave metrics. Examples of these events from different regions are summarized in various annual Explaining Extreme Events supplements of the Bulletin of the American Meteorological Society (Peterson et al., 2012, 2013a; Herring et al., 2014, 2015, 2016, 2018, 2019, 2020), including a number of approaches to examine extreme events (described in Easterling et al., 2016; Stott et al., 2016; Otto, 2017). Several studies of recent events from 2016 onwards have determined an infinite risk ratio (a fraction of attributable risk, or FAR, of 1), indicating that the occurrence probability for such events is close to zero in model simulations without anthropogenic influences (see Herring et al., 2018, 2019, 2020; Imada et al., 2019; Vogel et al., 2019). Though it is difficult to accurately estimate the lower bound of the uncertainty range of the FAR in these cases (Paciorek et al., 2018), the fact that those events are so far outside the envelop of the models with only natural forcing indicates that it is extremely unlikely for those events to occur without human influence.

Studies that focused on the attributable signal in observed cold extreme events show human influence reducing the probability of those events. Individual attribution studies on the extremely cold winter of 2011 in Europe (Peterson et al., 2012), in the eastern USA during 2014 and 2015 (Trenary et al., 2015, 2016; Wolter et al., 2015; Bellprat et al., 2016), in the cold spring of 2013 in the United Kingdom (Christidis et al., 2014), and of 2016 in eastern China (Qian et al., 2018; Y. Sun et al., 2018b) all showed a reduced probability due to human influence on the climate. An exception is the study of Grose et al. (2018), which found an increase in the probability of the severe western Australian frost of 2016 due to anthropogenically-driven changes in circulation patterns that drive cold outbreaks and frost probability.

Different event attribution studies can produce a wide range of changes in the probability of event occurrence because of different framing. The temperature event definition itself plays a crucial role in the attributable signal (Fischer and Knutti, 2015; Kirchmeier‐Young et al., 2019). Large-scale, longer-duration events tend to have notably larger attributable risk ratios (Angélil et al., 2014, 2018; Uhe et al., 2016; Harrington, 2017; Kirchmeier‐Young et al., 2019), as natural variability is smaller. While uncertainty in the best estimates of the risk ratios may be large, their lower bounds can be quite insensitive to uncertainties in observations or model descriptions, thus increasing confidence in conservative attribution statements (Jeon et al., 2016).

The relative strength of anthropogenic influences on temperature extremes is regionally variable, in part due to differences in changes in atmospheric circulation, land–surface feedbacks, and other external drivers such as aerosols. For example, in the Mediterranean and over western Europe, risk ratios on the order of 100 have been found (Kew et al., 2019; Vautard et al., 2020), whereas in the USA, changes are much less pronounced. This is probably a reflection of the land–surface feedback enhanced extreme 1930s temperatures that reduce the rarity of recent extremes, in addition to the definition of the events and framing of attribution analyses (e.g., spatial and temporal scales considered). Local forcing may mask or enhance the warming effect of greenhouse gases. In India, short-lived aerosols or an increase in irrigation may be masking the warming effect of greenhouse gases (Wehner et al., 2018c). Irrigation and crop intensification have been shown to lead to a cooling in some regions, in particular in North America, Europe, and India (high confidence) (N.D. Mueller et al., 2016; Thiery et al., 2017, 2020; Chen and Dirmeyer, 2019). Deforestation has contributed about one third of the total warming of hot extremes in some mid-latitude regions since pre-industrial times (Lejeune et al., 2018). Despite all of these differences, and larger uncertainties at the regional scale, nearly all studies demonstrated that human influence has contributed to an increase in the frequency or intensity of hot extremes and to a decrease in the frequency or intensity of cold extremes.

In summary, long-term changes in various aspects of long- and short-duration extreme temperatures, including intensity, frequency, and duration have been detected in observations and attributed to human influence at global and continental scales. It is extremely likely that human influence is the main contributor to the observed increase in the intensity and frequency of hot extremes and the observed decrease in the intensity and frequency of cold extremes on the global scale. It is very likely that this applies on continental scales as well. Some specific recent hot extreme events would have been extremely unlikely to occur without human influence on the climate system. Changes in aerosol concentrations have affected trends in hot extremes in some regions, with the presence of aerosols leading to attenuated warming, in particular from 1950 to 1980. Crop intensification, irrigation and no-till farming have attenuated increases in summer hot extremes in some regions, such as Central North America (medium confidence).

11.3.5 Projections

The AR5 (Chapter12, Collins et al., 2013) concluded that it is virtually certain there will be more frequent hot extremes and fewer cold extremes at the global scale and over most land areas in a future warmer climate, and it is very likely that heatwaves will occur with a higher frequency and longer duration.The SR1.5 (Chapter 3, Hoegh-Guldberg et al., 2018) assessment on projected changes in hot extremes at 1.5°C and 2°C global warming is consistent with the AR5 assessment, concluding that it is very likely a global warming of 2°C, when compared with a 1.5°C warming, would lead to more frequent and more intense hot extremes on land, as well as to longer warm spells, affecting many densely inhabited regions. The SR1.5 also assessed it is very likely that the strongest increases in the frequency of hot extremes are projected for the rarest events, while cold extremes will become less intense and less frequent, and cold spells will be shorter.

New studies since AR5 and SR1.5 confirm these assessments. New literature since AR5 includes projections of temperature-related extremes in relation to changes in mean temperatures, projections based on CMIP6 simulations, projections based on stabilized global warming levels, and the use of new metrics. Constraints for the projected changes in hot extremes were also provided (Borodina et al., 2017b; Sippel et al., 2017b; Vogel et al., 2017). Overall, projected changes in the magnitude of extreme temperatures over land are larger than changes in global mean temperature, over mid-latitude land regions in particular (Figures 11.3, 11.11; Fischer et al., 2014; Seneviratne et al., 2016; B.M. Sanderson et al., 2017; Wehner et al., 2018b; Di Luca et al., 2020b). Large warming in hot and cold extremes will occur, even at the 1.5°C GWL (Figure 11.11). At this level, widespread significant changes at the grid-box level occur for different temperature indices (Aerenson et al., 2018). In agreement with CMIP5 projections, CMIP6 simulations show that a 0.5°C increment in global warming will significantly increase the intensity and frequency of hot extremes, and decrease the intensity and frequency of cold extremes on the global scale (Figures 11.6, 11.8 and 11.12). It takes less than half of a degree for the changes in TXx to emerge above the level of natural variability (Figure 11.8) and the 66% ranges of the land medians of the 10-year or 50-year TXx events do not overlap between 1.0°C and 1.5°C in the CMIP6 multi-model ensemble simulations(Figure 11.6, Li et al., 2021).

Figure 11.11 | Projected changes in (a–c) annual maximum temperature (TXx) and (d–f) annual minimum temperature (TNn) at 1.5°C, 2°C, and 4°C of global warming compared to the 1850–1900 baseline. Results are based on simulations from the Coupled Model Intercomparison Project Phase 6 (CMIP6) multi-model ensemble under the Shared Socio-economic Pathways (SSPs) SSP1-1.9, SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios. The numbers in the top right indicate the number of simulations included. Uncertainty is represented using the simple approach: no overlay indicates regions with high model agreement, where ≥80% of models agree on the sign of change; diagonal lines indicate regions with low model agreement, where <80% of models agree on the sign of change. For more information on the simple approach, please refer to the Cross-Chapter Box (Atlas 1. For details on the methods see Supplementary Material 11.SM.2. Changes in TXx and TNn are also displayed in the Interactive Atlas. Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).
Figure 11.12 | Projected changes in the intensity of extreme temperature events under 1°C, 1.5°C, 2°C, 3°C, and 4°C global warming levels relative to the 1850–1900 baseline. Extreme temperature events are defined as the daily maximum temperatures (TXx) that were exceeded on average once during a 10-year period (10-year event, blue) and once during a 50-year period (50-year event, orange) during the 1850–1900 base period. Results are shown for the global land. For each box plot, the horizontal line and the box represent the median and central 66% uncertainty range, respectively, of the intensity changes across the multi-model ensemble, and the ‘whiskers’ extend to the 90% uncertainty range. The results are based on the multi-model ensemble from simulations of global climate models contributing to the Coupled Model Intercomparison Project Phase 6 (CMIP6) under different Shared Socio-economic Pathway forcing scenarios. Adapted from Li et al. (2021). Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

Projected warming is larger for TNn and exhibits strong equator-to-pole amplification, similar to the warming of boreal winter mean temperatures. The warming of TXx is more uniform over land and does not exhibit this behaviour (Figure 11.11). The warming of temperature extremes on global and regional scales tends to scale linearly with global warming (Section 11.1.4; Fischer et al., 2014; Seneviratne et al., 2016; Wartenburger et al., 2017; Li et al., 2021; see also SR1.5, Chapter 3). In the mid-latitudes, the rate of warming of hot extremes can be as large as twice the rate of global warming (Figure 11.11). In the Arctic winter, the rate of warming of the temperature of the coldest nights is about three times the rate of global warming (Appendix, Figure 11.A.1). Projected changes in temperature extremes can deviate from projected changes in annual mean warming in the same regions (Figures 11.3, 11.A.1 and 11.A.2; Di Luca et al., 2020b; Wehner, 2020) due to the additional processes that control the response of regional extremes, including, in particular, soil moisture–evapotranspiration–temperature feedbacks for hot extremes in the mid-latitudes and subtropical regions, and snow/ice–albedo–temperature feedbacks in high-latitude regions.

The probability of exceeding a certain hot extreme threshold will increase, while those for cold extreme will decrease with global warming (B. Mueller et al., 2016; Lewis et al., 2017b; Suarez-Gutierrez et al., 2020b). The changes tend to scale nonlinearly with the level of global warming, with larger changes for more rare events (Section 11.2.4; Cross-Chapter Box 11.11; Figures 11.6 and 11.12; Fischer and Knutti, 2015; Kharin et al., 2018; Li et al., 2021). For example, the CMIP5 ensemble projects the frequency of the present-day climate 20-year hottest daily temperature to increase by 80% at the 1.5°C GWL and by 180% at the 2.0°C GWL, and the frequency of the present-day climate 100-year hottest daily temperature to increase by 200% and more than 700% at the 1.5°C and 2.0°C warming levels, respectively (Kharin et al., 2018). CMIP6 simulations project similar changes (Li et al., 2021).

Tebaldi and Wehner (2018) showed that, at the middle of the 21st century, 66% of the land surface area would experience the present-day 20-year return values of TXx and the running three-day average of the daily maximum temperature every other year, on average, under the Representative Concentration Pathway 8.5 (RCP8.5) scenario, as opposed to only 34% under RCP4.5. By the end of the century, these area fractions increase to 92% and 62%, respectively. Such nonlinearities in the characteristics of future regional extremes are shown, for instance, for Europe (Dosio and Fischer, 2018; Spinoni et al., 2018b; Lionello and Scarascia, 2020), Asia (Guo et al., 2017; Harrington and Otto, 2018b; King et al., 2018), and Australia (Lewis et al., 2017a) under various global warming thresholds. The nonlinear increase in fixed-threshold indices (e.g., based on a percentile for a given reference period, or on an absolute threshold) as a function of global warming is consistent with a linear warming of the absolute temperature of the temperature extremes (e.g., Whan et al., 2015). Compared to the historical climate, warming will result in strong increases in heatwave area, duration and magnitude (Vogel et al., 2020b). These changes are mostly due to the increase in mean seasonal temperature, rather than changes in temperature variability, though the latter can have an effect in some regions (Brown, 2020; Di Luca et al., 2020b; Suarez-Gutierrez et al., 2020a).

Projections of temperature-related extremes in RCMs in the CORDEX regions demonstrate robust increases under future scenarios and can provide information on finer spatial scales than GCMs (e.g., Coppola et al., 2021b). Five RCMs in the CORDEX–East Asia region project increases in the 20-year return values of temperature extremes (summer maxima), with models that exhibit warm biases projecting stronger warming (Park and Min, 2019). Similarly, in the African domain, future increases in TX90p and TN90p are projected (Dosio, 2017; Mostafa et al., 2019). This regional-scale analysis provides fine-scale information, such as distinguishing the increase in TX90p over sub-equatorial Africa (Democratic Republic of the Congo, Angola, and Zambia) with values over the Gulf of Guinea, Central African Republic, South Sudan, and Ethiopia. Empirical statistical downscaling has also been used to produce more robust estimates for future heatwaves compared to RCMs based on large multi-model ensembles (Furrer et al., 2010; Keellings and Waylen, 2014; Wang et al., 2015; Benestad et al., 2018).

In all continental regions, including Africa (Table 11.4), Asia (Table 11.7), Australasia (Table 11.10), Central and South America (Table 11.13), Europe (Table 11.16), North America (Table 11.19) and at the continental scale, it is very likely that the intensity and frequency of hot extremes will increase and the intensity and frequency of cold extremes will decrease compared with the 1995–2014 baseline, even under 1.5°C global warming. Those changes are virtually certain to occur under 4°C global warming. At the regional scale, and for almost all AR6 regions, it is likely that the intensity and frequency of hot extremes will increase and the intensity and frequency of cold extremes will decrease compared with the 1995–2014 baseline, even under 1.5°C global warming. Those changes are virtually certain to occur under 4°C global warming. Exceptions include lower confidence in the projected decrease in the intensity and frequency of cold extremes compared with the 1995–2014 baseline under 1.5°C of global warming (medium confidence) and 4°C of global warming (very likely) in Northern Central America, Central North America, and Western North America.

In Africa (Table 11.4), evidence includes increases in the intensity and frequency of hot extremes, such as warm days, warm nights, and heatwaves, and decreases in the intensity and frequency of cold extremes, such as cold days and cold nights over the continent, as projected by CMIP5, CMIP6, and CORDEX simulations (Giorgi et al., 2014; Engelbrecht et al., 2015; Lelieveld et al., 2016; Russo et al., 2016; Dosio, 2017; Bathiany et al., 2018; Mba et al., 2018; Nangombe et al., 2018; Weber et al., 2018; Kruger et al., 2019; Coppola et al., 2021b; Li et al., 2021). Cold spells are projected to decrease under all RCPs, and even at low warming levels in Western and Central Africa (Diedhiou et al., 2018). The number of cold days is projected to decrease in East Africa (Ongoma et al., 2018b).

In Asia (Table 11.7), evidence includes increases in the intensity and frequency of hot extremes, such as warm days, warm nights, and heatwaves, and decreases in the intensity and frequency of cold extremes, such as cold days and cold nights over the continent, as projected by CMIP5, CMIP6, and CORDEX simulations (Sillmann et al., 2013b; Zhou et al., 2014; R. Zhang et al., 2015; Zhao et al., 2015; Pal and Eltahir, 2016; Singh and Goyal, 2016; Xu et al., 2017; Gao et al., 2018; Han et al., 2018; Shin et al., 2018; Sui et al., 2018; L. Li et al., 2019; Zhu et al., 2020). More intense heatwaves of longer durations and occurring at a higher frequency are projected over India (Murari et al., 2015; Mishra et al., 2017) and Pakistan (Nasim et al., 2018). Future mid-latitude warm extremes, similar to those experienced during the 2010 event, are projected to become more extreme, with temperature extremes increasing potentially by 8.4°C (RCP8.5) over north-west Asia (van der Schrier et al., 2018). Over West and East Siberia, and Russian Far East, an increase in extreme heat durations is expected in all scenarios (Sillmann et al., 2013b; Kattsov et al., 2017; Reyer et al., 2017). In the MENA regions (Arabian Peninsula and Western Central Asia), extreme temperatures could increase by almost 7°C by 2100 under RCP8.5 (Lelieveld et al., 2016).

In Australasia (Table 11.10), evidence includes increases in the intensity and frequency of hot extremes, such as warm days, warm nights, and heatwaves, and decreases in the intensity and frequency of cold extremes, such as cold days and cold nights over the continent, as projected by CMIP5, CMIP6, and CORDEX simulations (CSIROand BOM, 2015; Alexander and Arblaster, 2017; Lewis et al., 2017a; Herold et al., 2018; Coppola et al., 2021b; Evans et al., 2021). Over most of Australia, increases in the intensity and frequency of hot extremes are projected to be predominantly driven by the long-term increase in mean temperatures (Di Luca et al., 2020b). Future projections indicate a decrease in the number of frost days regardless of the region and season considered (Alexander and Arblaster, 2017; Herold et al., 2018).

In Central and South America (Table 11.13), evidence includes increases in the intensity and frequency of hot extremes, such as warm days, warm nights, and heatwaves, and decreases in the intensity and frequency of cold extremes, such as cold days and cold nights over the continent, as projected by CMIP5, CMIP6, and CORDEX simulations (Chou et al., 2014a; Cabré et al., 2016; López-Franca et al., 2016; Stennett-Brown et al., 2017; Coppola et al., 2021b; Li et al., 2021; Vichot-Llano et al., 2021). Over South-Eastern South America during the austral summer, the increase in the frequency of TN90p is larger than that projected for TX90p, consistent with observed past changes (López-Franca et al., 2016). Under RCP8.5, the number of heatwave days are projected to increase for the intra-Americas region for the end of the 21st century (Angeles-Malaspina et al., 2018). A general decrease in the frequency of cold spells and frost days is projected, as indicated by several indices based on minimum temperature (López-Franca et al., 2016).

In Europe (Table 11.16), evidence includes increases in the intensity and frequency of hot extremes, such as warm days, warm nights, and heatwaves, and decreases in the intensity and frequency of cold extremes, such as cold days and cold nights over the continent, as projected by CMIP5, CMIP6, and CORDEX simulations (Lau and Nath, 2014; Ozturk et al., 2015; Russo et al., 2015; Schoetter et al., 2015; Vogel et al., 2017; Winter et al., 2017; Jacob et al., 2018; Lhotka et al., 2018; Rasmijn et al., 2018; Suarez-Gutierrez et al., 2018; Cardoso et al., 2019; Lionello and Scarascia, 2020; Molina et al., 2020; Coppola et al., 2021b; Li et al., 2021). Increases in heatwaves are greater over the southern Mediterranean and Scandinavia (Forzieri et al., 2016; Abaurrea et al., 2018; Dosio and Fischer, 2018; Rohat et al., 2019). Thebiggest increases in the number of heatwave days are expected for southern European cities (Guerreiro et al., 2018a; Junk et al., 2019), and Central European cities will see the biggest increases in maximum heatwave temperatures (Guerreiro et al., 2018a).

In North America (Table 11.19), evidence includes increases in the intensity and frequency of hot extremes, such as warm days, warm nights, and heatwaves, and decreases in the intensity and frequency of cold extremes, such as cold days and cold nights over the continent, as projected by CMIP5, CMIP6, and CORDEX simulations (Grotjahn et al., 2016; Vose et al., 2017; Alexandru, 2018; C. Li et al., 2018, 2021; C. Yang et al., 2018; X. Zhang et al., 2019; Coppola et al., 2021b). Projections of temperature extremes for the end of the 21st century show that warm days and nights are very likely to increase, and cold days and nights are very likely to decrease in all regions. There is medium confidence in large increases in warm days and warm nights in summer, particularly over the USA, and in large decreases in cold days in Canada in autumn and winter (Grotjahn et al., 2016; Vose et al., 2017; Alexandru, 2018; C. Li et al., 2018, 2021; C. Yang et al., 2018; X. Zhang et al., 2019; Coppola et al., 2021b). Minimum winter temperatures are projected to rise faster than mean winter temperatures (Underwood et al., 2017). Projections for the end of the century under RCP8.5 showed the four-day cold spell that happens on average once every five years is projected to warm by more than 10°C. CMIP5 models do not project current 1-in-20-year annual minimum temperature extremes to recur over much of the continent (Wuebbles et al., 2014).

In summary, it is virtually certain that further increases in the intensity and frequency of hot extremes, and decreases in the intensity and frequency of cold extremes, will occur throughout the 21st century and around the world. It is virtually certain that the number of hot days and hot nights and the length, frequency, and/or intensity of warm spells or heatwaves compared to 1995–2014 will increase over most land areas. In most regions, changes in the magnitude of temperature extremes are proportional to global warming levels (high confidence). The highest increase of temperature of hottest days is projected in some mid-latitude and semi-arid regions, at about 1.5 times to twice the rate of global warming (high confidence). The highest increase of temperature of coldest days is projected in Arctic regions, at about three times the rate of global warming (high confidence). The probability of temperature extremes generally increases nonlinearly with increasing global warming levels (high confidence). Confidence in assessments depends on the spatial and temporal scales of the extreme in question, with high confidence in projections of temperature-related extremes at global and continental scales for daily to seasonal scales. There is high confidence that, on land, the magnitude of temperature extremes increases more strongly than global mean temperature.

11.4 Heavy Precipitation

This section assesses changes in heavy precipitation at global and regional scales. The main focus is on extreme precipitation at a daily scale where literature is most concentrated, though extremes of shorter (sub-daily) and longer (five-day or more) durations are also assessed to the extent the literature allows.

11.4.1 Mechanisms and Drivers

The SREX (Chapter 3, Seneviratne et al., 2012) assessed changes in heavy precipitation in the context of the effects of thermodynamic and dynamic changes. Box 11.1 assesses thermodynamic and dynamic changes in a warming world to aid the understanding of changes in observations and projections in some extremes and the sources of uncertainties (see also Section 8.2.3.2). In general, warming increases the atmospheric water-holding capacity following the Clausius–Clapeyron (C-C) relation. This thermodynamic effect results in an increase in extreme precipitation at a similar rate at the global scale. On a regional scale, changes in extreme precipitation are further modulated by dynamic changes (Box 11.1).

Large-scale modes of variability, such as the North Atlantic Oscillation (NAO), El Niño–Southern Oscillation (ENSO), Atlantic Multi-decadal Variability (AMV), and Pacific Decadal Variability (PDV) (Annex IV), modulate precipitation extremes through changes in environmental conditions or embedded storms (Section 8.3.2). Latent heating can invigorate these storms (Nie et al., 2018; Z. Zhang et al., 2019a); changes in dynamics can increase precipitation intensity above that expected from the C-C scaling rate (Sections 8.2.3.2 and 11.7; Box 11.1). Additionally, the efficiency of converting atmospheric moisture into precipitation can change as a result of cloud microphysical adjustment to warming,resulting in changes in the characteristics of extreme precipitation; but changes in precipitation efficiency in a warming world are highly uncertain (Sui et al., 2020).

It is difficult to separate the effect of global warming from internal variability inthe observed changes in the modes of variability (Section 2.4). Future projections of modes of variability are highly uncertain Section 4.3.3),resulting in uncertainty in regional projections of extreme precipitation. Future warming may amplify monsoonal extreme precipitation. Changes in extreme storms, including tropical/extratropical cyclones and severe convective storms, result in changes in extreme precipitation (Section 11.7). Also, changes in sea surface temperatures (SSTs) alter land–sea contrast, leading to changes in precipitation extremes near coastal regions. For example, the projected larger SST increase near the coasts of East Asia and India can result in heavier rainfall near these coastal areas from tropical cyclones (Mei and Xie, 2016) or torrential rains (Manda et al., 2014). The warming in the western Indian Ocean is associated with increases in moisture surges on the low-level monsoon westerlies towards the Indian subcontinent, which may lead to an increase in the occurrence of precipitation extremes over central India (Krishnan et al., 2016; Roxy et al., 2017).

Decreases in atmospheric aerosols results in warming and thus an increase in extreme precipitation (Samset et al., 2018; Sillmann et al., 2019). Changes in atmospheric aerosols also result in dynamic changes such as in tropical cyclones (Takahashi et al., 2017; Strong et al., 2018). Uncertainty in the projections of future aerosol emissions results in additional uncertainty in the heavy precipitation projections of the 21st century (Lin et al., 2016).

There has been new evidence of the effect of local land-use and land-cover change on heavy precipitation. There is a growing set of literature linking increases in heavy precipitation in urban centres to urbanization (Argüeso et al., 2016; Y. Zhang et al., 2019b). Urbanization intensifies extreme precipitation, especially in the afternoon and early evening, over the urban area and its downwind region (medium confidence) (Box 10.3). There are four possible mechanisms: (i) increases in atmospheric moisture due to horizontal convergence of air associated with the urban heat island effect (Shastri et al., 2015; Argüeso et al., 2016); (ii) increases in condensation due to urban aerosol emissions (Han et al., 2011; Sarangi et al., 2017); (iii) aerosol pollution that impacts cloud microphysics (Box 8.1; Schmid and Niyogi, 2017); and (iv) urban structures that impede atmospheric motion (Shepherd, 2013; Ganeshan and Murtugudde, 2015; Paul et al., 2018). Other local forcing, including reservoirs (Woldemichael et al., 2012), irrigation (Devanand et al., 2019), or large-scale land-use and land-cover change (Odoulami et al., 2019), can also affect local extreme precipitation.

In summary, precipitation extremes are controlled by both thermodynamic and dynamic processes. Warming-induced thermodynamic change results in an increase in extreme precipitation, at a rate that closely follows the C-C relationship at the global scale (high confidence). The effects of warming-induced changes in dynamic drivers on extreme precipitation are more complicated, difficult to quantify, and are an uncertain aspect of projections. Precipitation extremes are also affected by forcings other than changes in greenhouse gases, including changes in aerosols, land-use and land-cover change, and urbanization (medium confidence).

11.4.2 Observed Trends

Both SREX (Chapter 3, Seneviratne et al., 2012) and AR5 (IPCC, 2014 Chapter 2) concluded it was likely that the number of heavy precipitation events over land had increased in more regions than it had decreased, though there were wide regional and seasonal variations, and trends in many locations were not statistically significant. This assessment has been strengthened with multiple studies findingrobust evidence of the intensification of extreme precipitation at global and continental scales, regardless of spatial and temporal coverage of observations and the methods of data processing and analysis.

The average annual maximum precipitation amount in a day (Rx1day) has significantly increased since the mid-20th century over land (Du et al., 2019; Dunn et al., 2020) and in the humid and dry regions of the globe (Dunn et al., 2020). The percentage of observing stations with statistically significant increases in Rx1day is larger than expected by chance, while the percentage of stations with statistically significant decreases is smaller than expected by chance, over the global land as a whole and over North America, Europe, and Asia (Figure 11.13; Sun et al., 2021) and over global monsoon regions (Zhang and Zhou, 2019) where data coverage is relatively good. The addition of the past decade of observational data shows a more robust increase in Rx1day over the global land region (Sun et al., 2021). Light, moderate, and heavy daily precipitation has all intensified in a gridded daily precipitation dataset (Contractor et al., 2020a). Daily mean precipitation intensities have increased since the mid-20th century in a majority of land regions (high confidence) (Section 8.3.1.3). The probability of precipitation exceeding 50 mm/day increased during 1961–2018 (Benestad et al., 2019). The globally averaged annual fraction of precipitation from days in the top 5% (R95pTOT) has also significantly increased (Dunn et al., 2020). The increase in the magnitude of Rx1day in the 20th century is estimated to be at a rate consistent with C-C scaling with respect to global mean temperature (Fischer and Knutti, 2016; Sun et al., 2021). Studies on past changes in extreme precipitation of durations longer than a day are more limited, though there are some studies examining long-term trends in annual maximum five-day precipitation (Rx5day). On global and continental scales, long-term changes in Rx5day are similar to those of Rx1day in many aspects (Zhang and Zhou 2019; Sun et al., 2021). As discussed below, at the regional scale, changes in Rx5day are also similar to those of Rx1day where there are analyses of changes in both Rx1day and Rx5day.

Figure 11.13 | Signs and significance of the observed trends in annual maximum daily precipitation (Rx1day) during 1950–2018 at 8345 stations with suficient data. (a) Percentage of stations with statistically significant trends in Rx1day; green dots show positive trends and brown dots negative trends. Box and ‘whisker’ plots indicate the expected percentage of stations with significant trends due to chance estimated from 1000 bootstrap realizations under a no-trend null hypothesis. The boxes mark the median, 25th percentile, and 75th percentile. The upper and lower whiskers show the 97.5th and the 2.5th percentiles, respectively. Maps of stations with positive (b) and negative (c) trends. The light colour indicates stations with non-significant trends, and the dark colour stations with significant trends. Significance is determined by a two-tailed test conducted at the 5% level. Adapted from Sun et al. (2021). Figure copyright © American Meteorological Society (used with permission). Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

Overall, there is a lack of systematic analysis of long-term trends in sub-daily extreme precipitation at the global scale. Often, sub-daily precipitation data have only sporadic spatial coverage and are of limited length. Additionally, the available data records are far shorter than needed for a robust quantification of past changes in sub-daily extreme precipitation (C. Li et al., 2019a). Despite these limitations, there are studies in regions of almost all continents that generally indicate intensification of sub-daily extreme precipitation, although there remains low confidence in an overall increase at the global scale. Studies include an increase in extreme sub-daily rainfall in summer over South Africa (Sen Roy and Rouault, 2013), annually in Australia (Guerreiro et al., 2018b), over 23 urban locations in India (Ali and Mishra, 2018), in Peninsular Malaysia (Syafrina et al., 2015), and in eastern China in the summer season during 1971–2013 (Xiao et al., 2016). In some regions in Italy (Arnone et al., 2013; Libertino et al., 2019) and in the USA during 1950–2011 (Barbero et al., 2017), there is also an increase. In general, an increase in sub-daily heavy precipitation results in an increase in pluvial floods over smaller watersheds (Ghausi and Ghosh, 2020).

