Chapter 8: Water Cycle Changes

Coordinating Lead Authors:

Hervé Douville (France), Krishnan Raghavan (India), James Renwick (New Zealand)

Lead Authors:

Richard P. Allan (United Kingdom), Paola A. Arias (Colombia), Mathew Barlow (United States of America), Ruth Cerezo-Mota (Mexico), Annalisa Cherchi (Italy), ThianY. Gan (Canada/Malaysia), Joëlle Gergis (Australia), Dabang Jiang (China), Asif Khan (Pakistan), Wilfried Pokam Mba (Cameroon), Daniel Rosenfeld (Israel), Jessica Tierney (United States of America), Olga Zolina (The Russian Federation/France, The Russian Federation)


Pascale Braconnot (France), Arona Diedhiou (Côte d’Ivoire/Senegal)


Gabriel Abramowitz (Australia), Guðfinna Adalgeirsdottir (Iceland), Andrea Alessandri (Italy), Robert J. Allen (United States of America), Kevin Anchukaitis (United States of America), Richard A. Betts (United Kingdom), Céline J. W. Bonfils (United States of America/ France), Michael Bosilovich (United States of America), Olivier Boucher (France), Josephine Brown (Australia), Michael P. Byrne (United Kingdom/Ireland), Robin Chadwick (United Kingdom), Sarah Connors (France/United Kingdom), Benjamin Cook (United States of America), Erika Coppola (Italy), Alejandro Di Luca (Australia, Canada/Argentina), Aïda Diongue Niang (Senegal), Petra Döll (Germany), Ellen Douglas (United States of America), Paul J. Durack (United States of America/Australia), Hayley J. Fowler (United Kingdom), Alexander Gershunov (United States of America), Nicholas R. Golledge (New Zealand/ United Kingdom), James Kossin (United States of America), Won-Tae Kwon (Republic of Korea), Flavio Lehner (United States of America/Switzerland), Eric Maloney (United States of America), Vimal Mishra (India), Angeline Pendergrass (United States of America), Stefan Pfahl (Germany), Catherine Prigent (France), Catherine Rio (France), Alex C. Ruane (United States of America), Benjamin Sanderson (United Kingdom), Sonia I. Seneviratne (Switzerland), Shoichi Shige (Japan), Vijay Singh (United States of America), Abigail Swann (United States of America), Richard G. Taylor (United Kingdom/Canada, United Kingdom), Laurent Terray (France), Natalia Tilinina (The Russian Federation), Bart van den Hurk (The Netherlands), Sergio M. Vicente-Serrano (Spain), Michael Wehner (United States of America), Laura J. Wilcox (United Kingdom), Cunde Xiao (China), Prodromos Zanis (Greece), Xuebin Zhang (Canada)

Chapter Scientists:

Stéphane Sénési (France), Sabin Thazhe Purayil (India)

Figure 8.1

Figure 8.2

Figure 8.3

Figure 8.4

Figure 8.5

Figure 8.6

Figure 8.7

Figure 8.8

Figure 8.9

Figure 8.10

Figure 8.11

Figure 8.12

Figure 8.13

Figure 8.14

Figure 8.15

Figure 8.16

Figure 8.17

Figure 8.18

Figure 8.19

Figure 8.20

Figure 8.21

Figure 8.22

Figure 8.23

Figure 8.24

Figure 8.25

Figure 8.26

Figure 8.27

Box 8.1 Figure 1

Box 8.1 Figure 2

Box 8.2 Figure 1

FAQ 8.1 Figure 1

FAQ 8.2, Figure 1

FAQ 8.3, Figure 1

This chapter should be cited as:

Douville, H., K. Raghavan, J. Renwick, R.P. Allan, P.A. Arias, M. Barlow, R. Cerezo-Mota, A. Cherchi, T.Y. Gan, J. Gergis, D. Jiang, A. Khan, W. Pokam Mba, D. Rosenfeld, J. Tierney, and O. Zolina, 2021: Water Cycle Changes. In Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [Masson-Delmotte, V., P. Zhai, A. Pirani, S.L. Connors, C. Péan, S. Berger, N. Caud, Y. Chen, L. Goldfarb, M.I. Gomis, M. Huang, K. Leitzell, E. Lonnoy, J.B.R. Matthews, T.K. Maycock, T. Waterfield, O. Yelekçi, R. Yu, and B. Zhou (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, pp. 1055–1210, doi: 10.1017/9781009157896.010.

Executive Summary

This chapter assesses multiple lines of evidence to evaluate past, present and future changes in the global water cycle. It complements material in Chapters 2, 3 and 4 on observed and projected changes in the water cycle, and Chapters 10 and 11 on regional climate change and extreme events. The assessment includes the physical basis for water cycle changes, observed changes in the water cycle and attribution of their causes, future projections and related key uncertainties, and the potential for abrupt change. Paleoclimate evidence, observations, reanalyses and global and regional model simulations are considered. The assessment shows widespread, non-uniform human-caused alterations of the water cycle, which have been obscured by a competition between different drivers across the 20th century and that will be increasingly dominated by greenhouse gas forcing at the global scale.

Physical Basis for Water Cycle Changes

Modifications of Earth’s energy budget by anthropogenic radiative forcings drive substantial and widespread changes in the global water cycle. There is high confidence that global mean precipitation and evaporation increase with global warming, but the estimated rate is model-dependent (very likely range of 1–3% per 1°C). The global increase in precipitation is determined by a robust response to global mean surface air temperature (very likely 2–3% per 1°C) that is partly offset by fast atmospheric adjustments to atmospheric heating by greenhouse gases and aerosols. The overall effect of anthropogenic aerosols is to reduce global precipitation and alter large-scale atmospheric circulation patterns through their well-understood surface radiative cooling effect (high confidence). Land-use and land-cover changes also drive regional water cycle changes through their influence on surface water and energy budgets (high confidence). {8.2.1,,, Box 8.1}

A warmer climate increases moisture transport into weather systems, which, on average, makes wet seasons and events wetter (high confidence). An increase in near-surface atmospheric water holding capacity of about 7% per 1°C of warming explains a similar magnitude of intensification of heavy precipitation events (from sub-daily up to seasonal time scales) that increases the severity of flood hazards when these extremes occur (high confidence). The severity of very wet and very dry events increases in a warming climate (high confidence), but changes in atmospheric circulation patterns alter where and how often these extremes occur, with substantial regional differences and seasonal contrasts. A slowdown of tropical circulation with global warming partly offsets the warming-induced strengthening of precipitation in monsoon regions (high confidence). {8.2.2, 8.2.3,, 8.4.1, 8.5.1}

Warming over land drives an increase in atmospheric evaporative demand and the severity of droughts (high confidence). Greater warming over land than over the ocean alters atmospheric circulation patterns and, on average, reduces continental near-surface relative humidity, which contributes to regional drying (high confidence). Increasing atmospheric CO2 concentrations increase plant growth and water-use efficiency, but there is low confidence in how these factors drive regional water cycle changes. {8.2.2, 8.2.3}

Causes of Observed Changes

Human-caused climate change has driven detectable changes in the global water cycle since the mid-20th century (high confidence). Global warming has contributed to an overall increase in atmospheric moisture and precipitation intensity (high confidence), increased terrestrial evapotranspiration (medium confidence), influenced global patterns in aridity (very likely) , and enhanced contrasts in surface salinity and precipitation minus evaporation patterns over the oceans (high confidence). {3.4.2, 3.4.3, 3.5.2, 8.3.1, 9.2.2}

Greenhouse gas forcing has driven increased contrasts in precipitation amounts between wet and dry seasons and weather regimes over tropical land areas (medium confidence) and a detectable precipitation increase in the northern high latitudes (high confidence). Greenhouse gas forcing has also contributed to drying in dry summer climates, including the Mediterranean, south-western Australia, south-western South America, South Africa, and western North America (medium to high confidence). Earlier onset of spring snowmelt and increased melting of glaciers have already contributed to seasonal changes in streamflow in high-latitude and low-elevation mountain catchments (high confidence). {Box 8.2,, 8.3.1, 3.3.2, 3.3.3, 3.5.2}

Anthropogenic aerosols have driven detectable large-scale water cycle changes since at least the mid-20th century (high confidence). Shifts in the tropical rain belt are associated with the inter-hemispheric temperature response to the time-evolving radiative influence of anthropogenic aerosols and the ongoing warming influence of greenhouse gases (high confidence). Cooling in the Northern Hemisphere by sulphate aerosols explained a southward shift in the tropical rain belt and contributed to the Sahel drought from the 1970s to the 1980s (high confidence), subsequent recovery from which has been linked with greenhouse gas warming (medium confidence). Observed changes in regional monsoon precipitation, especially over South Asia, East Asia and West Africa, have been limited over much of the 20th century due to increases driven by warming from greenhouse gases being counteracted by decreases due to cooling from anthropogenic aerosols (high confidence). {,, Box 8.1}

Land-use change and water extraction for irrigation have influenced local and regional responses in the water cycle (high confidence). Large-scale deforestation haslikely decreased evapotranspiration and precipitation and increased runoff over the deforested regions. Urbanization has increased local precipitation (medium confidence) and resulting runoff intensity (high confidence). Increased precipitation intensities have enhanced groundwater recharge, most notably in tropical regions (medium confidence). There is high confidence that groundwater depletion has occurred since at least the start of the 21st century as a consequence of groundwater withdrawals for irrigation in agricultural areas in drylands (e.g., the southern High Plains and California Central Valley of the USA, North China Plain, and north-west India). {,, Box 10.3, FAQ 8.1}

Southern Hemisphere storm tracks and associated precipitation have shifted polewards since the 1970s, especially in the austral summer and autumn (high confidence). It is very likely that these changes are associated with a positive trend in the Southern Annular Mode, related to both stratospheric ozone depletion and greenhouse gas increases. There is medium confidence that the recent observed expansion of the Hadley circulation was caused by greenhouse gas forcing, especially in the Southern Hemisphere, but there is onlylow confidence in how it influences the drying of subtropical land areas. {8.2.2, 8.3.2, 3.3.3}

Future Water Cycle Changes

Without large-scale reduction in greenhouse gas emissions, global warming is projected to cause substantial changes in the water cycle at both global and regional scales (high confidence). Global annual precipitation over land is projected to increase on average by 2.4 [–0.2 to +4.7] % (likely range) in the SSP1-1.9 low-emissions scenario and by 8.3 [0.9 to 12.9] % in the SSP5-8.5 very high-emissions scenario by 2081–2100, relative to 1995–2014. It is virtually certain that evaporation will increase over the oceans and very likely that evapotranspiration will increase over land with regional exceptions in drying areas. There is low confidence in the sign and magnitude of projected changes in global land runoff in all Shared Socio-economic Pathway scenarios. Projected increases in precipitation amount and intensity will be associated with increased runoff in the northern high latitudes (high confidence). There is high confidence that mountain glaciers will diminish in all regions and that seasonal snow cover duration will generally decrease. Runoff from small glaciers will typically decrease through loss of ice mass, while runoff from large glaciers is likely to increase with increasing global warming until glacier mass becomes depleted (high confidence). {4.5.1, 8.4.1}

Increased evapotranspiration due to growing atmospheric water demand will decrease soil moisture over the Mediterranean, south-western North America, southern Africa, south-western South America, and south-western Australia (high confidence). In the Mediterranean, south-western South America, and western North America, future aridification will far exceed the magnitude of change seen in the last millennium (high confidence). Some tropical regions are also projected to experience increased aridity, including the Amazon basin and Central America (high confidence). {8.4.1}

Water cycle variability and extremes are projected to increase faster than average changes in most regions of the world and under all emissions scenarios (high confidence). In the tropics and in the extratropics of both hemispheres during summer/warm season, interannual variability of precipitation and runoff over land is projected to increase at a faster rate than changes in seasonal mean precipitation amount (medium confidence). It is very likely that rainfall variability related to the El Niño–Southern Oscillation will be amplified by the end of the 21st century. Sub-seasonal precipitation variability is also projected to increase, with fewer rainy days but increased daily mean precipitation intensity over many land regions (high confidence). Precipitation extremes will increase in almost all regions (high confidence), even where seasonal mean precipitation is projected to decrease (medium confidence). There is high confidence that heavy precipitation events associated with both tropical and extratropical cyclones will intensify. {,,, 8.4.1, 8.4.2, 8.5.2,}

There are contrasting projections in monsoon precipitation, with increases in more regions than decreases (medium confidence). Summer monsoon precipitation is projected to increase for the South, South East and East Asian monsoon domains, while North American monsoon precipitation is projected to decrease (medium confidence). West African monsoon precipitation is projected to increase over the Central Sahel and decrease over the far western Sahel (medium confidence). There is low confidence in projected precipitation changes in the South American and Australian monsoons (for both magnitude and sign). There is high confidence that the monsoon season will be delayed in North and South America and medium confidence that it will be delayed in the Sahel. {8.2.2,}

Precipitation associated with extratropical storms and atmospheric rivers will increase in the future in most regions (high confidence). A continued poleward shift of storm tracks in the Southern Hemisphere (likely) and the North Pacific (medium confidence) will lead to similar shifts in annual or seasonal precipitation. There is low confidence in projections of blocking and stationary waves and therefore their influence on precipitation for almost all regions. {8.4.2}

The seasonality of precipitation, water availability and streamflow will increase with global warming over the Amazon (medium confidence) and in the subtropics, especially in the Mediterranean and southern Africa (high confidence). The annual contrast between the wettest and driest month of the year is likely to increase by 3–5% per 1°C in most monsoon regions in terms of precipitation, precipitation minus evaporation, and runoff (medium confidence). There is high confidence in an earlier onset in spring snowmelt, with higher peak flows at the expense of summer flows in snow-dominated regions globally, butmedium confidence that reduced snow volume in lower-latitude regions will reduce runoff from snowmelt. {8.2.2, Box 8.2,,}

Confidence in Projections, Non-linear Responses and the Potential for Abrupt Changes

Representation of key physical processes has improved in global climate models, but they are still limited in their ability to simulate all aspects of the present-day water cycle and to agree on future changes (high confidence). Climate change studies benefit from sampling the full distribution of model outputs when considering future projections at regional scales. Increasing horizontal resolution in global climate models improves the representation of small-scale features and the statistics of daily precipitation (high confidence). High-resolution climate and hydrological models provide a better representation of land surfaces, including topography, vegetation and land-use change, which improve the accuracy of simulations of regional changes in the water cycle (high confidence). There is high confidence in the potential added value of regional climate models but only medium confidence that this potential is currently realized. {8.5.1}

Natural climate variability will continue to be a major source of uncertainty in near-term (2021–2040) water cycle projections (high confidence). Decadal predictions of water cycle changes should be considered with low confidence in most land areas because the internal variability of precipitation is difficult to predict and can offset or amplify the forced water cycle response. Water cycle changes that have already emerged from natural variability will become more pronounced in the near term, but the occurrence of volcanic eruptions (either single large events or clustered smaller ones) can alter the water cycle for several years, decreasing global mean land precipitation and altering monsoon circulation (high confidence). {8.5.2, Cross-Chapter Box 4.1}

Continued global warming will further amplifygreenhouse gas-induced changes in large-scale atmospheric circulation and precipitation patterns (high confidence), but in some cases regional water cycle changes are not linearly related to global warming. Non-linear water cycle responses are explained by the interaction of multiple drivers, feedbacks and time scales (high confidence). Non-linear responses of regional runoff, groundwater recharge and water scarcity highlight the limitations of simple pattern-scaling techniques (medium confidence). Water resources fed by melting glaciers are particularly exposed to non-linear responses (high confidence). {8.5.3}

Abrupt human-caused changes to the water cycle cannot be excluded. There is evidence of abrupt change in some high-emissions scenarios, but there is no overall consistency regarding the magnitude and timing of such changes. Positive land surface feedbacks, including vegetation and dust, can contribute to abrupt changes in aridity, but there is onlylow confidence that such changes will occur during the 21st century. Continued Amazon deforestation, combined with a warming climate, raises the probability that this ecosystem will cross a tipping point into a dry state during the 21st century (low confidence). The paleoclimate records show that a collapse in the Atlantic Meridional Overturning Circulation (AMOC) causes abrupt shifts in the water cycle (high confidence), such as a southward shift in the tropical rain belt, weakening of the African and Asian monsoons, strengthening of Southern Hemisphere monsoons, and drying in Europe. There is medium confidence that AMOC will not collapse before 2100, but should it collapse, it is very likely that there would be abrupt changes in the water cycle. {8.6.1, 8.6.2}

Solar radiation modification could drive abrupt changes in the water cycle (high confidence). It is very likely that abrupt water cycle changes will occur if solar radiation modification (SRM) techniques are implemented rapidly or terminated abruptly. The impact of SRM is spatially heterogeneous (high confidence), will not fully mitigate the greenhouse gas-forced water cycle changes (medium confidence), and can affect different regions in potentially disruptive ways (low confidence). {8.6.3}

8.1 Introduction

8.1.1 Scope and Overview Importance of Water for Human Societies and Ecosystems

Water is vital to all life on Earth. Seventy-one percent of the Earth is covered by water, with saline ocean water accounting for around 97% of total water availability (Figure 8.1). Terrestrial freshwater represents less than 2% of all water on Earth, and the remainder (around 1–2%) is primarily made up of saline groundwater and saline lakes (Durack, 2015; Abbott et al., 2019). Ice sheets, glaciers and snow pack account for approximately 96% of all freshwater, with less than 4% of freshwater considered easily accessible and available for essential ecosystem functioning and human society’s water resource needs (Durack, 2015; Abbott et al., 2019). This very small fraction of freshwater represents a total volume of about 835,000 km3, mostly contained in groundwater (630,000 km3), the remaining 205,000 km3being stored in lakes, rivers, wetlands and soils (Abbott et al., 2019). Although the natural cycling rate of this amount is theoretically enough to meet global human and ecosystem needs, there are large geographical and seasonal differences that influence the availability of freshwater to meet regional demands.

Freshwater is the most essential natural resource on the planet (Mekonnen and Hoekstra, 2016; Djehdian et al., 2019) and underpins almost all Sustainable Development Goals (SDGs), which require access to adequate and safe resources for drinking and sanitation (SDG 6) and many other purposes. Freshwater supports a range of human activities from irrigation to industrial processes including the generation of hydro-electricity and the cooling of thermoelectric power plants (Bates et al., 2008; Schewe et al., 2014). These activities require sufficient quantities of freshwater that can be drawn from rivers, lakes, groundwater stores, and in some cases, desalinated sea water (Schewe et al., 2014). Recent estimates of global water pools and fluxes suggest that half of global river discharge is redistributed each year by human water use (Abbott et al., 2019). This emphasizes the need to consider both anthropogenic climate change and direct human influences, such as population increase or migration, economic development, urbanization, and land use change, when planning water-related mitigation or adaptation strategies (Jiménez Cisneros et al., 2014).

Water scarcity occurs when there are insufficient freshwater resources to meet water demands, although water problems may also arise from water quality issues or from economic and institutional barriers (AR6 WGII Chapter 4). This affects the preservation of environmental flows that ultimately influence ecosystem functioning and services (Schewe et al., 2014; Mekonnen and Hoekstra, 2016; Djehdian et al., 2019). As such, water availability is a major constraint on human society’s ability to meet the future food and energy needs of a growing population (D’Odorico et al., 2018). Water plays a key role in the production of energy, including hydro-electricity, bioenergy, and the extraction of unconventional fossil fuels (Schewe et al., 2014; D’Odorico et al., 2018; Djehdian et al., 2019). These dependencies have resulted in increasing competition for water between the food and energy sectors. Pressures on this ‘food-energy-water nexus’ are further compounded by increasing globalization, which can transfer large-scale water demands to other regions of the world, raising serious concerns about local food and water security in regions that are highly dependent on agricultural exports or imports (D’Odorico et al., 2018).

The consequences of climate change on terrestrial ecosystems and human societies are primarily experienced through changes to the global water cycle (JiménezCisneros et al., 2014). Changes in the quantity and seasonality of water due to climate change have long been recognized by IPCC and global development agencies as heavily influencing the food security and economic prosperity of many countries, particularly in the arid and semi-arid areas of the world including Asia, Africa, Australia, Latin America, the Mediterranean, and small island developing states (Bates et al., 2008; Schewe et al., 2014; Mekonnen and Hoekstra, 2016). Having too much or too little water increases the likelihood of flooding and drought, as precipitation variability increases in a warming climate (Stockeret al., 2013; Hoegh-Guldberg et al., 2018). Climate change poses a threat to both regional water availability and global water security. Changes in precipitation and glacier runoff and snowmelt influence other hydroclimate variables like surface and subsurface runoff, and groundwater recharge, which are critical to the water, food and energy security of many regions (Oki andKanae, 2006; Jiménez Cisneros et al., 2014; Schewe et al., 2014; Mekonnen and Hoekstra, 2016).

Currently, around four billion people live under conditions of severe freshwater scarcity for at least one month of the year, with half a billion people in the world facing severe water scarcity all year round (Mekonnen and Hoekstra, 2016). The AR5 WGII reported that approximately 80% of the world’s population already suffers from high levels of threat to water security (Jiménez Cisneros et al., 2014). Given the vulnerability of the planet’s freshwater resources and the role of climate change in intensifying adverse impacts on human societies and ecosystems (Hoegh-Guldberg et al., 2018; IPCC, 2018), this chapter evaluates advances in the theoretical, observational and model based understanding of the global water cycle made since AR5 (IPCC, 2013) and AR6 Special Reports. Overview of the Global Water Cycle in the Climate System

As shown in Figure 8.1, the global water cycle is the continuous, naturally occurring movement of water through the climate system from its liquid, solid and gaseous forms among reservoirs of the ocean, atmosphere, cryosphere and land (Stocker et al., 2013). In the atmosphere, water primarily occurs as a gas (water vapour), but it is also present as ice and liquid water within clouds where it substantially affects Earth’s energy balance (Sections and The water cycle primarily involves the evaporation1 and precipitation of moisture at the Earth’s surface including transpiration associated with biological processes. Water that falls on land as precipitation, supplying soil moisture, groundwater recharge, and river flows, was once evaporated from the ocean or sublimated from ice-covered regions before being transported through the atmosphere as water vapour, or in some areas was generated over land through evapotranspiration (Gimeno et al., 2010; van der Ent and Savenije, 2013). In addition, the net flux of atmospheric and continental freshwater is a key driver of sea surface salinity, which in turn influences the density and circulation of the ocean (Chapter 9).

Figure 8.1 | Depiction of the present-day water cycle based on previous assessments (Trenberth et al., 2011; Rodell et al., 2015; Abbott et al., 2019) with adjustments for groundwater flows (Zhou et al., 2019c; Luijendijk et al., 2020), seasonal snow (Pulliainen et al., 2020) and ocean precipitation and evaporation (Stephens et al., 2012; Allan et al., 2020; Gutenstein et al., 2021). The net loss of frozen and liquid water from land to ocean is estimated from Chapter 9, Table 9.5. In the atmosphere, which accounts for only 0.001% of all water on Earth, water primarily occurs as a gas (water vapour), but it is also present as ice and liquid water within clouds. The ocean is the primary water reservoir on Earth: it comprises mostly liquid water across much of the globe but also includes areas covered by ice in polar regions. Liquid freshwater on land forms surface water (lakes, rivers) and, together with soil moisture and mostly unusable groundwater stores, accounts for less than 2% of global water (Stocker et al., 2013). Solid terrestrial water that occurs as ice sheets, glaciers, snow and ice on the surface, and permafrost currently represents nearly 2% of the planet’s water (Stocker et al., 2013). Water that falls as snow in winter provides soil moisture and streamflow after melting, which are essential for human activities and ecosystem functioning. Note that these best estimates do not lead to a perfectly closed global water budget and that this budget has no reason to be closed given the ongoing human influence through both climate change (e.g., melting of ice sheets and glaciers, see Chapter 9) and water use (e.g., groundwater depletion through pumping into fossil aquifers, see Figure 8.10).

Understanding the interactions between the water and energy cycles is one of the four core projects of the World Climate Research Programme (WCRP). Latent heat fluxes, released by condensation of atmospheric water vapour and absorbed by evaporative processes, are critical to driving the circulation of the atmosphere on scales ranging from individual thunderstorm cells to the global circulation of the atmosphere (Stocker et al., 2013; Miralles et al., 2019). Water vapour is the most important gaseous absorber in the Earth’s atmosphere, playing a key role in the Earth’s radiative budget (Schneider et al., 2010). As atmospheric water vapour content increases with temperature, it has a considerable influence on climate change (Section Additionally, a small fraction of the atmospheric water content is liquid or solid and has a major effect on both solar and longwave radiative fluxes, from the Earth’s surface to the top of the atmosphere. The cloud response to anthropogenic radiative forcings, both in the tropics and in the extratropics (Zelinka et al., 2020), is therefore also crucial for understanding climate change (Section

The terrestrial water and carbon cycles are also strongly coupled (Cross-Chapter Box 5.1). As atmospheric carbon dioxide (CO2) concentration increases, the physical environment in which plants grow is altered, including the availability of soil moisture necessary for plants’ CO2 uptake and, potentially, the effectiveness of CO2 removal techniques to mitigate climate change (Section Rising surface CO2 concentrations also modify stomatal (small pores at the leaf surface) regulation as well as the plants’ biomass, thus affecting ecosystem photosynthesis and transpiration rates and leading generally to a net increase in water use efficiency (Lemordant et al., 2018). These coupled changes have profound implications for the simulation of the carbon and water cycles (Gentine et al. , 2019; see also Section 5.4.1), which can be better assessed with the new generation Earth system models, although both the carbon concentration and carbon-climate feedbacks remain highly uncertain over land Section 5.4.5; Arora et al., 2020). The water constraints on the terrestrial carbon sinks are a matter of debate regarding the feasibility or efficiency of some land-based CO2 removal and sequestration techniques requested to comply with the Paris Agreement (Section5.; Fuss et al. , 2018; Belyazid and Giuliana, 2019).

8.1.2 Summary of Water Cycle Changes From AR5 and Special Reports

This Report is the first IPCC assessment to include a chapter specifically dedicated to providing an integrated assessment of the global water cycle changes, by building on many chapters from previous reports. This section summarizes observed and projected water cycle changes reported in AR5(IPCC, 2013) and in the recent IPCC Special Reports on Global Warming of 1.5°C (SR1.5), the Ocean and Cryosphere in a Changing Climate (SROCC), and Climate Change and Land (SRCCL). Summary of Observed and Projected Water Cycle Changes from AR5

Based on long-term observational evidence (Hartmann et al., 2013), AR5 concluded it was likely that anthropogenic influence has affected the water cycle since the 1960s (IPCC, 2018). Detectable human influ­ence on changes to the water cycle were found in atmospheric moisture content (medium confidence), global-scale changes of precipitation over land (medium confidence), intensification of heavy precipitation events over land regions where sufficient data networks exist (medium confidence), and very likely changes to ocean salinity through its connection with evaporation minus precipitation change patterns (Sections 2.5, 2.6, 3.3, 7.6, 10.3 and 10.4; Stocker et al., 2013). The AR5 also reported that it is very likely that global surface air specific humidity increased since the 1970s. There was low confidence in the observations of global-scale cloud variability and trends, medium confidence in reductions of pan-evaporation, and medium confidence in the non-monotonic changes of global evapotranspiration since the 1980s. In terms of streamflow and runoff, AR5 identified that there is low confidence in the observed increasing trends of global river discharge during the 20th century. Similarly, AR5 concluded that there is low confidence in any global-scale observed trend in drought or dryness (lack of rainfall) since the mid-20th century. Yet, the frequency and intensity of drought likely increased in the Mediterranean and West Africa, while theylikely decreased in central North America and north-western Australia since 1950.

Water cycle projections in AR5 (Collins et al., 2013) were considered primarily in terms of water vapour, precipitation, surface evaporation, runoff, and snowpack. Globally-averaged precipitation was projected to increase with global warming with virtual certainty (Chapter 12 Executive Summary and Section Regionally, precipitation in some areas of the tropics and polar regions could increase by more than 50% by the end of the 21st century under the RCP8.5 emissions scenario, while precipitation in large areas of the subtropics could decrease by 30% or more (AR5 FAQ 12.2, Figure 12.22). Overall, the contrast of annual mean precipitation between dry and wet regions and between dry and wet seasons (‘wet get wetter, dry get drier’) was projected to increase over most of the globe with high confidence (Chapter 12 Executive Summary and Section Globally, the frequency of intense precipitation events was projected to increase while the frequency of all precipitation events was projected to decrease, leading to the contradictory-seeming projection of a simultaneous increase in both droughts and floods (12.2 and Section in AR5 WGI). Surface evaporation change was projected to be positive over most of the ocean and to generally follow the pattern of precipitation change over land Chapter 12 Executive Summary, and Section Near-surface relative humidity reductions over many land areas were projected to be likely , with medium confidence (Section General decreases in soil moisture in present-day dry regions were considered likely , and projected with medium confidence under the RCP8.5 scenario (Section Soil moisture drying in the Mediterranean, south-west USA and southern African regions was considered likely , with high confidence by the end of this century under the RCP8.5 scenario (Section Projections for annual runoff included both decreases and increases. Decreases in Northern Hemisphere snow cover were assessed as very likely with continued global warming (Section As temperatures increase, snow accumulation was projected to begin later in the year and melting to start earlier, with related changes in snowmelt-driven river flows (FAQ 12.2 and Section in AR5 WGI). In terms of the potential for abrupt change in components of the water cycle, long-term droughts and monsoonal circulation were identified as potentially undergoing rapid changes, but the assessment was reported with low confidence (Sections and, and Table 12.4). Key Findings of AR6 Special Reports

The SR1.5 assessed the impacts of global warming of 1.5°C above pre-industrial levels. The dominant human influence on observed global warming and related water cycle changes was confirmed. Further evidence that anthropogenic global warming has caused an increase in the frequency, intensity and/or amount of heavy precipitation events at the global scale (medium confidence), as well as in drought occurrence in the Mediterranean region (medium confidence) was also reported. Chapter 3 of SR1.5 (Hoegh-Guldberg et al., 2018) highlights that each half degree of additional global warming influences the climate response. Heavy precipitation shows a global tendency to increase more at 2°C compared to 1.5°C, though there is low confidence in projected regional differences in heavy precipitation at 1.5°C compared to 2°C global warming, except at high latitudes or at high altitude where there is medium confidence. A key finding is that ‘limiting global warming to 1.5°C compared to 2°C would approximately halve the proportion of the world population expected to suffer water scarcity, although there is considerable variability between regions (medium confidence)’ (SR1.5). This is consistent with greater adverse impacts found at 2°C compared to 1.5°C for a number of dryness or drought indices (Schleussner et al., 2016; Lehner et al., 2017; Greve et al., 2018). There is also medium confidence that land areas with increased runoff and exposure to flood hazards will increase more at 2°C compared to 1.5°C of global warming.

The Special Report on the Ocean and Cryosphere in a Changing Climate (SROCC) provides a comprehensive assessment of recent and projected changes, specifically in snow and ice-covered areas that form a key component of the water cycle in high-elevation and high-latitude areas. High mountain regions have experienced significant warming since the early 20th century, resulting in reduced snowpack on average (Marty et al., 2017), with glaciers retreating globally since the mid-20th century (Marzeion et al., 2018; Zemp et al., 2019). Glacier shrinkage and snow cover changes have led to changes (both increases and decreases) in streamflow in many mountain regions in recent decades (Milner et al., 2017). Permafrost regions have undergone degradation and ground-ice loss due to recent warming (Lu et al., 2017). Glacier mass loss is projected to continue through the 21st century under all scenarios. In high mountain areas, low-elevation snow cover is also projected to decrease, regardless of emissions scenario. Widespread permafrost thaw is projected to continue through this century and beyond. River runoff in snow- or glacier-fed basins is projected to increase in winter and to decrease in summer (and in the annual mean) by 2100. In the oceans, the Atlantic Meridional Overturning Circulation (AMOC) will very likely weaken over the 21st century under all emissions scenarios (SROCC), with potential effects on atmospheric circulation and the water cycle at the regional scale (see also Section 8.6).

The Special Report on climate change, desertification, land degradation, sustainable management, food security, and greenhouse gas (GHG) fluxes in terrestrial ecosystems (SRCCL) has clear connections with the water cycle. This Report indicates that since 1850–1900, land surface temperature has risen nearly twice as much as global surface temperature (high confidence), with an increase in dry climates (high confidence). Land surface processes modulate the likelihood, intensity and duration of many extreme events including droughts (medium confidence) and heavy precipitation (medium confidence). The direction and magnitude of hydrological changes induced by land use change and land surface feedbacks vary with location and season (high confidence). Desertification exacerbates climate change through feedbacks involving vegetation cover, greenhouse gases and mineral dust aerosol (high confidence). Urbanization increases extreme rainfall events over or downwind of cities (medium confidence). Intensification of rainy events increase their consequences on land degradation.

8.1.3 Chapter Motivations, Framing and Preview

The AR5 report was a major step forward in the assessment of the human influence on the Earth’s water cycle, yet regional projections of precipitation and water resources often remained very uncertain for a range of reasons including modelling uncertainty and the large influence of internal variability (Sections 1.4.3 and 8.5.2; Hawkins and Sutton, 2011; Deser et al. , 2012). Since AR5, longer and more homogeneous observational and reanalysis datasets have been produced along with new ensembles of historical simulations driven by all or individual anthropogenic forcings. These factors, together with improved detection-attribution tools, has enabled a more comprehensive assessment and a better understanding of recent observed water cycle changes, including the competing effects of GHGs and aerosol emissions. New paleoclimate reconstructions have been also developed, particularly from the SH, that were not available at the time of AR5. There have also been advances in modelling clouds, precipitation, surface fluxes, vegetation, snow, floodplains, groundwater and other processes relevant to the water cycle. Convection permitting and cloud-resolving models have been implemented over increasingly large domains and can be used as benchmarks for the evaluation of the current-generation climate models. The added value of increased resolution in global or regional climate models can be also assessed more thoroughly based on dedicated model intercomparison projects (Sections 10.3.3 and 8.5.1). Ongoing research activities on decadal predictions and observational constraints are aimed at narrowing the plausible range of near-term (20212040) to long-term (20812100) water cycle changes.

This chapter assesses water cycle changes and considers climate change from the perspective of its effects on water availability (including streamflow and soil moisture, snow mass and glaciers, groundwater, wetlands and lakes) rather than only precipitation. The chapter highlights the sensitivity of the water cycle to multiple drivers and the complexity of its responses, depending on regions, seasons and time scales. Anthropogenic drivers include not only emissions of GHGs but also different species of aerosols, land and water management practices. Emphasis is placed on assessing the full range of projections, including ‘low likelihood, high impact’ climate trajectories such as the potential for abrupt changes in the water cycle.

The chapter starts with theoretical evidence that link small-scale processes and drivers, as well as global energy budget and large-scale circulation constraints to physically-understood changes in the global water cycle (Section 8.2). Observed and projected water cycle changes (Sections 8.3 and 8.4, respectively) are assessed in separate sections, but with a parallel structure to facilitate comparison of a specific topic across sections. Projections are primarily assessed on the basis of contrasted emissions scenarios to emphasize the water cycle response to mitigation. Unless otherwise specified, projected anomalies are estimated relative to the 19952014 baseline climatology and are assessed over 20-year time slices, 20212040, 20412060 and 20812100 for near-, mid- and long-term changes respectively. Beyond multi-model ensemble means, model response uncertainty, the influence of natural climate variability, and the potential non-linearities in the regional water cycle response are also considered (Section 8.5). Low likelihood but physically plausible high-impact scenarios are also assessed, especially the potential for abrupt climate change (Section 8.6). Final remarks about future studies on water cycle changes (Section 8.7) are also provided, and the chapter addresses three frequently asked questions (FAQs) on the water cycle’s sensitivity to land use change (FAQ 8.1), the projected occurrence and severity of floods (FAQ 8.2) and droughts (FAQ 8.3) at the global scale. This chapter outline is summarized with a schematic (Figure 8.2) which also provides a quick guide to the main topics addressed across the different sections.

