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}

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}

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}

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}

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}

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}.

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}.

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}

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.

Open figure

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., 90^th 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.

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.

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).

Open figure

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.

Open figure

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.

Open figure

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.

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.

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.

Open figure

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).

Open figure

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).

Open figure

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 .

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.

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.

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.

Open figure

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).

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.

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).

Open figure

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).

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).

Open figure

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.

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).

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).

Open figure

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).

Open figure

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.

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).

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).

Open figure

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.

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).

Open figure

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).

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.

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).

Open figure

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).

Open figure

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).

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).

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.

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.

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.

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.

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).

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).

Open figure

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.

11.6.5.3 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).

Open figure

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.

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).