There is a considerable body of literature examining scaling of sub-daily precipitation extremes, conditional on day-to-day air or dew-point temperatures (Westra et al., 2014; Fowler et al., 2021). This scaling, also termed ‘apparent scaling’ (Fowler et al., 2021), is robust when different methodologies are used in different regions, ranging between the C-C and two-times the C-C rate (e.g., Formayer and Fritz, 2017; Lenderink et al., 2017; Burdanowitz et al., 2019). This is confirmed when sub-daily precipitation data collected from multiple continents (Lewis et al., 2019) are analysed in a consistent manner using different methods (Ali et al., 2021). It has been hoped that apparent scaling might be used to help understand past and future changes in extreme sub-daily precipitation. However, apparent scaling samples multiple synoptic weather states, mixing thermodynamic and dynamic factors that are not directly relevant for climate change responses (Section 8.2.3.2; Prein et al., 2016b; Bao et al., 2017; X. Zhang et al., 2017; Drobinski et al., 2018; Sun et al., 2020). The spatial pattern of apparent scaling is different from those of projected changes over Australia (Bao et al., 2017) and North America (Sun et al., 2020) in regional climate model simulations. It thus remains difficult to use the knowledge about apparent scaling to infer past and future changes in extreme sub-daily precipitation according to observed and projected changes in local temperature.

In Africa (Table 11.5), evidence shows an increase in extreme daily precipitation for the late half of the 20th century over the continent where data are available; there is a larger percentage of stations showing significant increases in extreme daily precipitation than decreases (Sun et al., 2021). There are increases in different metrics relevant to extreme precipitation in various regions of the continent (Chaney et al., 2014; Harrison et al., 2019; Dunn et al., 2020; Sun et al., 2021). There is an increase in extreme precipitation events in Southern Africa (Weldon and Reason, 2014; Kruger et al., 2019) and a general increase in heavy precipitation over East Africa, the Greater Horn of Africa (Omondi et al., 2014). Over sub-Saharan Africa, increases in the frequency and intensity of extreme precipitation have been observed over the well-gauged areas during 1950–2013; however, this covers only 15% of the total area of sub-Saharan Africa (Harrison et al., 2019). There is medium confidence about the increase in extreme precipitation for some regions where observations are more abundant, but for Africa as whole, there is low confidence because of a general lack of continent-wide systematic analysis, the sporadic nature of available precipitation data over the continent, and spatially non-homogenous trends in places where dataare available (Donat et al., 2014a; Mathbout et al., 2018b; Alexander et al., 2019; Funk et al., 2020).

In Asia (Table 11.8), there is robust evidence that extreme precipitation has increased since the 1950s (high confidence), however, this is dominated by high spatial variability. Increases in Rx1day and Rx5day during 1950–2018 are found over two-thirds of stations. The percentage of stations with statistically significant trends is larger than can be expected by chance (Figure 11.13; Sun et al., 2021). An increase in extreme precipitation has also been observed in various regional studies based on different metrics of extreme precipitation and spatial and temporal coverage of the data. These include an increase in daily precipitation extremes over central Asia (Hu et al., 2016), most of South Asia (Zahid and Rasul, 2012; Pai et al., 2015; Sheikh et al., 2015; Adnan et al., 2016; Malik et al., 2016; Dimri et al., 2017; Priya et al., 2017; Roxy et al., 2017; Hunt et al., 2018; Kim et al., 2019; Wester et al., 2019), the Arabian Peninsula (Rahimi and Fatemi, 2019; Almazroui and Saeed, 2020; Atif et al., 2020), South East Asia (Siswanto et al., 2015; Supari et al., 2017; Cheong et al., 2018); the north-west Himalaya (Malik et al., 2016), parts of East Asia (Baeket al., 2017; Nayak et al., 2017; Ye and Li, 2017), the western Himalayas since the 1950s (Ridley et al., 2013; Dimri et al., 2015; Madhura et al., 2015), West and East Siberia, and Russian Far East (Donat et al., 2016a). A decrease was found over the eastern Himalayas (Sheikh et al., 2015; Talchabhadel et al., 2018). Increases have been observed over Jakarta (Siswanto et al., 2015), but Rx1day over most parts of the Maritime Continent has decreased (Villafuerte and Matsumoto, 2015). Trends in extreme precipitation over China are mixed with increases and decreases (G. Fu et al., 2013; Jiang et al., 2013; Ma et al., 2015; Yin et al., 2015; Xiao et al., 2016) and are not significant over China as whole (Jiang et al., 2013; Hu et al., 2016; Ge et al., 2017; Deng et al., 2018; He and Zhai, 2018; W. Li et al., 2018a; Tao et al., 2018; M. Liu et al., 2019b; Chen et al., 2021). With few exceptions, most South East Asian countries have experienced an increase in rainfall intensity, but with a reduced number of wet days (Donat et al., 2016a; Cheong et al., 2018; Naveendrakumar et al., 2019), though large differences in trends exists if the trends are estimated from different datasets, including gauge-based, remotely sensed, and reanalysis data, over a relatively short period (Kim et al., 2019). There is a significant increase in heavy rainfall (>100 mm day–1) and a significant decrease in moderate rainfall (5–100 mm day–1) in central India during the South Asian monsoon season (Deshpande et al., 2016; Roxy et al., 2017).

In Australasia (Table 11.11), available evidence has not shown an increase or a decrease in heavy precipitation over Australasia as a whole (medium confidence), but heavy precipitation tends to increase over Northern Australia (particularly the north-west) and decrease over the eastern and southernregions (e.g., Jakob and Walland, 2016; Guerreiro et al., 2018b; Dey et al., 2019b; Dunn et al., 2020; Sun et al., 2021). Available studies that used long-term observations since the mid-20th century showed nearly as many stations with an increase as those with a decrease in heavy precipitation (Jakob and Walland, 2016) or slightly more stations with a decrease than with an increase in Rx1day and Rx5day (Sun et al., 2021), or strong differences in Rx1day trends with increases over Northern Australia and Central Australia in general, but mostly decreases over Southern Australia and Eastern Australia (Dunn et al., 2020). Over New Zealand, decreases are observed for moderate–heavy precipitation events, but there are no significant trends for very heavy events (more than 64 mm in a day) for the period 1951–2012. The number of stations with an increase in very wet days is similar to that with a decrease during 1960–2019 (MfE and Stats NZ, 2020). Overall, there is low confidence in trends in the frequency of heavy rain days, with mostly decreases over New Zealand (Harringtonand Renwick, 2014; Caloiero, 2015).

In Central and South America (Table 11.14), evidence shows an increase in extreme precipitation, but in general there is low confidence; while continent-wide analyses produced wetting trends are not robust. Rx1day increased at more stations than it decreased in South America between 1950 and 2018 (Sun et al., 2021). Over the period 1950–2010, both Rx5day and R99p increased over large regions of South America, including North-Western South America, Northern South America, and South-Eastern South America (Skansi et al., 2013). There are large regional differences. A decrease in daily extreme precipitation is observed in north-eastern Brazil (Skansi et al., 2013; Bezerra et al., 2018; Dereczynski et al., 2020). Trends in extreme precipitation indices were not statistically significant over the period 1947–2012 within the São Francisco River basin in the Brazilian semi-arid region (Bezerra et al., 2018). An increase in extreme rainfall is observed in the Amazon with medium confidence (Skansi et al., 2013) and in South-Eastern South America with high confidence (Skansi et al., 2013; Valverde and Marengo, 2014; Barros et al., 2015; Ávila et al., 2016; Wu and Polvani, 2017; Lovino et al., 2018; Dereczynski et al., 2020). Among all sub-regions, South-Eastern South America shows the highest rate of increase for rainfall extremes, followed by the Amazon (Skansi et al., 2013). Increases in the intensity of heavy daily rainfall events have been observed in the southern Pacific and in the Titicaca basin (Skansi et al., 2013; Huerta and Lavado‐Casimiro, 2021). In Southern Central America, trends in annual precipitation are generally not significant, although small (but significant) increases are found in Guatemala, El Salvador, and Panama (Hidalgo et al., 2017). Small positive trends were found in multiple extreme precipitation indices over the Caribbean region over a short time period (1986–2010) (Stephenson et al., 2014; McLean et al., 2015).

In Europe (Table 11.17), there is robust evidence that the magnitude and intensity of extreme precipitation has very likely increased since the 1950s. There is a significant increase in Rx1day and Rx5day during 1950–2018 in Europe as a whole (Sun et al., 2021, also Figure 11.13). The number of stations with increases far exceeds those with decreases in the frequency of daily rainfall exceeding its 90th or 95th percentile in century-long series (Cioffi et al., 2015). The five-, 10-, and 20-year events of one-day and five-day precipitation during 1951–1960 became more common since the 1950s (van den Besselaar et al., 2013). There can be large discrepancies among studies and regions and seasons (Croitoru et al., 2013; Willems, 2013; Casanueva et al., 2014; Roth et al., 2014; Fischer et al., 2015); evidence for increasing extreme precipitation is more frequently observed for summer and winter, but not in other seasons (Madsen et al., 2014; Helama et al., 2018). An increase is observed in central Europe (Volosciuk et al., 2016; Zeder and Fischer, 2020), and in Romania (Croitoru et al., 2016). Trends in the Mediterranean region are in general not spatially consistent (Reale and Lionello, 2013), with decreases in the western Mediterranean and some increases in the eastern Mediterranean (Rajczak et al., 2013; Casanueva et al., 2014; de Lima et al., 2015; Gajić-Čapka et al., 2015; Sunyer et al., 2015; Pedron et al., 2017; Serrano-Notivoli et al., 2018; Ribes et al., 2019). In the Netherlands, the total precipitation contributed from extremes higher than the 99th percentile doubles per 1°C increase in warming (Myhre et al., 2019), though extreme rainfall trends in Northern Europe may differ in different seasons (Irannezhad et al., 2017).

In North America (Table 11.20), there is robust evidence that the magnitude and intensity of extreme precipitation has very likely increased since the 1950s. Both Rx1day and Rx5day have significantly increased in North America during 1950–2018 (Sun et al., 2021, also Figure 11.13). There is, however, regional diversity. In Canada, there is a lack of detectable trends in observed annual maximum daily (or shorter duration) precipitation (Shephard et al., 2014; Mekis et al., 2015; Vincent et al., 2018). In the USA, there is an overall increase in one-day heavy precipitation, both in terms of intensity and frequency (Villarini et al.,2012; Donat et al., 2013b; Wu, 2015; Easterling et al., 2017; H. Huang et al., 2017; Howarth et al., 2019; Sun et al., 2021), except for the southern USA (Hoerling et al., 2016) where internal variability may have played a substantial role in the lack of observed increases. In Mexico, increases are observed in R10mm and R95p (Donat et al., 2016a), very wet days over the cities (García-Cueto et al., 2019) and in total precipitation (PRCPTOT) and Rx1day (Donat et al., 2016b).

In Small Islands, there is a lack of evidence showing changes in heavy precipitation overall. There were increases in extreme precipitation in Tobago from 1985–2015 (Stephenson et al., 2014; Dookie et al., 2019) and decreases in south-western French Polynesia and the southern subtropics (low confidence) (Table 11.5; Atlas.10). Extreme precipitation leading to flooding in the Small Islands has been attributed in part to tropical cyclones, as well as being influenced by ENSO (Box 11.5; Khouakhi et al., 2016; Hoegh-Guldberg et al., 2018).

In summary, the frequency and intensity of heavy precipitation have likely increased at the global scale over a majority of land regions with good observational coverage. Since 1950, the annual maximum amount of precipitation falling in a day, or over five consecutive days, has likely increased over land regions with sufficient observational coverage for assessment, with increases in more regions than there are decreases. Heavy precipitation has likely increased on the continental scale over three continents (North America, Europe, and Asia) where observational data are more abundant. There is very low confidence about changes in sub-daily extreme precipitation due to the limited number of studies and available data.

11.4.3 Model Evaluation

Evaluating climate model competence in simulating heavy precipitation extremes is challenging due to a number of factors, including the lack of reliable observations and the spatial scale mismatch between simulated andobserved data (Avila et al., 2015; Alexander et al., 2019). Simulated precipitation represents areal means, but station-based observations are conducted at point locations and are often sparse. The areal-reduction factor, the ratio between pointwise station estimates of extreme precipitation and extremes of the areal mean, can be as large as 130% at CMIP6 resolutions (about 100 km) (Gervais et al., 2014). Hence, the order in which gridded station based extreme values are constructed (i.e., if the extreme values are extracted at the station first and then gridded, or if the daily station values are gridded and then the extreme values are extracted) represents different spatial scales of extreme precipitation and needs to be taken into account in model evaluation (Wehner et al. 2020). This aspect has been considered in some studies. Reanalysis products are used in place of station observations for their spatial completeness as well as spatial-scale comparability(Sillmann et al., 2013a; Kim et al., 2020; Li et al., 2021). However, reanalyses share similar parametrizations to the models themselves, reducing the objectivity of the comparison.

Different generations of CMIP models have improved over time, though quite modestly (Flato et al., 2013; Watterson et al., 2014). Improvements in the representation of the magnitude of the Expert Team on Climate Change Detection and Indices (ETCCDI) in CMIP5 over CMIP3(Sillmann et al., 2013a; Chen and Sun, 2015a) have been attributed to higher resolution, as higher-resolution models represent smaller areas at individual grid boxes. Additionally, the spatial distribution of extreme rainfall simulated by high-resolution models is generally more comparable to observations (Sillmann et al., 2013b; Kusunoki, 2017, 2018b; Scher et al., 2017) as these models tend to produce more realistic storms compared to coarser models (Section 11.7.2). Higher horizontal resolution alone improves simulation of extreme precipitation in some models (Wehner et al., 2014; Kusunoki, 2017, 2018b), but this is insufficient in other models (Bador et al., 2020) as parametrization also plays a significant role (M. Wu et al., 2020). A simple comparison of climatology may not fully reflect the improvements of the new models that have more comprehensive process formulations (Di Luca et al., 2015). Dittus et al. (2016) found that many of the eight CMIP5 models they evaluated reproduced the observed increase in the difference between areas experiencing an extreme high (90%) and an extreme low (10%) proportion of the annual total precipitation from heavy precipitation (R95p/PRCPTOT) for Northern Hemisphere regions. Additionally, CMIP5 models reproduced the relation between changes in extreme and non-extreme precipitation: an increase in extreme precipitation is at the cost of a decrease in non-extreme precipitation (Thackeray et al., 2018), a characteristic found in the observational record (Gu and Adler, 2018).

The CMIP6 models perform reasonably well in capturing large-scale features of precipitation extremes, including intense precipitation extremes in the intertropical convergence zone (ITCZ), and weak precipitation extremes in dry areas in the tropical regions (Li et al., 2021) but a double-ITCZ bias over the equatorial central and eastern Pacific that appeared in CMIP5 models remains (Section 3.3.2.3). There are also regional biases in the magnitude of precipitation extremes (Kim et al., 2020). The models also have difficulties in reproducing detailed regional patterns of extreme precipitation, such as over the north-east USA (Agel and Barlow, 2020), though they performed better for summer extremes over the USA (Akinsanola et al., 2020). The comparison between climatologies in the observations and in model simulations shows that the CMIP6 and CMIP5 models that have similar horizontal resolutions also have similar model evaluation scores, and their error patterns are highly correlated (Wehner et al., 2020). In general, extreme precipitation in CMIP6 models tends to be somewhat larger than in CMIP5 models (Li et al., 2021), reflecting smaller spatial scales of extreme precipitation represented by slightly higher-resolution models (Gervais et al., 2014). This is confirmed by Kim et al. (2020), who showed that Rx1day and Rx5day simulated by CMIP6 models tend to be closer to point estimates of HadEX3 data (Dunn et al., 2020) than those simulated by CMIP5. Figure 11.14 shows the multi-model ensemble bias in mean Rx1day over the period 1979–2014 from 21 available CMIP6 models when compared with observations and reanalyses. Measured by global land root-mean-square error, the model performance is generally consistent across different observed/reanalysis data products for the extreme precipitation metric (Figure 11.14). The magnitude of extreme area mean precipitation simulated by the CMIP6 models is consistently smaller than the point estimates of HadEX3, but the model values are more comparable to those of areal-mean values (Figure 11.14) of the ERA5 reanalysis or REGEN (Contractor et al., 2020b). Taylor-plot-based performance metrics reveal strong similarities in the patterns of extreme precipitation errors over land regions between CMIP5 and CMIP6 (Srivastava et al., 2020; Wehner et al., 2020) and between annual mean precipitation errors and Rx1day errors for both generations of models (Wehner et al., 2020).

Figure 11.14 | Multi-model mean bias in annual maximum daily precipitation (Rx1day, %) for the period 1979–2014. Calculated as the difference between the Coupled Model Intercomparison Project Phase 6 (CMIP6) multi-model mean and the average of available observational or reanalysis products including (a) ERA5, (b) HadEX3, and (c) REGEN. Bias is expressed as the percent error relative to the long-term mean of the respective observational data products. Brown indicates that models are too dry, while green indicates that they are too wet. Areas without sufficient observational data are shown in grey. Adapted from Wehner et al. (2020) under the terms of the Creative Commons Attribution licence. Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

In general, there is high confidence that historical simulations by CMIP5 and CMIP6 models of similar horizontal resolutions are interchangeable in their performance in simulating the observed climatology of extreme precipitation.

Studies using regional climate models (RCMs), for example, CORDEX (Giorgi et al., 2009) over Africa (Dosio et al., 2015; Klutse et al., 2016; Pinto et al., 2016; Gibba et al., 2019), Australia, East Asia (Park et al., 2016), Europe (Prein et al., 2016a; Fantini et al., 2018), and parts of North America (Diaconescu et al., 2018) suggest that extreme rainfall events are better captured in RCMs compared to their host GCMs due to their ability to address regional characteristics, for example, topography and coastlines. However, CORDEX simulations do not show good skill over South Asia for heavy precipitation, and do not add value with respect to their GCM source of boundary conditions (Mishra et al., 2014b; S. Singh et al., 2017). The evaluation of models in simulating regional processes is discussed in detail in Section 10.3.3.4. The high-resolution simulation of mid-latitude winter extreme precipitation over land is of similar magnitude to point observations. Simulation of summer extreme precipitation has a large bias when compared with observations at the same spatial scale. Simulated extreme precipitation in the tropics also appears to be too large, indicating possible deficiencies in the parametrization of cumulus convection at this resolution. Indeed, precipitation distributions at both daily and sub-daily time scales are much improved with a convection-permitting model (Belušić et al., 2020) over Western Africa (Berthou et al., 2019b), East Africa (Finney et al., 2019), North America and Canada (Cannon and Innocenti, 2019; Innocenti et al., 2019) and over Belgium in Europe (Vanden Broucke et al., 2019).

In summary, there is high confidence in the ability of models to capture the large-scale spatial distribution of precipitation extremes over land. The magnitude and frequency of extreme precipitation simulated by CMIP6 models are similar to those simulated by CMIP5 models (high confidence).

11.4.4 Detection and Attribution, Event Attribution

Both SREX (Chapter 3, Seneviratne et al., 2012) and AR5 (Chapter 10, IPCC, 2014) concluded with medium confidence that anthropogenic forcing has contributed to a global-scale intensification of heavy precipitation over the second half of the 20th century. These assessments were based on the evidence of anthropogenic influence on aspects of the global hydrological cycle, in particular, the human contribution to the warming-induced observed increase in atmospheric moisture that leads to an increase in heavy precipitation, and limited evidence of anthropogenic influence on extreme precipitation of durations of one and five days.

Since AR5 there has been new and robust evidence and improved understanding of human influence on extreme precipitation. In particular, detection and attribution analyses have provided consistent and robust evidence of human influence on extreme precipitation of one- and five-day durations at global to continental scales. The observed increases in Rx1day and Rx5day over the Northern Hemisphere land area during 1951–2005 can be attributed to the effect of combined anthropogenic forcing, including greenhouse gases and anthropogenic aerosols, as simulated by CMIP5 models and the rate of intensification with regard to warming is consistent with C-C scaling (Zhang et al., 2013). This is confirmed to be robust when an additional nine years of observational data and the CMIP6 model simulations were used (Cross-Chapter Box 3.2, Figure 1; Paik et al., 2020). The influence of greenhouse gases is attributed as the dominant contributor to the observed intensification. The global average of Rx1day in the observations is consistent with simulations by both CMIP5 and CMIP6 models under anthropogenic forcing, but not under natural forcing (Cross-Chapter Box 3.2, Figure 1). The observed increase in the fraction of annual total precipitation falling into the top fifth or top first percentiles of daily precipitation can also be attributed to human influence at the global scale (Dong et al., 2021). The CMIP5 models were able to capture the fraction of land experiencing a strong intensification of heavy precipitation during 1960–2010 under anthropogenic forcing, but not in unforced simulations (Fischer et al., 2014). But the models underestimated the observed trends (Borodina et al., 2017a). Human influence also significantly contributed to the historical changes in record-breaking one-day precipitation (Shiogama et al., 2016). There is also limited evidence of the influences of natural forcing. Substantial reductions in Rx5day and Simple Daily Intensity Index (SDII) for daily precipitation intensity over the global summer monsoon regions occurred during 1957–2000 after explosive volcanic eruptions (Paik and Min, 2018). The reduction in post-volcanic eruption extreme precipitation in the simulations is closely linked to the decrease in mean precipitation, for which both thermodynamic effects (moisture reduction due to surface cooling) and dynamic effects (monsoon circulation weakening) play important roles.

There has been new evidence of human influence on extreme precipitation at continental scales, including the detection of the combined effect of greenhouse gases and aerosol forcing on Rx1day and Rx5day over North America, Eurasia, and mid-latitude land regions (Zhang et al., 2013) and of greenhouse gas forcing in Rx1day and Rx5day in the mid-to-high latitudes, western and eastern Eurasia, and the global dry regions (Paik et al., 2020). These findings are corroborated by the detection of human influence in the fraction of extreme precipitation in the total precipitation over Asia, Europe, and North America (Dong et al., 2021). Human influence was found to have contributed to the increase in frequency and intensity of regional precipitation extremes in North America during 1961–2010, based on optimal fingerprinting and event attribution approaches (Kirchmeier-Young and Zhang, 2020). Tabari et al. (2020) found the observed latitudinal increase in extreme precipitation over Europe to be consistent with model-simulated responses to anthropogenic forcing.

Evidence of human influence on extreme precipitation at regional scales is more limited and less robust. In north-west Australia, the increase in extreme rainfall since 1950 can be related to increased monsoonal flow due to increased aerosol emissions, but cannot be attributed to an increase in greenhouse gases (Dey et al., 2019a). Anthropogenic influence on extreme precipitation in China was detected in one study (H. Li et al., 2017), but not in another using different detection and data-processing procedures (W. Li et al., 2018a), indicating the lack of robustness in the detection results. A still weak signal-to-noise ratio seems to be the main cause for the lack of robustness, as detection would become robust 20 years in the future (W. Li et al., 2018a). Krishnan et al. (2016) attributed the observed increase in heavy rain events (intensity >100 mm day–1) in the post-1950s over central India to the combined effects of greenhouse gases, aerosols, land-use and land-cover changes, and rapid warming of the equatorial Indian Ocean SSTs. Roxyet al. (2017) and Devanand et al. (2019) showed that the increase in widespread extremes over the South Asian Monsoon during 1950–2015 is due to the combined impacts of the warming of the Western Indian Ocean (Arabian Sea) and the intensification of irrigation water management over India.

Anthropogenic influence may have affected the large-scale meteorological processes necessary for extreme precipitation and the localized thermodynamic and dynamic processes, both contributing to changes in extreme precipitation events. Several new methods have been proposed to disentangle these effects by either conditioning on the circulation state or attributing analogues. In particular, the extremely wet winter of 2013–2014 in the UK can be attributed, approximately to the same degree, to both temperature-induced increases in saturation vapour pressure and changes in the large-scale circulation (Vautard et al., 2016; Yiou et al., 2017). There are multiple cases indicating that very extreme precipitation may increase at a rate more than the C-C rate (7% per 1°C of warming) (Pall et al., 2017; Risser and Wehner, 2017; van der Wiel et al., 2017; van Oldenborgh et al., 2017; S.-Y.S. Wang et al., 2018).

Event attribution studies found an influence of anthropogenic activities on the probability or magnitude of observed extreme precipitation events, including European winters (Schaller et al., 2016; Otto et al., 2018b), extreme 2014 precipitation over the northern Mediterranean (Vautard et al., 2015), parts of the USA for individual events (Knutson et al., 2014a; Szeto et al., 2015; Eden et al., 2016; van Oldenborgh et al., 2017), extreme rainfall in 2014 over Northland, New Zealand (Rosier et al., 2015) or China (Burke et al., 2016; Sun and Miao, 2018; Yuan et al., 2018b; Zhou et al., 2018). However, for other heavy rainfall events, studies identified a lack of evidence about anthropogenic influences (Imada et al., 2013; Schaller et al., 2014; Otto et al., 2015c; Siswanto et al., 2015). There are also studies where results are inconclusive because of limited reliable simulations (Christidis et al., 2013b; Angélil et al., 2016). Overall, both the spatial and temporal scales on which extreme precipitation events are defined are important for attribution; events defined on larger scales have larger signal-to-noise ratios and thus the signal is more readily detectable. At the current level of global warming, there is a strong enough signal to be detectable for large-scale extreme precipitation events, but the chance of detecting such signals for smaller-scale events decreases (Kirchmeier-Young et al., 2019).

In summary, most of the observed intensification of heavy precipitation over land regions is likely due to anthropogenic influence, for which greenhouse gases emissions are the main contributor. New and robust evidence since AR5 includes attribution to human influence of the observed increases in annual maximum one-day and five-day precipitation and in the fraction of annual precipitation falling in heavy events. The evidence since AR5 also includes a larger fraction of land showing enhanced extreme precipitation and a larger probability of record-breaking one-day precipitation than expected by chance, both of which can only be explained when anthropogenic greenhouse gas forcing is considered. Human influence has contributed to the intensification of heavy precipitation in three continents where observational data are more abundant (high confidence) (North America, Europe and Asia). On the spatial scale of AR6 regions, there is limited evidence of human influence on extreme precipitation, but new evidence is emerging; in particular, studies attributing individual heavy precipitation events found that human influence was a significant driver of the events, particularly in the winter season.

11.4.5 Projections

The AR5 concluded it is very likely that extreme precipitation events will be more frequent and more intense over most of the mid-latitude land masses and wet tropics in a warmer world (Collins et al., 2013). Post-AR5 studies provide more and robust evidence to support the previous assessments. These include an observed increase in extreme precipitation (Section 11.4.3) and human causes of past changes (Section 11.4.4), as well as projections based on either GCM and/or RCM simulations. The CMIP5 models project that the rate of increase in Rx1day with warming is independent of the forcing scenario (Section 8.5.3.1; Pendergrass et al., 2015) or forcing mechanism (Sillmann et al., 2017a). This is confirmed in CMIP6 simulations (Sillmann et al., 2019; Li et al., 2021). In particular, for extreme precipitation that occurs once a year or less frequently, the magnitudes of the rates of change per 1°C change in global mean temperature are similar, regardless of whether the temperature change is caused by increases in carbon dioxide (CO2), methane (CH4), solar forcing, or sulphate (SO4) (Sillmann et al., 2019). In some models – CESM1 in particular – the extreme precipitation response to warming may follow a quadratic relation (Pendergrass et al., 2019). Figure 11.15 shows changes in the 10- and 50-year return values of Rx1day at different warming levels as simulated by the CMIP6 models. The median value of the scaling over land, across all Shared Socio-economic Pathway (SSP) scenarios and all models, is close to 7% per 1°C of warming for the 50-year return value of Rx1day. It is just slightly smaller for the 10- and 50-year return values of Rx5day (Li et al., 2021). The 90% ranges of the multimodel ensemble changes across all land grid boxes in the 50-year return values for Rx1day and Rx5day do not overlap between 1.5°C and 2°C warming levels (Li et al., 2021), indicating that a small increment such as 0.5°C in global warming can result in a significant increase in extreme precipitation. Projected long-period Rx1day return value changes are larger than changes in mean Rx1day and with larger relative changes for more rare events (Pendergrass, 2018; Mizuta and Endo, 2020; Wehner, 2020). The rate of change of moderate extreme precipitation may depend more on the forcing agent, similar to the mean precipitation response to warming (Lin et al., 2016, 2018). Thus, there is high confidence that extreme precipitation that occurs once a year or less frequently increases proportionally to the amount of surface warming, and the rate of change in precipitation is not dependent on the underlying forcing agents of warming.