Figure 8.2 | Visual guide to Chapter 8.

(Chapter 8 has multiple links across all AR6 WGI chapters, so necessarily includes references to other chapter subsections and figures. Model evaluation of large-scale circulation and precipitation is mostly covered by Chapter 3, while hydrological extremes are covered by Chapter 11. Chapter 8 focuses on key processes relevant to the water cycle and their resolution-dependent representation in models. Observed and projected changes in large-scale circulation and precipitation are primarily assessed in Chapters 2, 3 and 4. Beyond global and regional mean precipitation amounts, Chapter 8 also focuses on other precipitation properties (e.g., frequency, intensity and seasonality) and other water cycle variables (evapotranspiration, runoff, soil moisture and aridity, solid and liquid freshwater reservoirs). Key regional phenomena (e.g., tropical overturning circulations, monsoons, extratropical stationary waves and storm tracks, modes of variability and related teleconnections) are also assessed given their major dynamical contribution to regional water cycle changes. Although the biosphere and the cryosphere are key components of the water cycle, a more comprehensive assessment of their responses can be found in Chapters 5 and 9, respectively. Further assessment on regional water cycle changes can be found in Chapters 10 to 12 and in the Atlas. The reader is also referred to the interactive (Atlas for a more detailed assessment of the range of model biases and responses at the regional scale. Beyond WGI, water is also a major topic for both adaptation and mitigation policies so has strong connections with both WGII and WGIII. Assessment of hydrological impacts at basin and catchment scales, including a broader discussion on adaptation and vulnerability, potential threats to water security, societal responses, improving resilience in water systems and related case studies is provided in WGII (Chapter 4).

8.2 Why Should We Expect Water Cycle Changes?

It is well understood that global precipitation and evaporation changes are determined by Earth’s energy balance (Section 8.2.1). At regional scales smaller than about 4000 km, water cycle changes become dominated by the transport of moisture (Dagan et al., 2019a; Jakob et al., 2019; Dagan and Stier, 2020), which depend on both thermodynamic and dynamical processes (Section 8.2.2). The constraints of energy budgets at global scales and moisture budgets at regional scales cause key water cycle characteristics such as precipitation intensity, duration and intermittence to alter as the climate warms (Pendergrass and Hartmann, 2014b; Döll et al., 2018). Future water availability is also determined by changes in evaporation, which is driven by a general increase in the atmospheric evaporative demand (Scheff and Frierson, 2014) and modulated by vegetation controls on evaporative losses (Milly and Dunne, 2016; Lemordant et al. , 2018; Vicente-Serrano et al. , 2020). At regional scales, water cycle changes result from the interplay between multiple potential drivers (CO2, aerosols, land use change and human water use; Section 8.2.3). This section assesses advances in physical understanding of global to regional drivers of water cycle changes.

8.2.1 Global Water Cycle Constraints

The Clausius–Clapeyron equation determines that low-altitude specific humidity increases by about 7% °C–1 of warming, assuming that relative humidity remains constant, which is approximately true at a global scale but not necessarily valid regionally. It is very likely that near surface specific humidity has increased since the 1970s (Section 2.3.1) and total atmospheric water vapour content (precipitable water) is very likely to increase at close to a thermodynamic rate on average globally with continued warming. Different radiative forcing mechanisms lead to some variation in the global mean thermodynamic response by altering the relative humidity distribution: the rate of global precipitable water increase with global surface temperature ranges2 from 6.4 ± 1.5% °C–1 for sulphate aerosol-induced changes to 9.8 ± 3.3% °C–1 for black carbon-induced changes based on idealized modelling (Hodnebrog et al., 2019b). Specific humidity increases at a lower rate over land due to decreasing relative humidity (Collins et al., 2013) as corroborated by observations and simple models (Byrne and O’Gorman, 2018). Prevalent increases in atmospheric water vapour drive powerful amplifying feedbacks (Section, intensify atmospheric moisture transport and heavy precipitation events (Section, and alter the surface and atmospheric energy balance, thereby influencing global evaporation and precipitation changes (Figure 8.3).

Figure 8.3 | Schematic representation of fast and slow responses of the atmospheric energy balance and global precipitation to radiative forcing. (‘Baseline’panel) The atmospheric energy budget: (1) responds instantaneously to radiative forcings; (2) leading to rapid atmospheric adjustments; and (3) slower semi‐rapid adjustments involving the land surface and vegetation that further modify atmospheric circulation patterns. (4) This slow precipitation response to global mean surface air temperature is quantified as (a) the hydrological sensitivity, η, and the total precipitation response, including initial rapid adjustments, is termed the apparent hydrological sensitivity, ηa. (b–d) The slow precipitation response over land and ocean develops over time. Large, filled arrows (in panels from ‘baseline’ to 4) depict fluxes or circulation change while small arrows (1–4) denote increases () or decreases () in variables (P is precipitation; L is atmospheric longwave radiative cooling; S is solar radiation absorption by the atmosphere; H is sensible heat flux; E is surface evaporative heat flux; and T is temperature). Adapted from Allan et al. (2020) with statistics taken from Figures 7.2 and Figure 8.1.

While thermodynamics exert a strong control on water vapour changes, global mean precipitation and evaporation are constrained by the balance of energy fluxes in the atmosphere and at the surface (Figure 8.3). Global mean precipitation increases of 13% per 1 °C of warming, as estimated in AR5 (Collins et al., 2013), are explained as a combination of rapid (or fast) atmospheric adjustments and slow temperature-driven responses (Figure 8.3, panels 14) to radiative forcings (Andrewset al., 2010; Bala et al., 2010; Cao et al., 2012). Fast atmospheric adjustments are caused by near-instantaneous (hours to days) changes in the atmospheric energy budget (Figure 8.3, panels 1–3) and atmospheric properties (e.g., temperature, clouds and water vapour) in direct response to the radiative effects of a forcing agent (Sherwoodet al., 2015). A further relatively fast (days to months) adjustment of the climate system involves interactions with vegetation and land surface temperature (Figure 8.3, panel 3), which respond more rapidly than ocean temperature to a radiative forcing (Cao et al., 2012; Dong et al., 2014). The slower temperature-dependent precipitation response is driven by the increased atmospheric radiative cooling rate of a warming atmosphere. Warming drives increases in precipitation intensity while frequency is dominated by rapid atmospheric adjustments to the radiative forcing based onabrupt 4×CO2CMIP6 simulations (Douville and John, 2021). Since AR5, many new studies applying the dual rapid adjustment and slow response framework show that global precipitation responses to different forcing agents are physically well understood (Fläschner et al. , 2016; MacIntosh et al. , 2016; Samset et al. , 2016; Myhre et al. , 2018a). Further confidence in the coupled processes involved are provided by simple models representing the energy budget and thermodynamic constraints that limit global mean evaporation to around 1.5% °C–1 (Siler et al., 2019). This strengthens the physical link between energy budget and thermodynamic drivers of the global water cycle (Section

Hydrological sensitivity (η) is defined as the linear change in global mean precipitation with global surface air temperature (GSAT) once rapid adjustments of the hydrological cycle to radiative forcings have occurred (Figure 8.3a). There is robust understanding and high agreement across idealized CO2 forcing CMIP5 and CMIP6 experiments (Fläschner et al., 2016; Samset et al., 2018b; Pendergrass, 2020b) thatη= 2.13.1% °C–1 (Figure 8.4). The magnitude ofηdepends primarily on atmospheric net radiative cooling which is controlled by thermal deepening of the troposphere (Jeevanjee and Romps, 2018) and limited by surface evaporation and consequent atmospheric latent heat release and warming (Webb et al., 2018). Climate feedbacks (e.g., temperature lapse rate and clouds) that vary across models (Sections 7.4 and 3.8.2) also modulate the magnitude ofη (O’Gorman et al. , 2012; Fläschner et al. , 2016; T.B. Richardson et al. , 2018a). Uncertainty in ηacross CMIP5 models relating to deficiencies in representing low‐altitude cloud feedbacks (Watanabe et al., 2018) and absorption of shortwave radiation by atmospheric water vapour (DeAngelis et al., 2015) do not apply well to CMIP6 simulations, the latter improvement explained by more accurate radiative transfer modelling (Pendergrass, 2020b).

Figure 8.4 | Estimate (5–95% range) of the increase in precipitation and its extremes with global mean surface warming. Global time-averaged precipitation changes (left) are based on responses to increasing CO2 (apparent hydrological sensitivity, ηa) and the temperature-dependent component (hydrological sensitivity, η), both of which are based on GCM experiments; the land (L) and ocean (O) components (Fläschner et al., 2016; T.B. Richardson et al., 2018a; Samset et al., 2018a; Pendergrass, 2020b; Rehfeld et al., 2020) and observational estimates (GPCP/HadCRUTv4.6) use trends (1988–2014) as a proxy forηa and interannual variability as a proxy forη, with 90% confidence range accounting for statistical uncertainty only (Adler et al., 2017; Allan et al., 2020). For extreme precipitation, assessment is for 24 hour, 99.9th percentile or annual maximum extremes from GCMs (Fischer and Knutti, 2015; Pendergrass et al., 2015; Borodina et al., 2017; Pfahl et al., 2017; Sillmann et al., 2017), regional climate models (RCMs) (Bao et al., 2017), an observationally-constrained tropical estimate (O’Gorman, 2012) and estimates from observed changes (Westra et al., 2013; Donat et al., 2016; Borodina et al., 2017; Zeder and Fischer, 2020; Sun et al., 2021). For hourly and sub-hourly extremes observed changes (Barbero et al., 2017; Guerreiro et al., 2018) and high-resolution models, including RCM and cloud-resolving models (CRMs) are assessed (Ban et al., 2015; Prein et al., 2017; Haerter and Schlemmer, 2018; Hodnebrog et al., 2019a; Lenderink et al., 2019). Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

Observed estimates of hydrological sensitivity (η= 3.2 ± 0.8% °C–1) based on interannual variability (Allan et al.,2020) or responses to El Niño–Southern Oscillation (ENSO) of 9% °C–1 (Adler et al., 2017) are not suitable to assess the magnitude of η (Figure 8.4). This is because these relationships depend on amplifying feedbacks associated with ENSO-related cloud changes (G.L. Stephens et al., 2018) that may not be relevant for longer term climate change. However, there is robust evidence and high agreement across observations, modelling and supporting physics that precipitation increases at a lower % °C–1rate than water vapour content in the global mean (Held and Soden, 2006; Collins et al., 2013; Allan et al., 2020), implying an increased residence time of atmospheric water vapour (Hodnebrog et al., 2019b; Dijk et al., 2020). Increasing globalprecipitation, evaporation and moisture fluxes with warming thereby drive an intensification but not acceleration of the global water cycle (Sections and

The overall global mean rate of precipitation change per 1 °C of GSAT increase, apparent hydrological sensitivity (ηa), is reduced compared to hydrological sensitivity by the direct influence of radiative forcing agents on the atmospheric energy balance. Rapid atmospheric adjustments that alter precipitation are primarily caused by GHGs and absorbing aerosols, withhigh agreement and medium evidence across idealized simulations (Fläschner et al.,2016; Samset et al., 2016). A range of rapid precipitation adjustments to CO2 between models are also attributed to vegetation responses leading to a re-partitioning of surface latent and sensible heat fluxes (DeAngelis et al.,2016). Values obtained from six CMIP5 models simulating the Last Glacial Maximum (LGM; 21,000–19,000 years ago) and pre-industrial period (ηa=1.63.0%°C–1) are larger than for each correspondingabrupt 4×CO2experiment (ηa=1.3–2.6% °C–1) due to differences in the mix of forcings, vegetation and land surface changes and a higher thermodynamic % °C–1evaporation scaling in the colder state (Figure 8.4, Section; G. Li et al., 2013). Updated estimates across comparable experiments from 22 CMIP5/CMIP6 models (Rehfeld et al., 2020) display a consistent range (ηa=1.7 ± 0.6%°C–1). Confirmingηain observations (Figure 8.4) is difficult due to measurement uncertainty, varying rapid adjustments to radiative forcing and unforced variability (Dai and Bloecker, 2019; Allan et al., 2020).

Climate drivers that instantaneously affect the surface much more than the atmospheric energy budget (such as solar forcing and sulphate aerosol) produce only a small rapid adjustment of the global water cycle and therefore largerηathan drivers that immediately modulate the atmospheric energy budget such as GHGs and absorbing aerosol (Salzmann, 2016; Samset et al. , 2016; Lin et al. , 2018; F. Liu et al. , 2018). Thus, global precipitation appears more sensitive to radiative forcing from sulphate aerosols (2.8 ± 0.7% °C–1; ηaη) than GHGs (1.4 ± 0.5% °C–1; ηa< η) while the response to black carbon aerosol can be negative (3.5 ± 5.0% °C–1; ηa<< η) due to strong atmospheric solar absorption (Samset et al., 2016). Therefore, artificially reducing surface-absorbed sunlight through solar radiation modification strategies to mitigate GHG warming will not mitigate precipitation changes (see Sections, 6.4.7 and 8.6.3). Aerosol-induced precipitation changes depend upon the type of aerosol species and their spatial distribution. Global mean precipitation increases after complete removal of present-day anthropogenic aerosol emissions (see also Section 4.4.4) in four different climate models (ηa= 1.65.5% °C–1) are mainly attributed to sulphate aerosol as opposed to other aerosol species (Samset et al., 2018b). Idealized modelling studies show that sulphate aerosol increases over Europe produce a larger global precipitation response than an equivalent increase in aerosol burden or radiative forcing overAsia, explained by differences in cloud climatology and cloud-aerosol interaction (Kasoar et al., 2018; L. Liu et al., 2018). The vertical profiles of black carbon and ozone further influence the magnitude of the rapid global precipitation response, yet are difficult to observe and simulate (Allen and Landuyt, 2014; MacIntosh et al., 2016; Stjern et al., 2017; Sand et al., 2020).

Hydrological sensitivity is generally lower over land but with a large uncertainty range (η=0.1 to 3.0% °C–1) relative to the oceans (η= 2.3 to 3.3% °C–1) based on multi-model 4 × CO2 CMIP6simulations (Pendergrass, 2020b), broadly consistent with comparable CMIP5 experiments (T.B. Richardson et al. , 2018a; Samset et al. , 2018a). Suppressed hydrological sensitivity over land (Figures 8.3d and 8.4) is associated with greater warming compared with the oceans, which alters atmospheric circulation and precipitation patterns (Saint-Lu et al., 2020). Also, since oceans supply much of the moisture to fuel precipitation over land, the slower ocean warming rate means there is insufficient moisture supplied to maintain continental relative humidity levels (Byrne and O’Gorman, 2018), which can inhibit convection (J. Chen et al., 2020a). Land surface feedbacks involving soil-vegetation-atmosphere coupling further drive continental drying (Berg et al., 2016; Kumar et al., 2016; Chandan and Peltier, 2020). The suppressed hydrological sensitivity is counteracted by rapid precipitation responses in most GHG-forced simulations, explained by increases in surface downward longwave radiation due to CO2 increases that rapidly warm the land, destabilize the troposphere and strengthen vertical motion in the short term (Chadwick et al., 2014; T.B. Richardson et al., 2016, 2018a). There is medium understanding of how land–sea warming contrast governs rapid precipitation responses based on idealized modelling that shows similar spatial patterns of precipitation response to radiative forcing from GHGs, solar forcing and absorbing aerosols (Xie et al., 2013; Samset et al., 2016; Kasoar et al., 2018). Rapid precipitation adjustments to CO2 have been counteracted by cooling from anthropogenic aerosol increases over land (Box 8.1) but this compensation is expected to diminish as aerosol forcing declines (T.B. Richardson et al., 2018a). Thefast and slow precipitation responses over global land combine on average during transient climate change (Figure 8.3d). This explains a consistent land and ocean mean precipitation increase in projections (Table 4.3) but this is determined by a complex and model-dependent evolution of continental water cycle changes over space and time.

Increases in global precipitation over time, as the climate warms, are partly offset by the overall cooling effects of anthropogenic aerosol and by rapid atmospheric adjustments to increases in GHGs and absorbing aerosol. This explains why multi-decadal trends in global precipitation responses in the satellite era (Adler et al., 2017; Allan et al., 2020) are small and difficult to interpret given observational uncertainty, internal variability and volcanic forcings. The delayed warming effect of rising CO2 concentration, combined with declining aerosol cooling, are expected to increase the importance of the slow temperature-related effects on the energy budget relative to the more rapid direct radiative forcing effects as transient climate change progresses (Shine et al., 2015; Salzmann, 2016; Myhre et al., 2018b).

In summary, there is high confidence that global mean evaporation and precipitation increase with global warming, but the estimated rate is model-dependent (very likely range of 13 % °C–1). The global increase in precipitation is determined by a robust response to global surface temperature only (very likely 2–3% °C–1) that is partly offset by fast atmospheric adjustments to the vertical profile of atmospheric heating by GHGs and aerosols. Global precipitation increases due to GHGs are offset by the well-understood overall surface radiative cooling effect by aerosols (high confidence). Over land, the average warming-related increase in precipitation is expected to be smaller than over the ocean due to increasing land ocean thermal contrast and surface feedbacks, but the overall precipitation increase over land is generally reinforced by fast atmospheric responses to GHGs that strengthens convergence of winds (medium confidence). Global mean precipitation and evaporation increase at a lower rate than atmospheric moisture per 1°C of global warming (high confidence), leading to longer water vapour lifetime in the atmosphere and driving changes in precipitation intensity, duration and frequency and an overall intensification but not acceleration of the global water cycle.

8.2.2 Constraints on the Regional Water Cycle Thermodynamic Constraints on Atmospheric Moisture Fluxes

A warming climate drives increases in atmospheric moisture and horizontal moisture transport from the divergent to the convergent portions of the atmospheric circulation (including storm systems, the tropical rain belt and monsoons) that on average amplifies existing precipitation minus evaporation (P–E) patterns (Held and Soden, 2006). Increased latent heat transports in high latitudes also contribute to polar amplification of warming (Section Although convergent parts of the atmospheric circulation are expected to become wetter (in terms of increasing P–E) and net evaporative regions drier (increasing E–P) these regions are not geographically and seasonally fixed and their location and timing are expected to alter (Section Atmospheric and ocean circulation changes overall decrease the amplification of P–E and salinity patterns. Paleoclimate evidence confirms that during the LGM zonal mean changes were roughly in agreement with thermodynamic expectations (G. Li et al., 2013). However regional changes can be dominated by dynamics, including responses to the large Northern Hemisphere (NH) ice sheets (DiNezio and Tierney, 2013; T. Bhattacharya et al., 2017; Scheff et al., 2017; D’Agostino et al., 2019; Lowry and Morrill, 2019) such that altered P–E patterns are not well described by thermodynamic drivers (Osteret al., 2015; Lora, 2018; Morrill et al., 2018).

There is robust evidence and high agreement across thermodynamics, detailed modelling and observations that amplification of P–E patterns occurs over the oceans (Figure 8.5a) with an associated ‘fresh gets fresher, salty gets saltier’ signature in ocean salinity (Sections and 3.5.2). This amplification is moderated by proportionally larger increases in subtropical ocean evaporation and weakening of the tropical circulation (Section, an expectation supported by observations (Skliriset al., 2016) and process understanding (Yang andRoderick, 2019). Thermodynamics explain a smaller low latitude evaporation increase (1% °C–1) than in high latitudes (5% °C–1) with changes in surface radiation, boundary layer adjustments and ocean heat uptake playing a secondary role, based on idealized modelling (Siler et al., 2019). Increased evaporation from warmer oceans and lakes is exacerbated by the loss of surface ice in some regions (Bintanja and Selten, 2014; Laîné et al., 2014; W. Wang et al., 2018; Sharma et al., 2019; Woolway et al., 2020). This can generate a more local moisture source for precipitation, for example in north-west Greenland during non-summer months since the 1980s (Nusbaumer et al., 2019), though moisture transport changes can counteract this effect (Nygård et al., 2020). Ocean stratification due to heating of the upper layers through radiative forcing has been identified as a mechanism that further amplifies surface salinity patterns beyond the responses driven by water cycle changes alone (Zika et al., 2018).

Figure 8.5 | Zonally-averaged annual mean changes in precipitation minus evaporation (P–E) over (a) ocean and (b) land between the historical (1995–2014) andSSP2-4.5 (2081–2100) CMIP6 simulations (blue lines, an average of the CanESM5 and MRI-ESM2-0 models). Dashed lines show estimated P–E changes using a simple thermodynamic scaling (Held and Soden, 2006); dotted lines show estimates using an extended scaling (Byrne and O’Gorman, 2016). All curves have been smoothed in latitude using a three grid-point moving-average filter. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

Since AR5, numerous studies have confirmed that changes in P–E with warming over land cannot be interpreted simply as a ‘wet regions get wetter, dry regions get drier’ response (Chadwick et al., 2013; Greve et al., 2014; Roderick et al., 2014; Byrne and O’Gorman, 2015; Scheff and Frierson, 2015). Firstly, P–E is a simplistic diagnostic of the water cycle that inadequately describes ‘dryness’ or aridity (Fu and Feng, 2014; Roderick et al., 2014; Greve and Seneviratne, 2015; Scheff and Frierson, 2015; Greve et al., 2019; Vicente-Serrano et al., 2020). Secondly, terrestrial P–E is generally positive and balanced by surface runoff and percolation into subsurface soils and aquifers (Figure 8.1). As a result, the simple thermodynamic scaling (Figure 8.5b) predicts that P–E over land will become more positive (wetter) with warming (Greve et al., 2014; Roderick et al., 2014; Byrne and O’Gorman, 2015). This is not necessarily true, however, in the dry seasons and regions where terrestrial water is lost to the atmosphere and exported (Sheffield et al., 2013; Kumar et al., 2015; Keune and Miralles, 2019). Thirdly, regional P–E patterns over land are affected by changes in atmospheric circulation, oceanic moisture supply and land surface feedbacks. As the land warms more than oceans, spatial gradients in temperature and relative humidity influence moisture supply and reduce P–E over some land regions, such as southern Chile and Argentina around 30°S50°S as captured by an extended thermodynamic scaling (Figure 8.5b). Drying of soils can be amplified by vegetation responses (Berg et al., 2016; Byrne and O’Gorman, 2016; Lambert et al., 2017) but limited by atmospheric circulation feedbacks (Zhou et al., 2021). Changes in soil moisture and rainfall intensity (Sections and can alter the partitioning of precipitation between evaporation and runoff, further complicating terrestrial P–E responses (Short Gianotti et al., 2020).

The strong physical basis for regionally and seasonally dependent responses of P–E and the expectation for an increasing contrast between wet and dry seasons and weather regimes is supported byhigh agreement across multiple observational and CMIP5/CMIP6 modelling studies (Liu and Allan, 2013; Kumar et al., 2015; Polson and Hegerl, 2017; Ficklin et al., 2019; Deng et al., 2020; Schurer et al., 2020). Increased moisture transports into storm systems, monsoons and high latitudes increase the intensity of wet events (Section, while stronger atmospheric evaporative demand with warming (Scheff and Frierson, 2014; Vicente-Serrano et al., 2018; Cook et al., 2019) is an important mechanism for intensifying dry events (Section and decreasing soil moisture over many subtropical land regions. However, aridification is modulated regionally by poleward migration of the subtropical dry zones and an increasing land ocean temperature contrast that drives declining relative humidity (Section

To summarize, increased moisture transport from evaporative oceans to high precipitation regions of the atmospheric circulation will drive amplified P–E and salinity patterns over the ocean (high confidence) while more complex regional changes are expected over land. Greater warming over land than ocean alters atmospheric circulation patterns and on average reduces continental near-surface relative humidity which along with vegetation feedbacks can contribute to regional decreases in precipitation (high confidence). Based on an improved understanding of thermodynamic drivers since AR5 and multiple lines of evidence, there is high confidence that very wet or dry seasons and weather patterns will intensify in a warming climate such that wet spells become wetter and dry spells drier. Large-scale Responses in Atmospheric Circulation Patterns

Responses of the large-scale atmospheric circulation to a warming climate are not as well understood as thermodynamic drivers (Shepherd, 2014). The AR5 identified robust features including a weakening and broadening of tropical circulation with poleward movement of tropical dry zones and mid-latitude jets (Collins et al., 2013). These can dominate regional water cycle changes, affecting the availability of freshwater and the occurrence of climate extremes. Atmospheric circulation changes generally dominate the spatial pattern of rapid precipitation adjustments (Section 8.2.1) to different forcing agents in the tropics (Bony et al., 2013; He and Soden, 2015; T.B. Richardson et al., 2016, 2018a; Tian et al., 2017; X. Li et al., 2018). Radiative forcing with heterogeneous spatial patterns such as ozone and aerosols (including cloud interactions; Section 6.4.1 and Box 8.1) drive substantial responses in regional atmospheric circulation through uneven heating and cooling effects(L. Liu et al., 2018; Dagan et al., 2019b; Wilcox et al., 2019). Changes in atmospheric circulation are also driven by slower, evolving patterns of warming and associated changes in temperature and moisture gradients (Bony et al., 2013; Samset et al., 2016, 2018a; Ceppi et al., 2018; Ma et al., 2018). There is strong evidence that large regional water cycle changes arise from the atmospheric circulation response to radiative forcings and associated SST pattern evolution butlow agreement in the sign and magnitude (Chadwick et al., 2016b). The role of prolonged weather regimes in determining wet and dry extremes is also better understood since AR5 (Kingston and McMecking, 2015; Schubert et al., 2016; D. Richardson et al., 2018; Barlow et al., 2019). Advances in knowledge of expected large-scale dynamical responses of the water cycle are further assessed in this section (see also Figure 8.21).

Long-term weakening of the tropical atmospheric overturning circulation is expected as climate warms in response to elevated CO2 (Collins et al., 2013). A weaker circulation is required to reconcile global mean low-level water vapour increases (around 7% °C–1) with the smaller global precipitation responses of about 1–3% °C–1 (Section 8.2.1). The slowdown can occur in both the Hadley and Walker circulations, but occurs preferentially in the Walker circulation in most climate models (Vecchi and Soden, 2007) but this response has been questioned on the basis of model bias in east Pacific SST (Seager et al., 2019a). Weakening is expected to drive P–E decreases over the western Pacific and increases over the eastern Pacific. However, the driving mechanisms for Walker circulation weakening differ to those involved in determining ENSO variability, so it is too simplistic to interpret changes as an El Niño pattern of regional hydrological cycle extremes (Sohn et al., 2019). Internal variability is also capable of temporarily strengthening the Walker circulation (Section; L’Heureux et al., 2013; Chung et al., 2019) while regional responses depend on the pattern of warming (Sandeep et al., 2014).

Model simulations show a stronger Pacific Walker circulation during the LGM in response to a cooler climate (consistent with an expected weakening in a warmer climate), but a weaker Indian Ocean east west circulation in response to the exposure of the Sunda and Sahul shelves due to lowered sea level (DiNezio et al., 2011). The latter effect is detectable in proxies for hydroclimate, as well as salinity and sea surface temperature (DiNezio and Tierney, 2013; DiNezio et al., 2018). More relevant to future warming is the mid-Pliocene period (3 million years ago), the last time the Earth experienced CO2 levels comparable to present (see Cross-Chapter Box 2.4). Sea surface temperature (SST) reconstructions show a weakening of the Pacific zonal gradient and a pattern of warmth consistent with a weaker Walker cycle response (Corvec and Fletcher, 2017; Tierney et al., 2019; McClymont et al., 2020). Although the Pliocene SST pattern and wet subtropics contrast with present conditions (Burls and Fedorov, 2017), the paleoclimate record strengthens evidence that a warmer climate is associated with a weaker Walker circulation (Cross-Chapter Box 2.4; Section 3.3.3).

Since AR5, weakening of the tropical circulation has been explained as a rapid response to increasing CO2 concentrations and slower response to warming and evolving SST patterns (He and Soden, 2017; Xia and Huang, 2017; Shaw and Tan, 2018; Chemke and Polvani, 2020). Large-scale tropical circulation weakens by 34% in a rapid response to a quadrupling of CO2 concentrations (Plesca et al., 2018), which suppresses tropospheric radiative cooling, particularly in subtropical ocean subsidence regions (Bony et al., 2013; Merlis, 2015; Richardson et al., 2016). The resulting increased atmospheric stability explains the rapid weakening of the Walker circulation (Wills et al., 2017) and Northern Hemisphere Hadley Cell (Chemke and Polvani, 2020). Subsequent surface warming contributes up to a 12% slowing of circulation for a uniform 4°C SST increase, driven by thermodynamic decreases in temperature lapse rate (Plesca et al., 2018).

The regional Inter-tropical Convergence Zone (ITCZ) position, width and strength determine the location and seasonality of the tropical rain belt. Since AR5, multiple studies have linked cross-equatorial energy transport to the mean ITCZ position (Donohoe et al., 2013; Frierson et al., 2013; Bischoff and Schneider, 2014; Boos and Korty, 2016; Loeb et al., 2016; Adam et al., 2018; Biasutti and Voigt, 2019). Multi-model studies agree that aerosol cooling in the NH led to a southward shift in the ITCZ and tropical precipitation after the 1950s up to the 1980s that is linked with the 1980s Sahel drought (Box 8.1; Section and In particular, aerosol-cloud interaction was identified as a potentially important driver of this shift (Chung and Soden, 2017) but this is uncertain since observations suggest that models may overestimate (Malavelle et al. , 2017; Toll et al. , 2017) or underestimate (Rosenfeld et al. , 2019) the aerosol cloud-mediated cooling effects. In addition, greenhouse gas forcing has been invoked in explaining much of the increase in Sahel precipitation since the 1980s through enhanced meridional temperature gradient, with only a secondary role for aerosol (Dong and Sutton, 2015).

Understanding of how ITCZ width and strength respond to a warming climate has improved since AR5 (Byrne and Schneider, 2016; Harrop and Hartmann, 2016; Popp and Silvers, 2017; Dixit et al., 2018; Zhou et al., 2020). Studies suggest that convection gets stronger and more focused within the core of the ITCZ (Lau and Kim, 2015; Byrne et al., 2018). This leads to drying on the equatorward edges of the ITCZ and a moistening tendency in the ITCZ core (Byrne and Schneider, 2016). Feedbacks involving clouds have been identified as an important mechanism leading to tightening and strengthening of the ITCZ (Popp and Silvers, 2017; Su et al., 2017, 2019, 2020; Talib et al., 2018). Stronger ascent in the core amplifies the ‘wet get wetter’ response while reduced moisture inflow near the ITCZ edges reduces this response below the 7% °C–1 Thermodynamic increase in moisture transport. Thus, there is a range of evidence and medium agreement for strengthening and contraction of the ITCZ with warming that sharpens contrasts between wet and dry regimes. However, understanding of how the regional ITCZ location responds in a warming climate is not robust (Section with limited evidence of distinct regional responses to GHG forcing including a northward shift over eastern Africa and the Indian Ocean and a southward shift in the eastern Pacific and Atlantic oceans (Mamalakis et al., 2021). Paleoclimate evidence highlights the distinct regional ITCZ responses to hemispheric asymmetry in volcanic and orbital forcing (McGee et al., 2014; Boos and Korty, 2016; Colose et al., 2016; Denniston et al., 2016; PAGES Hydro2K Consortium, 2017; Singarayer et al., 2017; Atwood et al., 2020) and rapid (>1° latitude over decades) shifts in the ITCZ and regional monsoons in response to AMOC collapse cannot be ruled out (Sections and 5.1.3).

Monsoons are key components of the tropical overturning circulation that can be understood as a balance between net energy input (e.g., radiative and turbulent fluxes) and the export of moist static energy. This is determined by contrasting surface heat capacity between ocean and land and modified through changes in atmospheric dynamics, tropical tropospheric stability and land surface properties (Jalihal et al., 2019). Thermodynamic increases in moisture transport are expected to increase monsoon strength and area (Christensen et al., 2013). Since AR5, evidence continues to demonstrate that monsoon circulation is sensitive to spatially varying radiative forcing by anthropogenic aerosols (Hwang et al. , 2013; R.J. Allen et al. , 2015; Z. Li et al. , 2016b) and GHGs (Dong andSutton, 2015). Changes in SST patterns also play a role (Guo et al., 2016; W. Zhou et al., 2019; Cao et al., 2020) by altering cross-equatorial energy transports and land–ocean temperature contrasts. This evidence continues to support a thermodynamic strengthening of monsoon precipitation that is partly offset by slowing of the tropical circulation but with weak evidence and low agreement for regional aspects of circulation changes. Disagreement between paleoclimate and modern observations, physical theory and numerical simulations of global monsoons have been partly reconciled (Section through improved understanding of regional processes (Harrison et al. , 2015; R. Bhattacharya et al. , 2017; Bhattacharya et al. , 2018; Biasutti et al. , 2018; D’Agostino et al. , 2019; Jalihal et al. , 2019; Seth et al. , 2019), although interpreting past changes in the context of future projections requires careful account of differing forcings and feedbacks (D’Agostinoet al., 2019). Assessment of past changes and future projections in regional monsoons are provided in Sections, and

Since AR5, understanding of poleward expansion of the Hadley Cells has improved (Section but its role in subtropical drying is limited to the zonal mean and dominated by ocean regions (Byrne and O’Gorman, 2015; Grise and Polvani, 2016; He and Soden, 2017; Schmidt and Grise, 2017; Siler et al., 2018; Chemke and Polvani, 2019; Grise and Davis, 2020). Over subtropical land, evolving SST patterns and land–ocean warming contrasts, that are partly explained by rapid responses to CO2 increases, can dominate aspects of the atmospheric circulation response (Byrne and O’Gorman, 2015; He and Soden, 2015; Chadwick et al., 2017; H. Yang et al., 2020) and resultant regional water cycle changes, particularly for projected drying in semi-arid, winter-rainfall dominated subtropical climates (Deitch et al. , 2017; Brogli et al. , 2019; Seager et al. , 2019b; Zappa et al. , 2020). Poleward expansion of the tropical belt is expected to drive a corresponding shift in mid-latitude storm tracks, but the controlling mechanisms differ between hemispheres. Southern Hemisphere expansion is driven by GHG forcing and amplified by stratospheric ozone depletion, while weaker Northern Hemisphere expansion in response to GHG forcing is modulated by tropospheric ozone and aerosol forcing, particularly black carbon (Davis et al. , 2016; Grise et al. , 2019; Watt-Meyer et al. , 2019; Zhao et al. , 2020). However, internal variability is found to dominate observed responses in the NH, precluding attribution to radiative forcing (D’Agostino et al., 2020a). Paleoclimate evidence of poleward expansion and weakening of westerly winds in both hemispheres in the warmer Pliocene is linked to reduced equator-to-pole thermal gradients and ice volume(Abell et al., 2021).

The influence of amplified Arctic warming on mid-latitude regional water cycles is not well understood based on simple physical grounds due to the large number of competing physical processes (Cross-Chapter Box 10.1). The thermal gradient between polar and lower latitude regions decreases at lowlevels due to Arctic warming amplification. However, at higher altitudes, the corresponding thermal gradient increases with warming due to cooling of the Arctic stratosphere and this is consistent with a strengthening of the winter jet stream in both hemispheres, yet there is low agreement on the precise mechanisms (Vallis et al., 2015; Vihma et al., 2016). Changes in the strength of the polar stratospheric vortex can also alter the mid-latitude circulation in winter, but responses are not consistent across models (Oudar et al., 2020a). Nevertheless, thermodynamic strengthening of moisture convergence into weather systems and polar regions is robust (Section and remains valid despite weak understanding of atmospheric circulation change.

In summary, there is high confidence that altered atmospheric wind patterns in response to radiative forcing and evolving surface temperature patterns will affect the regional water cycle in most regions. Mean tropical circulation is expected to slow with global warming (high confidence) but temporary multi-decadal strengthening is possible due to internal variability (medium confidence). Slowing of the tropical circulation reduces the meridional P–E gradient over the Pacific and can partly offset thermodynamic amplification of P–E patterns and strengthening of monsoons (high confidence) but regional characteristics of tropical rain belt changes are not well understood. There is medium confidence in processes driving strengthening and tightening of the ITCZ that increase the contrasts between wet and dry tropical weather regimes and seasons. There is high confidence in understanding of how radiative forcing and global warming drive a poleward expansion of the subtropics and mid-latitude storm tracks but onlylow confidence in how poleward expansion influences drying of subtropical and mid-latitude climates. There is low confidence in understanding how Arctic warming amplification affects mid-latitude regional water cycles but high confidence that thermodynamic strengthening of precipitation within weather systems and in monsoons and polar regions is robust to large-scale circulation changes.

8.2.3 Local-scale Physical Processes Affecting the Water Cycle

Processes operating at local scales are capable of substantially modifying the regional water cycle. This section assesses the development in understanding of processes affecting the atmosphere, surface and subsurface, including cryosphere and biosphere interactions and the direct impacts of human activities. Hydrological Processes Related to Ice and Snow

Declining ice-sheet mass, glacier extent and Northern Hemisphere (NH) sea ice, snow cover and permafrost (Collins et al., 2013; Vaughan et al., 2013) is an expected consequence of a warming climate (Sections 2.3.2, 3.4, and 9.39.5). A decline in mountain snow cover and increased snow and glacier melt will alter the amount and timing of seasonal runoff in mountain regions (Sections 3.4.2, 3.4.3 and 9.5). Earlier and more extensive winter and spring snowmelt (X. Zeng et al., 2018) can reduce summer and autumn runoff in snow-dominated river basins of mid–high latitudes of the NH (Rhoades et al., 2018; Blöschl et al., 2019). Since AR5, an earlier but less rapid snowmelt has been explained by reduced winter snowfall and less intense solar radiation earlier in the season (Musselman et al. , 2017; Wu et al. , 2018; Grogan et al. , 2020). Reduced snow cover also increases energy available for evaporation, which can dominate declining river discharge based on modelling of the Colorado River (Milly and Dunne, 2020). An increase in the fraction of precipitation falling as rain compared with snow can lead to declines in both streamflow and groundwater storage in regions where snowmelt is the primary source of recharge (Earman and Dettinger, 2011; Berghuijs et al., 2014). Such regions include western South America and western North America, semi-arid regions which rely on snowmelt from high mountain chains (Ragettli et al., 2016; Milly and Dunne, 2020). Rain-on-snow melt events reduce at lower altitudes due to declining snow cover but increase at higher altitudes where snow tends to be replaced by rain based on observations and modelling (Musselman et al., 2018; Pall et al., 2019), thereby altering seasonal and regional characteristics of flooding (Section 11.5).

Seasonal melt water from high mountain glaciers in Asia (see Cross-Chapter Box 10.4) supply the basic needs of 221 ± 97 million people (Pritchard, 2019; Immerzeel et al., 2020). Glacier-melt in response to warming can initially lead to increased runoff volumes, especially in peak summer flows, but they will eventually decline as most glaciers continue to shrink. SROCC concluded there is high confidence that the peak runoff has already been passed for some smaller glaciers (Hock et al., 2019a). Increased precipitation and glacier-melt can also contribute to rising lake levels and flood hazards in regions such as the inner Tibetan Plateau, Patagonia, Peru, Alaska and Greenland (Lei et al. , 2017; Shugar et al. , 2020; Stuart-Smith et al. , 2020). Since AR5, evidence from multiple locations (New Zealand, Greenland, Antarctica) shows that intrusions of warm, moist air are important in controlling glacier mass balance, the likelihood of extreme ablation or snowfall events depending on air temperature (Gorodetskaya et al. , 2014; Mackintosh et al. , 2017; Mattingly et al. , 2018; Little et al. , 2019; Oltmanns et al. , 2019; Wille et al. , 2019; Adusumilli et al. , 2021). Sensible heating from warm air and increased longwave radiation from atmospheric moisture and low clouds drive melt events (Stuecker et al., 2018).

Reductions in snow, freshwater ice and permafrost affect terrestrial hydrology. Permafrost degradation reduces soil ice and alters the extent of thermokarst lake coverage (Section 9.5.2; M. Meredith et al., 2019). A lag between current climate change and permafrost degradation is expected, given the slow response rates in frozen ground and the fact that snow cover insulates soil from sensible heat exchanges with the air above (Hoegh-Guldberg et al. , 2018; García-García et al. , 2019; Soong et al. , 2020). Post‐wildfire areas are also linked with permafrost degradation in the Arctic based on satellite observations (Yanagiya and Furuya, 2020). An increase in spring rainfall can increase heat advection by infiltration, exacerbating permafrost thaw and leading to increased methane emissions (Section 5.4.7; Neumann et al., 2019). Increased heat transport by Arctic rivers can also contribute to earlier sea ice melt (Park et al., 2020).

In summary, it is virtually certain that warming will cause a loss of frozen water stores, except in areas where temperatures remain below 0°C for most of the year. There is high confidence that warming and reduced snow volume drives an earlier snowmelt, leading to seasonally dependent changes in streamflow. There is medium confidence that weaker sunlight earlier in the season can reduce the rate of snowmelt. Melting of snowpack or glaciers can increase streamflow in high-latitude and high-altitude catchments until frozen water reserves are depleted (high confidence). There is high confidence that warm, moist airflows and associated precipitation dominate glacier mass balance in some regions (New Zealand, Greenland, Antarctica). Processes Determining Heavy Precipitation and Flooding

Evidence that heavy precipitation events (from sub-daily up to seasonal time scales) intensify as the planet warms has strengthened since AR5 (Section 11.4, Box 11.1 and Cross-Chapter Box 3.2) based on improved physical understanding, extensive modelling and increasing observational corroboration (O’Gorman, 2015; Fischer and Knutti, 2016; Neelin et al., 2017). There is robust evidence, with medium agreement across a range of modelling and observational studies, of thermodynamic intensification of wet seasons (Chou et al., 2013; Liu and Allan, 2013; Dunning et al., 2018; Lan et al., 2019; Zhang and Fueglistaler, 2019). Extreme daily precipitation is expected to increase at close to the 7% °C–1 increase in the near-surface atmospheric moisture-holding capacity determined by the Clausius–Clapeyron equation (Section 11.4, Figure 8.4), with limited evidence that higher rates apply for shorter duration precipitation events (Formayer and Fritz, 2017; Lenderink et al. , 2017; Ali et al. , 2018; Guerreiro et al. , 2018; Burdanowitz et al. , 2019; W. Zhang et al. , 2019a). However, observed estimates sample multiple synoptic weather states, mixing thermodynamic and dynamic factors, so are not directly relatable to climate change responses (Bao et al., 2017; Drobinski et al., 2018). The contrasting spatial scales sampled by the observations and models (from global to cloud resolving) explain the large range of daily and sub-daily precipitation scaling with temperature assessed in Figure 8.4.

Since AR5, advances in understanding the expected changes in intense rainfall at the sub-daily time scale (Section 11.4, Figure 8.4) are provided by idealized or high resolution model experiments and observations (Westra et al., 2014; Fowler et al., 2021). There is robust evidence from simplified calculations, convection resolving models and observations that thermodynamics drives an increase in convective available potential energy (CAPE) with warming and therefore the intensity of convective storms (Singh and O’Gorman, 2013; Romps, 2016; Barbero et al., 2019). Also, declining relative humidity over land (Sections and increases lifting condensation level, thereby delaying but intensifying convective systems (Louf et al., 2019; J. Chen et al., 2020a). Larger systems are linked with increasing tropopause height (Lenderink et al., 2017) that can also amplify storm precipitation (Prein et al., 2017). However, the heaviest rainfall is not necessarily associated with the most intense (deepest) storms based on satellite data (Hamada et al., 2015; Hamada and Takayabu, 2018). Precipitation intensification can exceed thermodynamic expectations where and when additional latent heating invigorates individual storms (Section 11.4.1) as implied by medium agreement across modelling and observational studies (Berg et al. , 2013; Molnar et al. , 2015; Scoccimarro et al. , 2015; Prein et al. , 2017; Zhou and Wang, 2017; Nie et al. , 2018; Kendon et al. , 2019; Z. Zhang et al. , 2019). This intensification depends on time of day, based on convection-permitting simulations (E.P. Meredith et al., 2019).

Intensification of sub-daily rainfall is inhibited in regions and seasons where available moisture is limited (Prein et al., 2017). However, a fixed threshold temperature above which precipitation is limited by moisture availability is not supported by modelling evidence (Neelin et al., 2017; Prein et al., 2017). Enhanced latent heating within storms can also suppress convection at larger scales due to atmospheric stabilization as demonstrated with high resolution, idealized and large ensemble modelling studies (Loriaux et al. , 2017; Chan et al. , 2018; Nie et al. , 2018; Tandon et al. , 2018; Kendon et al. , 2019). Stability is also increased by the direct radiative heating effect of higher CO2 concentrations (Baker et al., 2018) and influenced by aerosol effects on the atmospheric energy budget and cloud development (Box 8.1). Since AR5, modelling evidence shows increases in convective precipitation extremes are limited by droplet/ice fall speeds (Singh and O’Gorman, 2014; Sandvik et al., 2018) but these processes are only crudely represented (Tapiador et al., 2019a). Idealized regional and coupled global models combined with limitedobservational evidence shows that instantaneous precipitation extremes are sensitive to microphysical processes, while daily extremes are determined more by the degree of convective aggregation (Bao and Sherwood, 2019; Pendergrass, 2020a).

Dynamical changes modify and can dominate thermodynamic drivers of local rainfall and flood hazard change (Box 11.1). For example, increased land ocean temperature gradients (Section explain more intense rain from convective systems over the Sahel based on satellite data since the 1980s (Taylor et al., 2017) and dynamical feedbacks can invigorate active to break phase transition over India (Karmakar et al., 2017; Roxy et al., 2017). Satellite data shows long-lived, organized mesoscale convective systems contribute disproportionally to extreme tropical precipitation (Roca and Fiolleau, 2020). Since AR5, the spatial variability in soil moisture has been linked with the timing and location of convective rainfall by altering the partitioning between latent and sensible heating. This was demonstrated for the Sahel, Europe and India in observations (C.M. Taylor et al., 2013; Taylor, 2015; Petrova et al., 2018; Barton et al., 2020; Klein and Taylor, 2020) but depends on the moisture-convergence regime (Welty et al., 2020). Only high-resolution convection-permitting models can capture the sub-grid scale mechanisms for convective initiation (C.M. Taylor et al. , 2013; H. Moon et al. , 2019). There is medium evidence that greater tropical cyclone rainfall totals can be caused by dynamical feedbacks (Chauvin et al., 2017) and slower propagation speed as tropical circulation weakens (Kossin, 2018). These processes amplify the thermodynamic intensification of rainfall (Section, yet observational support is weak (Chan, 2019; Lanzante, 2019; I.J. Moon et al. , 2019; Knutson et al. , 2020). Slower decay following landfall, explained by larger stores of heat and moisture at higher SSTs, can also amplify rainfall amount based on observations and modelling (Li and Chakraborty, 2020). Rainfall intensity from the outer rain bands of tropical cyclones is also increased by aerosolcloud interactions (Box 8.1).

The amount and intensity of rainfall within extratropical storms is expected to increase with atmospheric moisture. This is particularly evident for atmospheric rivers (see Glossary) and research since AR5 has confirmed their link with flooding and terrestrial water storage (Froidevaux and Martius, 2016; Paltan et al. , 2017; Waliser and Guan, 2017; Adusumilli et al. , 2019; Ionita et al. , 2020; Payne et al. , 2020). There is robust evidence based on simple physics and detailed modelling that extratropical cyclone rainfall, including atmospheric river events, will intensify through increased atmospheric moisture flux (Lavers et al., 2013; Ramos et al., 2016; Yettella and Kay, 2017; V. Espinoza et al., 2018; Algarra et al., 2020; Xu et al., 2020; Zavadoff and Kirtman, 2020; Zhao, 2020), although changes in dynamical aspects will modify responses regionally (Section For example, stronger latitudinal temperature gradients in the high-latitude upper troposphere drive increased extratropical storm speed around 30°N70°N based on CMIP5 simulations (Dwyer and O’Gorman, 2017), causing reduced precipitation accumulation.

The response of flood hazard to changing rainfall characteristics depends on time and space scale and the nature of the land surface (Section 11.5.1 and FAQ 8.2). Sustained and heavy rainfall can lead to widespread flooding and landslides while intensification of short-duration intense rainfall can increase the severity and frequency of flash flooding (Marengo et al., 2013; Chan et al., 2016; Gariano and Guzzetti, 2016; Sandvik et al., 2018). Flooding events in many tropical regions (e.g., north-western South America, southern Africa and Australasia) are associated with ENSO variability (Emerton et al., 2017; Takahashi and Martínez, 2019; Pabón-Caicedo et al., 2020) and amplified by thermodynamic increases in water vapour. Flood hazard from heavy rainfall is modulated by snowmelt (Section, vegetation characteristics (Page et al., 2020; Murphy et al., 2021) and direct human intervention (Sections and FAQ 8.2) but also can be compounded by sea level rise (Sections and 9.6.4) in coastal and delta regions (Bevacqua et al., 2019; Ganguli and Merz, 2019; Eilander et al., 2020). Antecedent soil moisture conditions are an important modulator of flooding (Section 11.5.1) but become less important for smaller catchments and for more severe floods (Wasko and Nathan, 2019). Depleted soil moisture after more intense dry seasons (Section can allow greater uptake of wet season rainfall before soils saturate. Since AR5, evidence confirms that more intense rainfall increases the proportion of runoff and reservoir recharge relative to infiltration into the soil (Eekhout et al., 2018; Yin et al., 2018). More intense but less frequent storms (Kendon et al., 2019) favour focused groundwater recharge through leakage from surface waters (R.G. Taylor et al., 2013a; Cuthbert et al., 2019a) and runoff and flash flooding where the percolation capacity of the soil is exceeded (Yin et al., 2018).

Increased severity of flooding on larger, more slowly-responding rivers is expected as precipitation accumulations increase during persistent wet events over a season. This can occur where atmospheric blocking patterns repeatedly steer extratropical cyclones across large river catchments, as identified for NH mid-latitudes and Asia (Takahashi et al., 2015; Pfleiderer et al., 2018; Zhou et al., 2018; Blöschl et al., 2019; Lenggenhager et al., 2019; Nikumbh et al., 2019; Zanardo et al., 2019), although groundwater flooding and antecedent conditions including soil moisture and snowmelt also play a role (Muchan et al., 2015; Berghuijs et al., 2019). Increased atmospheric moisture amplifies the severity of these events when they occur in a warmer climate, yet drivers of change in the occurrence of blocking patterns, stationary waves and jet stream position are not well understood (Section and Cross-Chapter Box 10.1).

In summary, there is very high confidence that heavy precipitation events will become more intense in a warming climate. There is high confidence that increased moisture and its convergence within extratropical and tropical cyclones and storms will increase rainfall totals during wet events at close to the 7% °C–1 Thermodynamic response, with low confidence of higher rates for sub-daily intensities. There is medium confidence that more intense but less frequent rainfall increases the proportion of rainfall leading to surface runoff and focused groundwater recharge from temporary water bodies. There is low confidence in how the frequency of flooding will change regionally as it is strongly dependent on catchment characteristics, antecedent conditions and how atmospheric circulation systems respond to climate change, which is less certain than thermodynamic drivers (Section 11.5). However, there is high confidence that increases in precipitation intensity and amount during very wet events (from sub-daily up to seasonal time scales) will intensify severe flooding when these extremes occur. Drivers of Aridity and Drought

Regional changes in aridity – broadly defined as a deficit of moisture – are expected to occur in response to anthropogenic forcings as a consequence of shifting precipitation patterns, warmer temperatures, changes in cloudiness (affecting solar radiation), declining snowpack, changes in winds and humidity, and vegetation cover (Figure 8.6). Evapotranspiration (see Annex VII: Glossary) is a key component of aridity, and is composed of two main processes: evaporation from soil, water and vegetation surfaces; and transpiration, the exchange of moisture between plants and atmosphere through plant stomata. On a global level, warmer temperatures increase evaporative demand in the atmosphere, and thus (assuming sufficient soil moisture is available) increase moisture loss from evapotranspiration (high confidence) (Dai et al., 2018; Vicente-Serrano et al., 2020). On a regional level, aridity is further modulated by seasonal rainfall patterns, runoff, water storage, and interactions with vegetation.

Figure 8.6 | Climatic drivers of drought, effects on water availability, and impacts. Plus and minus signs denote the direction of change that drivers have on factors such as snowpack, evapotranspiration, soil moisture, and water storage. The three main types of drought are listed, along with some possible environmental and socio-economic impacts of drought (bottom).

Vegetation is a crucial interface between subsurface water storage (in soil moisture and groundwater) and the atmosphere. Plants alter evapotranspiration and the surface energy balance, and thus can have a large influence on regional aridity (Lemordant et al., 2018). SRCCL concluded there is high confidence that higher atmospheric CO2 increases the ratio of plant CO2 uptake to water loss (water-use efficiency; WUE) through the combined enhancement of photosynthesis and stomatal regulation (Section 5.4.1; DeKauwe et al. , 2013; C.D. Jones et al. , 2013; Deryng et al. , 2016; Swann et al. , 2016; Cheng et al. , 2017; Knauer et al. , 2017; Peters et al. , 2018; Guerrieri et al. , 2019). Modelling studies suggest that increasing WUE can partly counteract water losses from increased evaporative demand in a warmer atmosphere, potentially mitigating aridification (Milly and Dunne, 2016; Bonfils et al. , 2017; Cook et al. , 2018; Y. Yang et al. , 2018). However, observational studies suggest that this effect may be counter-balanced by the increase in plant growth in response to elevated CO2, which results in increased water consumption (De Kauwe et al. , 2013; Donohue et al. , 2013; Ukkola et al. , 2016b; Yang et al. , 2016; Guerrieri et al. , 2019; Mankin et al. , 2019; A. Singh et al. , 2020). In semi-arid regions, increased plant water consumption can reduce streamflow and exacerbate aridification (Ukkola et al. , 2016b; Mankin et al. , 2019; A. Singh et al. , 2020). Thus, there is low confidence that increased WUE in plants can counterbalance increased evaporative demand (Cross-Chapter Box 5.1).

A drought is a period of abnormally dry weather that persists for long enough to cause a serious hydrological imbalance (Glossary; Wilhite and Glantz, 1985; Wilhite, 2000; Cook et al., 2018). Most droughts begin as persistent precipitation deficits (‘meteorological drought’) that propagate over time into deficits in soil moisture, streamflow, and water storage (Figure 8.6), leading to a reduction in water supply (‘hydrological drought’). Increased atmospheric evaporative demand increases plant water stress, leading to ‘agricultural and ecological drought’ (Williams et al. , 2013; C.D. Allen et al. , 2015; Anderegg et al. , 2016; McDowell et al. , 2016; Grossiord et al. , 2020). Evaporative demand affects plants in two ways. It increases evapotranspiration, depleting soil moisture and stressing plants through lack of water (Teuling et al., 2013; Sperry et al., 2016), and also directly affects plant physiology, causing a decline in hydraulic conductance and carbon metabolism, leading to mortality (Figure 8.6; Breshears et al., 2013; Hartmann, 2015; McDowell and Allen, 2015; Fontes et al., 2018). While droughts are traditionally viewed as ‘slow moving’ disasters that typically take months or years to develop, rapidly evolving and often unpredictable flash droughtscan also occur (Otkin et al., 2016, 2018). Flash droughtscan develop within a few weeks, causing substantial disruption to agriculture and water resources (Pendergrass et al., 2020). Conversely, droughts that persist for a long time (usually a decade or more) are called megadroughts. Droughts span a large range of spatial and temporal scales, arise through a variety of climate system dynamics (e.g., internal atmospheric variability, ocean teleconnections), and can be amplified or alleviated by a variety of physical and biological processes. As such, droughts occupy a unique space within the framework of extreme climate and weather events, possessing no singular definition.

While the role of precipitation in droughts is obvious, other climatic drivers are also important, such as temperature, radiation, wind, and humidity (Figure 8.6). These factors have a strong influence on atmospheric evaporative demand, which affects evapotranspiration and soil moisture (Figure 8.6). In snow-dominated regions, high temperatures increase the fraction of precipitation falling as rain instead of snow and advance the timing of spring snowmelt (high confidence) (Vincent et al. , 2015; Mote et al. , 2016, 2018; Berg and Hall, 2017; Solander et al. , 2018). This can result in lower than normal snowpack levels (a ‘snow drought’), and thus reduced streamflow, even if total precipitation is at or above normal for the cold season (Harpold et al., 2017). Plants also affect the severity of droughts by modulating evapotranspiration (Figure 8.6). As discussed above, the effect of elevated CO2 on plants has the potential to both increase and reduce water loss through evapotranspiration via enhanced WUE and plant growth, respectively (Figure 8.6), but there is low confidence in whether one process dominates over another at the global scale.

Drought severity also depends on human activities and decision-making (AghaKouchak et al. , 2015; Van Loon et al. , 2016; Pendergrass et al. , 2020). Societies have developed a variety of strategies to manipulate the water cycle to increase resiliency in the face of water scarcity, including irrigation, creation of artificial reservoirs, and groundwater pumping. While potentially buffering water resource capacity, in some cases these interventions may unexpectedly increase vulnerability (medium confidence). For example, while increased irrigation efficiency may ensure more water is available to crops, the corresponding reduction in runoff and subsurface recharge may exacerbate hydrologic drought (Grafton et al., 2018). Furthermore, while building dams and increasing surface reservoir capacity can boost water resources, they may actually increase drought vulnerability if demands rise to take advantage of the increased supply or if over-reliance on these surface reservoirs is encouraged (Di Baldassarre et al., 2018). Interactions between adaptation, vulnerability, and drought impacts are discussed further in WGII (Chapters 2 and 4).

In summary, there is high confidence that a warming climate drives an increase in atmospheric evaporative demand, decreasing available soil moisture. There is high confidence that higher atmospheric CO2 increases plant water-use efficiency, but low confidence that this physiological effect can counterbalance water losses. Since drought can be defined in a number of ways, there are potentially different responses under a warming climate depending on drought type. Beyond a lack of precipitation, changes in evapotranspiration are critical components of drought, because these can lead to soil moisture declines (high confidence). Under very dry soil conditions, evapotranspiration becomes restricted and plants experience water stress in response to increased atmospheric demand (medium confidence). Human activities and decision-making have a critical impact on drought severity (high confidence). Direct Anthropogenic Influence on the Regional Water Cycle

Human activities influence the regional water cycle directly through modifying and exploiting stores and flows from rivers, lakes and groundwater and by altering land cover characteristics. These actions alter surface energy and water balances through changes in permeability, surface albedo, evapotranspiration, surface roughness and leaf area. Direct redistribution of water by human activities for domestic, agricultural and industrial use of about 24,000 km3yr–1 (Figure 8.1) is equivalent to half the global river discharge or double the global groundwater recharge each year (Abbott et al., 2019). Since AR5, both modelling studies and observations have demonstrated that land use change can drive local and remote responses in precipitation and river flow by altering the surface energy balance, moisture advection and recycling, land sea thermal contrast and associated wind patterns (Alter et al. , 2015; Wey et al. , 2015; De Vrese et al. , 2016; Pei et al. , 2016; Wang-Erlandsson et al. , 2018; Vicente-Serrano et al. , 2019). There is robust evidence that a warming climate combined with direct human demand for groundwater will deplete groundwater resources in already dry regions (Wada and Bierkens, 2014; D’Odorico et al., 2018; Jia et al., 2019).

The SRCCL presented evidence that extraction of water from the ground or river systems and intensive irrigation increases evaporation and atmospheric water vapour locally (Jia et al., 2019; Mishra et al., 2020). Irrigation can explain declining groundwater storage in some regions, including north-western India and North America (Asoka et al., 2017; G. Ferguson et al., 2018). Simulations spanning 1960–2010 indicate that approximately 30% of the present human water consumption is supplied from non-sustainable water resources (Wada and Bierkens, 2014). However, there is onlylimited evidence that groundwater extraction is lowering streamflow (Mukherjee et al., 2018; de Graaf et al., 2019). Model experiments show that irrigation can either aggravate or alleviate climate‐induced changes of surface or subsurface water (Lenget al., 2015). Widespread extraction of water from rivers can reduce flows and decrease the level and area of inland seas and lakes (Wurtsbaugh et al. , 2017; Torres-Batlló et al. , 2020; X. Wang et al. , 2020). Between 1985 and 2015, about 139,000 km2 of inland water areas have become land, while creation of dams has converted about 95,000 km2 of land to water, particularly in the Amazon and Tibetan Plateau (Donchyts et al., 2016). Direct management of river flow is comparable in magnitude to climate change effects for snow-fed rivers at a continental scale based on a global analysis and a study of 96 Canadian catchments (Tan and Gan, 2015; Arheimer et al., 2017).

The SRCCL assessed with medium confidence that mean and extreme precipitation is increased over and downwind of urban areas (Jia et al., 2019). There is medium confidence that altered thermodynamic and aerodynamic properties of the land surface from urbanization affects evaporation and increases precipitation over or downwind of cities (Box 10.3) due to altered stability and turbulence (Han et al. , 2014; Pathirana et al. , 2014; Jiang et al. , 2016; D’Odorico et al. , 2018; Sarangi et al. , 2018; Boyaj et al. , 2020). However, reduced biogenic aerosol, but increased anthropogenic aerosol emissions modify cloud microphysics and precipitation processes (Box 8.1; Schmidand Niyogi, 2017; D’Odorico et al. , 2018; Fan et al. , 2020; Zheng et al. , 2020). Urbanization also decreases permeability of the surface, leading to increased surface runoff (Chen et al., 2017; Jia et al., 2019). Large-scale infrastructure, such as the construction and operation of dikes, weirs, and hydropower plants, also alters surface energy and moisture fluxes, potentially influencing the regional water cycle. Limited modellingevidence suggests that large-scale solar and wind farms can increase precipitation locally (over the Sahel and North America) when dynamic vegetation responses are represented (Y. Li et al., 2018; Pryor et al., 2020), with remote effects also possible (Lu et al., 2021).

Changes in land use from forest to agriculture can exert profound regional effects on the water cycle (FAQ 8.1) by modifying the surface energy balance and moisture recycling (Krishnan et al. , 2016; Paul et al. , 2016; Llopart et al. , 2018; Singh et al. , 2019). There is medium evidence from modelling and observations over the Amazon and East Africa that deforestation drives increased streamflow (Dos Santos et al., 2018; Guzha et al., 2018; Levy et al., 2018) butlimited evidence that increases in global runoff due to deforestation are counterbalanced by decreases resulting from irrigation (Hoegh-Guldberg et al., 2018). Total Amazon deforestation drives reductions in precipitation but with a large 90% confidence range (38 to +5 %) based on 44 primarily pre-AR5 climate model simulations (Spracklen and Garcia-Carreras, 2015) with smaller reductions (2.3 to1.3 %) attributed to observed Amazon deforestation up to 2010. Climate model development has reduced this uncertainty range but has not altered the median change (Lejeune et al., 2015). Large-scale global deforestation (20 million km2) simulated by 9 CMIP6 models confirms a large range in precipitation amount reduction of37 ± 54 mm yr–1 over the deforested regions (Boysen et al., 2020). However, small-scale deforestation can increase precipitation locally (Lawrence and Vandecar, 2015). A 50–60% deforestation rate corresponded to a wet season delay of about one week and greater chance of dry spells of eight days or longer based on correlation analysis of rain gauge and land-use data for South America (Leite-Filhoet al., 2019). Forest and grassland fires can also modify hydrological response at the watershed scale (Havel et al., 2018). Afforestation or reforestation aimed at removing CO2 from the atmosphere can also alter the water cycle at the regional scale (Section 8.4.3 and Cross-Chapter Box 5.1).

In summary, there is high confidence that land-use change and water extraction for irrigation drive local, regional and remote responses in the water cycle. Large-scale deforestation is likely to decrease precipitation over the deforested regions but there is low confidence in the effects of limited deforestation. There is medium confidence that deforestation drives increased streamflow relative to the responses caused by climate change. Urbanization can increase local precipitation (medium confidence) and resulting runoff intensity (high confidence). A warming climate combined with direct human demand for water is expected to deplete groundwater resources in dry regions (high confidence).

Box 8.1 | Role of Anthropogenic Aerosols in Water Cycle Changes

Aerosols affect precipitation in two major pathways, by altering the shortwave and longwave radiation and influencing cloud microphysical properties.

Aerosol radiative effects on precipitation

Aerosols scatter and absorb solar radiation which reduces the energy available for surface evaporation and subsequent precipitation. In addition, cooling is incurred by the radiation that is reflected back to space directly by the aerosols and indirectly by the aerosol effect on cloud brightening. Northern Hemisphere (NH) station data indicate decreasing precipitation trends during the 1950s to the 1980s, which have since partially recovered (Wild, 2012; Bonfils et al., 2020). These changes are attributable with high confidence to anthropogenic aerosol emissions from North America and Europe causing dimming through reduced surface solar radiation. This peaked during the late-1970s and partially recovered thereafter following improved air quality regulations (Section 6.2.1; Box 8.1, Figure 1).

Box 8.1, Figure 1 | Northern Hemisphere surface downward radiation anomalies (W m–2; a) and precipitation anomalies (mm day–1; b) for1951–2014for summer season (May–September) monsoon region (Polson et al. , 2014) from CMIP6 DAMIP experiments. Observed solar radiation anomalies are from GEBA global data from 1961–2014 (Wild et al., 2017) and observed precipitation anomalies are from GPCC and CRU. CMIP6 multi-model mean anomalies are from all-forcings (ALL), greenhouse gas forcing (GHG) and anthropogenic aerosol forcing (AER) experiments. Anomalies are with respect to 1961–1990 and smoothed with a 11-year running mean. Red shading shows the ensemble spread of ALL forcing experiment (5–95% range). Models are masked to the GPCC data set. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

Dimming over the NH causes a relative cooling, compared to the Southern Hemisphere (SH), which induces a southward shift of the northern edge of the tropical rain belt (Section; Allen et al., 2014; Brönnimann et al., 2015). CMIP5 simulations show that most of the cooling is caused by the aerosol cloud-mediated effect (Chung and Soden, 2017). Dimming also weakens monsoon flow and precipitation, offsetting or even overcoming the expected precipitation increase due to increased GHGs (Ayantika et al., 2021). The oceanic response to a weakened monsoon cross-equatorial flow can further weaken the South Asian monsoon through an amplifying feedback loop (Swapna et al., 2012; Krishnan et al., 2016; Patil et al., 2019). These processes partially explain (medium confidence) the southward shift of the NH tropical edge of the tropical rain belt from the 1950s to the 1980s (Allen et al., 2014; Brönnimann et al., 2015) and the severe drought in the Sahel that peaked in the mid-1980s (Rotstayn et al., 2002; Undorf et al., 2018b). These processes also explain (high confidence) the observed decrease of South East Asian monsoon precipitation during the second half of the 20th century (Figure 8.7; Bollasina et al., 2011; Sanap et al., 2015; Krishnan et al., 2016; Lau and Kim, 2017; Lin et al., 2018; Undorf et al., 2018b).

Absorption of solar radiation by anthropogenic aerosols such as black carbon warms the lower troposphere and increases moist static energy, but also results in larger convection inhibition that suppresses light rainfall (Box 8.1, Figure 2; Y. Wang et al., 2013). Release of aerosol-induced instability, often triggered by topographical barriers, produces intense rainfall, flooding (Fan et al., 2015; Lee et al., 2016) and severe convective storms (medium confidence) (Saide et al., 2015). In particular, aerosols induce intense convection at the Himalaya foothills during the pre-monsoon season, which generates a regional convergence there (medium confidence). This mechanism is termed the ‘elevated heat pump hypothesis’ (Lau and Kim, 2006; D’Errico et al., 2015).

Box 8.1, Figure 2 | Schematic depiction of the atmospheric effects of light-absorbing aerosols on convection and cloud formation: (a) without and (b) with the presence of absorbing aerosols in the planetary boundary layer. The dashed and solid blue lines correspond to the vertical temperature profiles in the absence and presence of the absorbing aerosol layer, respectively, and the solid and dashed red lines denote the dry and moist adiabats, respectively. Absorbing aerosols result in an increasing temperature in the atmosphere but a reduced temperature at the surface. The reduced surface temperature and the increased temperature aloft led to a larger negative energy associated with convective inhibition (–) and a higher convection condensation level (CCL) under the polluted conditions. On the other hand, the absorbing aerosol layer induces a larger convective available potential energy (+) above CCL, facilitating more intensive vertical development of clouds, if lifting is sufficient to overcome the larger convective inhibition. Figure from Y. Wang et al. (2013).

Aerosol cloud microphysical effects

Cloud droplets nucleate on pre-existing aerosol particles which act as cloud condensation nuclei (CCN). Anthropogenic aerosols add CCN, compared to a pristine background, and produce clouds with more numerous and smaller droplets, slower to coalesce into raindrops and to freeze into ice hydrometeors at temperatures below 0°C. Adding CCN suppresses light rainfall from shallow and short-lived clouds, but it is compensated by heavier rainfall from deep clouds. Adding aerosols to clouds in extremely clean air invigorates them by more efficient vapour condensation on the added drop surfaces (Koren et al., 2014; Fan et al., 2018). Clouds forming in more polluted air masses (hence with more numerous and smaller drops) need to grow deeper to initiate rain (Freud and Rosenfeld, 2012; Konwar et al., 2012; Campos Braga et al., 2017). This leads to larger amount of cloud water evaporating aloft while cooling and moistening the air there at the expense of the lower levels, which leads to convective invigoration (Dagan et al., 2017; Chua and Ming, 2020), followed by convergence, air mass destabilization and added rainfall in an amplifying feedback loop

Box 8.1

(Abbott and Cronin, 2021). In addition, delaying rain initiation until greater altitudes are reached transports more cloud water above the 0°C altitude and leads to additional release of latent heat of freezing and/or vapour deposition, which in combination with the added latent heat of condensation enhances the cloud updrafts (Fan et al., 2018). The stronger updrafts invigorate mixed-phase precipitation and the resultant hail and cloud electrification (Rosenfeldet al., 2008; Thornton et al., 2017). This includes the outer convective rainbands of tropical cyclones. There is medium confidence that air pollution enhances flood hazard associated with the outer rain bands at the expense of the inner rain bands (Wang et al., 2014; C. Zhao et al., 2018; Souri et al., 2020).

The aerosol effect on invigoration and rainfall from deep convective clouds peaks at moderate levels (aerosol optical depth of 0.2 to 0.3), but reverses into suppression with more aerosols (H. Liu et al., 2019). More generally, the microphysical aerosol-related processes often compensate or buffer each other (Stevens and Feingold, 2009). For example, suppressed rain by slowing drop coalescence enhances mixed-phase precipitation. Therefore, despite the potentially large aerosol influence on the precipitation forming processes, the net outcome of aerosol microphysical effects on precipitation amount has generallylow confidence, especially when evaluated with respect to the background of high natural variability in precipitation (Tao et al., 2012).

Ice nucleating particle (INP) initiate ice precipitation from persistent supercooled water clouds that have cloud droplets too small for efficient warm rain, or expedite mixed-phase precipitation in short-lived supercooled rain clouds (Creamean et al., 2013). Most INPs are desert and soil dust particles, rather than air pollution aerosols (DeMott et al., 2010). Biogenic particles from terrestrial and marine origin are more rare, but important at temperatures above about15°C (Murray et al., 2012; DeMott et al., 2016). Dust particles from long-range transport across the Pacific were found to enhance snow-forming processes over the Sierra Nevada in California (Creamean et al., 2013; Fan et al., 2014). The impact of INPs was demonstrated by glaciogenic cloud seeding experiments, which enhanced orographic supercooled clouds with medium confidence of success (French et al., 2018; Rauber et al., 2019; Friedrich et al., 2020). There are still major gaps in understanding the effects of INPs mainly on deep convective clouds (Kanji et al., 2017; Stanford et al., 2017; Korolev et al., 2020).