Figure 11.15 | Projected changes in the intensity of extreme precipitation events under 1°C, 1.5°C, 2°C, 3°C, and 4°C global warming levels relative to the 1850–1900 baseline. Extreme precipitation events are defined as the annual maximum daily maximum precipitation (Rx1day) that was exceeded on average once during a 10-year period (10-year event, blue) and once during a 50-year period (50-year event, orange) during the 1850–1900 base period. Results are shown for the global land. For each box plot, the horizontal line and the box represent the median and central 66% uncertainty range, respectively, of the intensity changes across the multi-model median, and the ‘whiskers’ extend to the 90% uncertainty range. The results are based on the multi-model ensemble estimated from simulations of global climate models contributing to the Coupled Model Intercomparison Project Phase 6 (CMIP6) under different Shared Socio-economic Pathway forcing scenarios. Based on Li et al. (2021). Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

The spatial patterns of the projected changes across different warming levels are quite similar, as shown in Figure 11.16, and confirmed by near-linear scaling between extreme precipitation and global warming levels at regional scales (Seneviratne and Hauser, 2020). Internal variability modulates changes in heavy rainfall (Wood and Ludwig, 2020), resulting in different changes in different regions (Seneviratne and Hauser, 2020). Extreme precipitation nearly always increases across land areas with larger increases at higher global warming levels, except in very few regions, such as Southern Europe around the Mediterranean Basin at low warming levels (Table 11.17). The very likely ranges of the multi-model ensemble changes across all land grid boxes in the 50-year return values for Rx1day and Rx5day between 1.5°C and 1°C warming levels are above zero for all continents except Europe, with the lower bound of the likely range above zero over Europe (Li et al., 2021). Decreases in extreme precipitation are confined mostly to subtropical ocean areas and are highly correlated to decreases in mean precipitation due to storm track shifts. These subtropical decreases can extend to nearby land areas in individual realizations.

Figure 11.16 | Projected changes in annual maximum daily precipitation at (a) 1.5°C, (b) 2°C, and (c) 4°C of global warming compared to the 1850–1900 baseline. Results are based on simulations from the Coupled Model Intercomparison Project Phase 6 (CMIP6) multi-model ensemble under the Shared Socio-economic Pathway (SSP), SSP1-1.9, SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios. The numbers on the top right indicate the number of simulations included. Uncertainty is represented using the simple approach: no overlay indicates regions with high model agreement, where ≥80% of models agree on the sign of change; diagonal lines indicate regions with low model agreement, where <80% of models agree on the sign of change. For more information on the simple approach, please refer to the Cross-Chapter Box (Atlas 1. For details on the methods see Supplementary Material 11.SM.2. Changes in Rx1day are also displayed in the Interactive Atlas. Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

Projected increases in the probability of extreme precipitation of fixed magnitudes are nonlinear and show larger increases for more rare events (Figures 11.7 and 11.15; Fischer and Knutti, 2015; Kharin et al., 2018; Li et al., 2021).The CMIP5 model projected increases in the probability of high (99th and 99.9th) percentile precipitation between 1.5°C and 2°C warming scenarios are consistent with what can be expected based on observed changes (Fischer and Knutti, 2015), providing confidence in the projections. The CMIP5 model simulations show that the frequency for present-day climate 20-year extreme precipitation is projected to increase by 10% at the 1.5°C global warming level, and by 22% at the 2.0°C global warming level, while the increase in the frequency for present-day climate 100-year extreme precipitation is projected to increase by 20% and more than 45% at the 1.5°C and 2.0°C warming levels, respectively (Kharin et al., 2018). CMIP6 simulations with SSP scenarios show that the frequency of 10-year and 50-year events will be approximately doubled and tripled, respectively, at a very high warming level of 4°C (Figure 11.7; Li et al., 2021).

There is a limited number of studies on the projections of extreme hourly precipitation. The ability of GCMs to simulate hourly precipitation extremes is limited (Morrison et al., 2019) and very few modelling centres archive sub-daily and hourly precipitation prior to CMIP6 experiments. RCM simulations project an increase in extreme sub-daily precipitation in North America (C. Li et al., 2019b) and Sweden (Olsson and Foster, 2013), but these models still do not explicitly resolve convective processes that are important for properly simulating extreme sub-daily precipitation. Simulations by RCMs that explicitly resolve convective processes (convection-permitting models) are limited in length and only available in a few regions because of high computing costs. Yet, a majority of the available convection-permitting simulations project increases in the intensities of extreme sub-daily precipitation events, with the amount similar to or higher than the C-C scaling rate (Kendon et al., 2014, 2019; Ban et al., 2015; Prein et al., 2016b; Helsen et al., 2020; Fowler et al., 2021). An increase is projected in extreme sub-daily precipitation over Africa (Kendon et al., 2019); East Africa (Finney et al., 2020) and Western Africa (Berthou et al., 2019a; Fitzpatrick et al., 2020), even for areas where parametrized RCMs project a decrease; in Europe (Hodnebrog et al., 2019; Chan et al., 2020); as well as in the continental USA (Prein et al., 2016b). Overall, while limited, the available evidence points to an increase in extreme sub-daily precipitation in the future. Studies on future changes in extreme precipitation for a month or longer are limited. One study projects an increase in extreme monthly precipitation in Japan under 4°C global warming for around 80% of stations in the summer (Hatsuzuka and Sato, 2019).

In Africa (Table 11.5), extreme precipitation will likely increase under warming levels of 2°C or below (compared to pre-industrial values) and very likely increase at higher warming levels. Simulations by CMIP5, CMIP6 and CORDEX regional models project an increase in daily extreme precipitation between 1.5°C and 2.0°C warming levels. The pattern of change in heavy precipitation under different scenarios or warming levels is similar with larger increases for higher warming levels (e.g., Nikulin et al., 2018; Li et al., 2021). With increases in warming, extreme precipitation is projected to increase in the majority of land regions in Africa (Mtongori et al., 2016; Pfahl et al., 2017; Diedhiou et al., 2018; Dunning et al., 2018; Akinyemi and Abiodun, 2019; Giorgi et al., 2019). Over Southern Africa, heavy precipitation will likely increase by the end of the 21st century under RCP 8.5 (Dosio, 2016; Pinto et al., 2016; Abiodun et al., 2017; Dosio et al., 2019). However, heavy rainfall amounts are projected to decrease over western South Africa (Pinto et al., 2018) as a result of a projected decrease in the frequency of the prevailing westerly winds south of the continent that translates into fewer cold fronts and closed mid-latitudes cyclones (Engelbrecht et al., 2009; Pinto et al., 2018). Heavy precipitation will likely increase by the end of the century under RCP8.5 in West Africa (Diallo et al., 2016; Dosio, 2016; Sylla et al., 2016; Abiodun et al., 2017; Akinsanola and Zhou, 2019; Dosio et al., 2019) and is projected to increase (high confidence) in Central Africa (Fotso-Nguemo et al., 2018, 2019; Sonkoué et al., 2019) and eastern Africa (Thiery et al., 2016; Ongoma et al., 2018a). In north-east and central east Africa, extreme precipitation intensity is projected to increase across CMIP5, CMIP6 and CORDEX-CORE (high confidence) in most areas annually (Coppola et al., 2021a), but the trends differ from season to season in all future scenarios (Dosio et al., 2019). In northern Africa, there is low confidence in the projected changes in heavy precipitation, either due to a lack of agreement among studies on the sign of changes (Sillmann et al., 2013a; Giorgi et al., 2014) or due to insufficient evidence.

In Asia (Table 11.8), extreme precipitation will likely increase at global warming levels of 2°C and below, butvery likely increase at higher warming levels for the region as whole. The CMIP6 multi-model median projects an increase in the 10- and 50-year return values of Rx1day and Rx5day over more than 95% of regions, even at the 2°C warming level, with larger increases at higher warming levels, independent of emissions scenarios (Li et al., 2021, also Figure 11.7). The CMIP5 models produced similar projections. Both heavy rainfall and rainfall intensity are projected to increase (Zhou et al., 2014; Guo et al., 2016, 2018; Y. Xu et al., 2016; Endo et al., 2017; Han et al., 2018; G. Kim et al., 2018). A half-degree difference in warming between the 1.5°C and 2.0°C warming levels can result in a detectable increase in extreme precipitation over the region (Li et al., 2021), in the Asian–Australian monsoon region (Chevuturi et al., 2018), and over South Asia and China (D. Lee et al., 2018; W. Li et al., 2018b). While there are regional differences, extreme precipitation is projected to increase in almost all sub-regions, though there can be spatial heterogeneity within sub-regions, such as in India (Shashikanth et al., 2018) and South East Asia (Ohba and Sugimoto, 2019). In East and South East Asia, there is high confidence that extreme precipitation is projected to intensify (Seo et al., 2014; Zhou et al., 2014; Y. Xu et al., 2016; Nayak et al., 2017; X. Wang et al., 2017; Y. Wang et al., 2017; Guo et al., 2018; D. Li et al., 2018; Sui et al., 2018). Extreme daily precipitation is also projected to increase in South Asia (Xu et al., 2017; Han et al., 2018; Shashikanth et al., 2018). The extreme precipitation indices, including Rx5day, R95p, and days of heavy precipitation (i.e., R10mm), are all projected to increase under the RCP4.5 and RCP8.5 scenarios in central and northern Asia (Xu et al., 2017; Han et al., 2018). A general wetting across the whole Tibetan Plateau and the Himalayas is projected, with increases in heavy precipitation in the 21st century (Palazzi et al., 2013; Zhou et al., 2014; Rajbhandari et al., 2015; R. Zhang et al., 2015; Wu et al., 2017; Gao et al., 2018; Paltan et al., 2018). Agreement in projected changes by different models is low in regions of complex topography such as Hindu-Kush Himalayas (Roy et al., 2019), but CMIP5, CMIP6 and CORDEX-CORE simulations consistently project an increase in heavy precipitation in higher latitude areas, such as West and East Siberia, and Russian Far East (high confidence) (Coppola et al., 2021a).

In Australasia (Table 11.11), most CMIP5 models project an increase in Rx1day under RCP4.5 and RCP8.5 scenarios for the late 21st century (CSIRO and BOM, 2015; Alexander and Arblaster, 2017; Grose et al., 2020) and the CMIP6 multi-model median projects an increase in the 10- and 50-year return values of Rx1day and Rx5day at a rate between 5% and 6% per 1°C of near-surface global mean warming (Figure 11.7; Li et al., 2021). Yet, there is large uncertainty in the increase because projected changes in dynamic processes lead to a decrease in Rx1day that can offset the thermodynamic increase over a large portion of the region (Box 11.1, Figure 1; Pfahl et al., 2017). Projected changes in moderate extreme precipitation (the 99th percentile of daily precipitation) by RCMs under RCP8.5 for 2070–2099 are mixed, with more regions showing decreases than increases (Evans et al., 2021). It is likely that daily rainfall extremes such as Rx1day will increase at the continental scale for global warming levels at or above 3°C. Daily rainfall extremes are projected to increase at the 2.0°C global warming level (medium confidence), and there is low confidence in changes at the 1.5°C. Projected changes show important regional differences with very likely increases over Northern Australia (Alexander and Arblaster, 2017; Herold et al., 2018; Grose et al., 2020) and New Zealand (MfE, 2018) where projected dynamic contributions are small (Box 11.1 Figure 1; Pfahl et al., 2017) and medium confidence on increases over central, eastern, and Southern Australia where dynamic contributions are substantial and can affect local phenomena (CSIRO and BOM, 2015; Pepler et al., 2016; Bell et al., 2019; Dowdy et al., 2019).

In Central and South America (Table 11.14), extreme precipitation will likely increase at global warming levels of 2°C and below, butvery likely increase at higher warming levels for the region as whole. A larger increase in global surface temperature leads to a larger increase in extreme precipitation, independent of emissions scenarios (Li et al., 2021). But there are regional differences in the projection, and projected changes for more moderate extreme precipitation are also more uncertain. Extreme precipitation, represented by the number of days with daily precipitation exceeding 50 mm and the annual fraction of precipitation falling during days with the top 10% daily precipitation amount, is projected to increase on the eastern coast of Southern Central America, but to decrease along the Pacific coasts of El Salvador and Guatemala (Imbach et al., 2018). Chouet al. (2014b) and Giorgi et al. (2014) projected an increase in extreme precipitation over South-Eastern South America and the Amazon. Projected changes in moderate extreme precipitation represented by the 99th percentile of daily precipitation by different models under different emissions scenarios, even at high warming levels, are mixed: increases are projected for all regions by the CORDEX-CORE and CMIP5 simulations, while increases for some regions and decreases for other regions are projected by CMIP6 simulations (Coppola et al., 2021a). Extreme precipitation is projected to increase in the La Plata basin (Cavalcanti et al., 2015; Carril et al., 2016). Taylor et al. (2018) projected a decrease in days with intense rainfall in the Caribbean under 2°C global warming by the 2050s under RCP4.5 relative to 1971–2000.

In Europe (Table 11.17), extreme precipitation will likely increase at global warming levels of 2°C and below, butvery likely increase for higher warming levels for the region as whole. The CMIP6 multi-model median projects an increase in the 10- and 50-year return values of Rx1day and Rx5day over a majority of the region at the 2°C global warming level, with more than 95% of the region showing an increase at higher warming levels (Figure 11.7; C. Li et al., 2021). The most intense precipitation events observed today in Europe are projected to almost double in occurrence for each 1°C of further global warming (Myhre et al., 2019). Extreme precipitation is projected to increase in both boreal winter and summer over Europe (Madsen et al., 2014; Christensen et al., 2015; Nissen and Ulbrich, 2017). There are regional differences, with decreases or no change for the southern part of Europe, such as the southern Mediterranean (Tramblay and Somot, 2018; Lionello and Scarascia, 2020; Coppola et al., 2021a), uncertain changes over central Europe (Argüeso et al., 2012; Croitoru et al., 2013; Rajczak et al., 2013; Casanueva et al., 2014; Patarčić et al., 2014; Paxian et al., 2014; Roth et al., 2014; Fischer and Knutti, 2015; Monjo et al., 2016) and a strong increase in the remaining parts, including the Alps region (Gobiet et al., 2014; Donnelly et al., 2017), particularly in winter (Fischer et al., 2015), and in northern Europe. In a 3°C warmer world, there will be a robust increase in extreme rainfall over 80% of land areas in northern Europe (Madsen et al., 2014; Donnelly et al., 2017; Cardell et al., 2020).

In North America (Table 11.20), the intensity and frequency of extreme precipitation will likely increase at the global warming levels of 2°C and below, and very likely increase at higher warming levels. An increase is projected by CMIP6 model simulations (Li et al., 2021) and by previous model generations (Wu,2015; Easterling et al., 2017; Innocenti et al., 2019), as well as by RCMs (Coppola et al., 2021a). Projections of extreme precipitation over the southern portion of the continent and over Mexico are more uncertain, with decreases possible (Sillmann et al., 2013b; Alexandru, 2018; Coppola et al., 2021a).

In summary, heavy precipitation will generally become more frequent and more intense with additional global warming. At global warming levels of 4°C relative to the pre-industrial, very rare (e.g., one in 10 or more years) heavy precipitation events would become more frequent and more intense than in the recent past, on the global scale (virtually certain), and in all continents and AR6 regions: The increase in frequency and intensity is extremely likely for most continents and very likely for most AR6 regions. The likelihood is lower at lower global warming levels and for less-rare heavy precipitation events. At the global scale, the intensification of heavy precipitation will follow the rate of increase in the maximum amount of moisture that the atmosphere can hold as it warms (high confidence), of about 7% per 1°C of global warming. The increase in the frequency of heavy precipitation events will be non-linear with more warming and will be higher for rarer events (high confidence), with 10- and 50-year events to be approximately double and triple, respectively, at the 4°C warming level. Increases in the intensity of extreme precipitation events at regional scales will depend on the amount of regional warming as well as changes in atmospheric circulation and storm dynamics leading to regional differences in the rate of heavy precipitation changes (high confidence).

11.5 Floods

Floods are the inundation of normally dry land, and are classified into types (e.g., pluvial floods, flash floods, river floods, groundwater floods, surge floods, coastal floods) depending on the space and time scales and the major factors and processes involved (Section 8.2.3.2; Nied et al., 2014; Aerts et al., 2018). Flooded area is difficult to measure or quantify and, for this reason, many of the existing studies on changes in floods focus on streamflow. Thus, this section assesses changes in flow as a proxy for river floods, in addition to some types of flash floods. Pluvial and urban floods – types of flash floods resulting from the precipitation intensity exceeding the capacity of natural and artificial drainage systems – are directly linked to extreme precipitation. Because of this link, changes in extreme precipitation are the main proxy for inferring changes in pluvial and urban floods (see also Section 12.4), assuming there is no additional change in the surface condition. Changes in these types of floods are not assessed in this section, but can be inferred from the assessment of changes in heavy precipitation in Section 11.4. Coastal floods due to extreme sea levels and flood changes at regional scales are assessed in Section 12.4.

11.5.1 Mechanisms and Drivers

Since AR5, the number of studies on understanding how floods may have changed, and will change in the future, has substantially increased. Floods are a complex interplay of hydrology, climate, and human management, and the relative importance of these factors varies for different flood types and regions.

In addition to the amount and intensity of precipitation, the main factors for river floods include antecedent soil moisture (Paschalis et al., 2014; Berghuijs et al., 2016; Grillakis et al., 2016; Woldemeskel and Sharma, 2016) and snow water-equivalent in cold regions (Sikorska et al., 2015; Berghuijs et al., 2016). Other factors are also important, including stream morphology (Borga et al., 2014; Slater et al., 2015), river and catchment engineering (Pisaniello et al., 2012; Nakayama and Shankman, 2013; Kim and Sanders, 2016), land-use and land-cover characteristics (Aich et al., 2016; Rogger et al., 2017) and changes (Knighton et al., 2019), and feedbacks between climate, soil, snow, vegetation, etc. (Hall et al., 2014; Ortega et al., 2014; Berghuijs et al., 2016; Buttle et al., 2016; Teufel et al., 2019). Water regulation and management have, in general, increased resilience to flooding (Formetta and Feyen, 2019), masking effects of an increase in extreme precipitation on flood probability in some regions, even though they do not eliminate very extreme floods (Vicente-Serrano et al., 2017). This means that an increase in precipitation extremes may not always result in an increase in river floods (Sharma et al., 2018; Do et al., 2020). Yet, as very extreme precipitation can become a dominant factor for river floods, there can be some correspondence in the changes in very extreme precipitation and river floods (Ivancic and Shaw, 2015; Wasko and Sharma, 2017; Wasko and Nathan, 2019). This has been observed in the western Mediterranean (Llasat et al., 2016), in China (Q. Zhang et al., 2015a) and in the USA (Peterson et al., 2013b; Berghuijs et al., 2016; Slater and Villarini, 2016).

In regions with a seasonal snow cover, snowmelt is the main cause of extreme river flooding over large areas (Pall et al., 2019). Extensive snowmelt combined with heavy and/or long-duration precipitation can cause significant floods (D. Li et al., 2019; Krug et al., 2020). Changes in floods in these regions can be uncertain because of the compounding and competing effects of the responses of snow and rain to warming that affect snowpack size: warming results in an increase in precipitation, but also a reduction in the time period of snowfall accumulation (Teufel et al., 2019). An increase in atmospheric CO2 enhances water-use efficiency by plants (Roderick et al., 2015; Milly and Dunne, 2016; Swann et al., 2016; Swann, 2018); this could reduce evapotranspiration and contribute to the maintenance of soil moisture and streamflow levels under enhanced atmospheric CO2 concentrations (Yang et al., 2019). This mechanism would suggest an increase in the magnitude of some floods in the future (Kooperman et al., 2018). But this effect is uncertain as an increase in leaf area index, and vegetation coverage could also result in overall larger water consumption (Mátyás and Sun, 2014; Mankin et al., 2019; Teuling et al., 2019), and there are also other CO2 -related mechanisms that come into play (Cross-Chapter Box 5.1).

Various factors, such as extreme precipitation (Cho et al., 2016; Archer and Fowler, 2018), glacier lake outbursts (Schneider et al., 2014; Schwanghart et al., 2016), or dam breaks (Biscarini et al., 2016) can cause flash floods. Very intense rainfall, along with a high fraction of impervious surfaces can result in flash floods in urban areas (Hettiarachchi et al., 2018). Because of this direct connection, changes in very intense precipitation can translate to changes in urban flood potential (Rosenzweig et al., 2018), though there can be a spectrum of urban flood responses to this flood potential (Smith et al., 2013), as many factors, such as the overland flow rate and the design of urban (Falconer et al., 2009) and storm water drainage systems (Maksimović et al., 2009), can play an important role. Nevertheless, changes in extreme precipitation are the main proxy for inferring changes in some types of flash floods, (which are addressed in Section 12.4), given the relation between extreme precipitation and pluvial floods, the very limited literature on urban and pluvial floods (e.g., Skougaard Kaspersen et al., 2017), and limitations of existing methodologies for assessing changes in floods (Archer et al., 2016).

In summary, there is not always a one-to-one correspondence between an extreme precipitation event and a flood event, or between changes in extreme precipitation and changes in floods, because floods are affected by many factors in addition to heavy precipitation (high confidence). Changes in extreme precipitation may be used as a proxy to infer changes in some types of flash floods that are more directly related to extreme precipitation (high confidence).

11.5.2 Observed Trends

The SREX (Seneviratne et al., 2012) assessed low confidence for observed changes in the magnitude or frequency of floods at the global scale. This assessment was confirmed by AR5 (Hartmann et al., 2013). The SR1.5 (Hoegh-Guldberg et al., 2018) found increases in flood frequency and extreme streamflow in some regions, but decreases in other regions. While the number of studies on flood trends has increased since AR5, and there were also new analyses after the release of SR1.5 (Berghuijs et al., 2017; Blöschl et al., 2019; Gudmundsson et al., 2019), hydrological literature on observed flood changes is heterogeneous, focusing at regional and sub-regional basin scales, making it difficult to synthesize at the global and sometimes regional scales. The vast majority of studies focus on river floods using streamflow as a proxy, with limited attention to urban floods. Streamflow measurements are not evenly distributed over space, with gaps in spatial coverage, and their coverage in many regions of Africa, South America, and parts of Asia is poor (e.g., Do et al., 2017), leading to difficulties in detecting long-term changes in floods (Slater and Villarini, 2017). See also Section 8.3.1.5.

Peak flow trends are characterized by high regional variability and lack overall statistical significance of a decrease or an increase over the globe as a whole. Of more than 3500 streamflow stations in the USA, central and Northern Europe, Africa, Brazil, and Australia, 7.1% stations showed a significant increase, and 11.9% stations showed a significant decrease in annual maximum peak flow during 1961–2005 (Do et al., 2017). This is in direct contrast to the global and continental scale intensification of short-duration extreme precipitation (Section 11.4.2). There may be some consistency over large regions (see Gudmundsson et al., 2019), in high streamflows (>90th percentile), including a decrease in some regions (e.g., in the Mediterranean) and an increase in others (e.g., northern Asia), but gauge coverage is often limited. On a continental scale, a decrease seems to dominate in Africa (Tramblay et al., 2020) and Australia (Ishak et al., 2013; Wasko and Nathan, 2019), an increase in the Amazon (Barichivich et al., 2018), and trends are spatially variable in other continents (Q. Zhang et al., 2015b; Bai et al., 2016; Do et al., 2017; Hodgkins et al., 2017). In Europe, flow trends have large spatial differences (Hall et al., 2014; Mediero et al., 2015; Kundzewicz et al., 2018; Mangini et al., 2018), but there appears to be a pattern of increase in north-western Europe, and a decrease in southern and eastern Europe in annual peak flow during 1960–2000 (Blöschl et al., 2019). In North America, peak flow has increased in north-east USA and decreased in south-west USA (Peterson et al., 2013b; Armstrong et al., 2014; Mallakpour and Villarini, 2015; Archfield et al., 2016; Burn and Whitfield, 2016; Wehner et al., 2017; Neri et al., 2019). There are important changes in the seasonality of peak flows in regions where snowmelt dominates, such as northern North America (Burn and Whitfield, 2016; Dudley et al., 2017) and Northern Europe (Blöschl et al., 2017), corresponding to strong winter and spring warming.

In summary, the seasonality of floods has changed in cold regions where snowmelt dominates the flow regime in response to warming (high confidence). There is low confidence about peak flow trends over past decades on the global scale , but there are regions experiencing increases, including parts of Asia, Southern South America, north-east USA, north-western Europe, and the Amazon, and regions experiencing decreases, including parts of the Mediterranean, Australia, Africa, and south-western USA.

11.5.3 Model Evaluation

Hydrological models used to simulate floods are structurally diverse (Dankers et al., 2014; Mateo et al., 2017; Şen, 2018), often requiring extensive calibration since sub-grid processes and land-surface properties need to be parametrized, irrespective of the spatial resolutions (Döll et al., 2016; Krysanova et al., 2017). The data used to drive and calibrate the models are usually of coarse resolution, necessitating the use of a wide variety of downscaling techniques (Muerth et al., 2013). This adds uncertainty not only to the models but also to the reliability of the calibrations. The quality of the flood simulations also depends on the spatial scale, as flood processes are different for catchments of different sizes. It is more difficult to replicate flood processes for large basins, as water management and water use are often more complex for these basins.

Studies that use different regional hydrological models show a large spread in flood simulations (Dankers et al., 2014; Roudier et al., 2016; Trigg et al., 2016; Krysanova et al., 2017). Regional models reproduce moderate and high flows reasonably well (0.02–0.1 flow annual exceedance probabilities), but there are large biases for the most extreme flows (0–0.02 annual flow exceedance probability), independent of the climatic and physiographic characteristics of the basins (S. Huang et al., 2017a). Global-scale hydrological models have even more challenges, as they struggle to reproduce the magnitude of the flood hazard (Trigg et al., 2016). Also, the ensemble mean of multiple models does not perform better than individual models (Zaherpour et al., 2018).

The use of hydrological models for assessing changes in floods, especially for future projections, adds another dimension of uncertainty on top of uncertainty in the driving climate projections, including emissions scenarios, and in the driving climate models (both RCMs and GCMs) (Arnell and Gosling, 2016; Hundecha et al., 2016; Krysanova et al., 2017). The differences in hydrological models (Roudier et al., 2016; Thober et al., 2018), as well as post-processing of climate model output for the hydrological models (Muerth et al., 2013; Maier et al., 2018), add to uncertainty for flood projections.

In summary, there is medium confidence that simulations for the most extreme flows by regional hydrological models can have large biases. Global-scale hydrological models still struggle with reproducing the magnitude of floods. Projections of future floods are hampered by these difficulties and cascading uncertainties, including uncertainties in emissions scenarios and the climate models that generate inputs.

11.5.4 Detection and Attribution, Event Attribution

There are very few studies focused on the attribution of long-term changes in floods, but there are studies on changes in flood events. Most of the studies focus on flash floods and urban floods, which are closely related to intense precipitation events (Hannaford, 2015). In other cases, event attribution focused on runoff using hydrological models, and examples include river basins in the UK (Section 11.4.4; Schaller et al., 2016; Kay et al., 2018), the Okavango River in Africa (Wolski et al., 2014), and the Brahmaputra River in Bangladesh (Philip et al., 2019). Findings about anthropogenic influences vary between different regions and basins. For some flood events, the probability of high floods in the current climate is lower than in a climate without an anthropogenic influence (Wolski et al., 2014), while in other cases anthropogenic influence leads to more intense floods (Cho et al., 2016; Pall et al., 2017; van der Wiel et al., 2017; Philip et al., 2018a; Teufel et al., 2019). Factors such as land-cover change and river management can also increase the probability of high floods (Ji et al., 2020). These, along with model uncertainties and the lack of studies overall, suggest a low confidence in general statements to attribute changes in flood events to anthropogenic climate change. A few individual regions have been well studied, which allows for high confidence in the attribution of increased flooding in these cases. For example, flooding in the UK following increased winter precipitation (Schaller et al., 2016; Kay et al., 2018) can be attributed to anthropogenic climate change (Schaller et al., 2016; Vautard et al., 2016; Yiou et al., 2017; Otto et al., 2018b).

Attributing changes in heavy precipitation to anthropogenic activities (Section 11.4.4) cannot be readily translated to attributing changes in floods to human activities, because precipitation is only one of the multiple factors, albeit an important one, that affect floods. For example, Teufel et al. (2017) showed that, while human influence increased the odds of the flood-producing rainfall for the 2013 Alberta flood in Canada, it was not detected to have influenced the probability of the flood itself. Schaller et al. (2016) showed that human influence on the increase in the probability of heavy precipitation translated linearly into an increase in the resulting river flow of the Thames in the UK in winter 2014, but its contribution to the inundation was inconclusive.