8.3 How Is the Water Cycle Changing and Why?

This section focuses on the evaluation and attribution of past and recent water cycle changes using observational datasets, theoretical understanding and model simulations. Paleoclimate records and historical observations provide evidence for past water cycle changes caused both by natural variability and human activities (Haug et al., 2003; Buckley et al., 2010; Pederson et al., 2014). Key elements of the observed water cycle changes are assessed in this section, including flux and storage variations across the atmosphere, the continents and to a lesser extent the ocean and cryosphere, as well as related changes in large-scale atmospheric circulation and modes of variability. Particular emphasis is placed on assessing changes across regions and seasons (Box 8.2). Detailed regional assessments are presented in Chapters 10, 11, 12 and Atlas. Further information concerning large-scale observed water cycle changes and their attribution is available in Sections and 3.3.2.

8.3.1 Observed Water Cycle Changes Based on Multiple Datasets

This section provides a process-based evaluation and a comprehensive assessment of observed water cycle changes by integrating multiple lines of evidence including paleoclimate data, historical datasets, theoretical understanding (Section 8.2) and model simulations. Global Water Cycle Intensity and P–E Over Land and Oceans

The human influence on the global water cycle is often summarized as an intensification (Huntington, 2006; DeAngelis et al., 2015; W. Zhang et al., 2019b) or an overall strengthening which has been observed since at least 1980 (high confidence) (see Chapter 2). There is, however, no unique definition of the global water cycle intensity (Trenberth, 2011; Ficklin et al., 2019; Sprenger et al., 2019). One simple metric is the global and annual mean amount of precipitation. Although an increase in global precipitation is consistent with physical expectations (Section 8.2.1), it has not yet been detected and attributed to human activities given large observational uncertainties and low signal-to-noise ratio (Section Other metrics are more suitable to detect and attribute changes in the global water cycle, including the likely increase in global land precipitation since 1950 (Section which is likely due to a human influence (Section

The flux of freshwater between the ocean and atmosphere is determined by the difference between precipitation and evaporation (P–E). Evaporation is measured in very few locations across the global ocean, so that directly assessing P–E over the ocean is very challenging and relies on indirect reanalysis estimates (Robertson et al., 2020). The AR5 presented robust evidence ofan amplifiedoceanic pattern in P–E since the 1960s from both regional and global surface and subsurface salinity measurements and reanalyses. This pattern is consistent with our theoretical understanding of human-induced changes in the water cycle, leading to the conclusion that these changes are very likely the result of anthropogenic forcings (Section

In contrast, AR5 did not provide a conclusive assessment of observed changes in P–E over land. Continental P–E estimated from reanalyses and data-driven land surface models indicate that interannual variations are linked to ENSO (Robertson et al., 2014, 2020). Increasing trends in P–E since 1979 based on land models are not statistically significant. Observations and models show evidence that P–E increases in the wet parts and decreases in the dry parts of tropical circulation systems, which shift in location seasonally and from year to year, with increases in seasonality since 1979 (see Box 8.2; Chou et al., 2013; Liu and Allan, 2013; Fu and Feng, 2014).

In summary, a low signal-to-noise ratio, observational uncertainties and current data assimilation techniques limit the assessment of recent global trends in P–E over both land and ocean. It is likely that the global land P–E variations observed since the late 1970s were dominated by internal variability, mostly linked to ENSO teleconnections (medium confidence). In contrast, the attribution of changes in sea surface salinity (Section suggests that it is extremely likely that human influence has contributed to the regional changes in P–E observed over the global ocean since the mid-20th century. Water Vapour and Its Transport

The AR5 presented evidence of increases in global near-surface and tropospheric specific humidity since the 1970s but with medium confidence of a slowing of near-surface moistening trends over land associated with reduced relative humidity since the late 1990s. According to AR5, radiosonde, Global Positioning System (GPS) and satellite observations of tropospheric water vapour indicate very likely increases at near global scales since the 1970s occurring at a rate that is generally consistent with the Clausius–Clapeyron relation (about 7% °C–1 at low altitudes) and the observed atmospheric warming (Hartmann et al., 2013).

Since AR5, it is very likely that increases in global atmospheric water vapour were observed based on in situ, satellite and reanalysis data (with medium confidence in the magnitude; Section Satellite records show increases in upper tropospheric water vapour (constant relative humidity while temperatures have increased) since 1979 (E.-S. Chung et al. , 2014; Blunden and Arndt, 2020), to which human influence has likely contributed (Section Combined satellite and reanalysis estimates and CMIP6 atmosphere-only simulations (1988–2014) show global mean precipitable water vapour increases of 6.7 ± 0.3 % °C–1, very close to the Clausius–Clapeyron rate (Allan et al., 2020). Satellite-based products show increases close to the Clausius–Clapeyron rate over the ice-free oceans (about 7 to 9 % °C–1; 19982008), but reanalysis estimates outside this range (Schröder et al., 2019) are an expected consequence of their changing observing systems (Allan et al., 2014; Parracho et al., 2018). Increases in precipitable water vapour are found over the central and sub-Arctic based on multiple reanalyses with some corroboration from sparse, in situ data (Vihma et al., 2016; Rinke et al., 2019; Nygård et al., 2020).

Declining near-surface relative humidity over land areas (e.g., the USA, Mediterranean, South Asia, South America and southern Africa) is evident in surface observations (Willett et al., 2014, 2020; Dunn et al., 2017). This is consistent with a faster rate of warming over land than ocean (Sections and; Byrne and O’Gorman, 2018). CMIP5 simulations underestimate the observed decreases in relative humidity over much of global land during 1979–2015 (Douville and Plazzotta, 2017; Dunn et al., 2017)even when observed SSTs are prescribed (–0.05 to –0.25% per decade compared with an observed rate of –0.4 to –0.8% per decade). It is not yet clear if this discrepancy is related to internal variability or can be explained by deficiencies in models (Vannière et al., 2019; Douville et al., 2020)or observations (Willett et al., 2014). Over the NH mid-latitude continents, there is medium confidence that human influence has contributed to a decrease in near-surface relative humidity in summer(Sections and

Water vapour transport (or convergence) estimates from observations have substantial uncertainties even in regions of high quality radiosonde data. Consequently many studies use reanalyses for water transport estimates instead of instrumental observations. For example, increases in low-level (8001000 hPa) moisture convergence into the tropical wet regime with a smaller outflow increase in the mid-troposphere (400800 hPa) with warming was detected in one reanalysis (ERA-Interim; Allan et al., 2014). Modelling evidence combined with statistical analysis demonstrate consistency between reanalysis moisture convergence and P–E over land (Robertson et al., 2016). Advances in reanalysis representation of atmospheric moisture and winds in addition to new observational isotope analysis have improved the ability to identify the main sources of water vapour for key continental regions and quantify the relative contributions from moisture advection and recycling (Gimeno et al. , 2012; van Der Ent et al. , 2014; Joseph et al., 2016).

Observed changes in moisture transport can also arise from changes in atmospheric circulation as well as thermodynamics. For instance, moisture transport into the Arctic region estimated from reanalyses datasets is consistent with radiosonde data (Dufour et al., 2016), with increases since 1979 linked to atmospheric circulation (Nygård et al., 2020). Moisture transport into the Eurasian Arctic was identified to increase by 2.6% per decade during 19482008 based on a reanalysis estimate (X. Zhang et al., 2013). More intense moist intrusions associated with atmospheric rivers affecting the Arctic and Europe have been documented since 1979, but with a substantial influence from decadal internal variability (Ummenhofer et al., 2017; Mattingly et al., 2018). A recent strengthening of tropical circulation and associated moisture convergence has been identified since around 2000 for the Amazon region (Arias et al. , 2015; Barichivich et al. , 2018; J.C. Espinoza et al. , 2018; X.Y. Wang et al. , 2018). This was also strenghtened by increased moisture transport from the North Atlantic, driving more abundant latent heat release (Segura et al., 2020) and leading to an increased frequency of extreme floods in the northern Amazon (Barichivich et al., 2018; Heerspink et al., 2020). Overall, increased moisture transport has been linked to increased precipitation over wet tropical land areas (Gimeno et al., 2020) and to more extreme and persistent wet and dry weather events (Konapala et al., 2020) in many regions worldwide.

In summary, there is high confidence that human-caused global warming has led to an overall increase in water vapour and moisture transport throughout the troposphere, at least since the mid-1990s. In particular, there is high confidence that moisture transport into the Arctic has increased but only medium confidence in the attribution of such a trend to a human influence. There is medium confidence that human influence has contributed to a decrease in near-surface relative humidity over the Northern Hemisphere mid-latitude continents during summer(see also Sections and Precipitation Amount, Frequency and Intensity

This section assesses observed changes in precipitation at global and regional scales. Note that changes in precipitation seasonality are assessed in Box 8.2 and that changes in regional monsoons are assessed in section where observed changes in both circulation and rainfall are considered. Further assessment of regional changes in precipitation is presented in Chapters 10, 12 and Atlas, while extreme precipitation is presented in Chapter 11.

The AR5 concluded that it is likely there has been an overall increase in annual mean precipitation amount over mid-latitude land areas in the NH, with low confidence since 1901, butmedium confidence after 1951. There is further evidence of a faster increase since the 1980s (medium confidence) (Sections and Precipitation has increased from 1950 to 2018 over mid-high latitude Eurasia, most of North America, south-eastern South America, and north-western Australia, while it has decreased over most of Africa, eastern Australia, the Mediterranean region, the Middle East, and parts of East Asia, central South America, and the Pacific coasts of Canada, as simulated by the CMIP5 multi-ensemble mean (Dai, 2021). Since AR5, there have been updates of several precipitation datasets, including satellite estimates, reanalysis and merged products (Adler et al., 2017; Roca, 2019). However, observational uncertainties remain an issue for assessing regional trends in seasonal or annual mean precipitation amount (Hegerl et al. , 2015; Maidment et al. , 2015; Sarojini et al. , 2016; Beck et al. , 2017), as well as the convective and stratiform types of precipitation (e.g., Ye et al., 2017). Precipitation trends at regional scales are dominated by internal variability across much of the world (Knutson and Zeng, 2018). Regional changes in precipitation amounts can also be obscured by contrasting responses to GHG compared with aerosol forcings (Wu et al., 2013; Hegerl et al., 2015; Xie et al., 2016; Zhao and Suzuki, 2019; Zhao et al., 2020) and changes in precipitation intensity versus frequency (Shang et al., 2019).

Global and regional changes in precipitation frequency and intensity have been observed over recent decades. An analysis of 1875 rain gauge records worldwide over the period 1961–2018 indicates that there has been a general increase in the probability of precipitation exceeding 50 mm day–1, mostly due to an overall boost in rain intensity (Benestad et al., 2019). Such changes in precipitation intensity and frequency have not been formally attributed to human activities, but are consistent with the heating effect of increasing CO2 levels on the distribution of daily precipitation rates (Section and with a distinct overall intensification of heavy precipitation events found in both observations and CMIP5 models, though with an underestimated magnitude (Fischer and Knutti, 2014). Beyond amplified precipitation extremes (Section 11.4.2), CMIP5 models also indicate that anthropogenic forcings have increased temporal variability of annual precipitation amount over land from 1950 to 2005, which is most pronounced in annual mean daily precipitation intensity (Konapala et al., 2017).

Anthropogenic aerosols can alter precipitation intensities both through radiative and microphysical effects (Box 8.1 and Section Precipitation suppression through aerosol microphysical effects has been observed in shallow cloud regimes over South America and the south-eastern Atlantic, associated with local biomass burning (Andreae et al., 2004; Costantino and Bréon, 2010), and in industrial regions in Australia (Rosenfeld, 2000; Hewson et al., 2013; Heinzeller et al., 2016). In contrast, precipitation intensification through aerosol microphysical effects in deep convective clouds is seen in many regions such as the Amazon, southern USA, India, and Korea. This is associated with anthropogenic aerosols from cities (Hewson et al., 2013; Fan et al., 2018; S.S. Lee et al., 2018; Sarangi et al., 2018).

In the tropics, increases in precipitation amount are observed in convergence zones and decreases in the descending branches of the atmospheric circulation since 1979 (Chou et al., 2013; Liu and Allan, 2013; Gu et al., 2016; Polson et al., 2016; Polson and Hegerl, 2017), consistent with increased moisture transports with warming (Gimeno et al., 2020). Over tropical land areas, there is substantial variability in the ‘wet convergent regimes get wetter’ and ‘dry divergent regimes get drier’ pattern of trends observed since 1950 that are modulated by decadal changes in ENSO (Liu and Allan, 2013; Gu and Adler, 2018). CMIP6 models indicate an increased contrast between wet and dry regions in the tropics and subtropics (Figure 8.7; Schurer et al., 2020). This provides further evidence that rainfall has increased in wet regimes, and slightly decreased in dry regimes over the period 19882019 (Figure 3.14). This greater contrast is primarily attributable to greenhouse gas forcings, although the observed trends are statistically larger than the model responses (Section

Over the African continent, there are distinct precipitation trends observed in multiple datasets since the 1980s (Figure 8.7; Maidment et al., 2015; P. Nguyen et al., 2018). Increases in intense convective storms affecting the Sahel have been attributed to increased land ocean temperature gradients (Taylor et al., 2017), enhanced by intense heating of the Sahara (Dong and Sutton, 2015) rather than thermodynamics (Section 8.2.2). Changes in Sahel rainfall, with reduced precipitation amounts from the 1960s to the 1980s and a subsequent recovery, are assessed in Sections and In eastern Africa, decreasing precipitation amount (−2 to −7 % per decade for 19832010) was reported for the March to May ‘long rains’ season (Lyon and Dewitt, 2012; Viste et al., 2013; Liebmann et al., 2014; Maidment et al., 2015; Rowell et al., 2015) and evidence of a recovery since, with internal variability playing a large role in these decadal changes (Wainwright et al., 2019). In contrast, the second ‘short rains’ season in eastern Africa (October to December) does not exhibit significant precipitation trends (Rowell et al., 2015). Increases in annual southern African rainfall of 67% per decade during 19832010 are linked with the Pacific Decadal Oscillation (PDO; Maidment et al., 2015).

Figure 8.7 | Linear trends in annual mean precipitation (mm day–1per decade) for 1901–1984 (left) and 1985–2014 (right): (a, e) observational dataset, and the CMIP6 multi-model ensemble mean historical simulations driven by: (b, f) all radiative forcings; (c, g) GHG-only radiative forcings; (d, h) aerosol-only radiative forcings experiment. Colour shades without grey cross correspond to the regions exceeding 10% significant level. Grey crosses correspond to the regions not reaching the 10% statistically significant level. Nine CMIP6-DAMIP models have been used having at least three members. The ensemble mean is weighted per each model on the available and used members. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

(Section assesses changes in precipitation over the Mediterranean region and its connection with drought and aridity.

Rainfall increases have been observed over northern Australia since the 1950s, with most of the increases occurring in the north-west (Dey et al., 2019a, b; Dai, 2021) and decreases observed in the north-east (J. Li et al., 2012) since the 1970s. In contrast, there has been a decline in rainfall over southern Australia related to changes in the intensification and position of the subtropical ridge (CSIRO and BoM, 2015) and anthropogenic effects (Knutson and Zeng, 2018). The drying trend over south-west Australia is most pronounced during May to July, where rainfall has declined by 20% below the 1900–1969 average since 1970 and by about 28% since 2000 (BoM and CSIRO, 2020).

Over South America, there is observational and paleoclimate evidence of declining precipitation amount during the past 50 years over the Altiplano and central Chile, primarily explained by the PDO but with at least 25% of the decline attributed to anthropogenic influence (Morales et al., 2012; Neukom et al., 2015; Boisier et al., 2016; Seager et al., 2019b; Garreaud et al., 2020). In contrast, a significant rainfall increase has been detected over the Peruvian–Bolivian Altiplano (from observational data and satellite-based estimations) since the 1980s (Figure 8.7; Imfeld et al., 2020; Segura et al., 2020). Long-term (19022005) precipitation data indicate positive trends over south-eastern South America and negative trends over the southern Andes, with at least a partial contribution from anthropogenic forcing (Gonzalez et al., 2014; Vera and Díaz, 2015; Díaz and Vera, 2017; Boisier et al., 2018; Knutson and Zeng, 2018; see further assessment in Section and Atlas.7.2.2). The Peruvian Amazon has exhibited significant rainfall decreases during the dry season since 1980 (Lavado et al., 2013; Ronchail et al., 2018). Increases in wet season rainfall in the northern and central Amazon since the 1980s and decreases during the dry season in the southern Amazon (Barreiro et al., 2014; Gloor et al., 2015; Martín-Gómez and Barreiro, 2016; J.C. Espinoza et al., 2018; X.Y. Wang et al., 2018; Haghtalab et al., 2020) are not explained by radiative forcing based on CMIP6 experiments (Figure 8.7) and trends are insignificant over longer periods since 1930 (Kumar et al., 2013)or more recently, since 1973 (Almeida et al., 2017). See (Section for monsoon-related changes. For the tropical Andes region, trends in annual precipitation show heterogenous patterns, ranging between –4% per decade and +4% per decade in the northern and southern tropical Andes for a 30-year period at the end of the 20th century, although increases during 19651984 and decreases since 1984 have been registered in Bolivia (Carmona and Poveda, 2014; Pabón-Caicedo et al., 2020).

Over China, annual precipitation totals changed little from 1973 to 2016, but precipitation intensity significantly increased at a rate of 0.12 mm day–1 per decade, while the number of days with precipitation exceeding 0.1 mm day–1significantly decreased at a rate of 0.9 days per decade (Shang et al., 2019). There is consistency in trend estimates during 19982015 over mainland China among satellite-based products and station data, which show increased precipitation amounts in autumn and winter and decreases in summer (Chen and Gao, 2018), consistent with a decreased intensity of East Asian monsoon precipitation (Lin et al., 2014; Deng et al., 2018). Further assessment of precipitation changes over the South and South East Asian and the East Asian monsoon regions is presented in Section An increasing trend in the frequency of heavy rainfall occurrences at the expense of low and moderate rainfall occurrences is found over central India (Krishnan et al., 2016; Roxy et al., 2017) and over eastern China with the latter due to increasing high aerosol levels (Qian et al., 2009; J. Guo et al., 2017; Xu et al., 2017; Day et al., 2018), consistent with the effects of absorbing aerosol on stability and convective inhibition (Box 8.1).

Observed precipitation records since the early 1900s show increases in precipitation totals over central and north-eastern North America that are attributable to anthropogenic warming but larger in magnitude than found in CMIP5 simulations (Knutson and Zeng, 2018; Guo et al., 2019). Decreases in precipitation amount over the central and south-western USA and increases over the north-central USA during 19832015 (Cui et al., 2017; P. Nguyen et al., 2018), are not clearly associated with forced responses in CMIP6 simulations (Figure 8.7; see also Section Over Europe, precipitation trends since 1979 do not show coherence across datasets (Zolina et al., 2014; P. Nguyen et al., 2018). Longer records since 1910 show increases for much of Scandinavia, north-western Russia, and parts of north-western Europe/United Kingdom and Iceland (Knutson and Zeng, 2018). Records since 1930 show increases of annual preciptation amount over western Russia (see also Atlas.8.2). Widespread increases in daily precipitation intensity appear clearly over regions with a high density of rain gauges, such as Europe and North America over the 19512014 period (Alexander, 2016). Observations during 19662016 over northern Eurasia show increases in the contribution of heavy convective showers to total precipitation by 12% on average (with local trends of up to 5%) for all seasons except for winter (Chernokulsky et al., 2019). Increases in convective precipitation intensity have been identified, particularly on sub-daily time scales, using a range of modelling and observational data (Berg et al., 2013; Kanemaru et al., 2017; Pfahl et al., 2017).

Snowfall is an important component of precipitation in high-latitude and mountain watersheds. Reanalysis data indicate significant reductions in annual mean potential snowfall areas over NH land by 0.52 million km2 per decade, with the largest decline over the Alps, with snow water equivalent reductions of about 20 mm per decade (Tamang et al., 2020). In the Tibetan Plateau, region-wide winter snowfall has increased but summer snowfall has decreased during the 19602014 period (Deng et al., 2017). State-of-the-art model simulations indicate reduced mean annual snowfall in the Arctic, despite the strong precipitation increase, mainly in summer and autum, when temperatures are close to the melting point (Bintanja and Andry, 2017).

In summary, regional changes in precipitation amounts can be obscured by the contrasting responses to GHG and aerosol forcings across much of the 20th century and can thus be dominated by internal variability at decadal to multi-decadal time scales (high confidence). There is, however, a detectable increase in northern high-latitude annual precipitation over land which has been primarily driven by human-induced global warming (high confidence) (Section 3.3.2). Human influence has strengthened the zonal mean precipitation contrast between the wet tropics and dry subtropics since the 1980s (medium confidence), although regional studies suggest a more complex precipitation response to evolving anthropogenic forcings. There is high confidence that daily mean precipitation intensities have increased since the mid-20th century in a majority of land regions with available observations and it is likely that such an increase is mainly due to GHG forcing (see Section 11.4). Section assesses monsoon precipitation changes in detail. Evapotranspiration

The AR5 assessed that there was medium confidence that pan evaporation declined in most regions over the last 50 years, yetmedium confidence that evapotranspiration increased from the early 1980s to the late 1990s. Since AR5, these conflicting observations have been attributed to internal variability and by the fact that evapotranspiration is less sensitive to trends in wind speed and is partly controlled by vegetation greening (K. Zhang et al., 2015; Y. Zhang et al., 2016; Z. Zeng et al., 2018b). Observation-based estimates show a robust positive trend in global terrestrial evapotranspiration between the early 1980s and the early 2010s (Miralles et al., 2014b; Z. Zeng et al., 2014, 2018b; K. Zhang et al., 2015; Y. Zhang et al., 2016). The rate of increase varies among datasets, with an ensemble mean terrestrial average rate of 7.6 ± 1.3 mm yr1 per decade for 1882–2011 (Z. Zeng et al., 2018a). In addition, a decreasing trend in pan evaporation plateaued or reversed after the mid-1990s (C.M. Stephens et al., 2018) has been reported as due to a shift from a dominant influence of wind speed to a dominant effect of water vapour pressure deficit, which has increased sharply since the 1990s (Yuan et al., 2019). The absence of a trend in evapotranspiration in the decade following 1998 was shown to be at least partly an episodic phenomenon associated with ENSO variability (Miralles et al. , 2014b; K. Zhang et al. , 2015; Martens et al. , 2018). Thus, there is medium confidence that the apparent pause in the increase in global evapotranspiration from 1998 to 2008 is mostly due to internal variability. In contrast to AR5, there are now consistent trends in pan evaporation and evapotranspiration at the global scale, given the recent increase in both variables since the mid-1990s (medium confidence). Given the growing number of quantitative studies, there is high confidence that global terrestrial annual evapotranspiration has increased since the early 1980s.

Since AR5, the predominant contribution of transpiration to the observed trends in terrestrial evapotranspiration has been revisited and confirmed (Good et al., 2015; Wei et al., 2017). Using satellite and ecosystem models, Zhu et al. (2016) found a positive trend in leaf area index during 19822009, indicating that greening could contribute to the observed positive trend of evapotranspiration, in line withsimilar studies that focused on the 1981–2012(Y. Zhanget al. , 2016) and1982–2013(K. Zhanget al. , 2015) periods. Zeng et al. (2018) determined that the 8% global increase in satellite-observed leaf area index between the 1980s and the 2010s may explain an increase in evapotranspiration of 12.0 ± 2.4 mm yr–1 (about 55 ± 25% of the total observed increase). Forzieri et al. (2020) estimated that the recent increase in leaf area index led to 3.66 ± 0.45 W m–2 in latent heat flux (about 51 ± 6 mm yr–1) and that the sensitivity of energy fluxes to leaf area index increased by about 20% over the 1982–2016 period. Overall, there is medium confidence that greening has contributed to the global increase in evapotranspiration since the 1980s.

Plant water use efficiency(WUE)is expected to rise with CO2levels (high confidence) (Section and Box 5.2), andcan in theory counteract rising evapotranspiration in a warmer atmosphere (Section However, observational studies suggest that this may not be the case in some ecosystems. For example, Frank et al. (2015) found that while the WUE increased in European forests across the 20th century, transpiration also increased due to more plant growth, a lengthened growing season, and increased evaporative demand. Likewise Guerrieri et al. (2019) observed that while WUE and photosynthesis increased in North American forests, stomatal conductance experienced only modest declines that were restricted to moisture-limited forests. Other studies further suggest that in many ecosystems increased WUE will not compensate for increased plant growth, amplifying declines in surface water availability(De Kauweet al. , 2013; Ukkolaet al. , 2016b; A. Singhet al. , 2020), while drought conditions can also offset the CO2 fertilization effect and lead to a decline in WUE (N. Liu et al., 2020). There is low confidence regarding the impact of plant physiological effects on observed trends in evapotranspiration.

An increasing number of studies have identified signals of attribution in the recent observed trends in evapotranspiration. Douville et al. (2013) found that the post-1960 rise in evapotranspiration in both the mid-latitudes and northern high latitudes was related to anthropogenic radiative forcing. An analysis of CMIP5simulations suggests that anthropogenic forcing accounts for a large fraction of the global mean evapotranspiration trend from 1982 to 2010 (Dong and Dai, 2017). Padrón et al. (2020) determined that increases in evapotranspiration were responsible for the majority of the anthropogenic pattern in dry-season water availability that dominates global trends since 1984. These findings are further supported by CMIP6 model results (Figure 8.8) that show that the recent summer increase in evapotranspiration in the northern mid- and high latitudes is due to GHG forcing and decreasing anthropogenic aerosol emissions over Europe.

Figure 8.8 | Linear trends in annual mean evapotranspiration (mm day–1 per decade) for1901–1984 (left) and 1985–2014 (right): (a, e) Land Model Intercomparison Project (LMIP) and observational dataset, and the CMIP6 multi-model ensemble mean historical simulations driven by (b, f) all radiative forcings, (c, g) GHG-only radiative forcings, (d, h) aerosol-only radiative forcings experiment. Colour shade without grey cross correspond to the regions exceeding 10% significant level. Grey crosses correspond to the regions not reaching the 10% statistically significant level. Nine CMIP6-DAMIP models have been used having at least three members. The ensemble mean is weighted per each model on the available and used members. The Global Land Data Assimilation System (GLDAS) was not available over the early 20th century so was replaced by a multi-model off-line reconstruction, LMIP, which is consistent with GLDAS over the recent period but may be less reliable over the early 20th century given larger uncertainties in the atmospheric forcings. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

In summary, there is high confidence that terrestrial evapotranspiration has increased since the 1980s. There is medium confidence that this trend is driven by both increasing atmospheric water demand and vegetation greening, and high confidence that it can be partly attributed to anthropogenic forcing. There is low confidence about the extent to which increases in plant water use efficiency have influenced observed changes in evapotranspiration. Runoff, Streamflow and Flooding

The AR5 reported low confidence in the assessment of trends in global river discharge during the 20th century. This is because many streamflow observations have been impacted by land use and dam construction, and the largest river basins worldwide differ in many characteristics, including geography and morphology. In regions with seasonal snow storage, AR5 WGII assessed that there is robust evidence and high agreement that warming has led to earlier spring discharge maxima and robust evidence of earlier breakup of Arctic river ice, as well as indications that warming has led to increased winter flows and decreased summer flows where streamflows are lower and that the observed increases in extreme precipitation led to greater probability of flooding at regional scales with medium confidence. The SROCC found robust evidence and high agreement that discharge due to melting glaciers has already reached its maximum point and has begun declining with smaller glaciers, but onlylow confidence that anthropogenic climate change has already affected the frequency and magnitude of floods at the global scale.

Significant trends in streamflow and continental runoff were observed in 55 out of 200 large river basins during 19482012, with an even distribution of increasing and decreasing trends (Section; Dai, 2016). A global detection and attribution study shows that the simulation of spatially heterogeneous historical trends in streamflow is consistent with observed trends only if anthropogenic forcings are considered (Gudmundsson et al., 2019). Section assesses with medium confidence that anthropogenic climate change has altered regional and local streamflows, although a significant trend has not been observed in the global average (Sections and Multiple human-induced and natural drivers have been shown to play an important but variable role in observed regional trends of streamflow for several different areas (Fenta et al. , 2017; Ficklin et al. , 2018; Glas et al. , 2019; Vicente-Serrano et al. , 2019). For instance, decreasing runoff during the dry season has been observed over the Peruvian Amazon since the 1980s (Lavado et al., 2013; Ronchail et al., 2018). Up to 30–50% of the recent multi-decadal decline in streamflow across the Colorado River Basin can be attributed to anthropogenic warming and its impacts on snow and evapotranspiration (Woodhouse et al., 2016; McCabe et al., 2017; Udall and Overpeck, 2017; Xiao et al., 2018; Milly and Dunne, 2020). In the Upper Missouri River basin, Martin et al. (2020) found that warming temperatures have contributed to streamflow reductions since at least the late 20th century. Cold regions in the NH have experienced an earlier occurrence of snowmelt floods, an overall increase in water availability and streamflow during winter, and a decrease in water availability and streamflow during the warm season (Aygün et al., 2019).

Some studies have suggested that dam construction and water withdrawals can be the dominant drivers in observed trends in streamflow amount (Wada et al., 2013). Regionally, land-use and land cover changes have been identified as important factors for streamflow (H. Chen et al., 2020). The impact of surface dimming from aerosol emissions on evaporation was identified as a discernible influence in NH streamflows (Gedney et al., 2014). While changes in annual mean streamflow present a complicated picture, recent studies of changes in the timing of streamflow in snow-influenced basins continue to support a prominent influence from warming (Kang et al., 2016; Dudley et al., 2017; Kam et al., 2018). Global land runoff variations correlate significantly with ENSO variability (Miralles et al., 2014b; Schubert et al., 2016).

Observed changes in flooding are assessed in Section 11.5.2 and are summarized as follows. For changes in the magnitude of peak flow, recent studies show strong spatial heterogeneity in the sign, size and significance of trends. For changes in timing of peak flows, recent studies further support observed changes in snowmelt-driven rivers. Observed changes in runoff and flood magnitude cannot be explained by precipitation changes alone given the possible season- and region-dependent decreases in antecedent soil moisture and snowmelt, which can partly offset the increase in precipitation intensity (Sharma et al., 2018), or the expected effect of urbanization and deforestation which can, on the contrary, amplify the runoff response (Chen et al., 2017; Abbott et al., 2019; Cavalcante et al., 2019). Simulations of mean and extreme river flows are consistent with the observations only when anthropogenic radiative forcing is considered (Gudmundsson et al., 2021).

In summary, the assessment of observed trends in the magnitude of runoff, streamflow, and flooding remains challenging, due to the spatial heterogeneity of the signal and to multiple drivers. There is, however, high confidence that the amount and seasonality of peak flows have changed in snowmelt-driven rivers due to warming. There is also high confidence that land-use change, water management and water withdrawals have altered the amount, seasonality, and variability of river discharge, especially in small and human-dominated catchments. Aridity and Drought

The AR5 reported low confidence that changes in drought since the mid-20th century could be attributed to human influence, owing to observational uncertainties and difficulties in distinguishing decadal-scale variability from long-term trends. Changes in soil moisture, a metric of aridity, were not assessed thoroughly in AR5. Since AR5, new satellite products, land surface reanalyses, and land surface models have been used to document recent changes in soil moisture at the global scale. The science of detection and attribution has also progressed considerably (Trenberth et al., 2015; Easterling et al., 2016; Stott et al., 2016). Attribution efforts have further benefited from the increased use of paleoclimate information, which provides an important constraint on natural variability that is insufficiently sampled by short observational record (Cook et al., 2018; Kageyama et al., 2018).

Several studies have identified a persistent ‘fingerprint’ of anthropogenic forcing in global trends in aridity spanning the last 120 years. Using a combination of tree ring data, CMIP5 model simulations, and reanalysis products, Marvel et al. (2019) determined that the dominant trend in aridity since 1900, characterized by drying in North and Central America and the Mediterranean, is detectable and attributable to external forcing from 1900 to 1949. This trend weakens from 1950 to 1975, possibly due to aerosol forcing (Marvel et al., 2019), but then emerges again from 1981 to present, although it is not detectable in the GLEAM nor MERRA-2 soil moisture reanalysis products. Likewise, Bonfils et al. (2020) investigated changes in precipitation, temperature and continental aridity in CMIP5 historical simulations and found that the dominant multivariate fingerprint, an amplification of wet–dry latitudinal patterns and progressive continental aridification, was associated with greenhouse gas emissions (Figure 8.9a, d), and the second leading fingerprint was associated with anthropogenic aerosols (Figure 8.9e, h). This study found that the anthropogenic greenhouse gas signal is statistically detectable in reanalyses over the 1950–2014 period (signal-to-noise ratio above 1.96). Gu et al. (2019) found that a global trend in declining soil moisture is detectable in the GLDAS-2 reanalysis product and is attributable to greenhouse gas forcing. Padrón et al. (2020) reconstructed the global patterns of dry season water availability from 1902–2014, and found itextremely likely (99% range) that trends in the last three decades of the analysis period could be attributed to anthropogenic forcing, mainly due to increases in evapotranspiration. It is very likely (>90% range) that anthropogenic forcing has affected global patterns of soil moisture over the 20th century.

Figure 8.9 | Spatial expressions (a–c, e–g) of the leading multivariate fingerprints of temperature (°C), precipitation (mm day–1), and aridity (CMI; the Climate Moisture Index) in CMIP5 historical simulations and the corresponding temporal evolution in both CMIP5 and reanalysis products (d, h).

The first leading fingerprint is associated with greenhouse gas forcing (a–d) and the second leading fingerprint is associated with aerosol forcing (e–h). CMI is a dimensionless aridity indicator that combines precipitation and atmospheric evaporative demand. Figure after Bonfils et al. (2020). Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

On a regional scale, the robustness of trend attribution for drought and aridity varies widely. Key trends and their attributions are summarized here, while a complete regional assessment of observed trends in drought and aridity is in Chapter 11 (Sections 11.6.2, 12.3.2 and 12.4).

Several studies have analyzed CMIP5 and land surface models and detected a significant summer drying trend in the NH across the late 20th century that is attributable to anthropogenic forcings (Mueller and Zhang, 2016; Douville and Plazzotta, 2017). This trend is mainly driven by dryland areas such as the western USA and west-central Asia, where both reanalysis products and satellite data confirm there has been a persistent decline in soil moisture since 1990 (Y. Liu et al., 2019a). In the western USA, snow deficits have very likely contributed to recent drying (Mote et al., 2018). Spring snow water equivalent across the Sierra Nevada Mountains reached a record low in 2015 (Margulis et al., 2016; Mote et al., 2016), possibly the lowest of the last five hundred years (Belmecheri et al., 2016). Over the longer California drought (2011–2015) anthropogenic warming alone reduced snowpack levels in the Sierras by 25% (Berg and Hall, 2017). The north-western USA also experienced snow drought in 2015, despite near-normal levels of total cold season precipitation (Mote et al., 2016; Marlier et al., 2017). There is high confidence that anthropogenic warming contributed to these recent snow droughts (Belmecheri et al., 2016; Mote et al., 2016).

In the western USA, anthropogenic warming is amplifying drought and aridity by increasing evaporative demand and water loss to the atmosphere (Weiss et al., 2009; Overpeck, 2013; Cook et al., 2014; Griffin and Anchukaitis, 2014; Williams et al., 2020). For the California drought between 2012­–2014, Griffin and Anchukaitis (2014) used paleoclimate reconstructions to determine that while rainfall deficits were not unprecedented, record-high temperatures drove an exceptional decline in soil moisture relative to the last millennium. Williams et al. (2015) concluded that anthropogenic warming accounted for 8–27% of these soil moisture deficits. Robeson (2015) estimated that the California drought was a 1-in-10,000 year event. Tree ring reconstructions indicate that prolonged megadroughts have occurred in the western USA throughout the last 1200 years (Cook et al. , 2004, 2010; B.I. Cook et al. , 2015), forced by internal variability (Coats et al., 2016; Cook et al., 2016b). However, Williams et al. (2020) determined that 2000–2018 drought across the south-western USA was the second driest 19-year period since 800 CE, and attributed nearly half the magnitude of this event to anthropogenic forcing (see also Section Evidence for human signals in drought can also be found in western North American streamflow records, as noted above in Section There is high confidence that anthropogenic forcing has contributed to recent droughts and drying trends in western North America.

Large areas of east-central Asia experienced drying in the early 2000s as a result of warmer temperatures, lower humidity, and declining soil moisture(Wei and Wang, 2013; Z. Liet al. , 2017; Hesslet al. , 2018). Paleoclimate data from the Mongolian plateau suggest that this recent central Asian drought exceeds the 900-year return interval, but is not unprecedented in the last 2060 years (Hessl et al., 2018). There is low confidence due to limited evidence that recent droughts in central Asia can be attributed to anthropogenic forcing.

The Mediterranean region has experienced notable changes in drought and aridity. A number of studies have identified a decline in precipitation since 1960 and attributed this to anthropogenic forcing (Hoerling et al., 2012; Gudmundsson and Seneviratne, 2016; Knutson and Zeng, 2018; Seager et al., 2019b). Kelley et al. (2015) showed that climate change caused a three-fold increase in the likelihood of the 2007–2010 meteorological drought in the eastern Mediterranean. However, historical trends in precipitation across the Mediterranean are spatially variable and contain substantial decadal variability, such that an anthropogenic influence may not be detectable in all areas (Zittis, 2018; Vicente-Serrano et al., 2021). Records of soil moisture provide a clearer signal, indicating that higher temperatures and increased atmospheric demand have played a strong role in driving Mediterranean aridity (Vicente-Serrano et al., 2014). Hydrological modeling suggests that the recent decline in soil moisture in the Mediterranean is unprecedented in the last 250 years (Hanel et al., 2018). Paleoclimate evidence extends this view, additionally indicating that dryness in the Mediterranean is approaching an extreme condition compared to the last millennium (Markonis et al., 2018) and that the 15-year drought in the Levant (1998–2012) has an 89% likelihood of being the driest of the last 900 years (Cook et al., 2016a). Marvel et al. (2019) found that the Mediterranean region contributes strongly to the anthropogenic warming component of the global trend in aridity. There is high confidence that anthropogenic forcings are causing increased aridity and drought severity in the Mediterranean region.

Both central and north-eastern Africa have experienced a decline in rainfall since about 1980 (high confidence) (Lyon and Dewitt, 2012; Lyon, 2014; Hua et al., 2016; Nicholson, 2017). In Central Africa, the decline has been attributed to atmospheric responses to Indo-Pacific sea surface temperature variability (Hua et al., 2018). In north-eastern Africa, droughts have become longer and more intense in recent decades, continuing across rainy seasons (Hoell et al., 2017b; Nicholson, 2017), and this trend appears to be unusual in the context of the last 1500 years (Tierney et al., 2015). Knutson and Zeng (2018) attribute decreased annual precipitation over the Sudan to anthropogenic forcing, but other studies argue that the recent trend cannot yet be distinguished from natural variability, at least over parts of this region (Hoell et al., 2017b; Philip et al., 2018). There remains low confidence due to limited evidence that drying the north-eastern Africa is attributable to human influence. In the Western Cape region of South Africa, human influence increased the likelihood of the severe 2015–2017 drought by a factor of 3–6, depending on the analysis (Otto et al., 2018; Pascale et al., 2020). Anthropogenic forcing also contributed to the 2018 drought, mainly by increasing evapotranspiration (Nangombe et al., 2020). While some analysis of instrumental precipitation data in this region detect a slight long-term drying trend consistent with the simulated anthropogenic response (Seager et al., 2019b), there is strong multi-decadal variability in the data (Wolski et al., 2021). However, a study of streamflow in southern Africa detected a significant decline (Gudmundsson et al., 2019; see also Section 10.6.2). There is medium confidence in the long-term drying trend in this region and its attribution to anthropogenic forcing, and medium confidence that anthropogenic warming has contributed to recent severe drought events.

Several subtropical, semi-arid regions in the Southern Hemisphere have experienced long-term drying trends in the late 20th century. South-western South America (central Chile) experienced a multi-decadal decline in precipitation and streamflow culminating in a post-2010 megadrought that has been partly attributed to anthropogenic GHG emissions and ozone depletion (Boisier et al. , 2016, 2018; Saurral et al. , 2017; Knutson and Zeng, 2018; Seager et al. , 2019b; Garreaud et al. , 2020). There is medium confidence that drying in central Chile can be attributed to human influence. The tree-ring paleoclimate record demonstrates that the mid-century increase in exteme drought events in southern South America is unusual in the context of the last 600 years, suggesting an emerging influence of anthropogenic forcing (Morales et al., 2020).

There has been a 20% decrease in winter (May to July) rainfall in south-western Australia since 1970, with the decline increasing to around 28% since 2000 (Delworth and Zeng, 2014; BoM and CSIRO, 2020). There has also been a significant increase in the average intensity of seasonal droughts in the region since 1911in response to both lower precipitation and increased atmospheric evaporative demand (Gallant et al., 2013). Several studies attribute the precipitation declines in south-western Australia to anthropogenic changes in GHG and ozone (Delworth and Zeng, 2014; Knutson and Zeng, 2018; Seager et al., 2019b). There is high confidence that the observed drying in south-western Australia can be attributed to anthropogenic forcing.

In south-eastern Australia, the average length of droughts have increased significantly, lasting between 10 and 69% longer than droughts during the first half of the 20th century (Gallant et al., 2013). Paleoclimate reconstructions indicate a 97.1% probability that the decadal rainfall anomaly recorded during the 1997–2009 Millennium drought in south-eastern Australia was the worst experienced since 1783 (Gergis et al., 2012), and that the spatial extent and duration of cool season (April to September) rainfall anomalies were either very much below average or unprecedented over at least the last 400 years (Freund et al., 2017). Other paleoclimate studies suggest that the Millennium drought in eastern Australia was not unusual in the context of natural variability reconstructed over the past millennium (Palmer et al. , 2015; Cook et al. , 2016c; Kiem et al. , 2020). While there is currentlylow confidence that recent droughts in eastern Australia can be clearly attributed to human influence (Cai et al., 2014; Delworth and Zeng, 2014; Rauniyar and Power, 2020), there is emerging evidence that declines in April to October rainfall in south-eastern Australia since the 1990s would not have been as large without the influence of increasing levels of atmospheric GHGs (Rauniyar and Power, 2020).

In summary, it is very likely that anthropogenic factors have influenced global trends in aridity, mainly through competing changes in evapotranspiration and/or atmospheric evaporative demand due to anthropogenic emissions of GHG and aerosols. There is high confidence that the frequency and the severity of droughts has increased over the last decades in the Mediterranean, western North America, and south-western Australia and that this can be attributed to anthropogenic warming. There is medium confidence that recent drying and severe droughts in southern Africa and south-western South America can be attributed to human influence. In some regions of western North America and the Mediterranean, paleoclimate evidence suggests that recent warming has resulted in droughts that are of similar or greater intensity than those reconstructed over the last millennium (medium confidence). Freshwater Reservoirs Glaciers

The AR5 and SROCC found, with very high confidence, a general decline in glaciers due to climate change in recent decades. There is very high confidence that during the decade 2010–2019 glaciers lost more mass than in any other decade since the beginning of the observational record (Sections and 9.5.1). Human influence is very likely the main driver of the global, near-universal retreat of glaciers since the 1990s (Section In Table 9.5, the contribution of glaciers to sea level rise for different periods is presented; in 19712018 glacier mass loss contributed 20.9 [10.0 to 31.7] mm or 22.2% of the sea level rise during that period. The highest mass loss rates are observed in the southern Andes, New Zealand, Alaska, Central Europe and Iceland while the largest mass loss are observed in Alaska, the periphery of Greenland and Arctic Canada (Section 9.5.1 and Figure 9.20). Predominantly, runoff from small glaciers such as in Canada has decreased because of glacier mass loss, while runoff from larger glaciers such as in Alaska has typically increased (Bolch et al., 2010; Thomson et al., 2011; Tennant et al., 2012; WGMS, 2017; Huss and Hock, 2018). Asia contains the largest concentration of glaciers outside the polar regions where the total glacier mass change is –16.3 ± 3.5 Gt yr–1 over 20002016 with considerable intra-regional variability (Brun et al., 2017). Mass losses of glaciers in Asia between 2000 and 2018 are 19.0 ± 2.5 Gt yr–1 (Shean et al., 2020). The most negative changes were found in Nyainqentanglha with −4.0 ± 1.5 Gt yr–1, while glaciers in Kunlun, northern Tibetan Plateau, slightly gained mass at 1.4 ± 0.8 Gt yr–1.

There is some evidence that an increase of precipitation over high mountains can offset glacier ablation (melt; Farinotti et al., 2020). However, this process has only been described from the Karakoram region in the north-western Himalaya, where it is thought to be partly responsible to the advances of glacier changes in the last two decades, referred to as the ‘Karakoram Anomaly’ (Farinotti et al., 2020). In the Himalaya, Maurer et al. (2019) observed faster ice loss during 2000–2016 (7.5 ± 2.3 Gt yr–1) compared to 1975–2000 (–3.9 ± 2.2 Gt yr–1). In the Southern Hemisphere, the rate of glacier mass lost in South America is estimated at 19.4 ± 0.6 Gt yr–1based on surface elevation changes over 20002011, which include the North and South Patagonian Icefields of South America (Braun et al., 2019), and at −22.9 ± 5.9 Gt yr–1 over 20002018 (Dussaillant et al., 2019).

In summary, human-induced global warming has been the primary driver of a global glacier recession since the early 20th century (high confidence). Most glaciers have lost mass more rapidly since the 1960s and in an unprecedented way over the last decade, thereby contributing to increased glacier runoff, especially from larger glaciers until a maximum is reached, which tends to occur later in basins with larger glaciers and higher ice-cover fractions (high confidence). Seasonal snow cover

The AR5 assessed that Northern Hemisphere (NH) snow cover extent (SCE) has decreased since the late 1960s, especially in spring (very high confidence). This is confirmed by recent studies (Section; Kunkel et al., 2016). AR6 assesses that NH spring snow cover has been decreasing since 1978 (very high confidence) and that this trend extends back to 1950 (high confidence) (Section 9.5.3). Human-caused global warming is the dominant driver of this observed decline (Section 3.4.2; Estilow et al., 2015). Model simulations suggest that surface temperature responses at hemispheric/regional scales explain between 40% and 85% of the SCE trend variability (Mudryk et al., 2017). A decreasing trend in snowfall has also been detected in the NH (Figure 8.1; Rupp et al., 2013). Snowfall as a proportion of precipitation has decreased significantly in recent years (Berghuijs et al., 2014). However, a late-20th-century increase in snowfall in West Antarctica observed in ice cores has been linked to a combination of factors including the anthropogenically forced deepening of the Amundsen Sea Low (Thomas et al., 2015, 2017).

Observations show a rapid recent decrease of spring SCE in NH, mostly in Eurasia and North America, closely linked to temperature change, for example, March to April SCE is decreasing at 3.4% ± 1.1 % per decade (1979–2005; Brown and Robinson, 2011; Hernández-Henríquez et al., 2015). An overall increasing annual trend of the NH SCE since the late 1980s has been observed, in contrast to decreasing trends over 1960s to 1980s that are dominated by the autumn and winter seasons (Barry and Gan, 2020). Such recent positive trends in snow cover extent are however at odds with other surface and satellite datasets and with the negative trends simulated by most CMIP5 and CMIP6 models (Mudryk et al., 2017, 2020). Hernández-Henríquez et al. (2015) also detected positive trends in OctobertoNovember SCE in in the NOAA SCE Climate Data Record (NOAA-CDR), which are not replicated in other datasets (Section 9.5.3). Wu et al. (2018) found slower snowmelt rates over the NH in 1980–2017, with higher ablation rates in locations with deep snow water equivalent (SWE), but due to the reduction of SWE in deep snowpacks, moderate/high ablation rates showed decreasing trends. Santolaria-Otín and Zolina (2020) reported weak but significant decline in SCE in autumn over northern Eurasia and North America during 19792005, and similarly for spring, except for northern Siberia which showed higher spring SCE. Kapnick and Hall (2012) detected significant loss of spring mountain snowpack in western USA in 19502008. For Canada, extensive decreasing snow depths, SCE and duration were detected since mid-1970s, especially in western Canada during winter and spring (DeBeer et al. , 2016). Berghuijs et al. (2014) show that across the continental USA, catchments with more snowfall than rainfall generally have higher mean streamflow, which will probably decrease with smaller fractions of precipitation falling as snow because of climate warming.

In summary, a decline in the spring NH snow cover extent, snow depth and duration has been observed since the late 1960s and has been attributed to human influence (high confidence). Depending on the region and season, there is low-to-medium confidence in the main drivers of snow cover changes, although various regions exhibit a shortening of the snow cover season which is consistent with global warming. A more detailed assessment of observed changes in seasonal snow cover is provided in Section 9.5.3. Wetlands and lakes

Wetlands and lakes affect the climate through their impact on carbon and methane budgets (Section 5.2.2; e.g., Saunois et al., 2016; Zhan et al., 2019) and on surface heat fluxes, with coupled weather and climate effects(e.g., Zhan et al., 2019). Although these features are also affected by human activities and by climate change, AR5 did not specifically report on wetlands and lakes.

Inventories of surface water bodies are not systematically produced at national or regional levels. However, assessments are undertaken at the global scale (Ramsar Convention on Wetlands, 2018). Merging observations from multiple satellite sensors makes it possible to detect surface water even under vegetation and clouds over about 25 years, but with low spatial resolution (Prigent et al., 2016). Most recent multi-satellite products from visible, infrared, and microwave measurements, estimate a surface water area of about 12 to 14 million km2 (including permanent and transitory surfaces, e.g., Aires et al., 2018; Davidson et al., 2018), which is much higher than those provided by optical imagery (about 3 million km2). Inventories show a strong decrease in natural surface water of about 0.8% yr–1 in total from 1970 to the present (Ramsar Convention on Wetlands, 2018) but the sites are not evenly distributed. Multi-satellite estimates show a strong interannual variability in surface water extent over the period 19922015 with no clear long-term trend (Prigent et al., 2020).

Human-made water bodies represent approximately 10% of the total continental water surfaces (Figure 8.1; Ramsar Convention on Wetlands, 2018) and consist mainly of reservoirs and rice paddies. High resolution optical imagery over the period 19842015 (Donchyts et al., 2016; Pekel et al., 2016) shows a net increase of about 0.1 million km2 in artifical water surfaces, mainly due to the construction of reservoirs. Surfaces of rice paddies are also increasing, especially in South East Asia (Davidson et al., 2018).

In summary, there is high confidence that the extent of human-made surface water has increased over the 20th and early 21st centuries. In contrast, due to low agreement in the observational records at the global scale, there is onlylow confidence in the observed decline of the natural surface water extent in recent years (see also SRCCL). Groundwater

As the world’s most widespread store of freshwater (R.G. Taylor et al., 2013a), groundwater is estimated to supply between a quarter and a third of the world’s annual freshwater withdrawals to meet agricultural, industrial and domestic demands (Döll et al., 2012; Wada et al., 2014; Hanasaki et al., 2018).

Attribution of changes in groundwater storage, observed locally through piezometry (Figure 8.10; R.G. Taylor et al., 2013a) or estimated from GRACE satellite measurements (Rodell et al., 2018) at regional scales (>100,000 km2), is often complicated by non-climate influences that include land-use change (Favreau et al., 2009) and human withdrawals (Bierkens and Wada, 2019).

Figure 8.10 | Trends in Terrestrial Water Storage (TWS; in centimetres per year, cm yr–1) obtained on the basis of GRACE observations from April 2002 to March 2016. The cause of the trend in each outlined study region is briefly explained and colour-coded by category. The trend map was smoothed with a 150 km radius Gaussian filter for the purpose of visualization. However, all calculations were performed at the native 3° resolution of the data product. Figure from Rodell et al. (2018). Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

Following a global review of groundwater and climate change (R.G. Taylor et al., 2013a) and AR5 WGII, evidence of an association between heavy or extreme precipitation and groundwater recharge has continued to grow, especially in tropical (Asoka et al., 2018; Cuthbert et al., 2019a; Kotchoni et al., 2019) and subtropical regions (Meixner et al., 2016). Stable-isotope ratios of oxygen and hydrogen at 14 of 15 sites across the tropics trace groundwater recharge to intensive monthly rainfall, commonly exceeding the 70th intensity percentile, approximately (Jasechko and Taylor, 2015). Further, heavy rainfall recharging groundwater resources is often influenced by climate variability such as ENSO and PDO (R.G. Taylor et al., 2013b; Kuss and Gurdak, 2014; Asoka et al., 2017; Cuthbert et al., 2019b; Kolusu et al., 2019; Shamsudduha and Taylor, 2020). Additionally, increases in groundwater storage estimated from GRACE for 37 of the world’s large-scale aquifer systems from 2002 to 2016 are generally found to result from episodic recharge associated with extreme (>90th percentile) annual precipitation.

The overall underestimation of precipitation intensities in global climate models (Wehner et al., 2010, 2020; Goswami and Goswami, 2017) and of their sensitivity to warming temperatures (Borodina et al., 2017) may lead to underestimates of their recharging effect on groundwater (Mileham et al., 2009; Cuthbert et al., 2019b). The limited ability of global climate models to represent key controls on regional rainfall variability like ENSO (Technical (Annex VI and Section 3.7.3; R. Chen et al., 2020) may also underestimate observed recharge from such events that are of particular importance in drylands (R.G. Taylor et al. , 2013b; Cuthbert et al. , 2019b). Numerical representations of the impact of precipitation intensification on groundwater recharge in large-scale models remain constrained by the challenges of including key recharge pathways that consider preferential flowpaths in soils (Beven, 2018) and focused recharge through leakage from surface waters (Döll et al., 2014).

Increasing global freshwater withdrawals, primarily associated with the expansion of irrigated agriculture in drylands, have led to global groundwater depletion that has an estimated range of about 100 and about 300 km3yr–1 from hydrological models and volumetric-based calculations (Bierkens and Wada, 2019). The magnitude of this change is such that its estimated contribution to global sea level rise is in the order of 0.3 to 0.9 mm yr−1 (Wada et al. , 2010; Konikow, 2011; Döll et al. , 2014; Pokhrel et al. , 2015; de Graaf et al. , 2017; Hanasaki et al. , 2018). Groundwater depletion has been observed regionally in The USA High Plains, California’s Central Valley (Scanlon et al., 2012), north-west India(Rodellet al. , 2009; Asokaet al. , 2017), Upper Ganges in India(MacDonald et al., 2016), North China Plain (Feng et al., 2013), north-central Middle East region of Tigris–Euphrates–Western Iran (Voss et al., 2013), Central Asia (Hu et al., 2019), and North Africa (Bouchaou et al., 2013). The regional contribution of agricultural irrigation to groundwater depletion was previously highlighted by SRCCL but no formal assessment of observed changes in global or regional groundwater featured in AR5.

Quantification of changes in groundwater storage from GRACE is currently constrained by uncertainty in the estimation of changes in other terrestrial water stores using uncalibrated, global-scale Land Surface Models (Döll et al. , 2014; Scanlon et al. , 2018) and the limited duration of the period of GRACE observations (2002 to 2016). Centennial-scale piezometry in north-west India reveals that recent groundwater depletion traced by GRACE (Rodell et al., 2009; Chen et al., 2014), follows more than a century of groundwater accumulation through canal leakage (MacDonald et al., 2016). Further, groundwater depletion is often localized occurring below the footprint (200,000 km2) of GRACE, as has been well demonstrated by detailed modelling studies in the California Central Valley (Scanlon et al., 2012) and North China Plain (Cao et al., 2016).

Climate variability and drought affect groundwater depletion mainly due to amplified groundwater withdrawals. For instance, the depletion rate in Central Valley aquifer in the USA from 2006 to 2010 is estimated to range from 6 to 8 km3yr–1using GRACE data (Scanlon et al., 2012). In India, Asoka et al. (2017) show contrasting trends in groundwater storage in the north (declining at 2 cm yr–1) and south (increasing at 1–2 cm yr–1) that is explained by variations in human withdrawals and precipitation linked to Indian Ocean sea surface temperature variability.

Changes in meltwater regimes from glaciers and seasonal snow packs tend to reduce the seasonal duration and magnitude of recharge (Tague and Grant, 2009). Aquifers in mountain valleys show shifts in the timing and magnitude of: (i) peak groundwater levels due to an earlier spring melt; and (ii) low groundwater levels associated with lower baseflow periods (Allen et al., 2010; Dierauer et al., 2018; Hayashi, 2020). The effects of receding alpine glaciers on groundwater systems are not well understood but long-term loss of glacier storage is estimated to reduce summer baseflow (Gremaud et al., 2009). In permafrost regions, coupling between surface water and groundwater systems may be particularly enhanced by warming (Lamontagne-Hallé et al., 2018; Lemieux et al., 2020). In areas of seasonal or perennial ground frost, increased recharge is expected despite a decrease in absolute snow volume (Okkonen and Kløve, 2011; Walvoord and Kurylyk, 2016).

Coastal aquifers are the interface between the oceanic and terrestrial hydrological systems. Global sea level rise (SLR) causes interfaces between freshwater and saline-water to move inland. The extent of seawater intrusion into coastal aquifers depends on a variety of factors including coastal topography, recharge, and groundwater abstraction from coastal aquifers (Comte et al., 2016). Modelling results suggest that the impact of SLR on seawater intrusion is negligible compared to that of groundwater abstraction (Ferguson and Gleeson, 2012; Yu and Michael, 2019). Coastal aquifers under very low hydraulic gradients, such as the Asian mega-deltas, are theoretically sensitive to SLR but, according to evidence from Akter et al. (2019) in the Ganges-Brahmaputra-Megna basin, may be more severely and widely affected by changes in upstream river discharge. They argue further that saltwater inundation from storm surges will have the greatest localized effects.

In summary, there is medium confidence that increased precipitation intensities, partly due to human influence, have enhanced groundwater recharge, most notably in the tropics. There is high confidence that groundwater depletion has occurred since at least the start of the 21st century as a consequence of groundwater withdrawals for irrigation in some of the world’s most productive agricultural areas in drylands (e.g., southern High Plains and California Central Valley in the USA, the North China Plain, north-west India).

8.3.2 Observed Variations in Large-scale Phenomena and Regional Variability

Observed changes in large-scale circulation indicators (Cross-Chapter Box 2.2) are assessed in Chapters 2 and 3 (Sections and 3.3.3). In this chaper we focus on the influence of regional scale teleconnection variabililty on the water cycle and the attribution of these circulation changes. While observed changes in modes of variability are assessed in Chapters 2 and 4 (Sections 2.4 and 4.3.3), here focus on hydrological teleconnections of relevance to the water cycle. Inter-tropical Convergence Zone and Tropical Rain Belts

The AR5 concluded it is likely that the tropical belt, as delimited by the Hadley circulation, has widened since the 1970s. Observations in the satellite era indicate precipitation increases in the core of the Pacific Inter-tropical Convergence Zone (ITCZ) and decreases on the ITCZ margins (Gu et al., 2016; Su et al., 2017). As the satellite period has lengthened, observations have increasingly been used to assess trends in the ITCZ and tropical rain belt. Since AR5, significant narrowing and strengthening of the Pacific ITCZ after 1979 have been identified in atmospheric reanalyses (Wodzicki and Rapp, 2016), but no change in the ITCZ location (Byrne et al., 2018). Atmospheric model simulations suggest that with a narrower ITCZ, the subtropical jet becomes baroclinically unstable at a lower latitude and allows mid-latitude eddies to propagate farther equatorward (Watt-Meyer and Frierson, 2019). Observational analyses also show that the ITCZ narrowing (Zhou et al., 2020) is associated with increased precipitation in the ITCZ core region that is strongly coupled to increasing Outgoing Longwave Radiation (OLR) in the expanding dry zones, particularly over land regions in the subtropics and mid-latitudes (Lau and Tao, 2020). In addition, an eastward movement of the South Pacific Convergence Zone (SPCZ) between 1977 and 1999 has been reported, with associated significant precipitation trends in the South Pacific regions (Salinger et al., 2014).

ITCZ trends seen in satellites, precipitation measurements and reanalysis data are further supported by ocean surface-salinity observations. Long-term salinity observations show a freshening in the cores of the Atlantic and Pacific ITCZs and increased salinity on the ITCZ margins (Durack and Wijffels, 2010; Durack et al., 2012; Terray et al., 2012; Skliris et al., 2014). By investigating simultaneous changes in precipitation, temperature and continental aridity in CMIP5 historical simulations, Bonfils et al. (2020) found a secondary signal (Figure 8.9, right column) characterized by a robust inter-hemispheric temperature contrast (Section, a latitudinal shift in the ITCZ (in accordance with the theory of cross-equatorial energy transport; Section, and changes in aridity in the Sahel (Section These forced changes are statistically detectable in reanalyses datasets over the 19502014 period at the 95% confidence level.

Reconstructions in the Sahel (Carré et al., 2019) and Belize (Ridley et al., 2015) support the southward displacement of the tropical rain belt since 1850 and the narrowing trend of the tropical rainbelt detected in observations (Rotstayn et al., 2002; Hwang et al., 2013). Decreasing precipitation trends in the NH during the 1950s to 1980s have been attributed to anthropogenic aerosol emissions from North America and Europe, which peaked during the late1970s and declined thereafter following improved air quality regulations, causing dimming (brightening) through reduced (increased) surface solar radiation (Box 8.1 Figure 1), in agreement with model simulations (Chiang et al., 2013; Hwang et al., 2013). This is consistent with energetic constraints where tropical precipitation shifts are anti-correlated with cross-equatorial energy transport (Section6.3.3, Box 8.1). It also provides a physical mechanism for the severe drought in the Sahel that peaked in the mid-1980s (Sections and and the southward shift of the NH tropical edge from the 1950s to the 1980s (Allen et al., 2014; Brönnimann et al., 2015). However, CMIP5 and CMIP6 models still exhibit strong biases in representing the ITCZ, such as the simulation of a double ITCZ (Oueslati and Bellon, 2015; Adam et al., 2018; Tian and Dong, 2020). The impacts of aerosols and volcanic activity on the position of the ITCZ have been investigated but changes are difficult to characterize from observations (Section; Friedman et al. , 2013; J.M. Haywood et al. , 2013; Iles and Hegerl, 2014; Colose et al. , 2016; Chung and Soden, 2017). Such systematic shifts of the ITCZ can have important regional impacts like changes in precipitation (Figure 8.9).

In summary, there is medium confidence that the tropical rain belts over the oceans have been narrowing and strengthening in recent decades, leading to increased precipitation in the ITCZ core region (Section Decreasing precipitation trends in the NH during the 1950s1980s have been attributed to anthropogenic aerosol emissions from North America and Europe (high confidence). Hadley Circulation and Subtropical Belt

The AR5 reported low confidence in trends in the strength of the Hadley circulation (HC) due to uncertainties in reanalyses but high confidence on the widening of the tropical belt since 1979. In AR6, Chapter 2 (Section states that the HC has very likely widened and strengthened since at least the 1980s, mostly in the NH (medium confidence).

The poleward shift of the HC is closely related to migration of the location of tropical cyclone trajectories in bothhemispheres (Sharmila and Walsh, 2018; Studholme and Gulev, 2018), with avery likely poleward shift over the western North Pacific Oceans since the 1940s (Section Moreover, the Western North Pacific Subtropical High has extended westward since the 1970s, resulting in a monsoon rain band shift over China, with excessive rainfall along the middle and lower reaches of the Yangtze River valley along about 30°N over eastern China. At the same time, the effect of anthropogenic aerosols dominated the response to GHG increases over East Asia, resulting in a weakening of the East Asian summer monsoon and causing a drying trend in north-eastern China (Hu, 2003; Yu and Zhou, 2007; T. Wang et al., 2013; Z. Li et al., 2016b; Lau and Kim, 2017) and northern parts of South Asia (Section; Preethi et al., 2017). During 19772007, the precipitation variability over the eastern USA increased due to changes in the intensity and position of the western ridge of the North Atlantic Subtropical High (Li et al., 2011; Diem, 2013).

In the Southern Hemisphere (SH), the HC expansion has been associated with both the intensification and poleward shift of the subtropical high pressure belt (Nguyen et al., 2015), with consequences for precipitation amount over Africa, Australia, South America, and subtropical Pacific islands (Cai et al. , 2012; Grose et al. , 2015; Nguyen et al. , 2015; Sharmila and Walsh, 2018; McGree et al. , 2019). The subtropical ridge in Australia has intensified significantly since 1970, with marked declines observed in April to October rainfall across south-eastern and south-western Australia (Timbal and Drosdowsky, 2013).

The local tropical edges of the meridional overturning cells (as diagnosed from the horizontally divergent wind) are more closely associated with hydroclimate variations than the subtropical ridge (Staten et al., 2019). Poleward expansion of the tropical belt strongly contributes to precipitation decline in the poleward edge of the subtropics (Cai et al., 2012; Scheff and Frierson, 2012; Timbal and Drosdowsky, 2013; He and Soden, 2017; H. Nguyen et al., 2018; Tang et al., 2018), although recent modelling evidence suggests that subtropical precipitation declines are a response to direct CO2 radiative forcing mainly over ocean, irrespective of the HC expansion (He and Soden, 2017). Both reanalyses datasets and climate model simulations suggest that the HC expansion is not associated with widespread, zonally symmetric subtropical drying over land (Schmidt and Grise, 2017).

Since AR5, an improved understanding of the key drivers of the recent HC expansion has been achieved, identifying the role of both internal variability and anthropogenic climate change. Part of the recent expansion (19792005) of the HC has been driven by a swing from warm to cold phase of the Pacific Decadal Variability (PDV; Meehl et al., 2016; Grise et al., 2019). The presence of large multi-decadal variability in 20th-century reanalyses means there is limited evidence on the human influence on the recent HC strengthening, yet the southward shift of the southern edge and widening of the SH HC appeared as robust features in all reanalysis datasets, and their trends have accelerated during 19792010 (D’Agostino and Lionello, 2017). As assessed in Section, GHG increases and stratospheric ozone depletion have contributed to the expansion of the zonal mean HC in the SH since around 1980, and the expansion of the NH HC has not exceeded the range of internal variability (medium confidence). Moreover, Antarctic ozone depletion can cause a poleward shift in the SH mid-latitude jet and HC (Sections 3.3.3 and Further assessment of the attribution of recently observed changes in the HC extent and intensity is found in Section

In summary, it is very likely that the recent HC expansion was associated with poleward shifts of tropical cyclone tracks over the western North Pacific Ocean since the 1940s, and of extratropical storm tracks in the SH since the 1970s. Changes to the HC in the NH may have contributed to subtropical drying and a poleward expansion of aridity during the boreal summer, but there is low confidence due to limited evidence. GHG increases and stratospheric ozone depletion have contributed to expansion of the zonal mean HC in the SH since around 1970, while the expansion of the NH HC has not exceeded the range of internal variability (medium confidence). Walker Circulation

The AR5 concluded that the long-term weakening of the Pacific Walker circulation (WC) from the late 19th century to the 1990s has been largely offset by a recent strengthening (high confidence), though with low confidence in trends of the WC strength due to reanalysis uncertainties and large natural variability. The observed trends in the WC since 1980 are consistent with avery likely WC strengthening in the Pacific, similar to a La Niña pattern, with medium confidence in the magnitude of these changes due to differences between satellite observations and reanalyses.

The causes of the observed strengthening of the WC during 19802014 are not well understood due to competing influences from individual external forcings and since this strengthening is outside the range of variability simulated in coupled models (medium confidence), as assessed in Chapter 3 (Section Recent strengthening in the WC has been linked with internal variability (Chung et al., 2019), although one study argues that it could be a response forced by GHG that models do not capture because of common sea surface temperature (SST) biases in the equatorial Pacific (Seager et al., 2019a). It could be also related to an interbasin thermostat mechanism whereby the human-induced Indian Ocean warming emerged earlier than in the tropical Pacific (L. Zhang et al., 2018) and induced a transient strengthening of the zonal sea level pressure gradient and easterly trades in the tropical Pacific (L. Zhang et al., 2019).

The weakening of the PWC observed during most of the 20th century is associated with reductions in land rainfall over the Maritime Continent during 19501999 (Tokinaga et al., 2012; Yoden et al., 2017). In contrast, the recent strengthening of the WC has been associated with an intensification of extreme flooding (Barichivich et al., 2018) and an increased frequency of wet days (J.C. Espinoza et al., 2016, 2018) over the north-western Amazon, increased precipitation in South America (Yim et al., 2017), reduced precipitation over eastern Africa (Williams and Funk, 2011; Lyon and Dewitt, 2012), and increased rainfall in southern Africa (Maidment et al., 2015). Internal variability has been shown to have a dominant role in the recent strengthening of the WC (Chung et al., 2019).

In summary, there is high confidence that changes in the WC are associated with changes in the water cycle over regions like the Maritime Continent, South America and Africa. It is very likely that the WC has strengthened in the Pacific since the 1980s, with medium confidence that this strengthening is within the range of internal variability. Monsoons

The AR5 reported low confidence in the attribution of changes in monsoons to human influence, although a detailed attribution assessment of the observed changes in the regional monsoons was not presented.

Large human populations in the monsoon regions of the world heavily depend on freshwater supply for agriculture, water resources, industry, transport and various socio-economic activities. The effects of GHG forcing combined with water vapour feedback (R.J. Allen et al., 2015; Dong and Sutton, 2015; Evan et al., 2015; Dunning et al., 2018) and cloud feedbacks (Stephens et al., 2015; Potter et al., 2017) are fundamental to monsoon precipitation changes in a warming world. Since AR5 there has been improved understanding of precipitation changes associated with regional monsoons. Sections and provide an assessment of observed changes and attribution for the global monsoon. Here we provide an assessment of the observed changes in regional monsoons (see Annex V and Figure 8.11) and underlying causes. In AR6, the definition of regional monsoons slightly differs from AR5 and the rationale for it is provided in Annex V (see Glossary). Specific examples of regional monsoons are discussed further in Section 10.4.2, from the perspective of climate change attribution and in Section 10.6.3, from the viewpoint of constructing regional climate messages.