Gudmundsson et al. (2021) compared the spatial pattern of the observed regional trends in high river flows (>90th percentile) over 1971–2010 with that simulated by global hydrological models. The hydrological models were driven by outputs of climate model simulations under all historical forcing and pre-industrial forcing conditions. They found complex spatial patterns of extreme river flow trends. They also found the observed spatial patterns of trends can be reproduced only if anthropogenic climate change is considered, and that simulated effects of water and land management cannot reproduce the observed spatial pattern of trends. As there is only one study and multiple caveats associated with the study, including relatively poor observational data coverage, there is low confidence about human influence on the changes in high river flows on the global scale.

In summary there is low confidence in the human influence on the changes in high river flows on the global scale. In general, there is low confidence in attributing changes in the probability or magnitude of flood events to human influence because of a limited number of studies, differences in the results of these studies and large modelling uncertainties.

11.5.5 Future Projections

The SREX (Chapter 3, Seneviratne et al., 2012) stressed the low availability of studies on flood projections under different emissions scenarios, and concluded that there was low confidence in projections of flood events given the complexity of the mechanisms driving floods at the regional scale. The AR5 WGII report (Chapter 3, Jimenez Cisneros et al., 2014) assessed with medium confidence the pattern of future flood changes, including flood hazards increasing over about half of the globe (parts of southern and South East Asia, tropical Africa, north-east Eurasia, and South America) and flood hazards decreasing in other parts of the world, despite uncertainties in GCMs and their coupling to hydrological models. The SR1.5 (Chapter 3, Hoegh-Guldberg et al., 2018) assessed with medium confidence that global warming of 2°C would lead to an expansion of the fraction of global area affected by flood hazards, compared to conditions at 1.5°C of global warming, as a consequence of changes in heavy precipitation.

The majority of new studies that produce future flood projections based on hydrological models do not typically consider aspects that are also important to actual flood severity or damages, such as flood prevention measures (Neumann et al., 2015; Şen, 2018), flood control policies (Barraqué, 2017), and future changes in land cover (see also Section 8.4.1.5). At the global scale, Alfieri et al. (2017) used downscaled projections from seven GCMs as input to drive a hydrodynamic model. They found successive increases in the frequency of high floods in all continents except Europe, associated with increasing levels of global warming (1.5°C, 2°C, 4°C). These results are supported by Paltan et al. (2018), who applied a simplified runoff aggregation model forced by outputs from four GCMs. S. Huang et al. (2018) used three hydrological models forced with bias-adjusted outputs from four GCMs to produce projections for four river basins including the Rhine, Upper Mississippi, Upper Yellow, and Upper Niger under 1.5°C, 2°C, and 3°C global warming. This study found diverse projections for different basins, including a shift towards earlier flooding for the Rhine and the Upper Mississippi, a substantial increase in flood frequency in the Rhine only under the 1.5°C and 2°C scenarios, and a decrease in flood frequency in the Upper Mississippi under all scenarios.

At the continental and regional scales, the projected changes in floods are uneven in different parts of the world, but there is a larger fraction of regions with an increase than with a decrease over the 21st century (Hirabayashi et al., 2013; Dankers et al., 2014; Arnell and Gosling, 2016; Döll et al., 2018). These results suggestmedium confidence in flood trends at the global scale, but low confidence in projected regional changes. Increases in flood frequency or magnitude are identified for south-eastern and northern Asia and India (high agreement across studies), eastern and tropical Africa, and the high latitudes of North America (medium agreement), while decreasing frequency or magnitude is found for central and eastern Europe and the Mediterranean (high confidence), and parts of South America, southern and central North America, and south-west Africa (low confidence) (Hirabayashi et al., 2013; Dankers et al., 2014; Arnell and Gosling, 2016; Döll et al., 2018). Over South America, most studies based on global and regional hydrological models show an increase in the magnitude and frequency of high flows in the western Amazon (Guimberteau et al., 2013; Langerwisch et al., 2013; Sorribas et al., 2016; Zulkafli et al., 2016) and the Andes (Hirabayashi et al., 2013; Bozkurt et al., 2018). Section 12.4 provides a detailed assessment of regional flood projections.

In summary, global hydrological models project a larger fraction of land areas to be affected by an increase in river floods than by a decrease in river floods (medium confidence). There is medium confidence that river floods will increase in the western Amazon, the Andes, and south-eastern and northern Asia. Regional changes in river floods are more uncertain than changes in pluvial floods because complex hydrological processes and forcings are involved, including land cover change and human water management.

11.6 Droughts

Droughts refer to periods of time with substantially below-average moisture conditions, usually covering large areas, during which limitations in water availability result in negative impacts for various components of natural systems and economic sectors (Wilhite and Pulwarty, 2017; Ault, 2020). Depending on the variables used to characterize it and the systems or sectors being impacted, drought may be classified in different types (Figure 8.6 and Appendix Table 11.A.1) such as meteorological (precipitation deficits), agricultural(e.g., crop yield reductions or failure, often related to soil moisture deficits), ecological (related to plant water stress that causes e.g., tree mortality), or hydrologicaldroughts (e.g., water shortage in streams or storages such as reservoirs, lakes, lagoons, and groundwater; see Glossary). The distinction of drought types is not absolute, as drought can affect different sub-domains of the Earth system concomitantly, but sometimes also asynchronously, including propagation from one drought type to another (Brunner and Tallaksen, 2019). Because of this, drought cannot be characterized using a single universal definition (Lloyd-Hughes, 2014) or directly measured based on a single variable (SREX Chapter 3; Wilhite and Pulwarty, 2017). Drought can happen on a wide range of timescales – from ‘flash droughts’ on a scale of weeks, and characterized by a sudden onset and rapid intensification of drought conditions (Hunt et al., 2014; Otkin et al., 2018; Pendergrass et al., 2020) to multi-year or decadal rainfall deficits – sometimes termed ‘megadroughts’ (see Glossary; Ault et al., 2014; Cook et al., 2016b; Garreaud et al., 2017). Droughts are often analysed using indices that are measures of drought severity, duration and frequency (Sections 8.3.1.6, 8.4.1.6, 12.3.2.6 and 12.3.2.7, and Table 11.A.1). There are many drought indices published in the scientific literature, as also highlighted in SREX (SREX Chapter 3). These can range from anomalies in single variables (e.g., precipitation, soil moisture, runoff, evapotranspiration) to indices combining different atmospheric variables.

This assessment is focused on changes in physical conditions and metrics of direct relevance to droughts: (i) precipitation deficits; (ii) excess of atmospheric evaporative demand (AED); (iii) soil moisture deficits; (iv) hydrological deficits; and e) atmospheric-based indices combining precipitation and AED (Table 11.A.1). In the regional tables (Section 11.9), the assessment is structured by drought types, addressing: (i) meteorological, (ii) agricultural and ecological, and (iii) hydrological droughts. Note that the latter two assessments directly inform the Chapter 12 assessment on projected regional changes in these climatic impact-drivers (Section 12.4). The text refers to AR6 region acronyms (Section 11.9, and see Section 1.4.5).

11.6.1 Mechanisms and Drivers

Similar to many other extreme events, droughts occur as a combination of thermodynamic and dynamic processes (Box 11.1). Thermodynamic processes contributing to drought, which are modified by greenhouse gas forcing both at global and regional scales, are mostly related to heat and moisture exchanges, and are also partly modulated by plant coverage and physiology. They affect, for instance, atmospheric humidity, temperature, and radiation, which in turn affect precipitation and/or evapotranspiration in some regions and time frames. However, dynamic processes are particularly important to explain drought variability on different time scales, from a few weeks (flash droughts) to multiannual (megadroughts). There is low confidence in the effects of greenhouse gas forcing on changes in atmospheric dynamic (Section 2.4; Section 4.3.3), and on associated changes in drought occurrence. Thermodynamic processes are thus the main driver of drought changes in a warming climate (high confidence).

11.6.1.1 Precipitation Deficits

Lack of precipitation is generally the main factor controlling drought onset. There is high confidence that atmospheric dynamics, which vary on interannual, decadal and longer time scales, is the dominant contributor to variations in precipitation deficits in the majority of world regions (Dai, 2013; Miralles et al., 2014b; Seager and Hoerling, 2014; Burgman and Jang, 2015; Dong and Dai, 2015; Schubert et al., 2016; Raymond et al., 2018; Baek et al., 2019; Drumond et al., 2019; Herrera-Estrada et al., 2019; Gimeno et al., 2020; Mishra, 2020). Precipitation deficits are driven by dynamic mechanisms taking place on different spatial scales, including synoptic processes – atmospheric rivers and extratropical cyclones, blocking and ridges (Section 11.7; Sousa et al., 2017), dominant large-scale circulation patterns (Kingston et al., 2015), and global ocean–atmosphere coupled patterns such as inter-decadal Pacific Oscillation (IPO), Atlantic Multi-decadal Oscillation (AMO) and El Niño–Southern Oscillation (ENSO; Dai and Zhao, 2017). These various mechanisms occur on different scales, are not independent, and substantially interact with one another. Also regional moisture recycling and land–atmosphere feedbacks play an important role for some precipitation anomalies (see below).

There is high confidence that land–atmosphere feedbacks play a substantial or dominant role in affecting precipitation deficits in someregions (SREX, Chapter 3; Koster et al., 2011; Gimeno et al., 2012; Taylor et al., 2012; Guillod et al., 2015; Tuttle and Salvucci, 2016; Santanello Jr. et al., 2018; Haslinger et al., 2019; Herrera-Estrada et al., 2019). The sign of the feedbacks can be either positive or negative, as well as local or non-local (Taylor et al., 2012; Guillod et al., 2015; Tuttle and Salvucci, 2016). Earth system models (ESMs) tend to underestimate non-local negative soil-moisture–precipitation feedbacks (Taylor et al., 2012) and also show high variations in their representation in some regions (Berg et al., 2017b). Soil-moisture–precipitation feedbacks contribute to changes in precipitation in climate model projections in some regions, but ESMs display substantial uncertainties in their representation, and there is thus onlylow confidence in these contributions (Berg et al., 2017b; Vogel et al., 2017, 2018).

11.6.1.2 Atmospheric Evaporative Demand

Atmospheric evaporative demand (AED) quantifies the maximum amount of actual evapotranspiration (ET) that can happen from land surfaces if they are not limited by water availability (Table 11.A.1). AED is affected by radiative and aerodynamic components. For this reason, the atmospheric dryness, often quantified with the relative humidity or the vapour pressure deficit (VPD), is not equivalent to the AED, as other variables are also highly relevant, including solar radiation and wind speed (Hobbins et al., 2012; McVicar et al., 2012a; Sheffield et al., 2012). AED can be estimated using different methods (McMahon et al., 2013), and those solely based on air temperature (e.g., Hargreaves, Thornthwaite) usually overestimate it in terms of magnitude and temporal trends (Sheffield et al., 2012), in particular, in the context of substantial background warming. Physically-based combination methods such as the Penman-Monteith equation are more adequate and recommended since 1998 by the United Nations Food and Agriculture Oganization (Pereira et al., 2015). For this reason, the assessment of this Chapter, when considering atmospheric-based drought indices, only includes AED estimates using the latter (see also Section 11.9). AED is generally higher than ET, since AED represents an upper bound for ET. Hence, an AED increase does not necessarily lead to increased ET (Milly and Dunne, 2016), in particular under drought conditions given soil moisture limitation (Bonan et al., 2014; Berg et al., 2016; Konings et al., 2017; Stocker et al., 2018). In general, AED is highest in regions where ET is lowest (e.g., desert areas), further illustrating the decoupling between the two variables under limited soil moisture.

The influence of AED on drought depends on the drought type, background climate, the environmental conditions and the moisture availability (Hobbins et al., 2016, 2017; Vicente-Serrano et al., 2020a). This influence also includes effects not related to increased ET. Under low soil moisture conditions, increased AED increases plant stress, enhancing the severity of agricultural and ecological droughts (Williams et al., 2013; Allen et al., 2015; McDowell et al., 2016; Grossiord et al., 2020). Moreover, high VPD impacts overall plant physiology; it affects the leaf and xylem safety margins, and decreases the sap velocity and plant hydraulic conductance (Fontes et al., 2018). VPD also affects the plant metabolism of carbon and, if prolonged, it may cause plant mortality via carbon starvation (Breshears et al., 2013; Hartmann, 2015). Drought projections based exclusively on AED metrics overestimate changes in soil moisture and runoff deficits. Nevertheless, AED also directly impacts hydrological drought, as ET from surface waters is not limited (Wurbs and Ayala, 2014; Friedrich et al., 2018; Hogeboom et al., 2018; K. Xiao et al., 2018), and this effect increases under climate change projections (W. Wang et al., 2018; Althoff et al., 2020). In addition, high AED increases crop water consumptions in irrigated lands (García-Garizábal et al., 2014), contributing to intensifying hydrological droughts downstream (Fazel et al., 2017; Vicente-Serrano et al., 2017).

On subseasonal to decadal scales, temporal variations in AED are strongly controlled by circulation variability (Williams et al., 2014; Chai et al., 2018; Martens et al., 2018), but thermodynamic processes also play a fundamental role and, under human-induced climate change, dominate the changes in AED. Atmospheric warming due to increased atmospheric CO2 concentrations increases AED by means of enhanced VPD in the absence of other influences (Scheff and Frierson, 2015). Because of the greater warming over land than over oceans (Sections 2.3.1.1 and 11.3), the saturation pressure of water vapour increases more over land than over oceans; oceanic air masses advected over land thus contain insufficient water vapour to keep pace with the greater increase in saturation vapour pressure over land (Sherwood and Fu, 2014; Byrne and O’Gorman, 2018; Findell et al., 2019). Land–atmosphere feedbacks are also important in affecting atmospheric moisture content and temperature, with resulting effects on relative humidity and VPD (Box 11.1; Berg et al., 2016; Haslinger et al., 2019; S. Zhou et al., 2019).

11.6.1.3 Soil Moisture Deficits

Soil moisture shows an important correlation with precipitation variability (Khong et al., 2015; Seager et al., 2019), but ET also plays a substantial role in further depleting moisture from soils, in particular in humid regions during periods of precipitation deficits (Teuling et al., 2013; Padrón et al., 2020). In addition, soil moisture plays a role in drought self-intensification under dry conditions in which ET is decreased and leads to higher AED (Miralles et al., 2019), an effect that can also contribute to triggering flash droughts (Otkin et al., 2016, 2018; DeAngelis et al., 2020; Pendergrass et al., 2020). If soil moisture becomes limited, ET is reduced, which may decrease the rate of soil drying, but can also lead to further atmospheric dryness through various feedback loops (Seneviratne et al., 2010; Miralles et al., 2014a, 2019; Teuling, 2018; Vogel et al., 2018; S. Zhou et al., 2019; Liu et al., 2020). The process is complex since vegetation cover plays a role in modulating albedo and in providing access to deeper stores of water (both in the soil and groundwater). Also, changes in land cover and in plant phenology may alter ET (Sterling et al., 2013; Woodward et al., 2014; Frank et al., 2015; Döll et al., 2016; Ukkola et al., 2016; Trancoso et al., 2017; Hao et al., 2019; Lian et al., 2020). Snow depth has strong and direct impacts on soil moisture in many systems (Gergel et al., 2017; Williams et al., 2020).

Soil moisture directly affects plant water stress and ET. Soil moisture is the primary factor that controls xylem hydraulic conductance – that is, water uptake in plants (Sperry et al., 2016; Hayat et al., 2019; X. Chen et al., 2020). For this reason, soil moisture deficits are the main driver of xylem embolism, the primary cause of plant mortality (Anderegg et al., 2012, 2016; Rowland et al., 2015). Also carbon assimilation by plants strongly depends on soil moisture (Hartzell et al., 2017), with implications for carbon starvation and plant dying if soil moisture deficits are prolonged (Sevanto et al., 2014). These mechanisms explain that soil moisture deficits are usually more relevant than AED excess to explain gross primary production anomalies and vegetation stress, mostly in sub-humid and semi-arid regions (Stocker et al., 2018; Liu et al., 2020). High CO2 concentrations are shown to potentially decrease plant ET and increase plant water-use efficiency, affecting soil moisture levels, but this effect interacts with other CO2 physiological and radiative effects (Section 11.6.5.2 and Cross-Chapter Box 5.1), and has less relevance under low soil moisture (Morgan et al., 2011; Z. Xu et al., 2016; Nackley et al., 2018; Dikšaitytė et al., 2019). ESMs represent both surface (around 10cm) and total column soil moisture, whereby total soil moisture is of more direct relevance for root water uptake, in particular by trees. There is evidence that surface soil moisture projections are substantially drier than total soil moisture projections, and may overestimate drying of relevance for most vegetation (Berg et al., 2017a).

11.6.1.4 Hydrological Deficits

Drivers of streamflow and surface water deficits are complex and strongly depend on the hydrological system analysed (e.g., streamflows in the headwaters, medium course of the rivers, groundwater, highly regulated hydrological basins). Soil hydrological processes, which control the propagation of meteorological droughts throughout different parts of the hydrological cycle (Van Loon and Van Lanen, 2012), are spatially and temporally complex (Herrera-Estrada et al., 2017; S. Huang et al., 2017b) and difficult to quantify (Van Lanen et al., 2016; Apurv et al., 2017; Caillouet et al., 2017; Konapala and Mishra, 2017; Hasan et al., 2019). The physiographic characteristics of the basins also affect how droughts propagate throughout the hydrological cycle (Van Loon and Van Lanen, 2012; Van Lanen et al., 2013; Van Loon, 2015; Konapala and Mishra, 2020; Veettil and Mishra, 2020). In addition, the assessment of groundwater deficits is very difficult given the complexity of processes that involve natural and human-driven feedbacks and interactions with the climate system (Taylor et al., 2013). Streamflow and surface water deficits are affected by land cover, groundwater and soil characteristics (Van Lanen et al., 2013; Van Loon and Laaha, 2015; Barker et al., 2016; Tijdeman et al., 2018), as well as human activities (water management and demand, damming) and land-use changes (Section 11.6.4.3; Van Loon et al., 2016; He et al., 2017; Veldkamp et al., 2017; J. Wu et al., 2018; Y. Xu et al., 2019; Jehanzaib et al., 2020). Finally, snow and glaciers are relevant for water resources in some regions. For instance, warming affects snowpack levels (Dierauer et al., 2019; Huning and AghaKouchak, 2020), as well as the timing of snow melt, thus potentially affecting the seasonality and magnitude of low flows (Barnhart et al., 2016).

11.6.1.5 Atmospheric-based Drought Indices

Given the difficulties of drought quantification and data constraints, atmospheric-based drought indices combining both precipitation and AED have been developed, as they can be derived from meteorological data that is available in most regions (with few exceptions). These demand/supply indices are not intended to be metrics of soil moisture, streamflow or vegetation water stress. Because of their reliance on precipitation and AED, they are mostly related to the actual water balance in humid regions, in which ET is not limited by soil moisture and tends towards AED. In water-limited regions and in dry periods everywhere, they constitute an upper bound for overall water-balance deficits (e.g., of surface waters) but are also related to conditions conducive to vegetation stress, particularly under soil moisture limitation (Section 11.6.1.2).

Although there are many atmospheric-based drought indices, two are assessed in this chapter: the Palmer Drought Severity Index (PDSI) and the Standardized Precipitation Evapotranspiration Index (SPEI). The PDSI has been widely used to monitor and quantify drought severity (Dai et al., 2018), but is affected by some constraints (SREX Chapter 3; Mukherjee et al., 2018a). Although the calculation of the PDSI is based on a soil water budget, the PDSI is essentially a climate drought index that mostly responds to the precipitation and the AED (van der Schrier et al., 2013; Vicente-Serrano et al., 2015; Dai et al., 2018). The SPEI also combines precipitation and AED, being equally sensitive to these two variables (Vicente-Serrano et al., 2015). The SPEI is more sensitive to AED than the PDSI (Cook et al., 2014a; Vicente-Serrano et al., 2015), although under humid and normal precipitation conditions, the effects of AED on the SPEI are small (Tomas-Burguera et al., 2020). Given the limitations associated with temperature-based AED estimates (Section 11.6.1.2), only studies using the Penman-Monteith-based SPEI and PDSI (hereafter SPEI-PM and PDSI-PM) are considered in this assessment and in the regional tables in Section 11.9.

11.6.1.6 Relation of Assessed Variables and Metrics for Changes in Different Drought Types

This Chapter assesses changes in meteorological drought, agricultural and ecological droughts, and hydrological droughts. Precipitation-based indices are used for the estimation of changes in meteorological droughts, such as the Standardized Precipitation Index (SPI) and the number of consecutive dry days (CDD). Changes in total soil moisture and soil moisture-based drought events are used for the estimation of changes in agricultural and ecological droughts, complemented by changes in surface soil moisture, water-balance estimates (precipitation minus ET), and SPEI-PM and PDSI-PM. For hydrological droughts, changes in low flows are assessed, sometimes complemented by changes in mean streamflow.

In summary, different drought types exist and they are associated with different impacts and respond differently to increasing greenhouse gas concentrations. Precipitation deficits and changes in evapotranspiration govern net water availability. A lack of sufficient soil moisture, sometimes amplified by increased atmospheric evaporative demand, result in agricultural and ecological drought. Lack of runoff and surface water result in hydrological drought. Drought events are the result of dynamic and/or thermodynamic processes, with thermodynamic processes being the main driver of drought changes under human-induced climate change (high confidence).

11.6.2 Observed Trends

Evidence on observed drought trends was limited at the time of SREX (Chapter 3) and AR5 (Chapter 2). The SREX concluded: ‘There is medium confidence that since the 1950s some regions of the world have experienced a trend to more intense and longer droughts, in particular in southern Europe and west Africa, but in some regions droughts have become less frequent, less intense, or shorter, for example, in Central North America and north-western Australia.’ The assessment at the time did not distinguish between different drought types. This Chapter includes numerous updates on observed drought trends, associated with extensive new literature and longer datasets since AR5.

11.6.2.1 Precipitation Deficits

Strong precipitation deficits have been recorded in recent decades in the Amazon (2005, 2010), south-western China (2009–2010), south-western North America (2011–2014), Australia (1997–2009), California (2014), the middle East (2012–2016), Chile (2010–2015), the Great Horn of Africa (2011), among others (van Dijk et al., 2013; Mann and Gleick, 2015; Rowell et al., 2015; Marengo and Espinoza, 2016; Dai and Zhao, 2017; Garreaud et al., 2017, 2020; Marengo et al., 2017; Brito et al., 2018; Cook et al., 2018). Global studies generally show no significant trends in SPI time series (Orlowsky and Seneviratne, 2013; Spinoni et al., 2014), and in derived drought frequency and severity data (Spinoni et al., 2019), with very few regional exceptions (Section 11.9 and Figure 11.17). Long-term decreases in precipitation are found in some AR6 regions in Africa (Central Africa and East Southern Africa), and several regions in South America (North-Eastern South America, South American Monsoon, South-Western South America, and Southern South America) (Section 11.9). Evidence of precipitation-based drying trends is also found in Western Africa, consistent with studies based on CDD trends (Figure 11.17; Chaney et al., 2014; Donat et al., 2014b; Barry et al., 2018; Dunn et al., 2020), however, there is a partial recovery of the rainfall trends since the 1980s in this region (Section 10.4.2.1). Some AR6 regions show a decrease in meteorological drought, including Northern Australia, Central Australia, Northern Europe and Central North America (Section 11.9). Other regions either do not show substantial trends in long-term meteorological drought, or they display mixed signals depending on the considered time frame and sub-regions, such as in Southern Australia (Gallant et al., 2013; Delworth and Zeng, 2014; Alexander and Arblaster, 2017; Spinoni et al., 2019; Dunn et al., 2020; Rauniyar and Power, 2020) and the Mediterranean (Camuffo et al., 2013; Gudmundsson and Seneviratne, 2016; Spinoni et al., 2017; Stagge et al., 2017; Caloiero et al., 2018; Peña-Angulo et al., 2020b; see also Section 11.9 and Atlas.8.2).

Figure 11.17 | Observed linear trend for (a) consecutive dry days (CDD) during 1960–2018, (b) standardized precipitation index (SPI) and (c) standardized precipitation-evapotranspiration index (SPEI) during 1951–2016. CDD data are from the HadEx3 dataset (Dunn et al., 2020), trend calculation of CDD as in Figure 11.9. Drought severity is estimated using 12-month SPI (SPI-12) and 12-month SPEI (SPEI-12). SPI and SPEI datasets are from Spinoni et al. (2019). The threshold to identify drought episodes was set at -1 SPI/SPEI units. Areas without sufficient data are shown in grey. No overlay indicates regions where the trends are significant at the p = 0.1 level. Crosses indicate regions where trends are not significant. For details on the methods see Supplementary Material 11.SM.2. Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

11.6.2.2 Atmospheric Evaporative Demand

In several regions, AED increases have intensified recent drought events (Williams et al., 2014, 2020; Seager et al., 2015b; Basara et al., 2019; García-Herrera et al., 2019), enhanced vegetation stress (Allen et al., 2015; Sanginés de Cárcer et al., 2018; Yuan et al., 2019), or contributed to the depletion of soil moisture or runoff through enhanced ET (high confidence) (Teuling et al., 2013; Padrón et al., 2020). Trends in pan evaporation measurements and Penman-Monteith AED estimates provide an indication of possible trends in the influence of AED on drought. Given the observed global temperature increases (Sections 2.3.1.1 and 11.3) and dominant decrease in relative humidity over land areas (Simmons et al., 2010; Willett et al., 2014), VPD has increased globally (Barkhordarian et al., 2019; Yuan et al., 2019). Pan evaporation has increased as a consequence of VPD changes in several AR6 regions, such as East Asia (Li et al., 2013; Z. Sun et al., 2018; M.-Z. Yang et al., 2018), Western and Central Europe (Mozny et al., 2020), the Mediterranean, (Azorin-Molina et al., 2015) and Central and Southern Australia (Stephens et al., 2018). Nevertheless, there is an important regional variability in observed trends, and in other AR6 regions pan evaporation has decreased – for example, in North Central America (Breña-Naranjo et al., 2017) and in the Tibetan Plateau (C. Zhang et al., 2018)). Physical models also show an important regional diversity, with an increase in New Zealand (Salinger and Porteous, 2014) and the Mediterranean (Gocic and Trajkovic, 2014; Azorin-Molina et al., 2015; Piticar et al., 2016), a decrease in South Asia (Jhajharia et al., 2015), and strong spatial variability in North America (Seager et al., 2015b). This variability is driven by the role of other meteorological variables affecting AED. Changes in solar radiation as a consequence of solar dimming and brightening may affect trends (Section 7.2.2.2; Kambezidis et al., 2012; Wang and Yang, 2014; Sanchez-Lorenzo et al., 2015). Wind speed is also relevant (McVicar et al., 2012b), and studies suggest a reduction of the wind speed in some regions (Z. Zhang et al., 2019b) that could compensate the role of the VPD increase. Nevertheless, the VPD trend seems to dominate the overall AED trends, compared to the effects of trends in wind speed and solar radiation (Wang et al., 2012; Park Williams et al., 2017; Vicente-Serrano et al., 2020a).

11.6.2.3 Soil Moisture Deficits

There are limited long-term measurements of soil moisture from ground observations (Dorigo et al., 2011; Qiu et al., 2016; Quiring et al., 2016), which impedes their use in the analysis of trends. Among the few existing observational studies covering at least two decades, several studies have investigated trends in ground soil moisture in East Asia (Section 11.9; Chen and Sun, 2015b; Liu et al., 2015; Qiu et al., 2016). Alternatively, microwave-based satellite measurements of surface soil moisture have also been used to analyse trends (Dorigo et al., 2012; Jia et al., 2018). Although there is regional evidence that microwave-based soil moisture estimates can capture well drying trends in comparison with ground soil moisture observations (Jia et al., 2018), there is only medium confidence in the derived trends, since satellite soil moisture data are affected by inhomogeneities (Dorigo et al., 2015; Rodell et al., 2018; Preimesberger et al., 2021). Furthermore, microwave-based satellites only sense surface soil moisture, which differs from root-zone soil moisture (Berg et al., 2017a), although relationships can be derived between the two (Brocca et al., 2011). Several studies have also analysed long-term soil moisture time series from observation-driven land-surface or hydrological models, including land-based reanalysis products (Albergel et al., 2013; Jia et al., 2018; Gu et al., 2019b; Markonis et al., 2021). Such models have also been used to assess changes in land water availability, estimated as precipitation minus ET, which is equal to the sum of soil moisture and runoff (Greve et al., 2014; Padrón et al., 2020).