Figure 8.11 | Regional monsoon precipitation changes from observations and model attribution. Precipitation changes during 1951–2014 are shown as least-square linear trends in box-whisker plots (first and fourth rows) over the six regional monsoons, for example, North American monsoon (NAmerM, July–August–September, JAS), West African monsoon (WAfriM, June–July–August–September, JJAS), South and South East Asian monsoon (SAsiaM, June–July–August–September, JJAS), East Asian monsoon (EAsiaM, June–July–August), South American monsoon (SAmerM, December–January–February, DJF), Australian and Maritime Continent monsoon (AusMCM, December–January–February, DJF), and over the two land domains, for example, equatorial America (EqAmer, June–July–August, JJA) and South Africa (SAfri, December–January–February, DJF), as identified in the map shown in the middle and as described in Annex V. Precipitation changes are computed from observations and from Detection and Attribution Model Intercomparison Project (DAMIP) CMIP6 experiments over the historical period with all-forcing (ALL), GHG-only forcing (labelled GHG), Aerosol-only (AER) and Natural (NAT) forcings prescribed. Observations are based on the CRU (light green) and GPCC (light blue) datasets and the APHRODITE (light orange) dataset for SAsiaM and EAsiaM. CMIP6 simulations are taken from nine CMIP6 models contributing to DAMIP, with at least three members. Ensembles are weight-averaged for the respective model ensemble size. Observed trends are shown as coloured circles and the simulated trends from the CMIP6 multi-model experiments are shown as box-whisker plots. Precipitation anomaly time-series are shown in the second and third row. The thick black line is the multi-model ensemble-mean precipitation anomaly time-series from the ALL experiment and the grey shading shows the spread across the multi-model ensembles. An 11-year running mean has been applied on the precipitation anomaly time-series prior to calculating the multi-model ensemble mean. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1). South and South East Asian Monsoon

The AR5 reported a decreasing trend of global land monsoon precipitation over the last half-century, with primary contributions from the weakened summer monsoon systems in the Northern Hemisphere (NH). Since AR5, several studies have documented long-term variations and changes in the South and South East Asian summer monsoon (SAsiaM) rainfall. The SAsiaM strengthened during past periods of enhanced summer insolation in the NH, such as the early-to-mid Holocene warm period around 9000 to 6000 years before the present (BP) (Masson-Delmotte et al., 2013; Mohtadi et al., 2016; Braconnot et al., 2019) and weakened during cold periods (high confidence), such as the Last Glacial Maximum (LGM) and Younger Dryas (Shakun et al. , 2007; Cheng et al. , 2012; Dutt et al. , 2015; Chandana et al. , 2018; Hong et al. , 2018; E. Zhang et al. , 2018). These long-time scale changes in monsoon intensity are tightly linked to orbital forcing and changes in high-latitude climate (Braconnot et al. , 2008; Battisti et al. , 2014; Araya-Melo et al. , 2015; Rachmayani et al. , 2016; Bosmans et al. , 2018; E. Zhang et al. , 2018). A weakening trend of the SAsiaM during the last 200 years has been documented based on tree ring oxygen isotope chronology from the northern Indian subcontinent (Xu et al., 2018) and South East Asia (Xu et al., 2013), oxygen isotopes in speleothems from northern India (Sinha et al., 2015), and tree ring width chronologies from the Indian core monsoon region (Shi et al., 2017). Nevertheless, the detection of century-long decreases in regional monsoon rainfall is obscured by the presence of multi-decadal time scale precipitation variations (Turner and Annamalai, 2012; Knutson and Zeng, 2018) which are evident in long-term rain guage records extending back to the early 1800s (Sontakke et al., 2008) and emerge in long-term climate simulations (Braconnot et al., 2019).

A significant decline in summer monsoon precipitation is observed over India since the mid-20th century, which is accompanied by a weakening of the large-scale monsoon circulation (Mishra et al. , 2012; Abish et al. , 2013; Krishnan et al. , 2013, 2016; Saha et al. , 2014; Roxy et al. , 2015; Guhathakurta et al. , 2017; Samanta et al. , 2020). This precipitation decline is corroborated by a decreasing trend in the frequency of monsoon depressions that form over Bay of Bengal (Prajeesh et al., 2013; Vishnu et al., 2016), an increasing trend in the frequency and duration of monsoon breaks or ‘dry spells’ (Singh et al., 2014), significant decreases in soil moisture and increases in drought severity across different parts of India post-1950 (Niranjan Kumar et al. , 2013; Ramarao et al. , 2015, 2019; Krishnan et al. , 2016; Ganeshi et al. , 2020; Mujumdar et al. , 2020). While recent studies have reported an apparent recovery of the Indian summer monsoon over a relatively short period since 2003 (Jin and Wang, 2017; Hari et al., 2020), long-term trends for the period 19512015 indicate an overall decrease in the regional monsoon precipitation (Kulkarni et al., 2020; Ayantika et al., 2021). A case study on the Indian summer monsoon is provided in Section 10.6.3.

Evidence from several climate modelling studies indicates that the observed decrease in the regional monsoon precipitation during the second half of the 20th century is dominated by the radiative effects of NH anthropogenic aerosols, with smaller contributions due to volcanic aerosols from the Mount Pinatubo (1991) and El Chichón (1982) eruptions (Bollasina et al. , 2011; Polson et al. , 2014; Sanap et al. , 2015; Krishnan et al. , 2016; Liu et al. , 2016; Lau and Kim, 2017; Lin et al. , 2018; Takahashi et al. , 2018; Undorf et al. , 2018a, b; Patil et al. , 2019; M. Singh et al. , 2020; see Box 8.1, Figure 1 and Figure 8.11). Land-use changes over South and South East Asia and the rapid warming trend of the equatorial Indian Ocean during the recent few decades also appear to have contributed to the observed decrease in monsoon precipitation (Roxy et al., 2015; Krishnan et al., 2016; Singh, 2016). Overall, the magnitude of the precipitation response to anthropogenic forcing exhibits large spread across CMIP5 models pointing to the strong internal variability of the regional monsoon (Saha et al., 2014; Salzmann et al., 2014; Sinha et al., 2015), including variations linked to phase changes of the Pacific Decadal Variability (Section AVI.2.6; X. Huang et al., 2020a), uncertainties in representing aerosolcloud interactions (Takahashi et al., 2018), and the effects of local compared with remote aerosol forcing (Bollasina et al. , 2014; Polson et al. , 2014; Undorf et al. , 2018b). CMIP3 and CMIP5 models do not accurately reproduce the observed seasonal cycle of precipitation over the major river basins of South and South East Asia, limiting the attribution of observed regional hydroclimatic changes (Hasson et al., 2014, 2016; Biasutti, 2019). While warm rain processes and organized convection are known to dominate the heavy orographic monsoon rainfall over the Western Ghats mountains (Shige et al., 2017; Choudhury et al., 2018), in various parts of India (Konwar et al., 2012) and East Asia (Section, there are uncertainties in representing the regional physical processes of the monsoon environment, including cloudaerosol interactions (Sarangi et al., 2017), land atmosphere (e.g., Bartonet al., 2020) and oceanatmosphere coupling (Annamalai et al., 2017), in state-of-the-art climate models (see also Section 8.5.1).

In summary, there is high confidence in observational evidence for a weakening of the SAsiaM in the second half of the 20th century. Results from climate models indicate that anthropogenic aerosol forcing has dominated the recent decrease in summer monsoon precipitation, as opposed to the expected intensification due to GHG forcing (high confidence). On paleoclimate time scales, the SAsiaM strengthened in response to enhanced summer warming in the NH during the early-to-mid Holocene, while it weakened during cold intervals (high confidence). These changes are tightly linked to orbital forcing and changes in high-latitude climate (medium confidence). East Asian Monsoon

The AR5 reported low confidence in the observed weakening of the East Asian monsoon (EAsiaM) since the mid-20th century. Since AR5, there has been improved understanding of changes in the EAsiaM, based on paleoclimatic evidence, instrumental observations and climate modeling simulations. Rainfall reconstructions from the Loess Plateau in China indicate that the northern extent of the monsoon rain belts migrated at least 300 km to the north-west from the LGM to the mid-Holocene (Yang et al., 2015). Similarly, Pliocene reconstructions indicate stronger intensity of the EAsiaM with a more northward penetration of the monsoon rain belt (S. Yang et al., 2018a). EAsiaM variability has been related to Atlantic Meridional Overturning Circulation (AMOC) dynamics, especially during the last glacial period, but whether the relationship is negative or positive remains uncertain (Sun et al., 2012; Cheung et al., 2018; Kang et al., 2018).

Long-term precipitation observations from China indicate a trend of drying in the north and wetting in the central-eastern China along the Yangtze river valley since the 1950s (Qian and Zhou, 2014; Zhou et al., 2017b; Day et al., 2018), with a weakened EAsiaM low-level circulation that penetrates less far into northern China, increased surface pressure over north-east China and southward shift of the jet stream (Song et al., 2014). The southward shift and enhancement of the jet stream explains the increase of rainfall especially from the Meiyu front (Day et al., 2018) at the expense of drying over north-east China.

Anthropogenic factors such as GHGs and aerosols had an influence on the EAsiaM changes (Figure 8.11; T. Wang et al. , 2013; Song et al. , 2014; Xie et al. , 2016; Chen and Sun, 2017; Ma et al. , 2017; L. Zhang et al. , 2017; Day et al. , 2018; Tian et al. , 2018). Increased precipitation in the southern region has been linked to increased moisture flux convergence driven by GHG forcing while changes in anthropogenic aerosols have weakened the EAsiaM and reduced precipitation in the northern regions (Tian et al., 2018). Aerosol-induced cooling, associated atmospheric circulation changes and sea surface temperature (SST) feedbacks weaken the EAsiaM and favour the observed dry-north and wet-south pattern of rainfall anomalies (T. Wang et al. , 2013; Song et al. , 2014; L. Zhang et al. , 2017; G. Chen et al. , 2018; X. Chen et al. , 2018; Undorf et al., 2018b).

Internal variability and volcanic eruptions also contributed to the weakened EAsiaM (Hsu et al. , 2014; Qian and Zhou, 2014; Zhou et al. , 2017a; Knutson and Zeng, 2018). Since the late 1970s, the EAsiaM weakening has been also linked to SST changes in the Pacific Ocean with warm conditions in the central-eastern tropical part and cold ones in the north, similar to a positive phase of the Pacific Decadal Variability (PDV; Section AVI.2.6; Z. Li et al. , 2016b; Zhou et al. , 2017a). In the late 1990s the transition from a positive to a negative PDV has been associated with the recent recovery observed in the EAsiaM strength (Zhou et al., 2017a). Atlantic Multi-decadal Variability (AMV) also has an influence on the EAsiaM via the global teleconnection pattern propagating from the North Atlantic through the westerly jet (Zuo et al., 2013; Wu et al., 2016a, b). This North Atlantic influence has contributed to the increase of precipitation over the Huaihe-Huanghe valley since the late 1990s (Y. Li et al., 2017). When PDV and AMV are in opposite phase, the former has a larger influence in driving the southern flooding and northern drought pattern over the region (Q. Yang et al., 2017).

In summary, there is strong evidence of a stronger EAsiaM and northward migration of the rainbelt during warmer climates based on paleoclimate reconstructions. There is high confidence that anthropogenic forcing has been influencing historical EAsiaM changes with drying in the north and wetting in the south observed since the 1950s, but there is low confidence in the magnitude of the anthropogenic influence. The transition towards a positive PDV phase has been one of the main drivers of the EAsiaM weakening since the 1970s (high confidence). West African Monsoon

Since AR5, there has been improved understanding of the West African monsoon (WAfriM) response to natural and anthropogenic forcing. On paleoclimate time scales, enhanced summer insolation in the Northern Hemisphere (NH) intensified the WAfriM precipitation during the early-to-mid Holocene (high confidence), as seen in rainfall proxy records and climate model simulations (Masson-Delmotte et al. , 2013; Mohtadi et al. , 2016; Braconnot et al. , 2019). Despite improvements in model simulations of the present-day monsoons, CMIP5 and CMIP6 models underestimate mid-Holocene changes in the amount and spatial extent of the WAfriM precipitation (Section; Brierley et al., 2020).

During the recent past, long-term rain gauge observations display substantial variability in the WAfriM precipitation over the 20th century (Section The WAfriM experienced the wettest decade of the 20th century during the 1950s and early 1960s (high confidence), over much of the western and central Sahel region, followed abruptly by the driest years during 19701989 (Ali and Lebel, 2009; Nicholson, 2013; Descroix et al., 2015). The percentage deficit in the annual rainfall during 1970–1989, relative to the long-term mean, ranged from 60% in the north of Sahel to 2530% in the south (Le Barbé et al., 2002; Lebel et al., 2003). The long decline in annual rainfall is related to a decrease of rain occurrence over the Sahel (Le Barbé and Lebel, 1997; Frappart et al., 2009; Bodian et al., 2016) and the Soudano-Guinean sub-region of West Africa (Le Barbé et al., 2002), even though the interannual variability pattern is more complex (Balme et al., 2006). Decrease of rainfall occurrences resulted from decreases in large convective events in the core of the rainy season (Bell et al., 2006), that modulate interannual variability of the WAfriM (Panthou et al., 2018).

Wetter conditions of the WAfriM prevailed later from the mid-to-late 1990s, although the positive trend in precipitation started since the late 1980s (see also Section over the Sahel (high confidence) and in the Guinean coastal region (medium confidence), indicating the geographical variation in the wetting recovery (Descroix et al. , 2015; Sanogo et al. , 2015; Bodian et al. , 2016; Nicholson et al. , 2018). While the interannual and decadal variability of annual rainfall is not homogeneous over the entire Sahel, the rainfall recovery was stronger in the east than in the west of the region (Section; Nicholson et al., 2018). A shift in the seasonality of the Sahelian rainfall, including delayed cessation has also been reported (Section; Nicholson, 2013; Dunning et al., 2018).

In the Sahel region, the emergence of this new rainfall regime is reflected in increased number of heavy and extreme events, compared to the 1970s1980s, still not exceeding the values registered in the 1950s to 1960s (Descroix et al., 2013, 2015; Panthou et al., 2014, 2018; Sanogo et al., 2015), and in higher interannual variability (W. Zhang et al., 2017b; Akinsanola and Zhou, 2020) associated with SST variations in the tropical Atlantic, Pacific and Mediterranean Sea (Rodríguez-Fonseca et al., 2015; Diakhaté et al., 2019). Increased frequency of extreme rainfall events impacts high flow occurrences of the large Sahelian rivers as well as small to meso-scale catchments (Wilcox et al., 2018). Overall, extreme intense precipitation events are more frequent in the Sahel since the beginning of the 21st century (Giannini et al. , 2013; Panthou et al. , 2014, 2018; Sanogo et al. , 2015; Taylor et al. , 2017). Intensification of mesoscale convective systems associated with extreme rainfall in the WAfriM is favoured by enhancement of meridional temperature gradient by the warming of the Sahara desert (Taylor et al., 2017) at a pace that is two to four times greater than that of the tropical-mean temperature (K.H. Cook et al., 2015; Vizy et al., 2017). Periods of monsoon-breaks and the persistence of low rainfall events are still prominent, particularly after the onset, thus exposing West Africa simultaneously to the potential impacts of dry spells (W. Zhang et al., 2017b) and also extreme localized rains and floods (Engel et al., 2017; Lafore et al., 2017). Occurrence of extreme events is compounded by land use and land cover changes leading to increased runoff (Bamba et al., 2015; Descroix et al., 2018).

The Sahel drought from the 1970s until the early 1990s was related to anthropogenic emissions of sulphate aerosols in the Atlantic, which led to an inter-hemispheric pattern of SST anomalies and associated regional precipitation changes (Section and Box 8.1). Also the combined effects of anthropogenic aerosols and GHG forcing appear to have contributed to the late twentieth century drying of the Sahel through their effect on SST, by cooling the North Atlantic and warming the tropical oceans (Giannini and Kaplan, 2019; Hirasawa et al., 2020). Subsequent aerosol removal led to SST warming of the North Atlantic, shifting the ITCZ further northward and strengthening the WAfriM (Giannini and Kaplan, 2019). The recent recovery has been ascribed to prevailing positive SST anomalies in the tropical North Atlantic potentially associated with a positive phase of the Atlantic Multi-decadal Oscillation (Diatta and Fink, 2014; Rodríguez-Fonseca et al., 2015). The Sahel rainfall recovery has also been attributed to higher levels of GHG in the atmosphere and increases in atmospheric temperature (Dong and Sutton, 2015).

In summary, most regions of West Africa experienced a wet period in the mid-20th century followed by a very dry period in the 1970s and 1980s that is attributed to aerosol cooling of the NH (high confidence). Recent estimates provide evidence of a WAfriM recovery from the mid-to-late 1990s, with more intense extreme events partly due to the combined effects of increasing GHG and decreasing anthropogenic aerosols over Europe and North America (high confidence). On paleoclimate time scales, there is high confidence that the WAfriM strengthened during the early-to-mid Holocene in response to orbitally-forced enhancement of summer warming in the NH. North American Monsoon

Since AR5, there have been updates on the observed long-term variations and changes in the North American monsoon (NAmerM). During the Last Glacial Maximum (LGM; 21,00019,000 years ago), the NAmerM was substantially weaker due to cold, dry mid-latitude air associated with the Laurentide Ice Sheet (T. Bhattacharya et al. , 2017, 2018). The NAmerM strengthened until the mid-Holocene period, in response to ice-emsheet retreat and rising summer insolation, but probably did not exceed the strength of the modern system (low confidence), as indicated by model simulations (Metcalfe et al., 2015) and paleoclimatic reconstructions (Bhattacharya et al., 2018). Paleoclimatic evidence from proxy datasets and mid-Pliocene (PlioMIP1) simulations suggest a wetter south-western USA during that warmer period (A.M. Haywood et al., 2013; Pound et al., 2014; Ibarra et al., 2018) but it is not clear whether this is due to increases of precipitation associated with the monsoon or occurring during the winter season.

During 19482010, trends of boreal summer precipitation amount were significantly positive over New Mexico and the core NAmerM region, but significantly negative over south-western Mexico (Hoell et al., 2016). In addition, diverse datasets like CRU, CHIRPS and GPCP show significant decreases of precipitation in parts of the south-western USA and north-western Mexico, including the NAmerM region (Cavazos et al., 2020; Ashfaq et al., 2021). Other studies suggest a strengthening of the NAmerM upper level anticyclone since the mid-1970s, with a more frequent northward location (Diem et al., 2013). Between 19102010, the number of precipitation events increased across the northern Chihuahuan desert, within the NAmerM domain, despite a decrease in their magnitude, and the length of extreme dry and wet periods also increased (Petrie et al., 2014).

An increase in intense rainfall and severe weather events has been observed in several locations, especially in south-western Arizona since 1991, resulting from increases in atmospheric moisture content and instability; a change that has been confirmed by convective-permitting model simulations (Luong et al., 2017; Pascale et al., 2019). A dense network of 59 rain gauges located in south-eastern Arizona suggests an intensification of monsoon sub-daily rainfall since the mid-1970s (Demaria et al., 2019), as expected by a stronger global warming signature for sub-daily rather than daily or monthly precipitation accumulation (Section 11.4). Section provides further details on changes in precipitation in south-western North America. Evidence from multiple reanalyses suggests that increases in NAmerM rainfall have contributed to the increasing trend of global monsoon precipitation (Section; Lin et al., 2014). In addition, more frequent occurrence of earlier retreats of the NAmerM since 1979 is documented (Arias et al., 2012, 2015), in association with the positive phase of the Atlantic Multi-decadal Variability (AMV) and a westward expansion of the North Atlantic Subtropical High (W. Li et al., 2011, 2012).

Analyses from a 50-km resolution GCM indicate that the NAmerM response to CO2 is very sensitive to SST biases, showing reductions in summer NAmerM precipitation with increased CO2 when the SST biases are small (Pascale et al., 2017) in contrast to CMIP5 models (Cook and Seager, 2013; Maloney et al., 2014; Torres-Alavez et al., 2014; Hoell et al., 2016). The NAmerM has been shown to be also sensitive to sulphur dioxide (SO2) emissions (García-Martínez et al., 2020).

In summary, both paleoclimate evidence and observations indicate an intensification of the NAmerM in a warmer climate (medium confidence). The intensification recorded since about the 1970s has been partly driven by GHG emissions (medium confidence). South American Monsoon

Since AR5, there has been improved understanding of changes in the South American monsoon (SAmerM) as evidenced from paleoclimate records, instrumental observations and climate model simulations. However, general circulation models (GCMs) still exhibit difficulties in reproducing SAmerM precipitation amount (Rojas et al., 2016; D’Agostino et al., 2020b). Paleoclimate evidence suggests a relatively stronger SAmerM during the 1400–1600 period (Bird et al. , 2011b; Vuille et al. , 2012; Ledru et al. , 2013; Apaéstegui et al. , 2014; Novello et al. , 2016; Wortham et al. , 2017). Last millennium GCM simulations are able to reproduce stronger SAmerM during the 1400–1600 period in comparison with warmer epochs such as the 900–1100 period (Rojas et al., 2016) or the current warming period (Díaz and Vera, 2018). PMIP3/CMIP5 simulations indicate a consistent weaker SAmerM during the mid-Holocene (6000 years ago; see Cross-Chapter Box 2.1) in comparison to current conditions (Bird et al., 2011a; Mollier-Vogel et al., 2013; Prado et al., 2013a; D’Agostino et al., 2020b), thus favouring savannah/grassland-like vegetation (Smith and Mayle, 2018), in agreement with climate reconstructions from different proxies (Prado et al., 2013b). Signals of weak and strong SAmerM during mid-Holocene and LGM, respectively, are evident also in high-resolution long-term (i.e., more than about 22,000 years) rainfall reconstructions based on oxygen isotopes in speleothems from Brazil (Novello et al. , 2017; Stríkis et al. , 2018; Campos et al., 2019).

Isotope records from caves in the central Peruvian Andes show that the late Holocene (<3000 years ago) was characterized by multi-decadal and centennial-scale periods of significant decline in intensity of the SAmerM (Bird et al., 2011a; Vuille et al., 2012). This could be partly due to a reduction in the zonal SST gradient of the Pacific Ocean, favouring El Niño-like conditions (Kanner et al., 2013). Other studies suggest increased SAmerM precipitation amount during the Late Holocene, in association with the expansion of the tropical forest (Smith and Mayle, 2018). Well-dated equilibrium lines of glaciers during the deglaciation suggest that the AMOC enhances Atlantic moisture sources and precipitation amount increase over the tropical and southern Andes (Beniston et al., 2018).

Observations during 19792014 suggest that poleward shifts in the South Atlantic Convergence Zone (SACZ) noted in recent decades (Talento and Barreiro, 2018; Zilli et al., 2019), are associated with precipitation amount decrease along the equatorward margin and increase along the poleward margin of the convergenze zone (Zilli et al., 2019). Several observational studies identified delayed onsets of the SAmerM after 1978 related to longer dry seasons in the southern Amazon (Fu et al. , 2013; Yin et al. , 2014; Arias et al. , 2015; Debortoli et al. , 2015; Arvor et al. , 2017; Giráldez et al. , 2020; Haghtalab et al. , 2020; Correa et al. , 2021). In contrast, other studies indicate a trend toward earlier onsets of the SAmerM (Jones and Carvalho, 2013). These discrepancies are explained by the methodology used and the domain considered for the SAmerM, confirming the occurrence of delayed onsets of the SAmerM since 1978 (Correa et al., 2021). CMIP5 simulations show trends toward delayed onsets of the SAmerM in association with anthropogenic forcing, although the simulated trends underestimate the observed trends (Fu et al., 2013). Total rainfall reductions are observed in the southern Amazon during SeptemberOctoberNovember after 1978 (Fu et al., 2013; Bonini et al., 2014; Debortoli et al., 2015, 2016; Espinoza et al., 2019), consistent with reductions in river discharge in the region (Molina-Carpio et al. , 2017; Espinoza et al. , 2019; Heerspink et al., 2020).

Significant increases in precipitation have been observed over south-eastern Brazil during 19022005 while non-significant decreases have been found over central Brazil (Vera andDíaz, 2015). In Bolivia, increases were observed during 19651984, while reductions have occurred since then (Seiler et al., 2013). However, the Peruvian Amazon does not reveal significant changes in mean rainfall during 1965–2007 (Lavado et al., 2013; Ronchail et al., 2018). Historical simulations from CMIP5 ensembles adequately capture the observed summer precipitation amount over central and south-eastern Brazil, thereby providing high confidence in interpreting the observed variability of SAmerM for the period 19601999 (Gulizia and Camilloni, 2015; Pascale et al., 2019). Also, CMIP5 simulations indicate that the anthropogenic forcing associated with increased GHG emissions is necessary to explain the positive trends in upper-troposphere zonal winds observed over the South American Altiplano (Vera et al., 2019). However, the detection of anthropogenically-induced signals for precipitation is still ambiguous in monsoon regions, like the SAmerM (Hoegh-Guldberg et al., 2018).

In summary, there is high confidence that the SAmerM onset has been delayed since the late 1970s. This is reproduced by CMIP5 simulations that consider anthropogenic forcing. There is also high confidence that precipitation during the dry-to-wet transition season has been reduced over the southern Amazon. Paleoclimate reconstructions and simulations suggest a weaker SAmerM during warmer epochs such as the Mid-Holocene or the 900–1100 period, and stronger monsoon during colder epochs such as the LGM or the 1400–1600 period (high confidence). Australian and Maritime Continent Monsoon

Since AR5, several studies have examined observed variability and changes in the Australian and Maritime Continent monsoon (AusMCM) using paleoclimate records, instrumental observations and modeling studies (Denniston et al., 2016; Zhang and Moise, 2016). Paleoclimate reconstructions and modelling indicate that the Indo–Australian monsoon may vary in or out of phase with the EAsiaM, depending on whether there is a meridional displacement or expansion of the tropical rainfall belt (Ayliffe et al., 2013; Denniston et al., 2016). For instance, mid-Holocene simulations suggest that the AusMCM weakens and contracts due to a decreased net energy input and a weaker dynamic component (D’Agostino et al., 2020b).

Rainfall increases have been observed over northern Australia since the 1950s, with most of the increases occurring in the north-west (Dey et al., 2019a, b; Dai, 2021) and decreases observed in the north-east (J. Li et al., 2012) since the 1970s. There is also a trend towards more intense convective rainfall from thunderstorms over northern Australia (Dowdy, 2020). There is no consensus on the cause of the observed Australian monsoon rainfall trends, with some studies suggesting changes are due to altered circulation driving increased moisture transport or increased frequency of the wettest synoptic regimes (Catto et al., 2012; Clark et al., 2018). Other studies find that model simulations that include anthopogenic aerosols (Rotstayn et al., 2012; Dey et al., 2019a) are better able to capture observed Australian monsoon rainfall trends than simulations with natural or GHG forcing only (Knutson and Zeng, 2018).

The Maritime Continent (MC) experiences the influence of both the Asian and the Australian monsoons, with rainfall peaking during boreal winter/austral summer (Robertson et al., 2011). Reductions in land rainfall and marine cloudiness over the MC and weakening of surface moisture flux convergence have been observed in the period 19501999 (Tokinaga et al., 2012; Yoden et al., 2017). These trends are indicative of a slowdown of the Walker Circulation, with positive sea level pressure trends over the MC and negative trends over the central equatorial Pacific (Tokinaga et al., 2012). More recently (19812014), a trend of increasing annual rainfall over large areas of the MC has been identified (Hassim and Timbal, 2019). Given the large variability in MC rainfall on interannual time scales, the choice of time period may influence the calculated rainfall trend (Hassim and Timbal, 2019).

During 19512007 daily rainfall extremes did not increase over the MC, in contrast to the rest of South East Asia (Section 11.4.2; Villafuerte and Matsumoto, 2015). Rainfall extremes in Indonesia increased in austral summer, as evidenced from station weather observations for the period 19832012 (Supari et al., 2018).

In summary, notable rainfall increases have been observed in parts of northern Australia since the 1970s, although there is low confidence in the human contribution to these changes. Rainfall changes have been observed over the MC region but there is low confidence in the identification of trends because of large variability at interannual time scales. Tropical Cyclones

The AR5 assessed low confidence in centennial changes in tropical cyclone (TC) activity globally, and in the attribution of observed changes in TCs to anthropogenic forcing. Since AR5, there has been considerable progress in understanding the observed changes of TCs and an overall improved knowledge of the sensitivity of TCs to both GHG and aerosol forcing (Knutson et al., 2019; Sobel et al., 2019).

Although observational data limitations (Lau and Zhou, 2012) tend to limit detection of anthropogenic forced increases in TC precipitation (Knutson et al., 2019), there is medium confidence that anthropogenic forcing has contributed to observed heavy rainfall events over the USA associated with TCs (Kunkel et al., 2012) and other regions with sufficient data coverage (Section; Bindoff et al., 2013). There has been increased frequency of TC heavy rainfall events over several areas in the USA since the late 19th century that is greater than what would be expected solely from changes in US landfall frequency, suggesting the increasing role of TCs have in causing heavy rainfall events (Kunkel et al., 2010). For example, there is evidence for an anthropogenic contribution to the extreme rainfall of Hurricane Harvey in 2017 (Emanuel, 2017; Risser and Wehner, 2017; van Oldenborgh et al., 2017; Trenberth et al., 2018; S.-Y.S. Wang et al., 2018).

While TCs cause extreme local rainfall and flooding, they can be also an important contributor to annual precipitation and regional fresh water resources (Hristova-Veleva et al., 2020). Transport of moisture by TCs is an important contributor for precipitation over the coastal areas of East Asia mostly from July through October, with the TC rainfall accounting for nearly 10% to 30% of the total rainfall in the region (L. Guo et al., 2017). Local TC rainfall totals depend on rain-rate and translation speed (the speed of TC movement along the storm track) with slow TCs such as Hurricane Harvey (2017), providing a clear example of the effect of slow translation speed on local rainfall accumulation, with urbanization exacerbating the storm total rainfall and flooding (Section 11.7.1; W. Zhang et al., 2018).

In addition to evidence that rain-rates have increased, there is evidence that TC translation speed has slowed globally (Kossin, 2018) thus amplifying thermodynamic intensification of rainfall and may be linked to anthropogenic forcing (Gutmann et al., 2018). This is limited evidence however, so there is medium confidence of a detectable change in TC translation speed over the US. Since the 1900s, and there is low confidence for a global signal because of limited agreement among models and due to data heterogeneity. However, the slowdown is consistent with theoretical and modelling studies that indicate a general weakening of the tropical circulation with warming that reduces the speed of the TC system (Chauvin et al., 2017), though there is limitedobservational evidence (Sections and 11.7.1).

In summary, there is medium confidence of an observed increase in TC precipitation intensity in regions with sufficient data coverage Robust physical understanding (Section and detailed singular event attribution studies provide evidence that tropical cyclone rainfall has increased with a warming climate (high confidence, Section Stationary Waves

Stationary waves are planetary-scale waves that are approximately stable (stationary) in terms of geographic position, as opposed to propogating planetary waves, and are important both as part of the climatological general circulation and seasonal and shorter-term anomalies. They are related to surface features including land ocean contrasts and major mountain ranges, as well as atmospheric features including the jet stream, storm tracks, and blocking, which are considered separately in the following sections. While zonal mean changes in P–E (precipitation minus evaporation) are dominated by thermodynamic effects (Section, changes in stationary waves are of key importance in understanding zonal asymmetries in the water cycle response to global warming (Wills and Schneider, 2015; Wills et al., 2019). The AR5 did not explicitly assess stationary waves, but noted changes in related circulation features such as a likely poleward shift of the Northern Hemisphere (NH) storm tracks and an increase in frequency and eastward shift in North Atlantic blocking anticyclones, although there was low confidence in the global assessment of blocking.

Since AR5, several studies have demonstrated a link between stationary wave amplitude and wet and dry extremes in several different regions of the NH (Liu et al., 2012; Coumou et al., 2014; Screen and Simmonds, 2014; Yuan et al., 2015) with changes in moisture transport playing an important role (Yuan et al., 2015). A ‘resonance mechanism’ has been proposed for an increasing amplitude of stationary waves (Petoukhov et al., 2013, 2016; Coumou et al., 2014; Kornhuber et al., 2017)and several studies have linked increasing amplitude of stationary waves to Arctic warming (Francis and Vavrus, 2012, 2015; Liuet al. , 2012; Tanget al. , 2014) as well as to global warming (Mannet al. , 2017). However, other studies have not identified an increase in stationary wave amplitude (Barnes, 2013; Screen and Simmonds, 2013a, b).

There has been considerable work on linkages (teleconnections) between Arctic warming and the mid-latitude circulation (see also Cross-Chapter Box 10.1). The limited amount of research on Southern Hemisphere (SH) stationary waves suggests changes in high-latitude, mid-tropospheric stationary waves which influence Antarctic precipitation (Turner et al., 2017) and changes in stratospheric stationary waves that are associated with ozone depletion rather than increases in GHGs (L. Wang et al., 2013). The observed climatology of NH winter stationary waves is well-represented in the CMIP5 multi-model mean (Wills et al., 2019) but individual models have important deficiencies in reproducing stationary wave variability (Lee and Black, 2013). In the SH, the observed climatology of stationary waves in CMIP5 models has considerable bias in both phase and amplitude (Garfinkel et al., 2020). A comprehensive assessment is not yet available for CMIP6 models.

In summary, there is low confidence in strengthened winter stationary wave activity over the North Atlantic, associated with increased poleward moisture fluxes east of North America There is medium confidence in a recent amplification of the NH stationary waves in summer, but no formal attribution to anthropogenic climate change. Atmospheric Blocking

Atmospheric blocking refers to persistent, semi-stationary weather patterns characterized by a high-pressure (anticyclonic) anomaly that interrupts the westerly flow in the mid-latitudes of both hemispheres. By redirecting the pathways of mid-latitude cyclones, blocking can affect the water cycle and lead to negative precipitation anomalies in the region of the blocking anticyclone and positive anomalies in the surrounding areas (Sousa et al., 2017). In this way, blocking can also be associated with extreme events such as heavy precipitation (Lenggenhager et al., 2019), drought (Schubert et al., 2014) and heatwaves (Miralles et al., 2014a). The AR5 reported low confidence in global-scale changes in blocking, due to methodological differences between studies.

Currently no consensus exists on observed trends in blocking during 19792013. (Horton et al., 2015) identified increasing trends in anticyclonic circulation regimes based on geopotential height fields in the mid-troposphere, which may be partly related to the tropospheric warming itself and thus not represent real changes in the statistics of weather (Horton et al., 2015; Woollings et al., 2018). Hanna et al. (2018) and (Davini and D’Andrea, 2020) reported a significant increase in the frequency of summer blocking over Greenland. A weakening of the zonal wind, eddy kinetic energy and amplitude of Rossby waves in summer in the NH (Coumou et al., 2015, Kornhuber et al., 2019) and an increased ‘waviness’ of the jet stream associated with Arctic warming (Francis and Vavrus, 2015; Pfahl et al., 2015; Luo et al., 2019) have also been identified, which may be linked to increased blocking.

In contrast, it has been shown that observed trends in blocking are sensitive the choice of the blocking index, and that there is a large internal variability that complicates the detection of forced trends (Barnes et al., 2014; Cattiaux et al., 2016; Woollings et al., 2018), compromising the attribution of any observed changes in blocking. Many climate models still underestimate the occurrence of blocking, at least in winter over north-eastern Atlantic and Europe (Dunn-Sigouin and Son, 2013), which leads to caution in the interpretation of their results for these regions. However, over the Pacific Ocean there have been large improvements in the simulation of blocking for the last 20 years(Davini and D’Andrea, 2016; Patterson et al. , 2019). In the SH, increases in blocking frequency have occurred in the South Atlantic in austral summer (Dennison et al., 2016) and in the southern Indian Ocean in austral spring (Schemm, 2018). A reduced blocking frequency has been found over the south-western Pacific in austral spring (Sections and; Schemm, 2018).

In summary, no robust trend in atmospheric blocking has been detected in modern reanalyses and in CMIP6 historical simulations (medium confidence). The lack of trend is explained by strong internal variability and/or the competing effects of low-level Arctic amplification and upper-level tropical amplification of the equator-to-pole temperature gradient (medium confidence). Extratropical Cyclones, Storm Tracks and Atmospheric Rivers Extratropical cyclones and storm tracks

The AR5 indicated low confidence in long-term changes in the intensity of extratropical cyclones (ETC) over the 20th century derived from centennial reanalyses and storminess proxies based upon sea level pressure. This was confirmed by the SREX assessment that the main Northern Hemisphere (NH) and Southern Hemisphere (SH) extratropical storm tracks likely experienced a poleward shift during the last 50 years (Seneviratne et al., 2012) with low confidence, and inconsistencies within reanalysis datasets remain.

Since AR5 there has been considerable progress in quantifying storm track activity using multiple reanalysis products and different methodologies (Hodges et al. , 2011; Neu et al. , 2013; Tilinina et al. , 2013; X.L. Wang et al. , 2016). Over the NH increases in the total number of cyclones from 1979 show a large spread of trends across different estimates (Section; Neu et al., 2013; Z. Li et al., 2016a; Grieger et al., 2018) resulting in low confidence in any clear increase of in the total number of cyclones. However, starting from the early 1990s, most reanalyses show increases in the total cyclone number by about 25% per decade (Figure 8.12). Increasing trends in the total number of cyclones are dominated by the increase in the number of shallow and moderate cyclones (which are more dependent on the datasets and identification methods used) than with decreasing number of deep cyclones since the early 1990s (Tilinina et al., 2013; Chang, 2018). In the SH the variability of the total number of cyclones is characterized by strong inter-decadal variability preventing a clear assessment of trends. However, in contrast to the NH,there is a significant increasing trend in the number of deep cyclones (about 10% over 19792018) in ERA5, ERA-Interim, JRA55 and MERRA, and in the CFSR dataset after 2000 (Figure 8.12; Reboita et al., 2015; X.L. Wang et al., 2016).