Overall, evidence from global studies suggests that several land regions have been affected by increased soil moisture drying or water balance drying in past decades, despite some spread among products (Albergel et al., 2013; Greve et al., 2014; Gu et al., 2019b; Padrón et al., 2020). Drying has not only occurred in dry regions but also in humid regions (Greve et al., 2014). Some studies have specifically addressed changes in soil moisture at regional scale (Section 11.9). For AR6 regions, several studies suggest an increase in the frequency and areal extent of soil moisture deficits, with examples in East Asia (Cheng et al., 2015; Y. Qin et al., 2015; Jia et al., 2018), Western and Central Europe (Trnka et al., 2015b), and the Mediterranean (Hanel et al., 2018; Moravec et al., 2019; Markonis et al., 2021). Nonetheless, some analyses also show no long-term trends in soil drying in some AR6 regions – for example, in Eastern North America (Park Williams et al., 2017) and Central North America (Seager et al., 2019), as well as in North Eastern Africa (Kew et al., 2021). The soil moisture drying trends identified in both global and regional studies are generally related to increases in ET (associated with higher AED) rather than decreases in precipitation, as identified on global land for trends in water balance in the dry season (Padrón et al., 2020), as well as for some regions (Teuling et al., 2013; Cheng et al., 2015; Trnka et al., 2015a; van Der Linden et al., 2019; X. Li et al., 2020).

Evidence from observed or observations-derived trends in soil moisture and precipitation minus ET, are combined with evidence from SPEI and PDSI-PM studies to derive regional assessments of changes in agricultural and ecological droughts (Section 11.9). This assessment is summarized in Section 11.6.2.6.

11.6.2.4 Hydrological Deficits

There is evidence based on streamflow records of increased hydrological droughts in East Asia (D. Zhang et al., 2018) and southern Africa (Gudmundsson et al., 2019). In areas of Western and Central Europe and Northern Europe, there is no evidence of changes in the severity of hydrological droughts since 1950 based on flow reconstructions (Caillouet et al., 2017; Barker et al., 2019) and observations (Vicente-Serrano et al., 2019). In the Mediterranean region, there is high confidence in hydrological drought intensification (Section 11.9; Giuntoli et al., 2013; Lorenzo-Lacruz et al., 2013; Gudmundsson et al., 2019). In south-eastern South America there is a decrease in the severity of hydrological droughts (Rivera and Penalba, 2018). In North America, depending on the methods, datasets and study periods, there are differences between studies that suggest an increase (Shukla et al., 2015; Udall and Overpeck, 2017) versus a decrease in hydrological drought frequency (Mo and Lettenmaier, 2018), but in general there is strong spatial variability (Poshtiri and Pal, 2016). Streamflow observation reference networks of near-natural catchments have also been used to isolate the effect of climate trends on hydrological drought trends in a few regions, but these show limited trends in Northern Europe and Western and Central Europe (Stahl et al., 2010; Bard et al., 2015; Harrigan et al., 2018), North America (Dudley et al., 2020) and most of Australia, with the exception of Eastern and Southern Australia (X.S. Zhang et al., 2016). Given the low availability of observations, there are few studies analysing trends of drought severity in the groundwater. Nevertheless, some studies suggest a noticeable response of groundwater droughts to climate variability (Lorenzo-Lacruz et al., 2017) and increased drought frequency and severity associated with warming, probably as a consequence of enhanced ET induced by higher AED (Maxwell and Condon, 2016). This is supported by studies in Northern Europe (Bloomfield et al., 2019) and North America (Condon et al., 2020).

11.6.2.5 Atmospheric-based Drought Indices

Globally, trends in SPEI-PM and PDSI-PM suggest slightly higher increases of drought frequency and severity in regions affected by drying over the last decades in comparison to the SPI (Dai and Zhao, 2017; Spinoni et al., 2019; Song et al., 2020), mainly in regions of Western and Southern Africa, the Mediterranean and East Asia (Figure 11.17), which is consistent with observed soil moisture trends (Section 11.6.2.3). These indices suggest that AED has contributed to increase the severity of agricultural and ecological droughts compared to meteorological droughts (García-Herrera et al., 2019; Williams et al., 2020), reduce soil moisture during the dry season (Padrón et al., 2020), increase plant water stress (Allen et al., 2015; Grossiord et al., 2020; Solander et al., 2020) and trigger more severe forest fires (Abatzoglou and Williams, 2016; Turco et al., 2019; Nolan et al., 2020). A number of regional studies based on these drought indices have also shown stronger drying trends in comparison to trends in precipitation-based indices in the following AR6 regions (see also Section 11.9): NSA (R. Fu et al., 2013; Marengo and Espinoza, 2016), SCA (Hidalgo et al., 2017), WCA (Tabari and Aghajanloo, 2013; Sharafati et al., 2020), SAS (Niranjan Kumar et al., 2013), NEAF (Zeleke et al., 2017), WSAF (Edossa et al., 2016), NWN and NEN (Bonsal et al., 2013), EAS (Yu et al., 2014; Chen and Sun, 2015b; L. Li et al., 2020; Liang et al., 2020; Z. Wu et al., 2020) and MED (Kelley et al., 2015; Stagge et al., 2017; González-Hidalgo et al., 2018; Mathbout et al., 2018a).

11.6.2.6 Synthesis for Different Drought Types

Few AR6 regions show observed increases in meteorological drought (Section 11.9), mostly in Africa and South America (NES: high confidence; WAF, CAF, ESAF, SAM, SWS, SSA, SAS: medium confidence); a few others show a decrease (WSB, ESB, NAU, CAU, NEU, CNA: medium confidence). There are stronger signals indicating observed increases in agricultural and ecological drought (Section 11.9), which highlights the role of increased ET, driven by increased AED, for these trends (Sections 11.6.2.3 and11.6.2.5). Past increases in agricultural and ecological droughts are found on all continents and several regions (WAF, CAF, WSAF, ESAF, WCA, ECA, EAS, SAU, MED, WCE, NES: medium confidence), while decreases are found only in one AR6 region (NAU: medium confidence). The more limited availability of datasets makes it more difficult to assess historical trends in hydrological drought at regional scale (Section 11.9). Increasing (MED: high confidence; WAF, EAS, SAU: medium confidence) and decreasing (NEU, SES: medium confidence) trends in hydrological droughts have only been observed in a few regions.

In summary, there is high confidence that AED has increased on average on continents, contributing to increased ET and resulting water stress during periods with precipitation deficits, in particular during dry seasons. There is medium confidence in increases in precipitation deficits in a few regions of Africa and South America. Based on multiple evidence, there is medium confidence that agricultural and ecological droughts have increased in several regions on all continents (WAF, CAF, WSAF, ESAF, WCA, ECA, EAS, SAU, MED, WCE, NES: medium confidence), while there is only medium confidence in decreases in one AR6 region (NAU). More severe hydrological droughts are found in fewer regions (MED: high confidence; WAF, EAS, SAU: medium confidence).

11.6.3 Model Evaluation

11.6.3.1 Precipitation Deficits

ESMs generally show limited performance and large spread in identifying precipitation deficits and associated long-term trends in comparison with observations (Nasrollahi et al., 2015). Meteorological drought trends in the CMIP5 ensemble showed substantial disagreements compared with observations (Orlowsky and Seneviratne, 2013; Knutson and Zeng, 2018) including a tendency to overestimate drying, in particular in mid- to high latitudes (Knutson and Zeng, 2018). The CMIP6 models display a better performance in reproducing long-term precipitation trends or seasonal dynamics in some studies in Southern South America (Rivera and Arnould, 2020), East Asia (Xin et al., 2020), southern Asia (Gusain et al., 2020), and south-western Europe (Peña-Angulo et al., 2020b), but there is still too limited evidence to allow for an assessment of possible differences in performance between CMIP5 and CMIP6. Furthermore, ESMs are generally found to underestimate the severity of precipitation deficits and the dry day frequencies in comparison to observations (Fantini et al., 2018; Ukkola et al., 2018). This is probably related to shortcomings in the simulation of persistent weather events in the mid-latitudes (Section 10.3.3.3). ESMs also show a tendency to underestimate precipitation-based drought persistence at monthly to decadal time scales (Ault et al., 2014; Moon et al., 2018). The overall inter-model spread in the projected frequency of precipitation deficits is also substantial (Touma et al., 2015; Zhao et al., 2016; Engström and Keellings, 2018). Moreover, there are spatial differences in the spread, which is higher in the regions where enhanced drought conditions are projected and under high-emissions scenarios (Orlowsky and Seneviratne, 2013). Nonetheless, some event attribution studies have concluded that droughts at regional scales can be adequately simulated by some climate models (Schaller et al., 2016; Otto et al., 2018c).

11.6.3.2 Atmospheric Evaporative Demand

There is onlylimited evidence on the evaluation of AED in state-of-the-art ESMs, which is performed on externally computed AED, based on model output (Scheff and Frierson, 2015; Liu and Sun, 2016, 2017). An evaluation of average AED in 17 CMIP5 ESMs for 1981–1999 based on potential evaporation show that the models’ spatial patterns resemble the observations, but the magnitude of potential evaporation displays strong divergence among models globally and regionally (Scheff and Frierson, 2015). The evaluation of AED in 12 CMIP5 ESMs with pan evaporation observations in East Asia for 1961–2000 (Liu and Sun, 2016, 2017) show that the ESMs capture seasonal cycles well, but that regional AED averages are underestimated due to biases in the meteorological variables controlling the aerodynamic and radiative components of AED. The CMIP5 ESMs also show a strong underestimation of atmospheric drying trends compared to reanalysis data (Douville and Plazzotta, 2017).

11.6.3.3 Soil Moisture Deficits

The performance of climate models for representing soil moisture deficits shows more uncertainty than for precipitation deficits since, in addition to the uncertainties related to cloud and precipitation processes, there is uncertainty related to the representation of complex soil hydrological and boundary-layer processes (van den Hurk et al., 2011; Lu et al., 2019; Quintana-Seguí et al., 2020). Another limitation is the lack of observations, particularly for soil moisture, in most regions (Section 11.6.2.3) and the paucity of land surface property data to parametrize land surface models, in particular soil types, soil properties and depth (Xia et al., 2015). The spatial resolution of models is an additional limitation since the representation of some land–atmosphere feedbacks and topographic effects requires detailed resolution (Nicolai-Shaw et al., 2015; Van Der Linden et al., 2019). In addition to climate models, land surface and hydrological models are also used to derive historical and projected trends in soil moisture and related land water variables (Albergel et al., 2013; Cheng et al., 2015; Gu et al., 2019b; Padrón et al., 2020; Markonis et al., 2021; Pokhrel et al., 2021).

Overall, there are contrasting results on the performance of land surface models and climate models in representing soil moisture. Some studies suggest that soil moisture anomalies are well captured by land surface models driven with observation-based forcing (Dirmeyer et al., 2006; Albergel et al., 2013; Xia et al., 2014; Balsamo et al., 2015; Reichle et al., 2017; Spennemann et al., 2020), but other studies report limited agreement in the representation of interannual soil moisture variability (Stillman et al., 2016; Yuan and Quiring, 2017; Ford and Quiring, 2019) and noticeable seasonal differences in model skill in some regions (Xia et al., 2014, 2015). Models with good skill can nonetheless display biases in absolute soil moisture (Xia et al., 2014; Gu et al., 2019a), but these are not necessarily of relevance for the simulation of surface water fluxes and drought anomalies (Koster et al., 2009). There is also substantial inter-model spread (Albergel et al., 2013), particularly for the root-zone soil moisture (Berg et al., 2017a).

Regarding the performance of regional and global climate models, an evaluation of an ensemble of RCM simulations for Europe (Stegehuis et al., 2013) shows that these models display overly strong drying in early summer, resulting in an excessive decrease of latent heat fluxes, with potential implications for more severe droughts in dry environments (Teuling, 2018; van Der Linden et al., 2019). Compared with a range of observational ET estimates, CMIP5 models show an overestimation of ET on annual scale, but an ET underestimation in boreal summer in many Northern Hemisphere mid-latitude regions, also suggesting a tendency towards excessive soil drying (Mueller and Seneviratne, 2014), consistent with identified biases in soil-moisture–temperature coupling (Donat et al., 2018; Vogel et al., 2018; Selten et al., 2020). Land surface models used in ESMs display a bias in their representation of the sensitivity of interannual land carbon uptake to soil moisture conditions, which appears related to a limited range of soil moisture variations compared to observations (Humphrey et al., 2018).

For future projections, the spread of soil moisture outputs among different ESMs is more important than internal variability and scenario uncertainty, and the bias is strongly related to the sign of the projected change (Ukkola et al., 2018; Lu et al., 2019; Selten et al., 2020). The CMIP5 ESMs that project more drying and warming in mid-latitude regions show a substantial bias in soil-moisture–temperature coupling (Donat et al., 2018; Vogel et al., 2018). Although CMIP6 and CMIP5 simulations for soil moisture changes are similar overall, some differences are found in projections in a few regions (Section 11.9; Cook et al., 2020). There is still limited evidence to assess whether there are substantial differences in model performance in the two ensembles, but improvements in modelling aspects relevant for soil moisture have been reported for precipitation (Section 11.6.3.2), and a better performance has been found in CMIP6 for the representation of long-term trends in soil moisture in continental USA (Yuan et al., 2021). Despite the mentioned model limitations, the representation of soil moisture processes in ESMs uses physical and biological understanding of the underlying processes, which can well represent the temporal anomalies associated with temporal variability and trends in climate. In summary, there is medium confidence in the representation of soil moisture deficits in ESMs and related land surface and hydrological models.

11.6.3.4 Hydrological Deficits

Streamflow and groundwater are not directly simulated by ESMs, which only simulate runoff, but they are generally represented in hydrological models (Prudhomme et al., 2014; Giuntoli et al., 2015), which are typically driven in a stand-alone manner by observed or simulated climate forcing. The simulation of hydrological deficits is much more problematic than the simulation of mean streamflow or peak flows (Fundel et al., 2013; Stoelzle et al., 2013; Velázquez et al., 2013; Staudinger et al., 2015), since models tend to be too responsive to the climate forcing and do not satisfactorily capture low flows (Tallaksen and Stahl, 2014). Simulations of hydrological drought metrics show uncertainties related to the contribution of both GCMs and hydrological models (Bosshard et al., 2013; Giuntoli et al., 2015; Samaniego et al., 2017; Vetter et al., 2017), but hydrological models forced by the same climate input data also show a large spread (van Huijgevoort et al., 2013; Ukkola et al., 2018). At the catchment scale, the hydrological model uncertainty is higher than both GCM and downscaling uncertainty (Vidal et al., 2016), and the hydrological models show issues in representing drought propagation throughout the hydrological cycle (Barella-Ortiz and Quintana Seguí, 2019). A study on the evaluation of streamflow droughts in seven global (hydrological and land surface) models compared with observations in near-natural catchments of Europe showed a substantial spread among models, an overestimation of the number of drought events, and an underestimation of drought duration and drought-affected area (Tallaksen and Stahl, 2014).

11.6.3.5 Atmospheric-based Drought Indices

A number of studies have analysed the ability of models to capture drought severity and trends based on climatic drought indices. Given the limitations of ESMs in reproducing the dynamic of precipitation deficits and AED (11.6.3.1, 11.6.3.2), atmospheric-based drought indices derived from ESM data for these two variables are also affected by uncertainties and biases. A comparison of historical trends in PDSI-PM for 1950–2014 derived from CMIP3 and CMIP5, with respective estimates derived from observations (Dai and Zhao, 2017) show a similar behaviour at global scale (long-term decrease), but low spatial agreement in the trends except in a few regions (Mediterranean, South Asia, north-western USA). In future projections, there is an important spread in PDSI-PM and SPEI-PM among different models (Cook et al., 2014a).

11.6.3.6 Synthesis for Different Drought Types

The performance of ESMs used to assessed changes in variables related to meteorological droughts, agricultural and ecological droughts, and hydrological droughts, shows the presence of biases and uncertainties compared to observations, but there is medium confidence in their overall performance for assessing drought projections given process understanding. Given the substantial inter-model spread documented for all related variables, the consideration of multi-model projections increases the confidence of model-based assessments, with onlylow confidence in assessments based on single models.

In summary, the evaluation of ESMs, land surface and hydrological models for the simulation of droughts is complex, due to the regional scale of drought trends, their overall low signal-to-noise ratio, and the lack of observations in several regions, in particular for soil moisture and streamflow. There is medium confidence in the ability of ESMs to simulate trends and anomalies in precipitation deficits and AED, and also medium confidence in the ability of ESMs and hydrological models to simulate trends and anomalies in soil moisture and streamflow deficits, on global and regional scales.

11.6.4 Detection and Attribution, Event Attribution

11.6.4.1 Precipitation Deficits

There are only two AR6 regions where there is at least medium confidence that human-induced climate change has contributed to changes in meteorological droughts (Section 11.9). In South-Western South America, there is medium confidence that human-induced climate change has contributed to an increase in meteorological droughts (Boisier et al., 2016; Garreaud et al., 2020), while in Northern Europe, there is medium confidence that it has contributed to a decrease in meteorological droughts (Section 11.9; Gudmundsson and Seneviratne, 2016). In other AR6 regions, there is inconclusive evidence in the attribution of long-term trends, but a human contribution to single meteorological events or sub-regional trends has been identified in some instances (Section 11.9; see also below). In the Mediterranean region, some studies have identified a precipitation decline or increase in meteorological drought probability for time frames since the early or mid 20th century, and a possible human contribution to these trends (Hoerling et al., 2012; Gudmundsson and Seneviratne, 2016; Knutson and Zeng, 2018), also on sub-regional scale in Syria from 1930 to 2010 (Kelley et al., 2015). On the contrary, other studies have not identified precipitation and meteorological drought trends in the region for the long term (Camuffo et al., 2013; Paulo et al., 2016; Vicente-Serrano et al., 2021) and also from the mid 20th century (Norrant and Douguédroit, 2006; Stagge et al., 2017). There is evidence of substantial internal variability in long-term precipitation trends in the region (Section 11.6.2.1), which limits the attribution of human influence on variability and trends of meteorological droughts from observational records (Kelley et al., 2012; Peña-Angulo et al., 2020b). In addition, there are important sub-regional trends showing mixed signals (Section 11.9; MedECC, 2020). The evidence thus leads to an assessment of low confidence in the attribution of observed short-term changes in meteorological droughts in the region (Section 11.9). In North America, the human influence on precipitation deficits is complex (Wehner et al., 2017), with low confidence in the attribution of long-term changes in meteorological drought in AR6 regions (Section 11.9; Lehner et al., 2018). In Africa there is low confidence that human influence has contributed to the observed long-term meteorological drought increase in Western Africa (Sections 11.9 and 10.6.2). There is low confidence in the attribution of the observed increasing trends in meteorological drought in East Southern Africa, but evidence that human-induced climate change has affected recent meteorological drought events in the region (Section 11.9).

Attribution studies for recent meteorological drought events are available for various regions. In Western and Central Europe, a multi-method and multi-model attribution study on the 2015 Central European drought did not find conclusive evidence for whether human-induced climate change was a driver of the rainfall deficit, as the results depended on model and method used (Hauser et al., 2017). In the Mediterranean region, a human contribution was found in the case of the 2014 meteorological drought in the southern Levant based on a single-model study (Bergaoui et al., 2015). In Africa, there is some evidence of a contribution of human emissions to single meteorological drought events, such as the 2015–2017 southern African drought (Funk et al., 2018a; Yuan et al., 2018a; Pascale et al., 2020), and the three-year (2015–2017) drought in the western Cape Town region of South Africa (Otto et al., 2018c). An attributable signal was not found in droughts that occurred in different years with different spatial extents in the last decade in North and South Eastern Africa (Marthews et al., 2015; Uhe et al., 2017; Otto et al., 2018a; Philip et al., 2018b; Kew et al., 2021). However, an attributable increase in 2011 long rain failure was identified (Lott et al., 2013). Further studies have attributed some African meteorological drought events to large-scale modes of variability, such as the strong 2015 El Niño (Box 11.4; Philip et al., 2018b) and increased SSTs overall (Funk et al., 2015a, 2018b). Natural variability was dominant in the California droughts of 2011–2012 to 2013–2014 (Seager et al., 2015a). In Asia, no climate change signal was found in the record dry spell over Singapore and Malaysia in 2014 (Mcbride et al., 2015) or the drought in central south-west Asia in 2013–2014 (Barlow and Hoell, 2015). Nevertheless, the South East Asia drought of 2015 has been attributed to anthropogenic warming effects (Shiogama et al., 2020). Recent droughts occurring in South America, specifically in the southern Amazon region in 2010 (Shiogama et al., 2013) and in north-east South America in 2014 (Otto et al., 2015b) and 2016 (Martins et al., 2018) were not attributed to anthropogenic climate change. Nevertheless, the central Chile drought between 2010 and 2018 has been suggested to be partly associated to global warming (Boisier et al., 2016; Garreaud et al., 2020). The 2013 New Zealand meteorological drought was attributed to human influence by Harrington et al. (2014, 2016) based on fully coupled CMIP5 models, but no corresponding change in the dry end of simulated precipitation from a stand-alone atmospheric model was found by Angélil et al. (2017).

Event attribution studies also highlight a complex interplay of anthropogenic and non-anthropogenic climatological factors for some events. For example, anthropogenic warming contributed to the 2014 drought in North Eastern Africa by increasing east African and west Pacific temperatures, and increasing the gradient between standardized western and central Pacific SSTs, causing reduced rainfall (Funk et al., 2015a). As different methodologies, models and data sources have been used for the attribution of precipitation deficits, Angélil et al. (2017) re-examined several events using a single analytical approach and climate model and observational datasets. Their results showed a disagreement in the original anthropogenic attribution in a number of precipitation deficit events, which increased uncertainty in the attribution of meteorological droughts events.

11.6.4.2 Soil Moisture Deficits

There is a growing number of studies on the detection and attribution of long-term changes in soil moisture deficits. Mueller and Zhang (2016) concluded that anthropogenic forcing contributed significantly to soil moisture drying in the warm season in the Northern Hemisphere from 1951 to 2005 and also led to an increase in the land surface area affected by soil moisture deficits, which can be reproduced by CMIP5 models only if anthropogenic forcings are involved. Gu et al. (2019b) similarly identified a global-scale soil moisture drying tendency in land surface model data from the Global Land Data Assimilation System 2 over the time frame 1948–2005, which was attributed to anthropogenic forcing based on evaluation with CMIP5 models using optimal fingerprinting. Padrón et al. (2019) analysed long-term reconstructed and CMIP5 simulated dry season water availability, defined as precipitation minus ET (i.e., equivalent to soil moisture and runoff availability), also related to agricultural and ecological droughts. They found an intensification of dry-season precipitation minus evapotranspiration deficits over a predominant fraction of the land area in the last three decades, which can only be explained by anthropogenic forcing and is mostly related to increases in ET. Similarly, Williams et al. (2020) concluded that human-induced climate change contributed to the strong soil moisture deficits recorded in the last two decades in Western North America through VPD increases associated with higher air temperatures and lower air humidity. There are few studies analysing the attribution of particular episodes of soil moisture deficits to anthropogenic influence. Nevertheless, the available modelling studies coincide in supporting an anthropogenic attribution associated with more extreme temperatures, exacerbating AED and increasing ET, and thus depleting soil moisture, as observed in southern Europe in 2017 (García-Herrera et al., 2019) and in Australia in 2018 (Lewis et al., 2020) and 2019 (van Oldenborgh et al., 2021), the latter event having strong implications in the propagation of widespread megafires (Nolan et al., 2020).

11.6.4.3 Hydrological Deficits

It is often difficult to separate the role of climate trends from changes in land use, water management and demand for changes in hydrological deficits, especially on a regional scale. However, a global study based on a recent multi-model experiment with global hydrological models and covering several AR6 regions suggests a dominant role of anthropogenic radiative forcing for trends in low, mean and high flows, while simulated effects of water and land management do not suffice to reproduce the observed spatial pattern of trends (Gudmundsson et al., 2021). Regional studies also suggest that climate trends have been dominant compared to land use and human water management for explaining trends in hydrological droughts in some regions, for instance in Ethiopia (Fenta et al., 2017), China (Xie et al., 2015), and North America for the Missouri and Colorado basins, as well as in California (Shukla et al., 2015; Udall and Overpeck, 2017; Ficklin et al., 2018; K. Xiao et al., 2018; Glas et al., 2019; Martin et al., 2020; Milly and Dunne, 2020).

In other regions, the influence of human water uses can be more important to explain hydrological drought trends (Y. Liu et al., 2016; Mohammed and Scholz, 2016). There is medium confidence that human-induced climate change has contributed to an increase of hydrological droughts in the Mediterranean (Giuntoli et al., 2013; Vicente-Serrano et al., 2014; Gudmundsson et al., 2017), but also medium confidence that changes in land use and terrestrial water management contributed to these trends (Section 11.9; Teuling et al., 2019; Vicente-Serrano et al., 2019). A global study with a single hydrological model estimated that human water consumption has intensified the magnitude of hydrological droughts by 20–40% over the last 50 years, and that the human water use contribution to hydrological droughts was more important than climatic factors in the Mediterranean, and central USA, as well as in parts of Brazil (Wada et al., 2013). However, Gudmundsson et al. (2021) concluded that the contribution of human water use is smaller than that of anthropogenic climate change to explain spatial differences in the trends of low flows based on a multi-model analysis. There is still limited evidence and thus low confidence in assessing these trends at the scale of single regions, with few exceptions (Section 11.9).

11.6.4.4 Atmospheric-based Drought Indices

Different studies using atmospheric-based drought indices suggest an attributable anthropogenic signal, characterized by the increased frequency and severity of droughts (Cook et al., 2018), associated to increased AED (Section 11.6.4.2). The majority of studies are based on the PDSI-PM. Williams et al. (2015) and Griffin and Anchukaitis (2014) concluded that increased AED has had an increased contribution to drought severity over the last decades, and played a dominant role in the intensification of the 2012–2014 drought in California. The same temporal pattern and physical mechanism was stressed by Z. Li et al. (2017) in central Asia. Marvel et al. (2019) compared tree ring-based reconstructions of the PDSI-PM over the past millennium with PDSI-PM estimates based on output from CMIP5 models. The comparisons suggested a contribution of greenhouse gas forcing to the changes since the beginning of the 20th century, although characterized with temporal differences that could be driven by temporal variations in the aerosol forcing. This was in agreement with the dominant external forcings of aridification at global scale between 1950 and 2014 (Bonfils et al., 2020). In the Mediterranean region, there is medium confidence of drying attributable to antropogenic forcing as a consequence of the strong AED increase (Gocic and Trajkovic, 2014; Azorin-Molina et al., 2015; Liuzzo et al., 2016; Maček et al., 2018), which has enhanced the severity of drought events (Vicente-Serrano et al., 2014; Stagge et al., 2017; González-Hidalgo et al., 2018). In particular, this effect was identified to be the main driver of the intensification of the 2017 drought that affected south-western Europe, and was attributed to the human forcing (García-Herrera et al., 2019). Nangombe et al. (2020) and L. Zhang et al. (2020) concluded from differences between precipitation and AED that anthropogenic forcing contributed to the 2018 droughts that affected southern Africa and south-eastern China, respectively, principally as consequence of the high AED that characterized these two events.

11.6.4.5 Synthesis for Different Drought Types

The regional evidence on attribution for single AR6 regions generally shows low confidence for a human contribution to observed trends in meteorological droughts at regional scale, with few exceptions (Section 11.9). There is medium confidence that human influence has contributed to increases in agricultural and ecological droughts in the dry season in some regions and has led to an overall increase in the affected land area. At regional scales, there is medium confidence in a contribution of human-induced climate change to increases in agricultural and ecological droughts in the Mediterranean and Western North America (Section 11.9). There is medium confidence that human-induced climate change has contributed to an increase in hydrological droughts in the Mediterranean region, but also medium confidence in contributions from other human influences, including water management and land use (Section 11.9). Several meteorological and agricultural and ecological drought events have been attributed to human-induced climate change, even in regions where no long-term changes are detected (medium confidence). However, a lack of attribution to human-induced climate change has also been shown for some events (medium confidence).

In summary, human influence has contributed to increases in agricultural and ecological droughts in the dry season in some regions due to increases in evapotranspiration (medium confidence). The increases in evapotranspiration have been driven by increases in atmospheric evaporative demand induced by increased temperature, decreased relative humidity and increased net radiation over affected land areas (high confidence). There is low confidence that human influence has affected trends in meteorological droughts in most regions, butmedium confidence that they have contributed to the severity of some single events. There is medium confidence that human-induced climate change has contributed to increasing trends in the probability or intensity of recent agricultural and ecological droughts, leading to an increase of the affected land area. Human-induced climate change has contributed to global-scale change in low flow, but human water management and land-use changes are also important drivers (medium confidence).