Figure 8.12 | Annual anomalies (with respect to the reference period1979–2018) of the total number of extratropical cyclones (a, b) and of the number of deep cyclones (<980 hPa) (c, d) over the Northern (a, c) and the Southern (b, d) Hemispheres in different reanalyses (shown in colours in the legend).

Note different vertical scales for panels (a, b) and (c, d). Thin lines indicate annual anomalies and bold lines indicate five-year running averages. (e, f) The number of reanalyses (out of five) simultaneously indicating statistically significant (90% level) linear trends of the same sign during 1979–2018 for JFM (January–February–March) over the Northern Hemisphere (e) and over the Southern Hemisphere (f). Updated from Tilinina et al. (2013). Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

Changes in the number of deep storms, which are often associated with heavier precipitiation over the North Atlantic and North Pacific, exhibit strong seasonal differences and decadal variability (Colle et al. , 2015; Chang et al. , 2016; Matthews et al. , 2016; Priestley et al. , 2020a). An increase in the number of summer cyclones over the Atlantic-European sector (Tilinina et al. , 2013) is consistent with the increase in the strength of the strongest fronts over Europe (Schemm et al. , 2018). Chang et al. (2016) reported a decrease in the number of strong summer storms in the latitudinal band 40°N75°N over the last decades, however, the assessment of seasonal trends in the Atlantic-European sector is complicated by the choice of region, attribution of tracks to the region selected, and thresholds used to identify trajectories, leading to low confidence on regional seasonal trends. For the SH, Grieger et al. (2018) reported a growing number of cyclones over sub-Antarctic region in the austral-summer during 19792010, while statistically significant trends were absent during the austral winter.

Analysis of storm track activity over longer periods suffers from uncertainties associated with changing data assimilation and observations before and during the satellite era, resulting in in homogeneities and discontinuities in centennial reanalyses (Krueger et al. , 2013; X.L. Wang et al. , 2013, 2016; Chang and Yau, 2016; Varino et al. , 2019). Feser et al. (2015) reviewed multiple storm track records for the Atlantic-European sector and demonstrated growing storm activity north of 55°N from the 1970s to the mid-1990s with declining trend thereafter, sugesting strong inter-decadal variability in storm track activity. This was also confirmed by Krueger et al. (2019) from the analysis of geostrophic winds derived from sea level pressure gradients.

Poleward deflection of mostly oceanic winter storm tracks since 1979 was reported in both the North Atlantic and North Pacific (Tilinina et al., 2013; J. Wang et al., 2017). This large-scale tendency has regional variations and may be seasonally dependent. Wise and Dannenberg (2017) reported a southward shift in the east Pacific storm track from the 1950s to mid-1980s followed by northward deflection in the later decades. (King et al., 2019) reported an association of Atlantic storm track migrations with SSW events with Central and South European precipitation anomalies. Over centennial time scales, Gan and Wu (2014) reported an intensification of storm tracks in the poleward and downstream regions of the North Pacific and North Atlantic upper troposphere using the NOAA–CIRES–DOE Twentieth Century Reanalysis. Poleward migration of the SH storm tracks (Grise et al., 2014; X.L. Wang et al., 2016; Dowdy et al., 2019) was identified during the austral summer and is closely associated with cyclone-associated frontal activity (Solman and Orlanski, 2014, 2016) and cloud cover (Bender et al., 2012; Norris et al., 2016).

The representation of ETCs in both climate models and reanalyses is resolution-dependent, hence changes must be assessed with caution (Section In particular, CMIP5 models show a systematic underestimation of the intensity of ETCs (Zappa et al., 2014), a feature that is partially related to their relatively coarse resolution or other possible deficiencies such as an excess of dissipation (Chang et al., 2013). The best representation of ETCs and their intensity in the North Atlantic are provided by relatively high horizontal resolution CMIP5 models (Zappa et al., 2014). Using a single high-resolution climate model, (Hawcroft et al., 2016) showed that precipitation amount associated with ETCs was generally well simulated, though with too much precipitation during the strongest ECTs compared with observed estimations.

In summary, there is low confidence in recent changes in the total number of extratropical cyclones over both hemispheres. It is as likely as not that the number of deep cyclones over the NH has decreased after 1979 and it is likely that the number of deep extratropical cyclones increased over the same period in the SH. It is likely that extratropical cyclone activity in the SH has intensified during austral summer with no significant changes in austral winter. There is medium confidence that boreal-winter storm tracks during the last decades experienced poleward shifts over the NH and SH oceans. There is low confidence of changes in extratropical cyclone activity prior 1979 due to inhomogeneities in the intrumental records and modern reanalyses. Atmospheric rivers

Atmospheric rivers (ARs) are long, narrow (up to a few hundred kilometres wide), shallow (up to few kilometres deep) and transient corridors of strong horizontal water vapour transport that are typically associated with a low-level jet stream ahead of the cold front of an extratropical cyclone (Ralph et al., 2018). Atmospheric rivers were not assessed in AR5. ARs are associated with atmospheric moisture transport from the tropics to the mid- and high latitudes (Zhu and Newell, 1998), although the drivers of moisture transport relative to the different airstreams within extratropical cyclones remains a subject of current study (Dacre et al., 2019). While much previous research has focused on the west coast of North America, ARs occur throughout extratropical and polar regions (e.g., Guan and Waliser, 2015) and are often associated with locally-heavy precipitation, including a substantial fraction of all mid-latitude extreme precipitation events (e.g., Waliser and Guan, 2017). ARs also affect East Asia strongly during the period from late spring to summer (Kamae et al., 2017). ARs can be related to warming/melt events trough the intrusions of warm and moist air in Antarctica, Greenland and New Zealand (Bozkurt et al., 2018; Mattingly et al., 2018; Little et al., 2019), contributing about 4560% of total annual precipitation in subtropical South America (Viale et al., 2018). They alsotransport moisture from South America to the western and central South Atlantic, feeding the ARs that reach the west coast of South Africa(Ramos et al., 2019). However, the estimation of precipitation rate from ARs can have large uncertainties, especially as ARs hit topographically complex coastal regions (Behrangi et al., 2016), which can cause complexities in quantifying AR-related precipitation.

Analysis of observed trends in the characteristics of ARs has been limited. Gershunov et al. (2017) and Sharma and Déry (2019) have shown a rising trend in land-falling AR activity over the west coast of North American since 1948. (Gonzales et al., 2019) have also documented a seasonally-asymmetric warming of ARs affecting the West Coast of the USA since 1980, which has hydrological implications for the timing and magnitude of regional runoff. Longer-term paleoclimate analysis of ARs is even more limited, although Lora et al. (2017) reported that in the last glacial maximum, AR landfalls over the North American west coast were shifted southward compared to the present conditions.

In summary, it is likely that there was an increasing trend in the AR activity in the eastern North Pacific since the mid-20th century. However, there is low confidence in the magnitude of this trend and no formal attribution, although such an increase in activity is consistent with the expected and observed increase in precipitable water associated with human-induced global warming. Modes of Climate Variability and Regional Teleconnections

Following on from the assessment in Chapters 2 and 3, this section considers changes in modes of variability at seasonal to interannual time scales in terms of their implications on recent water cycle changes. These modes are described in details in Technical Annex IV. Tropical modes

The amplitude of the El Niño–Southern Oscillation (ENSO; Section AIV.2.3) variability has increased since 1950 (Section 2.4.2) but there is no clear evidence of human influence (Sections 2.4.2 and 3.7.3).

ENSO influences precipitation and evaporation dynamics, river flow and flooding at a global scale (Figure 3.37; Ward et al., 2014, 2016; Martens et al., 2018). Reconstruction (18042005) of Thailand’s Chao Praya River peak season streamflow displays a strong correlation with ENSO (Xu et al., 2019). Based on water storage estimates from 2002 to 2015, drought conditions over the Yangtze River basin followed La Niña events and flood conditions followed El Niño events (Z. Zhang et al., 2015). Strong correlation between ENSO and terrestrial water storage has been identified mostly in the subtropics but with diverse intensities and time lags depending on the region (Ni et al., 2018). The likelihood of increased/decreased flood hazard during ENSO events has a complex spatial pattern with large uncertainties (Emerton et al., 2017).

Tropical SSTs and associated global circulation may increase rainfall in West Africa, as observed in some years during 19502015, despite the presence of El Niño (Pomposi et al., 2020). During an El Niño summer, equatorial convective systems and the associated Walker circulation tend to shift eastward, leading to decreases in Indian summer monsoon rainfall (Li and Ting, 2015; Roy et al., 2019). This teleconnection is modulated by Indian Ocean Variability (Terray et al., 2021), as observed during the extreme positive IOD event in 2019 (Ratna et al., 2021). Since the end of the 19th century, synchronous hydroclimate changes (medium confidence) have been identified over south-eastern Australia and South Africa (Gergis and Henley, 2017) modulated by ENSO, as well as other regional fluctuations like the Botswana High over southern Africa (Driver and Reason, 2017). Over southern South America, the ENSO influence on precipitation (Cai et al., 2020; Poveda et al., 2020) interacts with the influence of SAM (Pedron et al., 2017), exhibiting large multi-decadal variations because of changes in the correlation between the two large-scale modes (Vera and Osman, 2018). Other processes underlying ENSO teleconnections of relevance for water cycle changes include water vapour and moisture transports, like over the Middle East (Sandeep and Ajayamohan, 2018), south-eastern China (S. Yang et al., 2018b), or central Asia (X. Chen et al., 2018), south-eastern South America (Martin-Gomez et al., 2016; Martín-Gómez and Barreiro, 2016), Australia (Rathore et al., 2020) and southern USA (Okumura et al., 2017).

There is no evidence of a trend in the Indian Ocean Dipole (IOD; Section AIV.2.4) mode and associated anthropogenic forcing (Sections 2.4.3 and 3.7.4). The AR5 concluded that the IOD is likely to remain active, affecting climate extremes in Australia, Indonesia and East Africa. Since the AR5, IOD teleconnections have been identified extending further to the Middle East (Chandran et al., 2016), to the Yangtze river (Xiao et al., 2015), where in boreal summer and autumn positive IOD events tend to increase the precipitation in the south-eastern and central part of the basin, and to the southern Africa extreme wet seasons (Hoell and Cheng, 2018). During the last millenium, the combined effect of a positive IOD and El Niño conditions have caused severe droughts over Australia (Abram et al., 2020). In the satellited period, it is found more effective in inducing significant decrease of rainfall over Indonesia, with the opposite occurring for negative IOD events (As-syakur et al., 2014; Nur’utami and Hidayat, 2016; Pan et al., 2018). Similarly, over the Ganges and Brahmaputra river basins major droughts have been recorded during co-occurring El Niño and positive IOD, while floods occurred during La Niña and negative IOD conditions (Pervez and Henebry, 2015). Over equatorial East Africa the IOD affects the short rain season (medium confidence) exacerbating flooding and inundations independently of ENSO (Behera et al., 2005; Conway et al., 2005; Ummenhofer et al., 2009; Hirons and Turner, 2018). Extreme conditions, like the 2019 Australian bushfires and African flooding, have been associated with strong positive IOD conditions (Cai et al., 2021).

Intraseasonal variability, like the Madden Julian Oscillation (MJO, Section AIV.2.8) and the Boreal Summer Intraseasonal Oscillation (BSISO), are highly relevant to the water cycle (Maloney and Hartmann, 2000; Lee et al., 2013; Yoshida et al., 2014; Nakano et al., 2015). Since AR5, studies on MJO teleconnections within the tropics and from the tropics to higher latitudes have continued (Guan et al., 2012;Mundhenk et al., 2018; Tseng et al., 2019; Aberson and Kaplan, 2020; Finney et al., 2020b; Fowler and Pritchard, 2020; Fromang and Rivière, 2020).

The strength and frequency of the MJO have increased over the past century (medium confidence) (Oliver and Thompson, 2012; Maloney et al., 2019; Cui et al., 2020) because of global warming (Arnold et al., 2015; Carlson and Caballero, 2016; Wolding et al., 2017; Maloney et al., 2019). A 20th century reconstruction suggests a 13% increase of the MJO amplitude (Oliver and Thompson, 2012), with differences in seasonal variability (Tao et al., 2015; Z. Wang et al., 2020). However, up to half of changes recorded during the second half of the 20th century could be due to internal variability (Schubert et al., 2013). Other observed changes in MJO characteristics include a decrease (by threetofour days) in the residence time over the Indian Ocean but an increase (by five to six days) over the Indo-Pacific and Maritime Continent sectors (Roxy et al., 2019).

Consequences of these changes are increased rainfall over South East Asia, northern Australia, south-west Africa and the Amazon, and drying over the west coast of the USA and Equador (Roxy et al., 2019). During the austral summer, airsea interactions and location of the MJO active phase are important to modulate the strength of the rainfall response in the South Atlantic Convergence Zone (Shimizu and Ambrizzi, 2016; Alvarez et al., 2017), including its southward shift (Barreiro et al., 2019). In the austral winter, the intraseasonal variability is mostly influential over regions of the Amazonian basin (Mayta et al., 2019). Some MJO phases are particularly effective in conjuction with tropical cyclones in enhancing westerly moisture fluxes towards East Africa (Finney et al., 2020b).

Simulated changes in MJO precipitation amplitude are extremely sensitive to the pattern of SST warming (Takahashi et al., 2011; Maloney and Xie, 2013; Arnold et al., 2015) and oceanatmosphere coupling (DeMott et al., 2019; Klingaman and Demott, 2020). In agreement with results from previous model generations, most CMIP5 models still underestimate MJO amplitude, and struggle to generate a coherent eastward propagation of precipitation and wind (Hung et al., 2013; Jiang et al., 2015; Ahn et al., 2017), affecting regional surface climate in the tropics and extratropics. In addition, most CMIP5 models simulate an MJO that propagates faster compared with observations, with a poorly represented intra-seasonal precipitation variability (Ahn et al., 2017). Over the Indian Ocean, the propagation speed of convection in some CMIP5 models tends to be slower than observed due to a strong persistence of equatorial precipitation (Hung et al., 2013; Jiang et al., 2015). Among other processes, improving the moisture-convection coupling, the representation of moist convection, the interaction between lower tropospheric heating and boundary layer convergence, and the topography of the Maritime Continent improve simulations of the MJO (Ahn et al. , 2017, 2020a; Kim and Maloney, 2017; Yang and Wang, 2019; H. Tan et al. , 2020; Y.-M. Yang et al. , 2020). In fact, CMIP6 models reproduce the amplitude and propagation of the MJO better than CMIP5 models due to increased horizontal moisture advection over the Maritime Continent (Ahn et al., 2020b). Despite the diverse theories of MJO evolution and processes that have been developed since its discovery, a better understanding of its dynamics is still needed (Jiang et al., 2020; Zhang et al., 2020). Furthermore, metrics based on dynamical processes are needed to assess model simulations of these events (Stechmann and Hottovy, 2017; B. Wang et al., 2018) as well as related teleconnections (J. Wang et al., 2020).

In summary, multiple water cycle changes related to ENSO and IOD teleconnections have been observed across the 20th century (high confidence), mostly dominated by interannual to multi-decadal variations. The MJO amplitude has increased in the second half of the 20th century partly because of anthropogenic global warming (medium confidence) altering regional precipitation signals. Extratropical modes

A positive trend has been observed in the Northern Annular Mode (NAM; Section AIV.2.1) in the second half of the 20th century, which partially reversed since the 1990s (Section, but the detection and attribution of these changes remain difficult (Section 3.7.1). The linkages of the NAM with weather and climate extremes in the northern extratropics are still unclear in models and observations (Vihma, 2014; Overland et al., 2016; Screen et al., 2018). However, robust links are identified between precipitation trends and variability in Europe and the phases of the Atlantic component of the NAM, that is, the NAO (Moore et al., 2013; Comas-Bru and McDermott, 2014). Reduced winter precipitation is well correlated with the NAO over Southern Europe and Mediterranean countries (Kalimeris et al. , 2017; Corona et al. , 2018; Vazifehkhah and Kahya, 2018; Neves et al. , 2019). NAO teleconnections in those regions include influences on groundwater and streamflow (Zamrane et al., 2016; Massei et al., 2017; Jemai et al., 2018). Remote teleconnections of the NAO have been identified over Northern China, the Yangtze River valley and India (Jin and Guan, 2017; Di Capua et al., 2020). The summer phase of the NAO is significantly correlated with variations in summer rainfall in East China, with the thermal forcing of the Tibetan Plateau providing a link to this Eurasian teleconnection (Z. Wang et al., 2018).

In the Southern Hemisphere (SH), an observed positive trend is identified in the strength of the Southern Annular Mode (SAM, Section AIV.2.2) since 1950, especially in austral summer (high confidence, Section While stratospheric ozone depletion and GHG increases largely contributed to this change, climate models still have trouble simulating the SAM and its response to ozone and GHGs (Section 3.7.2). Shifts in the south-westerly winds (Fletcher et al., 2018) and the expansion of the SH Hadley cell (Kang and Polvani, 2011; H. Nguyen et al., 2018) influence SAM-related rainfall anomalies in in southern South America and southern Australia during the austral spring–summer. Over New Zealand, large-scale SLP and zonal wind patterns associated with SAM phases modulate regional river flow (Li and McGregor, 2017). The SAM also influences precipitation and water vapour changes over Antarctica via moisture fluxes (Marshall et al., 2017; Oshima and Yamazaki, 2017; Grieger et al., 2018) but CMIP5 models are limited in their ability to simulate these regional teleconnections (Marshall and Bracegirdle, 2015; Palerme et al., 2017). SAM and its interaction with other large-scale modes of climate variability, like ENSO (Fogt et al., 2011) and the Indian Ocean Dipole (Hoell et al., 2017a), are responsible for fluctuations in southern African rainfall (Nash, 2017) and southern South America (Gergis and Henley, 2017). In May, the SAM can trigger a southern Indian Ocean Dipole SSTA favoring more or less precipitation over the Indian sub-continent and adjacent areas (Dou et al., 2017), also affecting subsequent summer monsoon in the South China Sea (T. Liu et al., 2018). Over South America, a positive SAM is associated with dry conditions (Holz et al., 2017) due to reduced frontal and orographic precipitation and weakening of moisture convergence. Regions particularly affected include Chile (Boisier et al., 2018) and the rivers of central Patagonia (Rivera et al., 2018).

In summary, while the attribution of 20th century variations of the NAM/NAO is still unclear, there is a strong relationship with precipitation changes over Europe and in the Mediterranean region (high confidence). SAM teleconnections are associated with changes in moisture transport and extend to South America, Australia and Antarctica (high confidence) with documented drying occurring as a result of the very likely human-induced SAM trend toward its positive phase observed from the 1970s until the 1990s (Section 3.7.2).

8.4 What Are the Projected Water Cycle Changes?

We consider global and regional climate projections of the water cycle, assessing projected changes in each component of the water cycle (Section 8.4.1) and the global-scale and regional phenomena that directly impact it (Section 8.4.2).

8.4.1 Projected Water Cycle Changes

Most projected changes in the water cycle are not expected to be uniform in space or time. They are driven by both dynamical and thermodynamical processes (Section 8.2) and have not necessarily emerged yet in the recent observational record (Section 8.3) as they are superimposed on substantial natural fluctuations in weather and climate. Therefore, projecting regional water cycle changes remains challenging. However, a number of physically understood responses can be evaluated using both CMIP5 and CMIP6 models, which are important for guiding decision making that anticipates, prepares for, and responds to water cycle changes. In this section, global maps of projected changes in water cycle variables are assessed using the WGI AR6 ‘simple method’ (see Cross-Chapter Box Atlas1), which uses hatching to highlight where less than 80% of the models agree on the sign of projected changes. This choice differs from Section 4.2.6 for a number of reasons. These include the weak signal-to-noise ratio of projected hydrological changes in low to medium emissions scenarios, the sensitivity of their statistical significance to the baseline reference period, and the non-Gaussian distribution of many water cycle variables (see Cross-Chapter Box Atlas.1 for more details on strengths and limitations of the hatching methods implemented within AR6). Global Water Cycle Intensity and P–E Over Land and Oceans

As discussed in, the definition of global water cycle intensity varies from the simple metric of increases in global mean precipitation to broader joint considerations of water vapour and its transport, precipitation minus evaporation (P–E) rates and continental runoff (Figure 8.1). The AR5 determined that globally averaged precipitation is virtually certain to increase with temperature and that there is high confidence that the contrast of annual mean precipitation between dry and wet regions and seasons will increase over most of the globe as temperatures and moisture transports increase (Collins et al., 2013). The AR5 also highlighted that continued ocean warming for a few decades after GHG forcing stabilizes or begins to decrease will also lead to further increases in global mean precipitation and evaporation.

Table 8.1 | Global and global land annual mean water cycle projections in the mid-term (2041–2060) and long term (2081–2100) relative to present day (1995–2014), showing present day mean and 90% confidence range across CMIP6 models (historical experiment) and projected mean changes and the 90% confidence range across the same set of models and a range of Shared Socio-economic Pathway scenarios. Note that the exact value of changes can vary slightly based on the number of models assessed, but not sufficiently to affect the assessment. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

Mid-term: 2041–2060Minus Reference Period

Long Term: 2081–2100Minus Reference Period

19952014reference period











Global Annual

Precipitation (mm day–1)

2.96 [2.76 to 3.17]

0.06 [0.03 to 0.11]

0.07 [0.03 to 0.12]

0.07 [0.04 to 0.12]

0.06 [0.03 to 0.11]

0.08 [0.03 to 0.14]

0.06 [0.02 to 0.11]

0.09 [0.04 to 0.17]

0.12 [0.07 to 0.21]

0.15 [0.08 to 0.24]

0.2 [0.1 to 0.33]

Precipitable Water (kg m2)

24.79 [23.06 to 26.82]

1.42 [0.7 to 2.26]

1.84 [1.03 to 2.62]

2.29 [1.6 to 3.09]

2.7 [1.92 to 3.92]

3.15 [2.13 to 4.38]

1.11 [0.28 to 2.13]

2.11 [0.98 to 3.15]

3.76 [2.41 to 5.08]

6.2 [4.24 to 8.83]

7.92 [5.21 to 10.69]

Global Land Annual

Precipitation (mm day–1)

2.27 [1.98 to 2.58]

0.07 [0.02 to 0.11]

0.07 [–0.0 to 0.13]

0.06 [0.01 to 0.13]

0.06 [0.02 to 0.12]

0.09 [0.01 to 0.16]

0.06 [0.01 to 0.1]

0.08 [0.02 to 0.16]

0.11 [0.02 to 0.19]

0.14 [0.03 to 0.22]

0.2 [0.07 to 0.32]

Precipitation – Evaporation (mm day–1)

0.87 [0.49 to 1.26]

0.02 [0.0 to 0.03]

0.02 [–0.01 to +0.05]

0.02 [–0.02 to +0.06]

0.03 [–0.0 to +0.06]

0.04 [0.0 to 0.1]

0.01 [–0.0 to +0.03]

0.03 [–0.01 to +0.08]

0.04 [–0.01 to 0.07]

0.07 [0.0 to 0.12]

0.1 [0.01 to 0.22]

Runoff (mm day–1)

0.79 [0.54 to 1.0]

0.02 [0.0 to 0.05]

0.04 [–0.0 to +0.1]

0.04 [–0.0 to +0.11]

0.04 [0.01 to 0.08]

0.06 [0.01 to 0.14]

0.02 [–0.0 to +0.03]

0.04 [–0.0 to +0.13]

0.06 [0.0 to 0.17]

0.1 [0.02 to 0.2]

0.15 [0.04 to 0.27]

Precipitable Water (kg m2)

18.86 [17.12 to 21.28]

1.23 [0.57 to 1.96]

1.58 [0.77 to 2.42]

1.96 [1.34 to 2.76]

2.33 [1.63 to 3.46]

2.72 [1.79 to 3.84]

0.95 [0.19 to 1.95]

1.78 [0.8 to 2.77]

3.18 [2.04 to 4.34]

5.33 [3.57 to 7.5]

6.81 [4.35 to 9.32]

In this Report, Chapter 4 provides an updated assessment of global annual precipitation (Section 4.3.1), finding that it is very likely that annual precipitation averaged over all land regions continuously increases as global surface temperatures increase in the 21st century (high confidence). CMIP6 projections for long-term changes in P–E (Figure 8.13) show that, for all scenarios, P–E increases over the tropics and high latitudes and decreases over the subtropics, resulting from a thermodynamically driven amplification of P–E patterns (Section Both the intensity of changes and the spread among the models is larger for the higher emissions scenarios. A less coherent latitudinal pattern and smaller magnitude of P–E changes over land reflect the complex influence of land–ocean warming contrast, atmospheric circulation change and vegetation feedbacks (Section However, stronger atmospheric moisture transport, increases in precipitation and evaporation over global land and ocean and larger continental runoff that is in part fed by melting of glaciers characterizes a more intense water cycle with global warming.

Figure 8.13 | Zonal and annual-mean projected long-term changes in the atmospheric water budget. Zonal and annual mean projected changes (mm day–1) in P (precipitation, left column), E (evaporation, middle column), and P–E (right column) over both land and ocean areas (coloured lines) and over land only (black lines) averaged across available CMIP6 models (number provided at the top left of each panel) in the SSP1-2.6 (top row), SSP2-4.5 (middle row) and SSP5-8.5 (bottom row) scenario, respectively. Shading denotes confidence intervals estimated from the CMIP6 ensemble under a normal distribution hypothesis. Colour shading denotes changes over both land and ocean. Grey shading represents internal variability derived from the pre-industrial control simulations. All changes are estimated for 2081–2100 relative to the 1995–2014 base period. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

Global and global land mean water cycle changes from CMIP6 projections are shown in Table 8.1. Increases in global and continental precipitation, P–E and runoff in both the mid-term and long-term illustrate the future intensification of the water cycle, with the magnitude of change increasing with emissions scenarios. Consistent with AR5, CMIP6 simulations of global mean precipitation show a systematic multi-model mean increase of 1.6 to 2.9 % °C–1 warming (apparent hydrological sensitivity; Section 8.2.1) by 20812100 relative to present day across the new SSP scenarios (using global surface air temperature change from Table 4.1). It is well understood that rising concentrations of CO2 drive a long-term increase in global precipitation with warming, but with the increase partly offset by rapid atmospheric adjustments to the direct atmospheric heating from radiative forcing agents (Section 8.2.1). The largest apparent hydrological sensitivity is found for SSP1-1.9, where the suppressing effects on precipitation from atmospheric heating by greenhouse gases (GHGs) rapidly reduce as their concentration falls. Additional warming due to reduced aerosol loadings under the SSP scenarios (Lund et al. , 2019) further increases global precipitation (Rotstayn et al., 2013; Wu et al., 2013; Salzmann, 2016; T.B. Richardson et al., 2018b; Samset et al., 2018b; Westervelt et al., 2018), with particularly strong contributions from increased monsoon rainfall over East and South Asia (Levy et al., 2013; Westervelt et al., 2015; Dwyer and O’Gorman, 2017).

Over global land there is a small range in global mean multi-model mean precipitation increase across scenarios in the mid-term (2.64.0 %), which widens (to 2.68.8 %) in the long-term (Table 8.1). The long-term projections are consistent with the Chapter 4 assessment that global annual precipitation over land is projected to increase on average by 2.4 [0.2 to +4.7] % (likely range) in the SSP1-1.9 low-emissions scenario and by 8.3 [0.9 to 12.9] % in the SSP5-8.5 high emissions scenario by 2081–2100 relative to 1995–2014. Small differences in assessed model mean changes in Chapter 4, Table 4.2 result from a slightly different set of models considered for Table 8.1. Over land, P–E increases by around 23% in the mid-term (apart from SSP5-8.5 where increases are almost 5%) and around 112% in the long-term, determined by increased moisture transport from the ocean to land Section Runoff increases are larger and less certain due to additional inputs from glacier melt and changes in groundwater storage (Section Overall, precipitation and runoff are very likely to increase over the global land in all scenarios in the mid- and long term. P–E is likely to increase over global land in the mid- and long term and very likely in SSP1-1.9, SSP3-7.0 and SSP5-8.5 pathways. The mid-term consistency in projections across scenarios is not apparent for precipitable water vapour, which increases over land by around 615% in the mid-term and 536% in the long-term across all scenarios. This implies that increases in extreme precipitation (closely related to atmospheric water vapour content; Section are dependent on mitigation pathway, even in the mid-term (Section 11.4.5). Water vapour residence time (computed as the ratio of precipitable water vapour to precipitation from values in Table 8.1) increases from eight days in the present to nine days in mid-term and up to about ten days in the long-term over land in SSP3-7.0, indicating a longer time to moisten the atmosphere between precipitation events. The CMIP6 projections are therefore consistent with anintensification but not acceleration of the global water cycle.

In summary, it is virtually certain that global water cycle intensity, considered in terms of global and continental mean precipitation, evaporation and runoff, will increase with continued global warming. Global annual precipitation over land is projected to increase on average by 2.4 [–0.2 to +4.7] % (likely range) in the SSP1-1.9 low-emissions scenario and by 8.3 [0.9 to 12.9] % in the SSP5-8.5 high emissions scenario by 2081–2100 relative to 1995–2014. Water Vapour and Its Transport

Globally, AR5 assessed that by the end of the 21st century, the average quantity of water vapour in the atmosphere could increase by 5–25%, depending on emissions. The AR5 assessed that increases in near-surface specific humidity over land are very likely , but that it was also likely that near-surface relative humidity would decrease over many land areas, although with only medium confidence. In terms of moisture transport, AR5 assessed that it was likely that moisture transport into the high latitudes would increase and that there was high confidence that, over the ocean, atmospheric moisture transport from the evaporative regions to the wet regions would increase.

CMIP6 climate models continue to project a steady increase in global mean column-integrated water vapour by around 613% by 20412060 and 532% by 20812100, depending on scenario (Table 8.1). This is consistent with projected atmospheric warming (Section and the Clausius–Clapeyron relationship (Section 8.2.1) where every degree Celsius of warming is associated with an approximate 7% increase in atmospheric moisture in the lower atmospheric layers where most of the water vapour is concentrated. This increase sustains a positive feedback on anthropogenic global warming (Section In contrast, the response of clouds is much more spatially heterogeneous, microphysically complex, and model-dependent so that the projected cloud feedbacks remain a key uncertainty for constraining climate sensitivity (Section

CMIP6 models project an overall decrease in near-surface relative humidity over land, although with some regional and seasonal variations in their response (Figure 4.26). Regional changes in near-surface humidity over land are dominated by thermodynamic processes and are primarily controlled by moisture transport from the warming ocean (Chadwick et al., 2016a). Increases in specific humidity lower than the thermodynamic rate are explained by greater warming over land than ocean and modulated by land atmosphere feedbacks such as soil moisture and plant stomatal changes (Section; Berg et al., 2017; Douville et al., 2020). This explains why climate models continue to project a contrasting response of near-surface relative humidity, with a slight and possibly overestimated increase over the oceans and a consistent but possibly underestimated decrease over land (Byrne and O’Gorman, 2016; Douville and Plazzotta, 2017; R. Zhang et al., 2018).

While projections of water vapour are well understood due to the constraints of the Clausius–Clapeyron relationship, projections of water vapour transport are complicated regionally by the role of changes in the wind field, which is influenced by a wide variety of factors. Additionally, there has been relatively little general evaluation of moisture transport in models. In CMIP5 models, both the mean and variability of the vertically-integrated moisture transport is projected to increase, largely due to increases in water vapour (Lavers et al., 2015), with substantial regional differences (Levang and Schmitt, 2015). Single-model studies have illustrated projected increases in low-altitude moisture transport into convergence regions (Allan et al., 2014) and from ocean to land (Zahn and Allan, 2013) that are consistent with present day trends. Increases in moisture transport have been linked to increases in large precipitation accumulations over land (Norris et al., 2019). Based on robust physics and supported by modelling studies, it is well understood that moisture transport increases into convergent parts of the atmospheric circulation such as storm systems, the tropical rain belt and high latitudes (Section, but changes in atmospheric circulation that are less well understood alter moisture transport regionally (Section Therefore, given the limited examination of moisture transport in models, regional projections should be considered with caution. Changes in moisture transport specifically associated with monsoons, atmospheric rivers, and other specific circulation features are discussed further in the following sections.

In summary, there is high confidence in continued increases in global mean column integrated water vapour and near-surface specific humidity over land. There is medium confidence in region and season-dependent decreases in near-surface relative humidity over land, due to the complex physical processes involved . In general, there will be increases in moisture transport into storm systems, monsoons and high latitudes (medium confidence). Precipitation Amount, Frequency and Intensity

This section assesses projected changes in precipitation at regional scales. Note that changes in precipitation seasonality are assessed in Box 8.2 and that changes in regional monsoons are assessed in Section, where both circulation and rainfall are considered. Further assessments of regional projections of precipitation are presented in Chapters 10, 12 and the Atlas, while a comprehensive assessment of changes in precipitation extremes is provided in Chapter 11.

The AR5 assessed that the contrast of mean precipitation amount between dry and wet regions and seasons is expected to increase over most of the globe as temperatures increase (high confidence), but with large regional variations. Precipitation over the high latitudes, equatorial Pacific Ocean, mid-latitude wet regions, and monsoon regions were assessed as likely to increase under the RCP8.5 scenario, and in many mid-latitude and subtropical dry regions as likely to decrease (AR5 Chapters 7, 12, and 14). Extreme precipitation over most mid-latitude land areas and wet tropical regions was assessed as very likely to become more intense and more frequent.