11.6.5 Projections

The SREX (Chapter 3) asssessed with medium confidence projections of increased drought severity in some regions, including southern Europe and the Mediterranean, central Europe, central America and Mexico, north-east Brazil, and southern Africa, and low confidence elsewhere given large inter-model spread. The AR5 (Chapters 11 and 12) also assessed large uncertainties in drought projections at the regional and global scales. The assessment of drought mechanisms under future climate change scenarios depends on the model used (Section 11.6.3). Moreover, uncertainties in drought projections are affected by the consideration of plant physiological responses to increasing atmospheric CO2 (Cross-Chapter Box 5.1; Milly and Dunne, 2016; Greve et al., 2019; Mankin et al., 2019; Yang et al., 2020), the role of soil-moisture–atmosphere feedbacks for changes in water balance and aridity (Berg et al., 2016; Zhou et al., 2021), and statistical issues related to considered drought time scales (Vicente-Serrano et al., 2020c). Nonetheless, the extensive literature available since AR5 allows a substantially more robust assessment of projected changes in droughts, also subdivided in different drought types (meteorological drought, agricultural and ecological drought, and hydrological drought). This includes assessments of projected changes in droughts, including changes at 1.5°C, 2°C and 4°C of global warming, for all AR6 regions (Section 11.9). Projected changes show increases in drought frequency and intensity in several regions as function of global warming (high confidence). There are also substantial increases in drought hazard probability from 1.5°C to 2°C global warming and for further additional increments of global warming (high confidence) (Figures 11.18 and 11.19). These findings are based on both CMIP5 and CMIP6 analyses (Section 11.9; Wartenburger et al., 2017; Greve et al., 2018; L. Xu et al., 2019), and strengthen the conclusions of SR1.5 Chapter 3.

11.6.5.1 Precipitation Deficits

Studies based on CMIP5, CMIP6 and Coordinated Regional Climate Downscaling Experiment (CORDEX) projections show a consistent signal in the sign and spatial pattern of projections of precipitation deficits. Global studies based on these multi-model ensemble projections (Orlowsky and Seneviratne, 2013; Martin, 2018; Spinoni et al., 2020; Ukkola et al., 2020; Coppola et al., 2021b) show particularly strong signal-to-noise ratios for increasing meteorological droughts in the following AR6 regions: MED, ESAF, WSAF, SAU, CAU, NCA, SCA, NSA and NES (Section 11.9). There is also substantial evidence of changes in meteorological droughts at 1.5°C versus 2°C of global warming from global studies (Wartenburger et al., 2017; L. Xu et al., 2019). The patterns of projected changes in mean precipitation are consistent with the changes in the drought duration, but they are not consistent with the changes in drought intensity (Ukkola et al., 2020). In general, CMIP6 projections suggest a stronger increase of the probability of precipitation deficits than CMIP5 projections (Cook et al., 2020; Ukkola et al., 2020). Projections for the number of CDDs in CMIP6 (Figure 11.19) for different levels of global warming relative to 1850–1900 show similar spatial patterns as projected precipitation deficits. The robustness of the patterns in projected precipitation deficits identified in the global studies is also consistent with results from regional studies (Giorgi et al., 2014; Marengo and Espinoza, 2016; Pinto et al., 2016; J. Huang et al., 2018; Maúre et al., 2018; Nangombe et al., 2018; Tabari and Willems, 2018; Abiodun et al., 2019; Dosio et al., 2019).

In Africa, a strong increase in the length of dry spells (CDD) is projected for 4°C of global warming over most of the continent, with the exception of central and eastern Africa (Section 11.9; Sillmann et al., 2013a; Giorgi et al., 2014; Han et al., 2019). In West Africa, a strong reduction of precipitation is projected (Sillmann et al., 2013a; Diallo et al., 2016; Akinsanola and Zhou, 2019; Han et al., 2019; Todzo et al., 2020) at 4°C of global warming, and CDD would increase with stronger global warming levels (Klutse et al., 2018). The regions most strongly affected are southern Africa (ESAF, WSAF) (Nangombe et al., 2018; Abiodun et al., 2019) and northern Africa (part of the MED region), with increases in meteorological droughts already at 1.5°C of global warming, and further increases with increasing global warming (Section 11.9). CDD is projected to increase more in the southern Mediterranean (northern Africa) than in the northern part of the Mediterranean region (Lionello and Scarascia, 2020).

In Asia, most AR6 regions showlow confidence in projected changes in meteorological droughts at 1.5°C and 2°C of global warming, with a few regions displaying a decrease in meteorological droughts at 4°C of global warming (RAR, ESB, RFE, ECA; medium confidence), although there is a projected increase in meteorological droughts in South East Asia at 4°C (medium confidence) (Section 11.9). In South East Asia, an increasing frequency of precipitation deficits is projected as a consequence of an increasing frequency of extreme El Niño (Cai et al., 2014b, 2015, 2018).

In Central America, projections suggest an increase in mid-summer meteorological drought (Imbach et al., 2018) and increased CDD (Chou et al., 2014a; Giorgi et al., 2014; Nakaegawa et al., 2014). In the Amazon, there is also a projected increase in dryness (Marengo and Espinoza, 2016), which is the combination of a projected increase in the frequency and geographic extent of meteorological drought in the eastern Amazon, and an opposite trend in the west (Duffy et al., 2015). In South-Western South America, there is a projected increase of CDD (Chou et al., 2014a; Giorgi et al., 2014) and in Chile, drying is projected to prevail (Boisier et al., 2018). In the South America monsoon region, an increase in CDD is projected (Chou et al., 2014a; Giorgi et al., 2014), but a decrease is projected in South-Eastern and Southern South America (Giorgi et al., 2014). In Central America, mid-summer meteorological drought is projected to intensify during 2071–2095 for the RCP8.5 scenario (Corrales‐Suastegui et al., 2020).

An increase in the frequency, duration and intensity of meteorological droughts is projected in south-west, south and east Australia (Kirono et al., 2020; Shi et al., 2020). In Canada and most of the USA, based on the SPI, Swain and Hayhoe (2015) identified drier summer conditions in projections over most of the region, and there is a consistent signal toward an increase in duration and intensity of droughts in southern North America (Pascale et al., 2016; Escalante-Sandoval and Nuñez-Garcia, 2017). In California, more precipitation variability is projected, characterized by increased frequency of consecutive drought and humid periods (Swain et al., 2018).

Substantial increases in meteorological drought are projected in Europe, in particular in the Mediterranean region, already at 1.5°C of global warming (Section 11.9). In southern Europe, model projections display a consistent drying among models (Russo et al., 2013; Hertig and Tramblay, 2017; Guerreiro et al., 2018a; Raymond et al., 2019). In Western and Central Europe there is some spread in CMIP5 projections, with some models projecting very strong drying, and others close to no trend (Vogel et al., 2018), although CDD is projected to increase in CMIP5 projections under the RCP 8.5 scenario (Hari et al., 2020). The overall evidence suggests an increase in meteorological drought at 4°C in the WCE region (medium confidence) (Section 11.9).

Overall, based on global and regional studies, several hot spot regions are identified, displaying more frequent and severe meteorological droughts with increasing global warming, including several AR6 regions at 1.5°C (WSAF, ESAF, SAU, MED, NES) and 2°C of global warming (WSAF, ESAF, EAU, SAU, MED, NCA, SCA, NSA, NES) (Section 11.9). At 4°C of global warming, there is also confidence in increases in meteorological droughts in further regions (WAF, WCE, ENA, CAR, NWS, SAM, SWS, SSA; Section 11.9), showing a geographical expansion of meteorological drought with increasing global warming. Only few regions are projected to have less intense or frequent meteorological droughts (Section 11.9).

11.6.5.2 Atmospheric Evaporative Demand

Effects of AED on droughts in future projections is under debate. The CMIP5 models project an increase in AED over the majority of the world with increasing global warming, mostly as a consequence of strong VPD increases (Scheff and Frierson, 2015; Vicente-Serrano et al., 2020a). However, ET is projected to increase less than AED in many regions due to plant physiological responses related to: i) CO2 effects on plant photosynthesis; and ii) soil moisture control on ET.

Several studies suggest that increasing atmospheric CO2 could lead to reduced leaf stomatal conductance, which would increase water-use efficiency and reduce plant water needs, thus limiting ET (Cross-Chapter Box 5.1; Roderick et al., 2015; Milly and Dunne, 2016; Swann et al., 2016; Greve et al., 2017; Scheff et al., 2017; Lemordant et al., 2018; Swann, 2018). The implemention of a CO2 -dependent land resistance parameter has been suggested for the estimation of AED (Yang et al., 2019). Nevertheless, there are other relevant mechanisms, as soil moisture deficits and VPD also play an important role in the control of the leaf stomatal conductance (Z. Xu et al., 2016; Menezes-Silva et al., 2019; Grossiord et al., 2020), and a number of ecophysiological and anatomical processes affect the response of plant physiology under higher atmospheric CO2 concentrations (Cross-Chapter Box 5.1; Mankin et al., 2019; Menezes-Silva et al., 2019). The benefits of the atmospheric CO2 for plant stress and agricultural and ecological droughts would be minimal precisely during dry periods given stomatal closure in response to limited soil moisture (Allen et al., 2015; Z. Xu et al., 2016). In addition, CO2 effects on plant stomatal conductance could not entirely compensate for the increased demand associated with warming (Liu and Sun, 2017); in large tropical and subtropical regions (e.g., southern Africa, the Amazon, the Mediterranean and southern North America), AED is projected to increase, even considering the possible CO2 effects on land resistance (Vicente-Serrano et al., 2020a). Moreover, these CO2 effects would not affect the direct evaporation from soil and water bodies, which is very relevant in the reservoirs of warm areas (Friedrich et al., 2018). Because of these uncertainties, there is low confidence whether increased CO2 -induced water-use efficiency in vegetation will substantially reduce global plant transpiration and will diminish the frequency and severity of soil moisture and streamflow deficits associated with the radiative effect of higher CO2 concentrations (Cross-Chapter Box 5.1).

Another mechanism reducing the ET response to increased AED in projections is the control of soil moisture limitations on ET, which leads to reduced stomatal conductance under water stress (Berg and Sheffield, 2018; Stocker et al., 2018; Zhou et al., 2021). This response may be further amplified through VPD-induced decreases in stomatal conductance (Anderegg et al., 2020). However, the decreased stomatal conductance in response to soil moisture limitation and enhanced CO2 would further enhance AED (Sherwood and Fu, 2014; Berg et al., 2016; Teuling, 2018; Miralles et al., 2019), whereby the overall effects on AED in ESMs are found to be of similar magnitude for soil moisture limitation and CO2 physiological effects on stomatal conductance (Berg et al., 2016). Increased AED is thus both a driver and a feedback with respect to changes in ET, complicating the interpretation of its role on drought changes with increasing CO2 concentrations and global warming.

11.6.5.3 Soil Moisture Deficits

Areas with projected soil moisture decreases do not fully coincide with areas that have projected precipitation decreases, although there is substantial consistency in the respective patterns (Dirmeyer et al., 2013; Berg and Sheffield, 2018). However, there are more regions affected by increased soil moisture deficits (Figure 11.19) than precipitation deficits (Figures 2a,b,c and Cross-Chapter Box 11.1) as a consequence of enhanced AED and the associated increased ET, as highlighted by some studies (Orlowsky and Seneviratne, 2013; Dai et al., 2018; Section 8.2.2.1). Moisture in the top soil layer is projected to decrease more than precipitation at all warming levels (Lu et al., 2019), extending the regions affected by severe soil moisture deficits over most of south and central Europe (Lehner et al., 2017; Ruosteenoja et al., 2018; Samaniego et al., 2018; van Der Linden et al., 2019), southern North America (Cook et al., 2019), South America (Orlowsky and Seneviratne, 2013), southern Africa (Lu et al., 2019), East Africa (Rowell et al., 2015), Southern Australia (Kirono et al., 2020), India (Mishra et al., 2014a) and East Asia (Figure 11.19; Cheng et al., 2015). Projected changes in total soil moisture display less widespread drying than those for surface soil moisture (Berg et al., 2017a), but still more than for precipitation (Cross-Chapter Box 11.1, Figures 2a,b,c). The severity of droughts based on surface soil moisture in future projections is stronger than projections based on precipitation and runoff (Dai et al., 2018; Vicente-Serrano et al., 2020c). Nevertheless, in many parts of the world where soil moisture is projected to decrease, the signal-to-noise ratio among models is low; only the projections in the Mediterranean, Europe, the south-western USA, and southern Africa show a high signal-to-noise ratio in soil moisture projections (Figure 11.19; Lu et al., 2019). Increases in soil moisture deficits are found to be statistically signicant at regional scale in the Mediterranean region, southern Africa and western South America for changes as small as 0.5°C in global warming, based on differences between +1.5°C and +2°C of global warming (Wartenburger et al., 2017). Several other regions are affected when considering changes in droughts for higher changes in global warming (Section 11.9 and Figure 11.19). Seasonal projections of drought frequency for boreal winter (December–January–February) and summer (June–July–August), from CMIP6 multi-model ensemble for 1.5°C, 2°C and 4°C global warming levels, show contrasting trends (Figure 11.19). In the boreal winter in the Northern Hemisphere, the areas affected by drying show high agreement with those characterized by an increase in meteorological drought projections (Figures 8.14 and 12.4). On the contrary, in the boreal summer, the drought frequency increases worldwide in comparison to meteorological drought projections, with large areas of the Northern Hemisphere displaying a high signal-to-noise ratio (low spead between models). This stresses the dominant influence of ET (as a result of increased AED) in intensifying agricultural and ecological droughts in the warm season in many locations, including mid- to high latitudes.

Increased soil moisture limitation and associated changes in droughts are projected to lead to increased vegetation stress affecting the global land carbon sink in ESM projections (Green et al., 2019), with implications for projected global warming (Cross-Chapter Box 5). There is high confidence that the global land sink will become less efficient due to soil moisture limitations and associated agricultural and ecological drought conditions in some regions in higher-emissions scenarios, specially under global warming levels above 4°C; however, there is low confidence in how these water cycle feedbacks will play out in lower-emissions scenarios (at 2°C global warming or lower; Cross-Chapter Box 5.1).

Figure 11.18 | Projected changes in (a) the intensity and (b) the frequency of drought under 1°C, 1.5°C, 2°C, 3°C, and 4°C global warming levels relative to the 1850–1900 baseline. (c) Summaries are computed for the AR6 regions in which there is at least medium confidence in an increase in agriculture/ecological drought at the 2°C global warming level (‘drying regions’), including Western North America, Central North America, North Central America, Southern Central America, Northern South America, North-Eastern South America, South American Monsoon, South-Western South America, Southern South America, West and Central Europe, Mediterranean, West Southern Africa, East Southern Africa, Madagascar, Eastern Australia, Southern Australia. Caribbean is not included in the calculation because the number of land grid points was too small. A drought event is defined as a 10-year drought event whose annual mean soil moisture was below its 10th percentile from the 1850–1900 base period. For each box plot, the horizontal line and the box represent the median and central 66% uncertainty range, respectively, of the frequency or the intensity changes across the multi-model ensemble, and the ‘whiskers’ extend to the 90% uncertainty range. The line of zero in (a) indicates no change in intensity, while the line of one in (b) indicates no change in frequency. The results are based on the multi-model ensemble estimated from simulations of global climate models contributing to the Coupled Model Intercomparison Project Phase 6 (CMIP6) under different Shared Socio-economic Pathway (SSP) forcing scenarios. Intensity changes in (a) are expressed as standard deviations of the interannual variability in the period 1850–1900 of the corresponding model. For details on the methods see Supplementary Material 11.SM.2. Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

11.6.5.4 Hydrological Deficits

Some studies support wetting tendencies as a response to a warmer climate when considering globally averaged changes in runoff over land (Roderick et al., 2015; Greve et al., 2017; Y. Yang et al., 2018), and streamflow projections respond to enhanced CO2 concentrations in CMIP5 models (Yang et al., 2019). Nevertheless, when focusing regionally on low-runoff periods, model projections also show an increase of hydrological droughts in large world regions (Wanders and Van Lanen, 2015; Dai et al., 2018; Vicente-Serrano et al., 2020c). In general, the frequency of hydrological deficits is projected to increase over most of the continents, although with regionally and seasonally differentiated effects (Section 11.9), with medium confidence of increase in the following AR6 regions: WCE, MED, SAU, WCA, WNA, SCA, NSA, SAM, SWS, SSA, WSAF, ESAF and MDG (Section 11.9; Forzieri et al., 2014; Prudhomme et al., 2014; Giuntoli et al., 2015; Wanders and Van Lanen, 2015; Roudier et al., 2016; Marx et al., 2018; Cook et al., 2019; Zhao et al., 2020). However, there are large uncertainties related to the hydrological/impact model used (Prudhomme et al., 2014; Schewe et al., 2014; Gosling et al., 2017), limited signal-to-noise ratio (due to model spread) in several regions (Giuntoli et al., 2015), and also uncertainties in the projection of future human activities, including water demand and land cover changes, which may represent more than 50% of the projected changes in hydrological droughts in some regions (Wanders and Wada, 2015).

Regions dependent on mountainous snowpack as a temporary reservoir may be affected by severe hydrological droughts in a warmer world. In the southern European Alps, both winter and summer low flows are projected to be more severe, with a 25% decrease in the 2050s (Vidal et al., 2016). In western USA, a 22% reduction in winter snow water equivalent is projected at around 2°C of global warming, with a further decrease of a 70% reduction at 4°C global warming (Rhoades et al., 2018). This decline would cause less predictable hydrological droughts in snowmelt-dominated areas of North America (Livneh and Badger, 2020). The exact magnitude of the influence of higher temperatures on snow-related droughts is, however, difficult to estimate (Mote et al., 2016), since the streamflow changes could affect the timing of peak streamflows but not necessarily their magnitude. In addition, projected changes in hydrological droughts downstream of declining glaciers can be very complex to assess (Chapter 9, see also SROCC).

11.6.5.5 Atmospheric-based Drought Indices

Studies show a stronger drying in projections based on atmospheric-based drought indices compared to ESM projections of changes in soil moisture (Berg and Sheffield, 2018) and runoff (Yang et al., 2019). It has been suggested that this difference is due to physiological CO2 effects (Section 11.6.5.2; Roderick et al., 2015; Milly and Dunne, 2016; Swann et al., 2016; Lemordant et al., 2018; Scheff, 2018; Swann, 2018; Greve et al., 2019; Yang et al., 2020). Nonetheless, there is evidence that differences in projections between atmospheric-based drought indices and water-balance metrics from ESMs are not alone due to CO2 -plant effects (Berg et al., 2016; Scheff et al., 2021). Differences can also be related to the fact that AED is an upper bound for ET in dry regions and conditions (Section 11.6.1.2) and that soil moisture stress limits increases in ET in projections (Section 11.6.5.2; Berg et al., 2016; Zhou et al., 2021). In general, atmospheric-based indices show more drying than total column soil moisture (Berg and Sheffield, 2018; Cook et al., 2020; Scheff et al., 2021), but are more consistent with projected increases in surface soil moisture deficits (Dirmeyer et al., 2013; Dai et al., 2018; Lu et al., 2019; Cook et al., 2020; Vicente-Serrano et al., 2020c).

Atmospheric-based drought indices are not metrics of soil moisture or runoff (Section 11.6.1.5) so their projections may not necessarily reflect the same trend of online simulated soil moisture and runoff. Independently of effects on the land water balance, atmospheric-based drought indices will reflect the potential vegetation stress resulting from deficits between available water and enhanced AED, even in conditions with no or low ET. Under dry conditions, the enhanced AED associated with human forcing would increase plant water stress (Brodribb et al., 2020), with effects on widespread forest dieback and mortality (Anderegg et al., 2013; Williams et al., 2013; Allen et al., 2015; McDowell and Allen, 2015; McDowell et al., 2016, 2020), and stronger risk of megafires (Flannigan et al., 2016; Podschwit et al., 2018; Clarke and Evans, 2019; Varela et al., 2019). For these reasons, there is high confidence that the future projections of enhanced drought severity showed by the PDSI-PM and the SPEI-PM are representative of more frequent and severe plant stress episodes and more severe agricultural and ecological drought impacts in some regions.

Global tendencies towards more severe and frequent agricultural and ecological drought conditions are identified in future projections when focusing on atmospheric-based drought indices such as the PDSI-PM or the SPEI-PM. They expand the spatial extent of drought conditions compared to meteorological drought to most of North America, Europe, Africa, Central and East Asia and Southern Australia (Cook et al., 2014a; Chen and Sun, 2017a, b; Gao et al., 2017b; Lehner et al., 2017; Zhao and Dai, 2017; Dai et al., 2018; Naumann et al., 2018; Potopová et al., 2018; Gu et al., 2020; Vicente-Serrano et al., 2020c; Dai, 2021). Projections in PDSI-PM and SPEI-PM are used to complement total soil moisture projections in assessing projected changes in agricultural and ecological drought (Section 11.9).

11.6.5.6 Synthesis for Different Drought Types

The tables in Section 11.9 provide assessed projected changes in metorological drought, agricultural and ecological drought, and hydrological droughts. The assessment shows that several regions will be affected by more severe agricultural and ecological droughts even if global warming is stabilized at 2°C, including MED, WSAF, SAM and SSA (high confidence), and ESAF, MDG, EAU, SAU, SCA, CAR, NSA, NES, SWS, WCE, NCA, WNA and CNA (medium confidence). Some regions are also projected to be affected by more severe agricultural and ecological droughts at 1.5°C (MED, WSAF, ESAF, SAU, NSA, SAM, SSA, can; medium confidence) At 4°C of global warming, even more regions would be affected by agricultural and ecological droughts (WCE, MED, CAU, EAU, SAU, WCA, EAS, SCA, CAR, NSA, NES, SAM, SWS, SSA, NCA, CNA, ENA, WNA, WSAF, ESAF and MDG). NEAF, SAS are also projected to experience less agricultural and ecological drought with global warming (medium confidence). Projected changes in meteorological droughts are, overall, less extended but also affect several AR6 regions, at 1.5°C and 2°C (MED, EAU, SAU, SCA, NSA, NCA, WSAF, ESAF, MDG) and 4°C of global warming (WCE, MED, EAU, SAU, SEA, SCA, CAR, NWS, NSA, NES, SAM, SWS, SSA, NCA, ENA, WAF, WSAF, ESAF, MDG). Several regions are also projected to be affected by more hydrological droughts at 1.5°C and 2°C (WCE, MED, WNA, WSAF, ESAF) and 4°C of global warming (NEU, WCE, EEU, MED, SAU, WCA, SCA, NSA, SAM, SWS, SSA, WNA, WSAF, ESAF, MDG). To illustrate the changes in both intensity and frequency of drought in the regions where strongest changes are projected, Figure 11.18 displays changes in the intensity and frequency of soil moisture drought under different global warming levels (1.5°C, 2°C, 4°C) relative to the 1851-1900 baseline based on CMIP6 simulations under different SSP forcing scenarios averaged over “drying regions”, i.e. AR6 regions for which there is at least medium confidence in increase in agricultural and ecological drought at 2°C of global warming. The 90% uncertainty ranges for the projected changes in both intensity and frequency are above zero, indicating significant increase in both intensity and frequency of drought in these regions as whole.

In summary, more regions are affected by increases in agricultural and ecological droughts with increasing global warming (high confidence). New evidence strengthens the SR1.5 conclusion that even relatively small incremental increases in global warming (+0.5°C) cause a worsening of droughts in some regions (high confidence). Some regions are projected to be affected by more severe agricultural and ecological droughts at 1.5°C of global warming (MED, WSAF, ESAF, SAU, NSA, SAM, SSA, can; medium confidence). A larger number of regions are projected to be affected by more severe agricultural and ecological droughts at 2°C of global warming, including MED, WSAF, SAM and SSA (high confidence), and ESAF, MDG, EAU, SAU, SCA, CAR, NSA, NES, SWS, WCE, NCA, WNA and CNA (medium confidence). At 4°C of global warming, even more regions would be affected by agricultural and ecological droughts (WCE, MED, CAU, EAU, SAU, WCA, EAS, SCA, CAR, NSA, NES, SAM, SWS, SSA, NCA, CNA, ENA, WNA, WSAF, ESAF and MDG). Some regions are also projected to experience less agricultural and ecological drought with global warming (medium confidence;NEAF, SAS). There is high confidence that the projected increases in agricultural and ecological droughts are strongly affected by AED increases in a warming climate, although ET increases are projected to be smaller than those in AED due to soil moisture limitations and CO2 effects on leaf stomatal conductance. Enhanced atmospheric CO2 concentrations lead to enhanced water-use efficiency in plants (medium confidence), but there is low confidence that it can alleviate agricultural and ecological droughts, or hydrological droughts, at higher global warming levels characterized by limited soil moisture and enhanced AED.

Projected changes in meteorological droughts are overall less extended than for agricultural and ecological droughts, but also affect several AR6 regions, even at 1.5°C and 2°C of global warming. Several regions are also projected to be more strongly affected by hydrological droughts with increasing global warming (NEU, WCE, EEU, MED, SAU, WCA, SCA, NSA, SAM, SWS, SSA, WNA, WSAF, ESAF, MDG). Increased soil moisture limitation and associated changes in droughts are projected to lead to increased vegetation stress in many regions, with implications for the global land carbon sink (Cross-Chapter Box 5). There is high confidence that the global land carbon sink will become less efficient due to soil moisture limitations and associated drought conditions in some regions in higher-emissions scenarios, especially under global warming levels above 4°C; however, there is low confidence on how these water cycle feedbacks will play out in lower-emissions scenarios (at 2°C global warming or lower; Cross-Chapter Box5.1).

11.7 Extreme Storms

Extreme storms, such as tropical cyclones (TCs), extratropical cyclones (ETCs), and severe convective storms often have substantial societal impacts. Quantifying the effect of climate change on extreme storms is challenging, partly because extreme storms are rare, short-lived, and local, and individual events are largely influenced by stochastic variability. The high degree of random variability makes detection and attribution of extreme storm trends more uncertain than detection and attribution of trends in other aspects of the environment in which the storms evolve (e.g., larger-scale temperature trends). Projecting changes in extreme storms is also challenging because of constraints in the models’ ability to accurately represent the small-scale physical processes that can drive these changes. Despite the challenges, progress has been made since AR5.

The SREX (Chapter 3) concluded that there is low confidence in observed long-term (40 years or more) trends in TC intensity, frequency, and duration, and any observed trends in phenomena such as tornadoes and hail; it is likely that extratropical storm tracks have shifted poleward in both the Northern and Southern Hemispheres, and that heavy rainfalls and mean maximum wind speeds associated with TCs will increase with continued greenhouse gas warming; it is likely that the global frequency of TCs will either decrease or remain essentially unchanged, while it is more likely than not that the frequency of the most intense storms will increase substantially in some ocean basins; there is low confidence in projections of small-scale phenomena such as tornadoes and hail storms; and there is medium confidence that there will be a reduced frequency and a poleward shift of mid-latitude cyclones due to future anthropogenic climate change.

Figure 11.19 | Projected changes in (a–c) the number of consecutive dry days (CDD), (d–f) annual mean soil moisture over the total column, and (g–l) the frequency and intensity of 1-in-10-year soil moisture drought for the June-to-August and December-to-February seasons at 1.5°C, 2°C, and 4°C of global warming compared to the 1850–1900 baseline. The unit for soil moisture change is the standard deviation of interannual variability in soil moisture during 1850–1900. Standard deviation is a widely used metric in characterizing drought severity. A projected reduction in mean soil moisture by one standard deviation corresponds to soil moisture conditions typical of about 1-in-6-year droughts during 1850–1900 becoming the norm in the future. Results are based on simulations from the Coupled Model Intercomparison Project Phase 6 (CMIP6) multi-model ensemble under the Shared Socio-economic Pathway (SSP), SSP1-1.9, SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios. The numbers in the top right indicate the number of simulations included. Uncertainty is represented using the simple approach: no overlay indicates regions with high model agreement, where ≥80% of models agree on the sign of change; diagonal lines indicate regions with low model agreement, where <80% of models agree on the sign of change. For more information on the simple approach, please refer to the Cross-Chapter Box (Atlas 1. For details on the methods see Supplementary Material 11.SM.2. Changes in CDDs are also displayed in the Interactive Atlas. Further details on data sources and processing are available in the chapter data table (Table 11.SM.9).

Since SREX, several IPCC Reports also assessed storms. The AR5 (Chapter 2, Hartmann et al., 2013) assessment observed with low confidence long-term trends in TC metrics, but revised the statement from SREX to state that it is virtually certain that there are increasing trends in North Atlantic TC activity since the 1970s, with medium confidence that anthropogenic aerosol forcing has contributed to these trends. The AR5 concluded that it is likely that TC precipitation and mean intensity will increase and more likely than not that the frequency of the strongest storms will increase with continued greenhouse gas warming. confidence in projected trends in overall TC frequency remained low. confidence in observed and projected trends in hail storm and tornado events also remained low. The SROCC (Chapter 6, Collins et al., 2019) assessed past and projected TCs and ETCs, supporting the AR5 conclusions with some additional detail. Literature subsequent to AR5 adds support to the likelihood of increasing trends in TC intensity, precipitation, and frequency of the most intense storms, while some newer studies have added uncertainty to projected trends in overall frequency. A growing body of literature since AR5 on the poleward migration of TCs led to a new assessment in SROCC of low confidence that the migration in the western North Pacific represents a detectable climate change contribution from anthropogenic forcing. The SR1.5 (Chapter 3, Hoegh-Guldberg et al., 2018) essentially confirmed the AR5 assessment of TCs and ETCs, adding that heavy precipitation associated with TCs is projected to be higher at 2°C compared to 1.5°C global warming (medium confidence).

The SREX, AR5, SROCC, and SR1.5, do not provide assessments of the atmospheric rivers, and SROCC and SR1.5 do not assess severe convective storms and extreme winds. This section assesses the state of knowledge on the four phenomena of TCs, ETCs, severe convective storms, and extreme winds. Atmospheric rivers are addressed in Chapter 8. In this respect, this assessment closely mirrors the SROCC assessment of TCs and ETCs, while updating SREX and AR5 assessments of severe convective storms and extreme winds.

11.7.1 Tropical Cyclones

11.7.1.1 Mechanisms and Drivers

The genesis, development, and tracks of TCs depend on conditions of the larger-scale circulations of the atmosphere and ocean (Christensen et al., 2013). Large-scale atmospheric circulations, such as the Hadley and Walker circulations and the monsoon circulations can significantly affect TCs, as can internal variability acting on various time scales (Annex IV), from intra-seasonal (e.g., the Madden–Julian and Boreal Summer Intraseasonal oscillations and equatorial waves) and interannual (e.g., the El Niño–Southern Oscillation and Pacific and Atlantic Meridional Modes), to inter-decadal (e.g., Atlantic Multidecadal Variability and Pacific Decadal Variability). This broad range of natural variability makes detection of anthropogenic effects difficult, and uncertainties in the projected changes of these modes of variability increase uncertainty in the projected changes in TC activity. Aerosol forcing also affects sea surface temperature (SST) patterns and cloud microphysics, and it is likely that observed changes in TC activity are partly caused by changes in aerosol forcing (Evan et al., 2011; Ting et al., 2015; Sobel et al., 2016, 2019; Takahashi et al., 2017; Zhao et al., 2018; Reed et al., 2019). Among possible changes from these drivers, there is medium confidence that the Hadley cell has widened and will continue to widen in the future (Sections 2.3, 3.3 and 4.5). This likely causes latitudinal shifts of TC tracks (Sharmila and Walsh, 2018). Regional TC activity changes are also strongly affected by projected changes in SST warming patterns (Yoshida et al., 2017), which are highly uncertain (Chapters 4 and 9).

11.7.1.2 Observed Trends

Identifying past trends in TC metrics remains a challenge due to the heterogeneous character of the historical instrumental data, which are known as ‘best-track’ data (Schreck et al., 2014). There is low confidence in most reported long-term (multi-decadal to centennial) trends in TC frequency- or intensity-based metrics due to changes in the technology used to collect the best-track data. This should not be interpreted as implying that no physical (real) trends exist, but rather as indicating that either the quality or the temporal length of the data is not adequate to provide robust trend detection statements, particularly in the presence of multi-decadal variability.

There are previous and ongoing efforts to homogenize the best-track data (Elsner et al., 2008; Kossin et al., 2013, 2020; Choy et al., 2015; Landsea, 2015; Emanuel et al., 2018) and there is substantial literature that finds positive trends in intensity-related metrics in the best-track during the ‘satellite period’, which is generally limited to around the past 40 years (Kang and Elsner, 2012; Kishtawal et al., 2012; Kossin et al., 2013, 2020; Mei and Xie, 2016; Zhao et al., 2018; Tauvale and Tsuboki, 2019). When best-track trends are tested using homogenized data, the intensity trends generally remain positive, but are smaller in amplitude (Kossin et al., 2013; Holland and Bruyère, 2014). Kossin et al. (2020) extended the homogenized TC intensity record to the period 1979–2017 and identified significant global increases in major TC exceedance probability of about 6% per decade. In addition to trends in TC intensity, there is evidence that TC intensification rates and the frequency of rapid intensification events have increased within the satellite era (Kishtawal et al., 2012; Balaguru et al., 2018; Bhatia et al., 2018). The increase in intensification rates is found in the best-track and the homogenized intensity data.

A subset of the best-track data corresponding to hurricanes that have directly impacted the USA since 1900 is considered to be reliable, and shows no trend in the frequency of USA landfall events (Knutson et al., 2019). However, an increasing trend in normalized USA hurricane damage, which accounts for temporal changes in exposed wealth (Grinsted et al., 2019), and a decreasing trend in TC translation speed over the USA (Kossin, 2019) have also been identified in this period. A similarly reliable subset of the data representing TC landfall frequency over Australia shows a decreasing trend in Eastern Australia since the 1800s (Callaghan and Power, 2011), as well as in other parts of Australia since 1982 (Chand et al., 2019; Knutson et al., 2019). A paleoclimate proxy reconstruction shows that recent levels of TC interactions along parts of the Australian coastline are the lowest in the past 550–1500 years (Haig et al., 2014). Existing TC datasets show substantial inter-decadal variations in basin-wide TC frequency and intensity in the western North Pacific, but a statistically significant north-westward shift in the western North Pacific TC tracks since the 1980s (T.-C. Lee et al., 2020). Inthe case of the North Indian Ocean, analyses of trends are highly dependent on the details of each analysis (e.g., pre- and/or post-monsoon season period, or Bay of Bengal and/or Arabian Sea region). The most consistent trends are an increase in the occurrence of the most intense TCs, and a decrease in the overall TC frequency, in particular in the Bay of Bengal (Sahoo and Bhaskaran, 2016; Balaji et al., 2018; Singh et al., 2019; Baburaj et al., 2020). In the South Indian Ocean (SIO), an increase in the occurrence of the most intense TCs has been noted; however, there are well-known data quality issues there (Kuleshov et al., 2010; Fitchett, 2018). When the SIO data are homogenized, a significant increase is found in the fractional proportion of global Category 3–5 TC instances (6-hourly intensity estimates during the lifetime of each TC) to all Category 1–5 instances (Kossin et al., 2020).

Figure 11.20 | Summary schematic of past and projected changes in tropical cyclone (TC), extratropical cyclone (ETC), atmospheric river (AR), and severe convective storm (SCS) behaviour. Global changes (blue shading) from top to bottom: (i) Increased mean and maximum rain rates in TCs, ETCs, and ARs [past (low confidence due to lack of reliable data) and projected (high confidence)]; (ii) Increased proportion of stronger TCs [past (medium confidence) and projected (high confidence)]; (iii) Decrease or no change in global frequency of TC genesis [past (low confidence due to lack of reliable data) and projected (medium confidence)]; and (iv) Increased and decreased ETC wind speed, depending on the region, as storm tracks change [past (low confidence due to lack of reliable data) and projected (medium confidence)]. Regional changes, from left to right: (i) Poleward TC migration in the western North Pacific and subsequent changes in TC exposure [past (medium confidence) and projected (medium confidence)]; (ii) Slowdown of TC forward translation speed over the contiguous USA and subsequent increase in TC rainfall [past (medium confidence) and projected (low confidence due to lack of directed studies)]; and (iii) Increase in mean and maximum SCS rain rate and increase in spring SCS frequency and season length over the contiguous USA [past (low confidence due to lack of reliable data) and projected (medium confidence)].

As with all confined regional analyses of TC frequency, it is generally unclear whether any identified changes are due to a basin-wide change in TC frequency, or to systematic track shifts (or both). From an impacts perspective, however, these changes over land are highly relevant and emphasize that large-scale modifications in TC behaviour can have a broad spectrum of impacts on a regional scale.

Subsequent to AR5, two metrics have been analysed that are argued to be comparatively less sensitive to data issues than frequency- and intensity-based metrics. Trends in these metrics have been identified over the past 70 years or more (Knutson et al., 2019). The first metric – the mean latitude where TCs reach their peak intensity – exhibits a global and regional poleward migration during the satellite period (Kossin et al., 2014). The poleward migration can influence TC hazard exposure and risk (Kossin et al., 2016a) and is consistent with the independently observed expansion of the tropics (Lucas et al., 2014). The migration has been linked to changes in the Hadley circulation (Altman et al., 2018; Sharmila and Walsh, 2018; Studholme and Gulev, 2018). The migration is also apparent in the mean locations where TCs exhibit eyes (Knapp et al., 2018), which is when TCs are most intense. Part of the Northern Hemisphere poleward migration is due to basin-wide changes in TC frequency (Kossin et al., 2014, 2016b; Moon et al., 2015, 2016) and the trends, as expected, can be sensitive to the time period chosen (Tennille and Ellis, 2017; Kossin, 2018; Song and Klotzbach, 2018) and to subsetting of the data by intensity (Zhan and Wang, 2017). The poleward migration is particularly pronounced and well-documented in the western North Pacific basin (Kossin et al., 2016a; Oey and Chou, 2016; Liang et al., 2017; Nakamura et al., 2017; Altman et al., 2018; Daloz and Camargo, 2018; J. Sun et al., 2019; T.-C. Lee et al., 2020; Yamaguchi and Maeda, 2020a; Kubota et al., 2021).

A second metric that is argued to be comparatively less sensitive to data issues than frequency- and intensity-based metrics is TC translation speed (Kossin, 2018), which exhibits a global slowdown in the best-track data over the period 1949–2016. TC translation speed is a measure of the speed at which TCs move across the Earth’s surface, and is very closely related to local rainfall amounts (i.e., a slower translation speed causes greater local rainfall). TC translation speed also affects structural wind damage and coastal storm surge by changing the hazard event duration. The slowdown is observed in the best-track data from all basins except the Northern Indian Ocean, and is also found in a number of regions where TCs interact directly with land. The slowing trends identified in the best-track data by Kossin (2018) have been argued to be largely due to data heterogeneity. Moon et al. (2019) and Lanzante (2019) provide evidence that meridional TC track shifts project onto the slowing trends, and argue that these shifts are due to the introduction of satellite data. Kossin (2019) provides evidence that the slowing trend is real by focusing on Atlantic TC track data over the contiguous USA in the 118-year period 1900–2017, which are generally considered reliable. In this period, mean TC translation speed has decreased by 17%. The slowing TC translation speed is expected to increase local rainfall amounts, which would increase coastal and inland flooding. In combination with slowing translation speed, abrupt TC track direction changes – that can be associated with track ‘meanders’ or ‘stalls’ – have become increasingly common along the North American coast since the mid-20th century, leading to more rainfall in the region (Hall and Kossin, 2019).

In summary, there is mounting evidence that a variety of TC characteristics have changed over various time periods. It is likely that the global proportion of Category 3–5 tropical cyclone instances and the frequency of rapid intensification events have increased globally over the past 40 years. It is very likely that the average location where TCs reach their peak wind intensity has migrated poleward in the western North Pacific Ocean since the 1940s. It is likely that TC translation speed has slowed over the USA since 1900.

11.7.1.3 Model Evaluation

Accurate projections of future TC activity have two principal requirements: accurate representation of changes in the relevant environmental factors (e.g., SSTs) that can affect TC activity, and accurate representation of actual TC activity in given environmental conditions.In particular, models’ capacity to reproduce historical trends or interannual variabilities of TC activity is relevant to the confidence in future projections. One test of the models is to evaluate their ability to reproduce the dependency of the TC statistics in the different basins in the real world, in addition to their capability of reproducing atmospheric and ocean environmental conditions. For the evaluation of projections of TC-relevant environmental variables, AR5 confidence statements were based on global surface temperature and moisture, but not on the detailed regional structure of SST and atmospheric circulation changes such as steering flows and vertical shear, which affect characteristics of TCs (genesis, intensity, tracks, etc.). Various aspects of TC metrics are used to evaluate how capable models are of simulating present-day TC climatologies and variability (e.g., TC frequency, wind intensity, precipitation, size, tracks, and their seasonal and interannual changes) (Walsh et al., 2015; Camargo and Wing, 2016; Knutson et al., 2019, 2020). Other examples of TC climatology/variability metrics are spatial distributions of TC occurrence and genesis (Walsh et al., 2015), seasonal cycles and interannual variability of basin-wide activity (Zhao et al., 2009; Shaevitz et al., 2014; Kodama et al., 2015; Murakami et al., 2015; Yamada et al., 2017) or landfalling activity (Lok and Chan, 2018), as well as newly developed process-diagnostics designed specifically for TCs in climate models (D. Kim et al., 2018; Wing et al., 2019; Moon et al., 2020).

Confidence in the projection of intense TCs, such as those of Category 4–5, generally becomes higher as the resolution of the models becomes higher. The Coupled Model Intercomparison Project Phases 5 and 6 (CMIP5/6) class climate models (around 100–200 km grid spacing) cannot simulate TCs of Category 4–5 intensity. They do simulate storms of relatively high vorticity that are at best described as ‘TC-like’, but metrics such as storm counts are highly dependent on tracking algorithms (Camargo, 2013; Wehner et al., 2015; Zarzycki and Ullrich, 2017; Roberts et al., 2020a). High-resolution GCMs (around 10–60 km grid spacing), as used in HighResMIP (Haarsma et al., 2016; Roberts et al., 2020a), begin to capture some structures of TCs more realistically, as well as produce intense TCs of Category 4–5 despite the effects of parametrized deep cumulus convection processes (Murakami et al., 2015; Wehner et al., 2015; Yamada et al., 2017; Roberts et al., 2018; Moon et al., 2020). Convection-permitting models (around 1–10 km grid-spacing), such as used in some dynamical downscaling studies, provide further realism with capturing TC eye-wall structures (Tsuboki et al., 2015). Model characteristics besides resolution, especially details of convective parametrization, can influence a model’s ability to simulate intense TCs (Reed and Jablonowski, 2011; Zhao et al., 2012; He and Posselt, 2015; D. Kim et al., 2018; Zhang and Wang, 2018; Camargo et al., 2020). However, models’ dynamical cores and other physics also affect simulated TC properties (Reed et al., 2015; Vidale et al., 2021). Both wide-area regional and global convection-permitting models without the need for parameterized convection are becoming more useful for TC regional model projection studies (Tsuboki et al., 2015; Kanada et al., 2017a; Gutmann et al., 2018) and global model projection studies (Satoh et al., 2015, 2017; Yamada et al., 2017), as they capture more realistic TC eye wall structures (Kinter III et al., 2013) and are becoming more useful for investigating changes in TC structures (Kanada et al., 2013; Yamada et al., 2017). Large ensemble simulations of GCMs with 60 km grid spacing provide TC statistics that allow more reliable detection of changes in the projections, which are not well captured in any single experiment (Yoshida et al., 2017; Yamaguchi et al., 2020). Variable resolution global models offer an alternative to regional models for individual TC or basin-wide simulations (Yanase et al., 2012; Zarzycki et al., 2014; Harris et al., 2016; Reed et al., 2020; Stansfield et al., 2020). Computationally less intense than equivalent uniform resolution global models, they also do not require lateral boundary conditions, thus reducing this source of error (Hashimoto et al., 2016). Confidence in the projection of TC statistics and properties is increased by the use of higher-resolution models with more realistic simulations.

Operational forecasting models also reproduce TCs, and their use for climate projection studies shows promise. However, there is limited application for future projections as they are specifically developed for operational purposes, and TC climatology is not necessarily well evaluated. Intercomparison of operational models indicates that enhancement of horizontal resolution can provide more credible projections of TCs (Nakano et al., 2017). Likewise, high-resolution climate models show promise as TC forecast tools (Zarzycki and Jablonowski, 2015; Reed et al., 2020), further narrowing the continuum of weather and climate models, and increasing confidence in projections of future TC behaviour. However, higher horizontal resolution does not necessarily lead to an improved TC climatology (Camargo et al., 2020).

Atmosphere–ocean interaction is an important process in TC evolution. Atmosphere–ocean coupled models are generally better than atmosphere-only models at capturing realistic processes related to TCs (Murakami et al., 2015; Ogata et al., 2015, 2016; Zarzycki, 2016; Kanada et al., 2017b; Scoccimarro et al., 2017). However, the basin-scale SST biases commonly found in atmosphere–ocean models can introduce substantial errors in the simulated TC number (Hsu et al., 2019). Higher-resolution ocean models improve the simulation of TCs by reducing the SST climatology bias (Li and Sriver, 2018; Roberts et al., 2020a). Coarse resolution atmospheric models may degrade coupled model performance as well. For example, in a case study of Hurricane Harvey, Trenberth et al. (2018) suggested that the lack of realistic hurricane frequency and intensity within coupled climate models hampers the models’ ability to simulate SST and ocean heat content and their changes.

Even with higher-resolution atmosphere–ocean coupled models, TC projection studies still rely on assumptions in experimental design that introduce uncertainties. Computational constraints often limit the number of simulations, resulting in relatively small ensemble sizes and incomplete analyses of possible future SST magnitude and pattern changes (Zhao and Held, 2011; Knutson et al., 2013). Uncertainties in aerosol forcing also are reflected in TC projection uncertainty (Wang et al., 2014).

Regional climate models (RCM) with grid spacing around 15–50 km can be used to study the projection of TCs. RCMs are run with lateral and surface boundary conditions, which are specified by the atmospheric state and SSTs simulated by GCMs. Various combinations of the lateral and surface boundary conditions can be chosen for RCM studies, and uncertainties in the projection can be further examined in general. They are used for studying changes in TC characteristics in a specific area, such as Vietnam (Redmond et al., 2015) and the Philippines (Gallo et al., 2019).

Less computationally expensive downscaling approaches that allow larger ensembles and long-term studies are also used in the projection of TCs (Emanuel et al., 2006; C.Y. Lee et al., 2018). A statistical–dynamical TC downscaling method requires assumptions of the rate of seeding of random initial disturbances, which are generally assumed to not change with climate change (Emanuel et al., 2008; Emanuel, 2013). The results with the downscaling approach might depend on the assumptions, which are required for the simplification of the methods.

In summary, various types of models are useful to study how TCs change in response to climate changes, and there is no unique solution for choosing a model type. However, higher-resolution models generally capture TC properties more realistically (high confidence). In particular, models with horizontal resolutions of 10–60 km are capable of reproducing strong TCs with Category 4–5 and those of 1–10 km are capable of the eye wall structure of TCs. Uncertainties in TC simulations come from details of the model configuration of both dynamical and physical processes. Models with realistic atmosphere–ocean interactions are generally better than atmosphere-only models at reproducing realistic TC evolutions (high confidence).

11.7.1.4 Detection and Attribution, Event Attribution

There is general agreement in the literature that anthropogenic greenhouse gases and aerosols have measurably affected observed oceanic and atmospheric variability in TC-prone regions (see Chapter 3). This underpinned the SROCC assessment of medium confidence that humans have contributed to the observed increase in Atlantic hurricane activity since the 1970s (Chapter 5, Bindoff et al., 2013). Literature subsequent to AR5 lends further support to this statement (Knutson et al., 2019). However, there is still no consensus on the relative magnitude of human and natural influences on past changes in Atlantic hurricane activity, and particularly on which factor has dominated the observed increase (Ting et al., 2015) and it remains uncertain whether past changes in Atlantic TC activity are outside the range of natural variability. A recent result using high-resolution dynamical model experiments suggested that the observed spatial contrast in TC trends cannot be explained only by multi-decadal natural variability, and that external forcing plays an important role (Murakami et al., 2020).Observational evidence for significant global increases in the proportion of major TC intensities (Kossin et al., 2020) is consistent with both theory and numerical modelling simulations, which generally indicate an increase in mean TC peak intensity and the proportion of very intense TCs in a warming world (Knutson et al., 2015, 2020; Walsh et al., 2015, 2016). In addition, high-resolution coupled model simulations provide support that natural variability alone is unlikely to explain the magnitude of the observed increase in TC intensification rates and upward TC intensity trend in the Atlantic basin since the early 1980s (Bhatia et al., 2019; Murakami et al., 2020).

The cause of the observed slowdown in TC translation speed is not yet clear. Yamaguchi et al. (2020) used large ensemble simulations to argue that part of the slowdown is due to actual latitudinal shifts of TC tracks, rather than data artefacts, in addition to atmospheric circulation changes. G. Zhang et al. (2020) used large ensemble simulations to show that anthropogenic forcing can lead to a robust slowdown, particularly outside of the tropics at higher latitudes. Yamaguchi and Maeda (2020b) found a significant slowdown in the western North Pacific over the past 40 years and attributed the slowdown to a combination of natural variability and global warming. The slowing trend since 1900 over the USA is robust and significant after removing multi-decadal variability from the time series (Kossin, 2019). Among the hypotheses discussed is the physical linkage between warming and slowing circulation (Held and Soden, 2006; see also Section 8.2.2.2), with expectations of Arctic amplification and weakening circulation patterns through weakening meridional temperature gradients (Coumou et al., 2018; see also Cross-Chapter Box 10.1), or through changes in planetary wave dynamics (Mann et al., 2017). The tropics expansion and the poleward shift of the mid-latitude westerlies associated with warming is also suggested as the reason of the slowdown (G. Zhang et al., 2020). However, the connection of these mechanisms to the slowdown has not been robustly shown. Furthermore, slowing trends have not been unambiguously observed in circulation patterns that steer TCs, such as the Walker and Hadley circulations (Section 2.3.1.4), although these circulations generally slow down in numerical simulations under global warming (Sections 4.5.1.6 and 8.4.2.2).

The observed poleward trend in western North Pacific TCs remains significant after accounting for the known modes of dominant interannual to decadal variability in the region (Kossin et al., 2016a), and is also found in CMIP5 model-simulated TCs (in the recent historical period 1980–2005), although it is weaker than observed and is not statistically significant (Kossin et al., 2016a). However, the trend is significant in 21st-century CMIP5 projections under the RCP8.5 scenario, with a similar spatial pattern and magnitude to the past observed changes in that basin over the period 1945–2016, supporting a possible anthropogenic greenhouse gas contribution to the observed trends (Kossin et al., 2016a; Knutson et al., 2019).

The recent active TC seasons in some basins have been studied to determine whether there is anthropogenic influence. For 2015, Murakami et al. (2017b) explored the unusually high TC frequency near Hawaii and in the eastern Pacific basin. W. Zhang et al. (2016b) considered unusually high Accumulated Cyclone Energy (ACE) in the western North Pacific; and S.-H. Yang et al. (2018) and Yamada et al. (2019) looked at TC intensification in the western North Pacific. These studies suggest that the anomalous TC activity in 2015 was not solely explained by the effect of an extreme El Niño (see Box 11.4) and that there was also an anthropogenic contribution, mainly through the effects of SSTs in subtropical regions. In the post-monsoon seasons of 2014 and 2015, tropical storms with lifetime maximum winds greater than 46 m s−1 were first observed over the Arabian Sea, and Murakami et al. (2017a) showed that the probability of late-season severe tropical storms is increased by anthropogenic forcing compared to the preindustrial era. Murakami et al. (2018) concluded that the active 2017 Atlantic hurricane season was mainly caused by pronounced SSTs in the tropical North Atlantic and that these types of seasonal events will intensify with projected anthropogenic forcing. The trans-basin SST change, which might be driven by anthropogenic aerosol forcing, also affects TC activity. Takahashi et al. (2017) suggested that a decrease in sulphate aerosol emissions caused about half of the observed decreasing trends in TC genesis frequency in the south-eastern region of the western North Pacific during 1992–2011.

Event attribution is used in TC case studies to test whether the severities of recent intense TCs are explained without anthropogenic effects. In a case study of Hurricane Sandy (2012), Lackmann (2015) found no statistically significant impact of anthropogenic climate change on storm intensity, while projections in a warmer world showed significant strengthening. However, Magnusson et al. (2014) found that, in European Centre for Medium-Range Weather Forecast (ECMWF) simulations, the simulated cyclone depth and intensity, as well as precipitation, were larger when the model was driven by the warmer actual SSTs than the climatological average SSTs. In Super Typhoon Haiyan, which struck the Philippines on 8 November 2013, Takayabu et al. (2015) took an event attribution approach with cloud system-resolving (around 1 km) downscaling ensemble experiments to evaluate the anthropogenic effect on typhoons, and showed that the intensity of the simulated worst-case storm in the actual conditions was stronger than that in a hypothetical condition without historical anthropogenic forcing in the model. However, in a similar approach with two coarser parametrized convection models, Wehner et al. (2019) found conflicting human influences on Haiyan’s intensity. Patricola and Wehner (2018) found little evidence of an attributable change in intensity of hurricanes Katrina (2005), Irma (2017), and Maria (2017) using a regional climate model configured between 3 km and 4.5 km resolution. They did, however, find attributable increases in heavy precipitation totals. These results imply that higher resolution, such as in a convective permitting 5 km or less mesh model, is required to obtain a robust anthropogenic intensification of a strong TC by simulating realistic rapid intensification (Kanada and Wada, 2016; Kanada et al., 2017a), and that whether the TC intensification can be attributed to the recent warming depends on the case.

The dominant factor in the extreme rainfall amounts during Hurricane Harvey’s passage onto the USA in 2017 was its slow translation speed. But studies published after the event have argued that anthropogenic climate change contributed to an increase in rain rate, which compounded the extreme local rainfall caused by the slow translation. Emanuel (2017) used a large set of synthetically-generated storms and concluded that the occurrence of extreme rainfall as observed in Harvey was substantially enhanced by anthropogenic changes to the larger-scale ocean and atmosphere characteristics; Trenberth et al. (2018) linked Harvey’s rainfall totals to the anomalously large ocean heat content from the Gulf of Mexico; and van Oldenborgh et al. (2017) and Risser and Wehner (2017) applied extreme value analysis to extreme rainfall records in the Houston, Texas region, both attributing large increases to climate change. Large precipitation increases during Harvey due to global warming were also found using climate models (van Oldenborgh et al., 2017; S.-Y.S. Wang et al., 2018). Harvey precipitation totals were estimated in these papers to be three to 10 times more probable due to climate change. A best estimate from a regional climate and flood model is that urbanization increased the risk of the Harvey flooding by a factor of 21 (W. Zhang et al., 2018), using a regional climate and flood model, found that surface roughness from urbanization increased the risk of the Harvey flooding by a factor of 21. Anthropogenic effects on precipitation increases were also predicted in advance from a forecast model for Hurricane Florence in 2018 (Reed et al., 2020).

In summary, it is very likely that the recent active TC seasons in the North Atlantic, the North Pacific, and Arabian basins cannot be explained without an anthropogenic influence. The anthropogenic influence on these changes is principally associated to aerosol forcing, with stronger contributions to the response in the North Atlantic. It is more likely than not that the slowdown of TC translation speed over the USA has contributions from anthropogenic forcing. It is likely that the poleward migration of TCs in the western North Pacific and the global increase in TC intensity rates cannot be explained entirely by natural variability. Event attribution studies of specific strong TCs provide limited evidence for anthropogenic effects on TC intensifications so far, but high confidence for increases in TC heavy precipitation. There is high confidence that anthropogenic climate change contributed to extreme rainfall amounts during Hurricane Harvey (2017) and other intense TCs.

11.7.1.5 Projections

A summary of studies on TC projections for the late 21st century, particularly studies since AR5, is given by Knutson et al. (2020), which is an assessment report mandated by the World Meteorological Organization (WMO). Studies subsequent to Knutson et al. (2020) are generally consistent, and the confidence assessments here closely follow theirs (Cha et al., 2020), although there are some differences due to the varying confidence calibrations between the IPCC and WMO reports.

There is not an established theory for the drivers of future changes in the frequency of TCs. Most, but not all, high-resolution global simulations project significant reductions in the total number of TCs, with the bulk of the reduction at the weaker end of the intensity spectrum as the climate warms (Knutson et al., 2020). Recent exceptions based on high-resolution coupled model results arenoted in Bhatia et al. (2018) and Vecchi et al. (2019). Vecchi et al. (2019) showed that the representation of synoptic-scale seeds for TC genesis in their high-resolution model causes different projections of global TC frequency, and there is evidence for a decrease in cyclone seeds in some projected TCsimulations (Sugi et al., 2020; Yamada et al., 2011). However, other research indicates that TC seeds are not an independent control on climatological TC frequency, rather the seeds covary with the large-scale controls on TCs (Patricola et al., 2018). While empirical genesis indices derived from observations and reanalysis describe well the observed subseasonal and interannual variability of current TC frequency (Camargo et al., 2007, 2009; Tippett et al., 2011; Menkes et al., 2012), they fail to predict the decreased TC frequency found in most high-resolution model simulations (Zhang et al., 2010; Camargo, 2013; Wehner et al., 2015), as they generally project an increase as the climate warms. This suggests a limitation of the use of the empirical genesis indices for projections of TC genesis, in particular due to their sensitivity to the humidity variable considered in the genesis index for these projections (Camargo et al., 2014). In a different approach, a statistical–dynamical downscaling framework assuming a constant seeding rate with warming (Emanuel, 2013, 2021) exhibits increases in TC frequency consistent with genesis indices-based projections, while downscaling with a different model leads to two different scenarios depending on the humidity variable considered (C.-Y. Lee et al., 2020). This disparity in the sign of the projected change in global TC frequency, and the difficulty in explaining the mechanisms behind the different signed responses, further emphasize the lack of process understanding of future changes in tropical cyclogenesis (Walsh et al., 2015; Hoogewind et al., 2020). Even within a single model, uncertainty in the pattern of future SST changes leads to large uncertainties (including the sign) in the projected change in TC frequency in individual ocean basins, although global TCs would appear to be less sensitive (Yoshida et al., 2017; Bacmeister et al., 2018).

Changes in SST and atmospheric temperature and moisture play a role in tropical cyclogenesis (Walsh et al., 2015). Reductions in vertical convective mass flux due to increased tropical stability have been associated with a reduction in cyclogenesis (Held and Zhao, 2011; Sugi et al., 2012). Satoh et al. (2015) further posit that the robust simulated increase in the number of intense TCs, and hence increased vertical mass flux associated with intense TCs, must lead to a decrease in overall TC frequency because of this association. The Genesis Potential Index can be modified to mimic the TC frequency decreases of a model by altering the treatment of humidity (Camargo et al., 2014). This supports the idea that increased mid-tropospheric saturation deficit (Emanuel et al., 2008) controls TC frequency, but the approach remains empirical. Other possible controlling factors, such as a decline in the number of seeds (held constant in Emanuel’s downscaling approach, or dependent on the genesis index formulation in the approach proposed by C.-Y. Lee et al., 2020) caused by increased atmospheric stability have been proposed, but questioned as an important factor (Patricola et al., 2018). The resolution of atmospheric models affects the number of seeds, hence TC genesis frequency (Vecchi et al., 2019; Sugi et al., 2020; Yamada et al., 2021). The diverse and sometimes inconsistent projected changes in global TC frequency by high-resolution models indicate that better process understanding and improvement of the models are needed to raise confidence in these changes.

Most TC-permitting model simulations (10–60 km or finer grid spacing) are consistent in their projection of increases in the proportion of intense TCs (Category 4–5), as well as an increase in the intensity of the strongest TCs defined by maximum wind speed or central pressure fall (Murakami et al., 2012; Tsuboki et al., 2015; Wehner et al., 2018a; Knutson et al., 2020). The general reduction in the total number of TCs, which is concentrated in storms weaker than or equal to Category 1, contributes to this increase. The models are somewhat less consistent in projecting an increase in the frequency of Category 4–5TCs (Wehner et al., 2018a; Knutson et al., 2020). The projected increase in the intensity of the strongest TCs is consistent with theoretical understanding (e.g., Emanuel, 1987) and observations (e.g., Kossin et al., 2020). For a 2°C global warming, the median proportion of Category 4–5 TCs increases by 13%, while the median global TC frequency decreases by 14%, which implies that the median of the global Category 4–5 TC frequency is slightly reduced by 1% or almost unchanged (Knutson et al., 2020). Murakami et al. (2020) projected a decrease in TC frequency over the coming century in the North Atlantic due to greenhouse warming, as consistent with Dunstone et al. (2013), and a reduction in TC frequency almost everywhere in the tropics in response to +1% CO2 forcing. Exceptions include the central North Pacific (Hawaii region), east of the Philippines in the North Pacific, and two relatively small regions in the northern Arabian Sea and Bay of Bengal. These projections can vary substantially between ocean basins, possibly due to differences in regional SST warming and warming patterns (Sugi et al., 2017; Yoshida et al., 2017; Bacmeister et al., 2018). A summary of projections of TC characteristics is schematically shown by Figure 11.20.

The increase in global TC maximum surface wind speeds is about 5% for a 2°C global warming across a number of high-resolution multi-decadal studies (Knutson et al., 2020). This indicates the deepening in global TC minimum surface pressure under the global warming conditions. A regional cloud-permitting model study shows that the strongest TC in the western North Pacific can be as strong as 857 hPa in minimum surface pressure with a wind speed of 88 m s–1under warming conditions in 2074–2087 (Tsuboki et al., 2015). TCs are also measured by quantities such as ACE and the power dissipation index (PDI), which conflate TC intensity, frequency, and duration (Murakami et al., 2014). Several TC modelling studies (Yamada et al., 2010; H.S. Kim et al., 2014; Knutson et al., 2015) project little change or decreases in the globally accumulated value of PDI or ACE, which is due to the decrease in the total number of TCs.

A projected increase in global average TC rain rates of about 12% for a 2°C global warming is consistent with the Clausius–Clapeyron scaling of saturation-specific humidity (Knutson et al., 2020). Increases substantially greater than Clausius–Clapeyron scaling are projected in some regions, which is caused by increased low-level moisture convergence due to projected TC intensity increases in those regions (Knutson et al., 2015; Phibbs and Toumi, 2016; Patricola and Wehner, 2018; M. Liu et al., 2019a). Projections of TC precipitation using large-ensemble experiments (Kitoh and Endo, 2019) show that the annual maximum one-day precipitation total is projected to increase, except for the western North Pacific where only a small change (or even a reduction) is projected, mainly due to a projected decrease of TC frequency. They also show that the 10-year return value of extreme Rx1day associated with TCs will greatly increase in a region extending from Hawaii to the south of Japan. TC tracks and the location of topography relative to TCs significantly affect precipitation, thus, in general, areas on the eastern and southern faces of mountains have more impacts of TC precipitation changes (Hatsuzuka et al., 2020). Projection studies using variable-resolution models in the North Atlantic (Stansfield et al., 2020) indicate that TC-related precipitation rates within North Atlantic TCs and the amount of hourly precipitation due to TC are projected to increase by the end of the century compared to a historical simulation. However, the annual average TC-related Rx5day over the eastern USA is projected to decrease because of a reduction in landfalling TCs. RCM studies with around 25–50 km grid spacing are used to study projected changes in TCs. The projected changes of TCs in South East Asia simulated by RCMs are consistent with those of most GCMs, showing a decrease in TC frequency and an increase in the amount of TC-associated precipitation or an increase in the frequency of intense TCs (Redmond et al., 2015; Gallo et al., 2019).

Projected changes in TC tracks or TC areas of occurrence in the late 21st century vary considerably among available studies, although there is better agreement in the western North Pacific. Several studies project either poleward or eastward expansion of TC occurrence over the western North Pacific region, and more TC occurrence in the central North Pacific (Yamada et al., 2017; Yoshida et al., 2017; Wehner et al., 2018a; Roberts et al., 2020b). The observed poleward expansion of the latitude of maximum TC intensity in the western North Pacific is consistently reproduced by the CMIP5 models and downscaled models, and these models show further poleward expansion in the future; the projected mean migration rate of the mean latitude where TCs reach their lifetime-maximum intensity is 0.2±0.1° from CMIP5 model results, while it is 0.13±0.04° from downscaled models in the western North Pacific (Kossin et al., 2014, 2016a). In the North Atlantic, while the location of TC maximum intensity does not show clear poleward migration observationally (Kossin et al., 2014), it tends to migrate poleward in projections (Garner et al., 2017). The poleward migration is less robust among models and observations in the Indian Ocean, eastern North Pacific, and South Pacific (e.g., Tauvale and Tsuboki, 2019; Ramsay et al. 2018; Cattiaux et al. 2020). There is presently no clear consensus in projected changes in TC translation speed (Knutson et al., 2020), although recent studies suggest a slowdown outside of the tropics (Kossin, 2019; Yamaguchi et al., 2020; G. Zhang et al., 2020), but regionally there can even be an acceleration of the storms (Hassanzadeh et al., 2020).

The spatial extent, or ‘size’, of the TC wind field is an important determinant of storm surge and damage. No detectable anthropogenic influences on TC size have been identified to date, because TCs in observations vary in size substantially (Chan and Chan, 2015) and there is no definite theory on what controls TC size, although this is an area of active research (Chavas and Emanuel, 2014; Chan and Chan, 2018). However, projections by high-resolution models indicate future broadening of TC wind fields when compared to TCs of the same categories (Yamada et al., 2017), while Knutson et al. (2015) simulate a reasonable interbasin distribution of TC size climatology, but project no statistically significant change in global average TC size. A plausible mechanism is that, as the tropopause height becomes higher with global warming, the eye wall areas become wider because the eye walls are inclined outward with height to the tropopause. This effect is only reproduced in high-resolution convection-permitting models capturing eye walls, and such modelling studies are not common. Moreover, the projected TC size changes are generally on the order of 10% or less, and these size changes are still highly variable between basins and studies. Thus, the projected change in both magnitude and sign of TC size is uncertain.

The coastal effects of TCs depend on TC intensity, size, track, and translation speed. Projected increases in sea level, average TC intensity, and TC rainfall rates each generally act to further elevate future storm surge and fresh-water flooding (see Section 9.6.4.2). Changes in TC frequency could contribute toward increasing or decreasing future storm surge risk, depending on the net effects of changes in weaker vs stronger storms. Several studies (McInnes et al., 2014, 2016; Little et al., 2015; Garner et al., 2017; Timmermans et al., 2017, 2018) have explored future projections of storm surge in the context of anthropogenic climate change with the influence of both sea level rise and future TC changes. Garner et al. (2017) investigated the near-future changes in the New York City coastal flood hazard, and suggested a small change in storm-surge height because effects of TC intensification are compensated by the offshore shifts in TC tracks, but concluded that the overall effect due to the rising sea levels would increase the flood hazard. Future projection studies of storm surge in East Asia, including China, Japan and Korea, also indicate that storm surges due to TCs become more severe (J.A. Yang et al., 2018; Mori et al., 2019, 2021; J. Chen et al., 2020b). For the Pacific Islands, McInnes et al. (2014) found that the future projected increase in storm surge in Fiji is dominated by sea level rise, and projected TC changes make only a minor contribution. Among various storm surge factors, there is high confidence that sea level rise will lead to a higher possibility of extreme coastal water levels in most regions, with all other factors assumed equal.

In the North Atlantic, vertical wind shear, which inhibits TC genesis and intensification, varies in a quasi-dipole pattern, with one centre of action in the tropics and another along the south-east USA coast (Vimont and Kossin, 2007). This pattern of variability creates a protective barrier of high shear along the USA coast during periods of heightened TC activity in the tropics (Kossin, 2017), and appears to be a natural part of the Atlantic ocean–atmosphere climate system (Ting et al., 2019). Greenhouse gas forcing in CMIP5 and the Community Earth System Model Large Ensemble (Kay et al., 2015) simulations, however, erodes the pattern and degrades the natural shear barrier along the USA coast. Following the RCP8.5 emissions scenario, the magnitude of the erosion of the barrier equals the amplitude of past natural variability (time of emergence) by the mid-21st century (Ting et al., 2019). The projected reduction of shear along the USA East Coast with warming is consistent among studies (e.g., Vecchi and Soden, 2007).

In summary, average peak TC wind speeds and the proportion of Category 4–5 TCs will very likely increase globally with warming. It is likely that the frequency of Category 4–5 TCs will increase in limited regions over the western North Pacific. It is very likely that average TC rain rates will increase with warming, and likely that the peak rain rates will increase at rate greater than the Clausius–Clapeyron scaling rate of 7% per 1°C of warming in some regions due to increased low-level moisture convergence caused by regional increases in TC wind intensity. It is likely that the average location where TCs reach their peak wind intensity will migrate poleward in the western North Pacific Ocean as the tropics expand with warming, and that the global frequency of TCs over all categories will decrease or remain unchanged.

11.7.2 Extratropical Storms

This section focuses on extratropical cyclones (ETCs) that are either classified as strong or extreme by using some measure of their intensity, or by being associated with the occurrence of extremes in variables such as precipitation or near-surface wind speed (Seneviratne et al., 2012). Since AR5, the high relevance of ETCs for extreme precipitation events has been well established (Pfahl and Wernli, 2012; Catto and Pfahl, 2013; Utsumi et al., 2017), with 80% or more of hourly and daily precipitation extremes being associated with either ETCs or fronts over oceanic mid-latitude regions, and somewhat smaller, but still very large, proportions of events over mid-latitude land regions (Utsumi et al., 2017). The emphasis in this section is on individual ETCs that have been identified using some detection and tracking algorithms. Mid-latitude atmospheric rivers are assessed in Section 8.3.2.8.

11.7.2.1 Observed Trends

Section 2.3.1.4.3 concluded that there is overall low confidence in recent changes in the total number of ETCs over both hemispheres, and that there is medium confidence in a poleward shift of the storm tracks over both hemispheres since the 1980s. Overall, there is also low confidence in past-century trends in the number and intensity of the strongest ETCs due to the large interannual and decadal variability (Feser et al., 2015; Reboita et al., 2015; Wang et al., 2016; Varino et al., 2019) and due to temporal and spatial heterogeneities in the number and type of assimilated data in reanalyses, particularly before the satellite era (Krueger et al., 2013; Tilinina et al., 2013; Befort et al., 2016; Chang and Yau, 2016; Wang et al., 2016). There is medium confidence that the agreement among reanalyses and detection and tracking algorithms is higher when considering stronger cyclones (Neu et al., 2013; Pepler et al., 2015; Wang et al., 2016). Over the Southern Hemisphere, there is high confidence that the total number of ETCs with low central pressures (<980 hPa) has increased between 1979 and 2009, with all eight reanalyses considered by Wang et al. (2016) showing positive trends, and five of them showing statistically significant trends. Similar results were found by Reboita et al. (2015) using a different detection and tracking algorithm and a single reanalysis product. Over the Northern Hemisphere, there is high agreement among reanalyses that the number of cyclones with low central pressures (<970 hPa) has decreased in summer and winter during the period 1979–2010 (Tilinina et al., 2013; Chang et al., 2016). However, changes exhibit substantial decadal variability and do not show monotonic trends since the 1980s. For example, over the Arctic and North Atlantic, Tilinina et al. (2013) showed that the number of cyclones with very low central pressure (<960 hPa) increased from 1979 to 1990 and then declined until 2010 in all five reanalyses considered. Over the North Pacific, the number of cyclones with very low central pressure reached a peak around 2000 and then decreased until 2010 in the five reanalyses considered (Tilinina et al., 2013). Overall, however, it should be noted that characterising trends in the dynamical intensity of ETCs (e.g., wind speeds) using the absolute central pressure is problematic because the central pressure depends on the background mean sea level pressure, which varies seasonally and regionally (e.g., Befort et al., 2016).

11.7.2.2 Model Evaluation

There is high confidence that coarse-resolution climate models (e.g., CMIP5 and CMIP6) underestimate the dynamical intensity of ETCs, including the strongest ETCs, as measured using a variety of metrics, including mean pressure gradient, mean vorticity and near-surface wind speeds, over most regions (Colle et al., 2013; Zappa et al., 2013a; Govekar et al., 2014; Di Luca et al., 2016; Trzeciak et al., 2016; Seiler et al., 2018; Priestley et al., 2020). There is also high confidence that most current climate models underestimate the number of explosive systems (i.e., systems showing a decrease in mean sea level pressure of at least 24 hPa in 24 hours) over both hemispheres (Seiler and Zwiers, 2016a; Gao et al., 2020;Priestley et al., 2020). There is high confidence that the underestimation of the intensity of ETCs is associated with the coarse horizontal resolution of climate models, with higher horizontal resolution models, including HighResMIP and CORDEX, usually showing better performance (Colle et al., 2013; Zappa et al., 2013a; Di Luca et al., 2016; Trzeciak et al., 2016; Seiler et al., 2018; Gao et al., 2020; Priestley et al., 2020). The improvement by higher-resolution models is found, even when comparing models and reanalyses after post-processing data to a common resolution (Zappa et al., 2013a; Di Luca et al., 2016; Priestley et al., 2020). The systematic bias in the intensity of ETCs has also been linked to the inability of current climate models to resolve diabatic processes, particularly those related to the release of latent heat (Willison et al., 2013; Trzeciak et al., 2016) and the formation of clouds (Govekar et al., 2014). There is medium confidence that climate models simulate well the spatial distribution of precipitation associated with ETCs over the Northern Hemisphere, together with some of the main features of the ETC life cycle, including the maximum in precipitation occurring just before the peak in dynamical intensity (e.g., vorticity) as observed in a reanalysis and observations (Hawcroft et al., 2018). There is, however, large observational uncertainty in ETC-associated precipitation (Hawcroft et al., 2018) and limitations in the simulation of frontal precipitation, including overly low rainfall intensity over mid-latitude oceanic areas in both hemispheres (Catto et al., 2015).

11.7.2.3 Detection and Attribution, Event Attribution

(Section 3.3.3.3 concluded that there is low confidence in the attribution of observed changes in the number of ETCs in the Northern Hemisphere and high confidence that the poleward shift of storm tracks in the Southern Hemisphere is linked to human activity, mostly due to emissions of ozone-depleting substances. Specific studies attributing changes in the most extreme ETCs are not available. The human influence on individual extreme ETC events has been considered only a few times and there is overall low confidence in the attribution of these changes (NASEM, 2016; Vautard et al., 2019).

11.7.2.4 Projections

The frequency of ETCs is expected to change, primarily following a poleward shift of the storm tracks as discussed in Section 4.5.1.6 (see also Figure 4.31) and Section 8.4.2.8. There is medium confidence that changes in the dynamical intensity (e.g., wind speeds) of ETCs will be small, although changes in the location of storm tracks can lead to substantial changes in local extreme wind speeds (Zappa et al., 2013b; Chang, 2014; Li et al., 2014; Seiler and Zwiers, 2016b; Yettella and Kay, 2017; Barcikowska et al., 2018; Kar-Man Chang, 2018). Yettella and Kay (2017) detected and tracked ETCs over both hemispheres in an ensemble of 30 Community Earth System Model Large Ensemble simulations, differing only in their initial conditions, and found that changes in mean wind speeds around ETC centres are often negligible between present (1986–2005) and future (2081–2100) periods. Using 19 CMIP5 models, Zappa et al. (2013b) found an overall reduction in the number of cyclones associated with low-troposphere (850-hPa) wind speeds larger than 25 m s–1 over the North Atlantic and Europe with the number of the 10% strongest cyclones decreasing by about 8% and 6% in December–January–February and June–July–August according to the RCP4.5 scenario (2070–2099 vs. 1976–2005). Over the North Pacific, Chang (2014) showed that CMIP5 models project a decrease in the frequency of ETCs, with the largest central pressure perturbation (i.e., the depth, strongly related with low-level wind speeds) by the end of the century according to simulations using the RCP8.5 scenario. Using projections from CMIP5 GCMs under the RCP8.5 scenario (1981–2000 to 2081–2100), Seiler and Zwiers (2016b) projected a northward shift in the number of explosive ETCs in the northern Pacific, with fewer and weaker events south, and more frequent and stronger events north of 45°N. Using 19 CMIP5 GCMs under the RCP8.5 scenario, Kar-Man Chang (2018) found a significant decrease in the number of ETCs associated with extreme wind speeds (2081–2100 vs. 1980–99) over the Northern Hemisphere (average decrease of 17%) and over some smaller regions, including the Pacific and Atlantic regions.

Over the Southern Hemisphere, future changes (RCP8.5 scenario; 1980–1999 to 2081–2100) in extreme ETCs were studied by Chang (2017) using 26 CMIP5 models, and a variety of intensity metrics (850-hPa vorticity, 850-hPa wind speed, mean sea level pressure and near-surface wind speed). They found that the number of extreme cyclones is projected to increase by at least 20% and as much as 50%, depending on the specific metric used to define extreme ETCs. Increases in the number of strong cyclones appear to be robust across models and for most seasons, although they show strong regional variations, with increases occurring mostly over the southern flank of the storm track, consistent with a shift and intensification of the storm track. Overall, there is medium confidence that projected changes in the dynamical intensity of ETCs depend on the resolution and formulation (e.g., explicit or implicit representation of convection) of climate models (Booth et al., 2013; Michaelis et al., 2017; Zhang and Colle, 2017).

As reported in AR5 and in Section 8.4.2.8, despite small changes in the dynamical intensity of ETCs, there is high confidence that the precipitation associated with ETCs will increase in the future (Zappa et al., 2013b; Marciano et al., 2015; Pepler et al., 2016; Michaelis et al., 2017; Yettella and Kay, 2017; Zhang and Colle, 2017; Barcikowska et al., 2018; Hawcroft et al., 2018; Zarzycki, 2018; Kodama et al., 2019; Bevacqua et al., 2020a; Reboita et al., 2021). There is high confidence that increases in precipitation will follow increases in low-level water vapour (i.e., about 7% per 1°C of surface warming; see Box 11.1) and will be larger for higher warming levels (Zhang and Colle, 2017). There is medium confidence that precipitation changes will show regional and seasonal differences due to distinct changes in atmospheric humidity and dynamical conditions (Zappa et al., 2015; Hawcroft et al., 2018), with decreases in some specific regions such as the Mediterranean (Zappa et al., 2015; Barcikowska et al., 2018). There is high confidence that snowfall associated with winter ETCs will decrease in the future, because increases in tropospheric temperatures lead to a lower proportion of precipitation falling as snow (O’Gorman, 2014; Rhoades et al., 2018; Zarzycki, 2018). However, there is medium confidence that extreme snowfall events associated with winter ETCs will change little in regions where snowfall will be supported in the future (O’Gorman, 2014; Zarzycki, 2018).

In summary, there is low confidence in past changes in the dynamical intensity (e.g., maximum wind speeds) of ETCs and medium confidence that, in the future, these changes will be small, although changes in the location of storm tracks could lead to substantial changes in local extreme wind speeds. There is high confidence that average and maximum ETC precipitation-rates will increase with warming, with the magnitude of the increases associated with increases in atmospheric water vapour. There is medium confidence that projected changes in the intensity of ETCs, including wind speeds and precipitation, depend on the resolution and formulation of climate models.

11.7.3 Severe Convective Storms

Severe convective storms are convective systems that are associated with extreme phenomena such as tornadoes, hail, heavy precipitation (rain or snow), strong winds, and lightning. The assessment of changes in severe convective storms in SREX (Chapter 3, Seneviratne et al., 2012) and AR5 (Chapter 12, Collins et al., 2013) is limited and focused mainly on tornadoes and hail storms. The SREX assessed that there is low confidence in observed trends in tornadoes and hail because of data inhomogeneities and inadequacies in monitoring systems. Subsequent literature assessed in the Climate Science Special Report (Kossin et al., 2017) led to the assessment of the observed tornado activity over the 2000s in the USA, with a decrease in the number of days per year with tornadoes and an increase in the number of tornadoes on these days (medium confidence). However, there is low confidence in past trends for hail and severe thunderstorm winds. Climate models consistently project environmental changes that would support an increase in the frequency and intensity of severe thunderstorms that combine tornadoes, hail, and winds (high confidence), but there is low confidence in the details of the projected increase. Regional aspects of severe convective storms and details of the assessment of tornadoes and hail are also assessed in Section 12.3.3.2 (tornadoes), Section 12.3.4.5 (hail), Section 12.4.5.3 (Europe), Section 12.4.6.3 (North America), and Section 12.7.2 (regional gaps and uncertainties).

11.7.3.1 Mechanisms and Drivers

Severe convective storms are sometimes embedded in synoptic-scale weather systems, such as TCs, ETCs, and fronts (Kunkel et al., 2013). They are also generated as individual events as mesoscale convective systems (MCSs) and mesoscale convective complexes (MCCs, a special type of a large, organized and long-lived MCS), without being clearly embedded within larger-scale weather systems. In addition to the general vigorousness of precipitation, hail, and winds associated with MCSs, characteristics of MCSs are viewed in new perspectives in recent years, probably because of both the development of dense mesoscale observing networks and advances in high-resolution mesoscale modelling (Sections 11.7.3.2 and 11.7.3.3). The horizontal scale of MCSs is discussed with their organization of the convective structure, and it is examined with a concept of ‘convective aggregation’ in recent years (Holloway et al., 2017). MCSs sometimes take a linear shape and stay almost stationary with successive production of cumulonimbus on the upstream side (back-building type convection), and cause heavy rainfall (Schumacher and Johnson, 2005). Many of the recent severe rainfall events in Japan are associated with band-shaped precipitation systems (Kunii et al., 2016; Oizumi et al., 2018; Tsuguti et al., 2018; Kato, 2020), suggesting common characteristics of severe precipitation, at least in East Asia. The convective modes of severe storms in the USA can be classified into rotating or linear modes and preferable environmental conditions for these modes, such as vertical shear, have been identified (Trapp et al., 2005; Smith et al., 2013; Allen, 2018). Cloud microphysics characteristics of MCSs were examined and the roles of warm rain processes on extreme precipitation were emphasized recently (Sohn et al., 2013; Hamada et al., 2015; Hamada and Takayabu, 2018). Idealized studies also suggest the importance of ice and mixed-phase processes of cloud microphysics on extreme precipitation (Sandvik et al., 2018; Bao and Sherwood, 2019). However, it is unknown whether the types of MCS are changing in recent periods or observed ubiquitously all over the world.

Severe convective storms occur under conditions preferable for deep convection, that is, conditionally unstable stratification, sufficient moisture, both in lower and middle levels of the atmosphere, and a strong vertical shear. These large-scale environmental conditions are viewed as necessary conditions for the occurrence of severe convective systems, or the resulting tornadoes and lightning, and the relevance of these factors strongly depends on the region (e.g., Antonescu et al., 2016a; Allen, 2018; Tochimoto and Niino, 2018). Frequently used metrics are atmospheric static stability, moisture content, convective available potential energy (CAPE) and convective inhibition, wind shear or helicity, including storm-relative environmental helicity (Tochimoto and Niino, 2018; Elsner et al., 2019). These metrics, largely controlled by large-scale atmospheric circulations or synoptic weather systems, such as TCs and ETCs, are then generally used to examine severe convective systems. In particular, there is high confidence that CAPE in the tropics and the subtropics increases in response to global warming (M.S. Singh et al., 2017), as supported by theoretical studies (Singh and O’Gorman, 2013; Seeley and Romps, 2015; Romps, 2016; Agard and Emanuel, 2017). The uncertainty, however, arises from the balance between factors affecting severe storm occurrence. For example, the warming of mid-tropospheric temperatures leads to an increase in the freezing level, which leads to increased melting of smaller hailstones, while there may be some offset by stronger updrafts driven by increasing CAPE, which would favour the growth of larger hailstones, leading to less melting when falling (Allen, 2018; Mahoney, 2020).

There are few studies on relations between changes in severe convective storms and those of the large-scale circulation patterns. Tornado outbreaks in the USA are usually associated with ETCs with their frontal systems and TCs (Fuhrmann et al., 2014; Tochimoto and Niino, 2016). In early June to late July in East Asia, associated with the Baiu/Changma/Mei-yu, severe precipitation events are frequently caused by MCSs. Severe precipitation events are also caused by remote effects of TCs, known as predecessor rain events (Galarneau et al., 2010). Atmospheric rivers and other coherent types of enhanced water vapour flux also have the potential to induce severe convective systems (Kamae et al., 2017a; Waliser and Guan, 2017; Ralph et al., 2018; see Section 8.3.2.8.2). Combined with the above drivers, topographic effects also enhance the intensity and duration of severe convective systems and the associated precipitation (Ducrocq et al., 2008; Piaget et al., 2015). However, the changes in these drivers are not generally significant, so their relations to severe convective storms are unclear.

In summary, severe convective storms are sometimes embedded in synoptic-scale weather systems, such as TCs, ETCs and fronts, and modulated by large-scale atmospheric circulation patterns. The occurrence of severe convective storms and the associated severe events, including tornadoes, hail, and lightning, is affected by environmental conditions of the atmosphere, such as CAPE and vertical shear. The uncertainty, however, arises from the balance between these environmental factors affecting severe storm occurrence.

11.7.3.2 Observed Trends

Observed trends in severe convective storms or MCSs are not well documented, but the climatology of MCSs has been analysed in specific regions (North America, South America, Europe, Asia; regional aspects of convective storms are separately assessed in Chapter 12). As the definition of severe convective storms varies depending on the literature, it is not straightforward to make a synthesizing view of observed trends in severe convective storms in different regions. However, analysis using satellite observations provides a global view of MCSs (Kossin et al., 2017). The global distribution of thunderstorms is captured (Zipser et al., 2006; Liu and Zipser, 2015) by using the satellite precipitation measurements by the Tropical Rainfall Measuring Mission (TRMM) and Global Precipitation Mission (GPM) (Hou et al., 2014). The climatological characteristics of MCSs are provided by satellite analyses in South America (Durkee and Mote, 2010; Rasmussen and Houze, 2011; Rehbein et al., 2018) and those of MCCs in the Maritime Continent by Trismidianto and Satyawardhana (2018). Analysis of the environmental conditions favourable for severe convective events indirectly indicates the climatology and trends of severe convective events (Allen et al., 2018; Taszarek et al., 2018, 2019), though favourable conditions depend on the location, such as the difference for tornadoes associated with ETCs between the USA and Japan (Tochimoto and Niino, 2018).

Observed trends in severe convective storms are highly regionally dependent. In the USA, it is indicated that there is no significant increase in convective storms, and hail and severe thunderstorms (Kunkel et al., 2013; Kossin et al., 2017). There is an upward trend in the frequency and intensity of extreme precipitation events in the USA (high confidence) (Kunkel et al., 2013; Easterling et al., 2017), and MCSs have increased in occurrence and precipitation amounts since 1979 (limited evidence) (Feng