Geographical patterns of projected precipitation changes show substantial seasonal contrasts and regional differences, including over land (Figure 8.14 and Figure 4.27). Projections for 20812100under the SSP2-4.5scenariosuggest increased precipitation over the tropical oceans, north-eastern Africa, the Arabian Peninsula, India, south-eastern Asia and the Polar regions while decreased precipitation is projected mainly over the subtropical regions (Section Precipitation changes contrast regionally in the tropics with wetter wet seasons over South Asia, central Sahel and eastern Africa, but less precipitation over Amazonia and coastal West Africa (Section These large-scale responses are associated with stronger moisture transports in a warmer climate that are modulated by the greater warming over land than ocean, atmospheric circulation responses and land surface feedbacks (Section 8.2.2). There is agreement across CMIP5 and CMIP6 modelling studies that precipitation increases in wet parts of the atmospheric circulation and decreases in dry parts (Liu and Allan, 2013; Kumar et al., 2015; Deng et al., 2020; Schurer et al., 2020) although these regions shift with atmospheric circulation changes. The overall pattern is robust across different model scenarios and time horizons (Tebaldi and Knutti, 2018), but some deviations from the mean pattern cannot be excluded due to the multiple time scales and non-linear atmospheric or land surface processes involved (Section 8.5.3). Near-term regional changes in precipitation are more uncertain because of a stronger sensitivity to natural variability (Section 8.5.2) and non-GHG anthropogenic forcings (Section and

Figure 8.14 | Projected long-term relative changes in seasonal mean precipitation. Global maps of projected relative changes (%) in seasonal mean of precipitation averaged across available CMIP6 models (number provided at the top right of each panel) in the SSP2-4.5 scenario. All changes are estimated for 2081–2100 relative to the 1995–2014 base period. Uncertainty is represented using the simple approach. No overlay indicates regions with high model agreement, where ≥80% of models agree on sign of change; diagonal lines indicate regions with low model agreement, where <80% of models agree on sign of change. For more information on the simple approach, please refer to the Cross-Chapter Box Atlas.1. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

Projected changes in regional precipitation also arise as a response to changes in large-scale atmospheric circulation (Section and 8.4.2), both in the tropics (Chadwick et al. , 2016b; Byrne et al. , 2018) and extratropics (Shaw, 2019; Oudar et al., 2020b). Despite variability in simulated changes, CMIP5 climate models consistently project large rainfall changes (of varying sign) over considerable proportions of tropical land during the 21st century (Chadwick et al., 2016b). Since AR5, some robust responses in large-scale circulation patterns have been identified. For example, and as further assessed in Section 8.4.2, CMIP6 models project a northward shift in the tropical rain belt over eastern Africa and the Indian Ocean and a southward shift in the eastern Pacific and Atlantic oceans (Mamalakis et al., 2021). A projected strengthening and tightening of the tropical rain belt increases the contrasts between wet and dry tropical weather regimes and seasons. It is less clear how the well understood poleward expansion of the subtropics and mid-latitude storm tracks influences precipitation over subtropical and mid-latitude continents (Section

An ensemble of 31 CMIP6 models under the SSP5-8.5 scenario projects increases precipitation by 10–30% over much of the USA and decreases by 10–40% over Central America and the Caribbean by 20802099 (Almazroui et al., 2021). This CMIP6 ensemble also projects an increase in annual precipitation over the southern Arabian Peninsula and a decrease over the northern Arabian Peninsula, as also projected by CMIP3 and CMIP5 models (Almazroui et al., 2020a). Annual mean precipitation is projected to increase over South Asia during the 21st century under all scenarios, although the rate of change varies within the region based on 27 CMIP6 models (Almazroui et al., 2020c). CMIP6 projections also display a reduction in annual mean precipitation over northern and southern Africa while increases are projected over Central Africa, under the SSP1-2.6, SSP2-4.5 and SSP5-8.5 scenarios (Almazroui et al., 2020b). The AR6 (Atlas assesses that regions where annual mean rainfall is likely to increase include the Ethiopian Highlands, East, South and North Asia, south-eastern South America, northern Europe, northern and eastern North America, and the Polar Regions. In contrast, regions where annual mean rainfall is likely to decrease include southern Africa, coastal West Africa, Amazonia, south-western Australia, Central America, south-western South America, and the Mediterranean.

The AR5 identified that high-latitude precipitation increase may lead to an increase in snowfall in the coldest regions and a decrease of snowfall in warmer regions due to a decreased number of freezing days. The fraction of precipitation falling as snow and the duration of snow cover was projected to decrease. Heavy snowfall events globally are not expected to decrease significantly with warming as they occur close to the water freezing point, which will migrate poleward and in altitude (O’Gorman, 2014; Turner et al., 2019). There are only a small number of studies evaluating the implications of this mechanism in specific regions. A study for the north-eastern USA indicates smaller reductions for major snowfall events against the broader decline in snowfall expected from thermodynamic effects (Bintanja and Andry, 2017). Arctic snowfall is projected to decrease as rainfall makes up more of the precipitation (Zarzycki, 2018).

Beyond annual or seasonal mean precipitation amounts, an implication of the parallel intensification of the global water cycle and of the increased residence time of atmospheric water vapour (Section 8.2.1) is that the distribution of daily and sub-daily precipitation intensities will experience significant changes (Pendergrass and Hartmann, 2014b; Pendergrass et al., 2015; Bador et al., 2018; Douville and John, 2021), with fewer but potentially stronger events (high confidence) (Section 4.3.3). CMIP6 projections show that in the long-term more drier days but more intense single events of precipitation are expected, regardless of scenario (Figure 8.15). Over almost all land regions, it is very likely that extreme precipitation will intensify at a rate close to the 7% °C–1 of global warming, but with large spatial differences (Sections 11.4 and The projected increase in precipitable water is expected to lead to an increase in the highest possible precipitation intensities and an increase in the probability of occurrence of extreme precipitation events on the global scale (Neelin et al., 2017), regardless of how annual-mean precipitation changes (O’Gorman and Schneider, 2009; O’Gorman, 2015). The projected increase in heavy precipitation intensity is also found for daily mean precipitation intensity though at a lower rate (Pendergrass and Hartmann, 2014a).

Figure 8.15 | Projected long-term relative changes in daily precipitation statistics. Global maps of projected seasonal mean relative changes (%) in the number of dry days (i.e., days with less than 1 mm of rain) and daily precipitation intensity (in mm day1, estimated as the mean daily precipitation amount at wet days – for example, days with intensity above 1 mm day1) averaged across available CMIP6 models (number provided at the top right of each panel) in the SSP1-2.6 (a, b), SSP2-4.5 (c, d) and SSP5-8.5 (e, f) scenario respectively. Uncertainty is represented using the simple approach. No overlay indicates regions with high model agreement, where ≥80% of models agree on sign of change; diagonal lines indicate regions with low model agreement, where <80% of models agree on sign of change. For more information on the simple approach, please refer to the Cross-Chapter Box Atlas.1. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

An increase in the number of dry days is also projected in several regions of the world (Polade et al., 2014; Berthou et al., 2019a), which can dominate the annual precipitation change at least in the subtropics (Polade et al., 2014; Douville and John, 2021). These findings are supported by CMIP6 projections showing a widespread increase in daily mean precipitation intensity over land (Figure 8.15b,d,f) as well as an increase in the number of dry days in the subtropics and over Amazonia and Central America (Figure 8.15a,b,c). Such changes in precipitation regimes, as well as the general increase in the frequency and intensity of precipitation extremes (Section 11.4.5), contribute to an overall increase in precipitation variability (Polade et al., 2014; Pendergrass et al., 2017; Douville and John, 2021). This is also found in CMIP6 models, which show a stronger increase of interannual variability than in seasonal mean precipitation changes, apart from in the winter extratropics where both quantities increase at the same rate with increasing global warming levels (Figure 8.16).

Figure 8.16 | Rate of change in components of water cycle mean and variability across increasing global warming levels. Relative change (%) in seasonal mean total precipitable water (grey line), precipitation (red solid lines), runoff (blue solid lines), as well as in standard deviation of precipitation (red dashed lines) and runoff (blue dashed lines) averaged over extratropical land in (c) summer and (d) winter, and tropical land in (a) June–July–August (JJA) and (b) December–January–February (DJF) as a function of global mean surface temperature for the CMIP6 multi-model mean across the SSP5-8.5 scenario. Extratropical winter refers to DJF for Northern Hemisphere and JJA for Southern Hemisphere (and the reverse for extratropical summer). Each marker indicates a 21-year period centred on consecutive decades between 2015 and 2085 relative to the 1995–2014 base period. Precipitation and runoff variability are estimated by their standard deviation after removing linear trends from each time series. Error bars show the 5–95% confidence interval for the warmest 5°C global warming level. Figure adapted from Pendergrass et al. (2017) and updated with CMIP6 models. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

In summary, it is virtually certain that global precipitation will increase with warming due to increases in GHG concentrations and decreases in air pollution. There is high confidence that total precipitation will increase in the high latitudes, with a shift from snowfall to rainfall except in the coldest regions and seasons. There is also high confidence that precipitation will decrease over the Mediterranean, southern Africa, Amazonia, Central America, south-western South America, south-western Australia and coastal West Africa and that monsoon precipitation will increase over South Asia, East Asia and central-eastern Sahel. See (Section for a more detailed assessment of changes in regional monsoons. Daily mean precipitation intensities, including extremes, are projected to increase over most regions (high confidence). The number of dry days is projected to increase over the subtropics, Amazonia, and Central America (medium confidence). There is high confidence in an overall increase in precipitation variability over most land areas.

Box 8.2 | Changes in Water Cycle Seasonality

Observed changes

The AR5 did not highlight observed changes in water cycle seasonality and SRCCL mostly emphasized changes in vegetation seasonality. Since AR5, a number of relevant studies have been published, but often with conflicting results. Based on three in situdatasets, reduced precipitation seasonality was identified over 62% of the terrestrial ecosystems analysed from 19502009 (Murray-Tortarolo et al. 2017). In contrast, both in situ and satellite data show a general increase in the annual range of precipitation from 1979 to 2010, which is dominated by wetter wet seasons (Chou et al., 2013). This paradox may be partly explained by a larger aerosol radiative forcing in the middle of the 20th century as well as by internal variability (Kumar et al., 2015; see also Box 8.1). For instance, the ‘long rains’ over East Africa experienced declining trends in the 1980s and 1990s (Nicholson, 2017), which was linked to anthropogenic aerosols and SST patterns (Rowell et al., 2015), followed by a recent recovery that was linked to internal variability (Wainwright et al., 2019). Two satellite datasets revealed decreased rainfall seasonality in the tropics but an increased seasonality in the subtropics and mid-latitudes since 1979, without clear attribution (Marvel et al., 2017).

Large differences have been found across seven global precipitation datasets, with no region showing a consistent, statistically significant, positive or negative trend over the last three decades (X. Tan et al., 2020). Regional studies suggest that observed changes in precipitation seasonality are neither uniform nor stable across the 20th century (X. Li et al. , 2016; Mallakpour and Villarini, 2017; Sahany et al. , 2018; Deng et al. , 2019). Since the 1980s, there is growing evidence that contrasts between wet and dry regimes, including seasonality, have increased (Liu and Allan, 2013; Polson et al. , 2013; Murray-Tortarolo et al. , 2016; Tapiador et al. , 2016; Gallego et al. , 2017; Polson and Hegerl, 2017; Barkhordarian et al. , 2018; Lan et al. , 2019; Liang et al. , 2020; Schurer et al., 2020).

Additional changes in seasonality may manifest in the timing and duration of wet seasons. A later monsoon onset trend was reported throughout India from 1901 to 2013 (Sahany et al., 2018). Conversely, an earlier rainfall onset was implicated in increased springtime rainfall over the Tibetan Plateau in recent decades (W. Zhang et al., 2017a). Winter and early spring precipitation over the north-western Himalaya for the period 19512007 shows an increasing trend of daily precipitation extremes in association with enhanced amplitude variations of extratropical synoptic-scale systems known as ‘Western Disturbances’ (Madhura et al. , 2014; Cannon et al. , 2015; Krishnan et al. , 2019). In China, an earlier onset was observed during 1961-2012 (Deng et al., 2019). In the African Sahel, rainfall has been most concentrated in the peak of the rainy season since the end of the 20th century (Biasutti, 2019). A shift in the seasonality of Sahelian rainfall, including delayed cessation has also been reported (Section; Nicholson, 2013; Dunning et al., 2018). Over southern Africa, an observed earlier onset (19852007) is in contrast to a simulated historical and projected future delay in the wet season (Maidment et al., 2015; Dunning et al., 2018). An increasingly early onset of the North American monsoon has been observed from 1978 to 2009 (Arias et al. , 2015). Seasonality changes in the South American monsoon indicate delayed onsets since 1978 (Fu et al. , 2013; Yin et al. , 2014; Arias et al. , 2015; Debortoli et al. , 2015; Arvor et al. , 2017; Giráldez et al. , 2020; Haghtalab et al. , 2020; Correa et al., 2021).

In northern high latitudes, a shorter snow season (X. Zeng et al., 2018) is mainly due to an earlier onset of spring snowmelt (Peng et al., 2013) which has been attributed to anthropogenic climate change (Najafi et al., 2016). Changes in snow seasonality affect streamflow at the regional scale, with an earlier peak in spring and a possible decrease of low-level flow in summer (Berghuijs et al., 2014; Kang et al., 2016; Dudley et al., 2017), while glacier shrinking can also alter the low-level flow in mountain catchments (Lutz et al., 2014; Milner et al., 2017; Huss and Hock, 2018). This can be partly ameliorated by water management in regulated catchments (Arheimer et al., 2017), but not in large river basins such as the Amazon which also shows an increased seasonality of discharge since 1979 (Liang et al., 2020).

Increasing aridity contrasts between wet and dry seasons over the late 20th century have been suggested (Kumar et al., 2015), with a human-induced decrease of water availability during the dry season over Europe, western North America, northern Asia, southern South America, Australia and eastern Africa (Padrón et al., 2020). Seasonal contrasts in microwave surface soil moisture measurements have also increased over 19792016 (Pan et al., 2019). Terrestrial water storage variations derived from gravimetric measurements since 2003 show a strong seasonality which is underestimated by global hydrological models (Scanlon et al., 2019) and whose multi-decadal trends are difficult to interpret given the direct effect of enhanced water use (Rodell et al., 2018; Scanlon et al., 2018).

In summary, there is medium confidence that the annual range of precipitation has increased since the 1980s, at least in subtropical regions and over the Amazon. There is low confidence that this increase is due to human influence and that GHG forcing has already altered the timing or duration of wet seasons . There is high confidence that the human-induced retreat of the springtime snow cover and melting of glaciers have already contributed to changes in streamflow seasonality in high-latitude and low-elevation mountain catchments, and medium confidence that human activities have also contributed to an increased seasonality of water availability, including a drier dry season, in the extratropics.

Project ed changes

The AR5 reported with high confidence that the contrast between wet and dry seasons will generally increase with global warming and that monsoon onset dates will likely become earlier or show little change, while monsoon retreat dates will likely be delayed, resulting in a lengthening of the wet season in many regions.

Since AR5, several studies have further documented a projected increase in rainfall seasonality and the understanding of the underlying mechanisms has been improved (Sections 8.2.1 and 8.3.2). CMIP5 models show that the seasonal concentration of annual precipitation will increase over many regions by the end of the 21st century, with robust model agreement in most subtropical regions where an increase in the mean number of dry days was also reported in the RCP8.5 scenario (Pascale et al., 2016). The semi-arid, winter rainfall dominated subtropical climate is projected to shift poleward and eastward, with the equatorward margins replaced by a more arid climate type. However, evolving SST patterns and land ocean warming contrasts cause more complex responses (Alessandri et al. , 2015; Polade et al. , 2017; Brogli et al. , 2019; Zappa et al. , 2020). Projections over California show a stronger and shorter wet season (Polade et al., 2017; Dong et al., 2019). Decreases in future winter and spring rainfall are projected over south-western Australia (Hope et al., 2015). Central Asia is projected to experience wetter winters, associated with an increase in snow depth in the north-eastern regions (Y. Li et al., 2019). Even in a +2°C climate, both extreme precipitation and dryness will increase significantly in the extratropics, amplifying the seasonal precipitation range (Fujita et al., 2019). A single-model study shows that the annual range of precipitation increases globally by 2.6% per 1°C of global warming in stabilized low-warming scenarios (Z. Chen et al., 2020a).

In the tropics, an amplified annual cycle (by about 3–5% °C–1) of global land monsoon hydroclimates (precipitation, precipitation minus evaporation (P–E), and runoff) is projected by CMIP5 models under the RCP8.5 scenario, mostly due to a more intense wet season (W. Zhang et al., 2019b). A longer rainy season is projected by CMIP6 models over most regional monsoon areas except in the Americas (Moon and Ha, 2020). A delayed onset and cessation of the wet season over West Africa and the Sahel (Dunning et al., 2018) and a slightly delayed onset of South Asian monsoon rainfall (Hasson et al., 2016) are projected by CMIP5 models. CMIP5 projections suggest a strengthening of the annual cycle and a lengthening of the dry season in Southern Amazonia (Fu et al. , 2013; Reboita et al. , 2014; Boisier et al. , 2015; Pascale et al. , 2016; Sena and Magnusdottir, 2020). This is further verified by the projections from six CMIP6 models (Moon and Ha, 2020). A wet season shorter by 510 days by the end to the 21st century is projected for southern Africa (Dunning et al., 2018).

An increase in streamflow seasonality is projected over several large rivers in the low-mitigation RCP8.5 scenario, but with only small changes in the seasonality timing, except in northern high latitudes due to the earlier but potentially slower snowmelt in a warmer world (Eisner et al., 2017; Musselman et al., 2017). At the end of the century in a high-emissions scenario, peak snowmelt timing is projected to occur one month earlier and peak water volume is 79% lower in the eastern USA (Rhoades et al., 2018). Earlier snowmelt is projected, for example, by 30 days at the end of the 21st century in RCP4.5 for the Sierra Nevada in the western USA (F. Sun et al., 2018). Sub-seasonal changes in water availability were found in many regions in the RCP8.5 scenario. However, these should be considered with caution given the magnitude of model errors (C.R. Ferguson et al., 2018). Increases in the seasonality of water availability has been found to be more pronounced in areas with high atmospheric evaporative demand, giving rise to a pattern of seasonally variable regimes becoming even more variable (Konapala et al., 2020). RCP4.5 and RCP8.5 projections show a pronounced soil drying in summer and autumn over western Europe, and a springtime drying over northern Europe due to an earlier snowmelt (Ruosteenoja et al., 2018).

A simple relative seasonality metric (Walsh and Lawler, 1981) applied to global projections based on CMIP6 models and SSP scenarios supports previous CMIP5 findings, especially the amplified seasonality of precipitation around the Mediterranean, and across southern Africa, California, southern Australia and the Amazon (Box 8.2, Figure 1). While such changes are not significant in the low-emissions SSP1-2.6 scenario, they are consistent with the increased frequency of dry days projected over the same regions (Figure 8.16). In monsoon regions outside the Americas, rainfall seasonality does not show a significant increase even in high-emissions scenarios. This challenges previous CMIP5 findings based on the difference between maximum and minimum monthly precipitation in a year (W. Zhang et al., 2019b) and higher sensitivity to the projected increase in precipitation extremes (Section 11.4.5). In the northern high latitudes, milder winters are associated with wetter conditions and a decrease in precipitation seasonality.

Box 8.2, Figure 1 | Projected long-term changes in precipitation seasonality. Global maps of projected changes in precipitation seasonality (simply defined as the sum of the absolute deviations of mean monthly rainfalls from the overall monthly mean, divided by the mean annual rainfall as in Walsh and Lawler, 1981) averaged across available CMIP6 models (number provided at the top right of each panel) in the SSP1-2.6 (b), SSP2-4.5 (c) and SSP5-8.5 (d) scenario respectively. The simulated 1995–2014 climatology is shown in panel (a). All changes are estimated in 2081–2100 relative to 1995–2014. Uncertainty is represented using the simple approach. No overlay indicates regions with high model agreement, where ≥80% of models agree on sign of change. Diagonal lines indicate regions with low model agreement, where <80% of models agree on sign of change. For more information on the simple approach, please refer to the Cross-Chapter Box Atlas.1. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

In summary, the annual range of precipitation, water availability and streamflow will increase with global warming over subtropical regions and the Amazon (medium confidence), especially around the Mediterranean and across southern Africa (high confidence). The contrast between the wettest and driest month of the year is likely to increase by 3–5% °C–1 with global warming in most monsoon regions, in terms of precipitation, water availability (P–E) and runoff (medium confidence). There is medium confidence that the monsoon season could be delayed in a warmer climate in the Sahel. There is high confidence of earlier snowmelt. Evapotranspiration

Since AR5, there is a growing body of evidence suggesting that future projections in evapotranspiration are driven by changes in temperature and relative humidity (Laîné et al. , 2014; Pan et al. , 2015; Ukkola et al. , 2016a), as well as precipitation patterns, as found in AR5.

Analysis of CMIP5 models suggests that atmospheric evaporative demand will increase over most areas of the world in high-emissions scenarios (virtually certain), mostly as a consequence of an increase in vapour pressure deficit (Scheff and Frierson, 2014, 2015; Greve and Seneviratne, 2015; Vicente-Serrano et al., 2020). CMIP5 models also project an increase in evapotranspiration over most land areas (medium confidence) (Laîné et al., 2014). However, regional changes in evapotranspiration can also be influenced by changes in soil moisture and vegetation, which modulate the moisture flux from the land to the atmosphere. Several studies of CMIP5 projections suggest that increases in plant water use efficiency will limit or counteract rising evapotranspiration (Milly and Dunne, 2016; Swann et al. , 2016; Lemordant et al. , 2018; Y. Yang et al. , 2018). However, other studies have found that transpiration increases due to the impact of climate change on growing season length, leaf area, and evaporative demand (Section; Frank et al. , 2015; Mankin et al. , 2017, 2018, 2019; Guerrieri et al. , 2019; S. Zhou et al. , 2019; Vicente-Serrano et al. , 2020). The parametrizations accounting for these complex physiological processes in global climate models may also be insufficient (Franks et al., 2017; Peters et al., 2018; Peano et al., 2019). Thus, there is currentlylow confidence in the role of vegetation physiology in modulating future projections of evapotranspiration.

CMIP6 models project a geographical pattern of changes in evapotranspiration similar to previous generation models (Figure 8.17), although the magnitude is generally larger than found for CMIP5 projections (X. Liu et al., 2020). There is, however, a strong seasonality in many regions, with a larger relative increase in the winter season of the Northern Hemisphere (NH) and smaller relative changes in the summer (Figure 8.17). Evapotranspiration increases in most land regions, except in areas that are projected to become moisture-limited (due to reduced precipitation and increased evaporative demand), such as the Mediterranean, South Africa, and the Amazonian basin (medium confidence). The patterns of change increase in magnitude from low to high-emissions SSP scenarios (medium confidence).

Figure 8.17 | Projected long-term relative changes in seasonal mean evapotranspiration. Global maps of projected relative changes (%) in seasonal mean of surface evapotranspiration for December–January–February (DJF; left panels) and June–July–August (JJA; right panels) averaged across available CMIP6 models (number provided at the top right of each panel) for SSP1.2-6 (a, b) SSP2-4.5 (c, d) and SSP5-8.5 (e, f) scenario respectively. All changes are estimated in 2081–2100 relative to 1995–2014. Uncertainty is represented using the simple approach. No overlay indicates regions with high model agreement, where ≥80% of models agree on sign of change; diagonal lines indicate regions with low model agreement, where <80% of models agree on sign of change. For more information on the simple approach, please refer to the Cross-Chapter Box Atlas.1. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

In summary, future projections indicate that anthropogenic forcings will drive an increase in global mean evaporation over most oceanic areas (high confidence) (Figure 8.17), an increase in global atmospheric demand (virtually certain) and an increase in evapotranspiration over most land areas, with the exception of moisture-limited regions (medium confidence). However, substantial uncertainties in projections of evapotranspiration, especially at seasonal and regional scales, remain (see also Section and Cross-Chapter Box 5.1). Runoff, Streamflow and Flooding

The AR5 assessed that projected changes in runoff had low confidence over the period 2016–2035; however, under the RCP8.5 scenario, runoff increases by 2100 are likely in high northern latitudes. This is consistent with projected regional precipitation increases, based on consistency of changes across different generations of models and different forcing scenarios, and with runoff decreases being likely in southern Europe, the Middle East and southern Africa. There was considerable uncertainty in the magnitude and direction of change for some regions, largely driven by the uncertainty in projected precipitation changes, particularly across south Asia. For flooding, AR5 assessed with medium confidence that flooding would increase over parts of South and South East Asia, tropical Africa, north-east Eurasia, and South America, and decrease for parts of Northern and Eastern Europe, Anatolia, Central Asia, Central North America, and southern South America. The SR1.5 assessed with medium confidence that warming of 2°C would increase the fraction of global area affected by flood hazard relative to warming of 1.5°C. Projected climate-driven changes to runoff, streamflow, and flooding will occur in the context of potential human-caused land-use and land-cover changes, which can have a large influence on surface water (Sterling et al., 2013) but which have considerable uncertainty in projections (Prestele et al., 2016).

Since AR5, studies confirm that global mean annual runoff increases with global surface temperature increase (X. Zhang et al. , 2014, 2018; Lehner et al. , 2019), but varies regionally (Chen et al. , 2017; H. Yang et al. , 2017; Cook et al. , 2020). CMIP5 models display a large spread in the ratio of runoff to precipitation for the present-day climate, which applies also to future runoff changes under global warming (Lehner et al., 2019). In studies of CMIP6 projections, runoff increases in most parts of the northern high latitudes and Asia and north and eastern Africa, and decreases in the Mediterranean region, southern Africa, southern Australia and in parts of western Africa, as well as in Central and South America (Greve et al., 2018; Cook et al., 2020). Projected changes in runoff also vary seasonally. In the Northern Hemisphere (NH), runoff increases during winter since more precipitation falls as rain than snow and decreases in the summer as less snow is available to contribute to runoff during the warm season (Cook et al., 2020). Global maps of projected changes for December–January–February and June–July–August are shown in Figure 8.18, showing projected changes becoming larger and more consistent in the higher emissions scenarios. Runoff projections for CMIP6 are also shown in Figure 8.16 for tropical and extratropical averages at a range of global mean warming levels and in Table 8.1 for global land in different future scenarios. In the tropics, both the mean and interannual variability of runoff increase with warming. The increase in variability is roughly twice as large as the increase in the mean, and has a large spread across models. In the extratropics, changes are small in the summer but there are large increases in the winter, with the mean increasing much more than the variability, in contrast to the tropics.

Figure 8.18 | Projected long-term relative changes in seasonal mean runoff. Global maps of projected relative change (%) in runoff seasonal mean for December–January–February (DJF; left panels) and June–July­–August (JJA; right panels) averaged across available CMIP6 models (number provided at the top right of each panel) SSP1.2-6 (a, b), SSP2-4.5 (c, d) and SSP5-8.5 (e, f) scenario respectively. All changes are estimated in 2081–2100 relative to 1995–2014. Uncertainty is represented using the simple approach. No overlay indicates regions with high model agreement, where ≥80% of models agree on sign of change, diagonal lines indicate regions with low model agreement, where <80% of models agree on sign of change. For more information on the simple approach, please refer to the Cross-Chapter Box Atlas.1. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

Changes in streamflow vary regionally and increase in magnitude with emissions scenarios, as with runoff (although the two are not equivalent, as runoff includes both surface runoff and streamflow). Streamflow projections additionally require the use of hydrologic models forced by the output from climate models and have not been as widely explored as they are not variables directly included in climate models. On an annual basis, streamflows have been projected to increase in the northern high latitudes and tropical Asia and Africa, and to decrease in the Mediterranean, tropical South America, and South Africa (Döll et al., 2018). For a 4°C global warming, half of the global land area is projected to be exposed to increased high flows (average increase 25%), while about 60% may be exposed to decreased low flows (average decrease 50%) (Asadieh and Krakauer, 2017).

Changes in the seasonality of runoff and streamflow are assessed in Box 8.2. The seasonality of runoff and streamflow (calculated as the annual difference between the wettest and driest months of the year), is expected to increase with global warming in the subtropics, especially in the Mediterranean and southern Africa with high confidence, and in the Amazon with medium confidence. For regions where snowmelt is an important contributor to streamflow, there is high confidence that snowmelt occurring earlier in the year will result in peak flows also occurring earlier in the year, and medium confidence that reduced snow volume and the weaker solar radiation earlier in the year will reduce the most intense flows (see Section In roughly half of 56 large-scale glacierized drainage basins, projected runoff changes show an increase until a maximum is reached, beyond which runoff steadily declines because of limited ice volumes (Huss and Hock, 2018).

As future changes in flood events are assessed in Chapters 9, 11 and 12, only a summary is presented here. There are a number of complicating factors for projecting both pluvial (overland) and fluvial (river) flooding that limit confidence in their assessment. In addition to precipitation, flooding also depends on basin and river characteristics such as permeability, antecedent soil moisture, and antecedent flow levels for river flooding, so projections of extreme precipitation and flooding are not always closely linked (Section Possible changes in water resources management and land use add another layer of complexity to future changes. There is medium confidence in a general increase in pluvial and fluvial flooding, although there are large regional variations, discussed further in Sections 11.5.5, and 12.4. There is medium confidence in a substantial increase in the frequency of extreme sea level events for coastal regions (Section and the associated coastal flooding is regionally assessed in Section 12.4. The risk of glacier lake outburst floods (GLOFs) is expected to increase with glacier melting in some high mountain regions (Section 12.4).

In summary, there is medium confidence that global runoff will increase with global warming, but with large regional and seasonal variations. There is high confidence that runoff will increase in the northern high latitudes and decrease in the Mediterranean region and southern Africa. There is medium confidence that runoff will increase in regions of central and eastern Africa, and decrease in Central America and parts of southern South America, with the magnitude of the change increasing with emissions. There is medium confidence that the seasonality of runoff and streamflow will increase with global warming in the subtropics. In snow-dominated regions, there is high confidence that peak flows associated with spring snowmelt will occur earlier in the year and medium confidence that snowmelt-induced runoff will decrease with reduced snow, except in glacier-fed basins where runoff may increase in the near term. There is medium confidence that flooding in general will increase, although with considerable variation based on geographic region and flood type. These projected climate-related changes will occur in the context of human-caused land-use and land-cover changes, which may also have a large influence. Aridity and Drought

The AR5 concluded that regional to global-scale projections of aridity and drought remained relatively uncertain compared to other aspects of the water cycle. It reported that there is a likely increase in drought occurrence (medium confidence) by 2100 in regions that are currently drought-prone under the RCP8.5 scenario due to projected decreases in soil moisture. It stated that it is likely that the most prominent projected decreases in soil moisture would occur in the Mediterranean, south-western USA, and southern Africa, consistent with projected changes in the Hadley circulation and increased surface temperatures. These AR5 conclusions are generally supported by more recent analyses of CMIP5 models (Feng and Fu, 2013; Berg et al., 2017; Cook et al., 2018).

Results from the latest generation of models in CMIP6 are largely congruent with CMIP5. Consistent with the coherent nature of warming in future projections, increases in vapour pressure deficit and evaporative demand are widespread and consistent across regions, seasons, and models, increasing in magnitude in accordance with the emissions scenario (high confidence) (Figure 8.19; Scheff and Frierson, 2014, 2015; Vicente-Serrano et al., 2020). Even under a low-emissions scenario (SSP1-2.6), projections of soil moisture show significant decreases in the Mediterranean, southern Africa, and the Amazonian basin (high confidence) (Figure 8.19). Under mid- and high-emissions scenarios (SSP2-4.5 and SSP5-8.5), coherent declines emerge across Europe, westernmost North Africa, south-western Australia, Central America, south-western North America, and south-western South America (high confidence) (Figure 8.19; Cook et al., 2020). Compared to CMIP5 results, CMIP6 models exhibit more consistent drying in the Amazonian basin (Parsons, 2020), more extensive declines in total soil moisture in Siberia (Cook et al., 2020), and stronger declines in westernmost North Africa and south-western Australia (Figure 8.19).

Figure 8.19 | Projected long-term relative changes in annual mean soil moisture and vapour pressure deficit. Global maps of projected relative changes (%) in annual mean vapor pressure deficit (left), surface soil moisture (top 10cm, middle) and total column soil moisture (right) from available CMIP6 models (number provided at the top right of each panel) for the SSP1.2-6 (a, b, c), SSP2-4.5 (d, e, f) and SSP5-8.5 (g, h, i) scenarios respectively. All changes are estimated for 2081–2100 relative to a 1995–2014 base period. Uncertainty is represented using the simple approach. No overlay indicates regions with high model agreement (‘Robust change’), where ≥80% of models agree on sign of change, diagonal lines indicate regions with low model agreement, where <80% of models agree on sign of change. For more information on the simple approach, please refer to the Cross-Chapter Box Atlas.1. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

Soil moisture in the top soil layer (10 cm) shows more widespread drying than total soil moisture, reflecting a greater sensitivity of the upper soil layer to increasing evaporative demand (Figure 8.19; Berg et al., 2017). Conversely, total column soil moisture represents the carry-over of moisture from previous seasons deeper in the soil column, and potentially higher sensitivity to vegetation processes (Berg et al., 2017; Kumar et al., 2019). Central America, the Amazonian basin, the Mediterranean region, southern Africa, and south-western Australia are projected to experience significant declines in total soil moisture, whereas declines in Europe (north of the Mediterranean), western Siberia, and north-eastern North America are limited to the surface (Figure 8.19). It should be noted that because models differ in their number of hydrologically active layers, there is less confidence in total soil moisture projections than surface soil moisture projections. Based on surface soil moisture projections, more than 40% of global land areas (excluding Antarctica and Greenland) are expected to experience robust year-round drying, even under lower emissions scenarios (Cook et al., 2020). The percentage of land area experiencing drying is slightly lower when runoff is used as an aridity metric instead (20–30%); taking this into consideration, it is estimated that about a third of global land areas will experience at least moderate drying in response to anthropogenic emissions, even under SSP1-2.6 (medium confidence) (Cook et al., 2020).

Although there are regions where multiple models predict consistent and significant changes in soil moisture, as with evapotranspiration (Section, there is still uncertainty in these projections related to the response of plants to elevated CO2. Most models project increases in two variables that have opposite effects on surface water availability: plant water use efficiency (WUE) and leaf area index (LAI; see Section As discussed in Sections, and, there is low confidence in how these changes in plant physiology will affect future projections of evapotranspiration, and likewise, drought and aridity.

Changes in meteorological (precipitation-based) drought duration and intensity in CMIP6 models are more robust than projected changes in mean precipitation, more than found in CMIP5 projections (Ukkola et al., 2020). Significant increases in drought duration are expected in Central America, the Amazonian basin, south-western South America, the Mediterranean, westernmost North Africa, southern Africa, and south-western Australia, on the order of 0.5 to 1 month for a moderate emissions scenario (SSP2-4.5) and two months for a high-emissions scenario (SSP5-8.5; Ukkola et al., 2020). Drought intensity is projected to increase in the tropics, mainly in the Amazonian basin, Central Africa, and southern Asia, as well as in Central America and south-western South America (Ukkola et al., 2020). The CORDEX South Asia multi-model ensemble projections indicate an increase in the frequency and severity of droughts over central and northern India during the 21st century, under the RCP4.5 and RCP8.5 scenarios (medium confidence) (Mujumdar et al., 2020). Under intermediate or high-emissions scenarios, the likelihood of extreme droughts (events that have magnitudes equal to or less than the 10th percentile of the 1851–1880 baseline period) increases by 200–300% in the Amazonian basin, south-western North America, Central America, the Mediterranean, southern Africa, and south-western South America (Cook et al., 2020). Even under a low-emissions scenario (SSP1-2.6), the likelihood of extreme droughts increases by 100% in south-western North America, south-western South America, the Amazon, the Mediterranean, and southern Africa (Cook et al., 2020). Thus, there is high confidence that drought severity and intensity will increase in the Mediterranean, southern Africa, south-western South America, south-western North America, south-western Australia, Central America and the Amazonian basin.

Paleoclimate records provide context for these future expected changes in drought and aridity. In the Mediterranean, western North America, and Central Chile, there is high confidence that climate change will shift soil moisture (as represented by the Palmer Drought Severity Index) outside the range of observed and reconstructed values spanning the last millennium (Figure 8.20; Cook et al., 2014; Otto-Bliesner et al., 2016). Warmer temperatures, leading to increased evaporative losses, are clearly implicated in the projected future drying in these semi-arid regions (Dai et al., 2018), emphasizing the central role that warming plays in driving increased evaporative demand (Vicente-Serrano et al., 2020). In contrast, future trajectories are more uncertain in regions like Central Asia and eastern Australia–New Zealand where projected changes in precipitation and soil moisture are less coherent (Figure 8.19 and 8.20; Hessl et al., 2018). More information on projected changes in drought, including specific categories or drought, can be found in Section 11.6.5 and Section 12.4.

Figure 8.20 | Past-to-future drought variability in paleoclimate reconstructions and models for select regions. On the left (a, c, e, g, i), tree-ring reconstructed Palmer Drought Severity Index (PDSI) series (black line) for the Mediterranean (10°W–45°E, 30°–47°N; E.R. Cook et al., 2015; Cook et al., 2016a), central Chile (70°W–74°W, 32°S–37°S; Morales et al., 2020), western North America (117°W–124°W, 32°N–38°N; Cook et al., 2010; Griffin and Anchukaitis, 2014), Eastern Australia and New Zealand (136°E–178°E, 46°S–11°S; Palmer et al., 2015), and Central Asia (99°E–107°E, 47°N–49°N; Pederson et al., 2014; Hessl et al., 2018) plotted in comparison to the past-to-future, fully-forced simulations from four ensemble members (thin blue lines) from the NCAR CESM Last Millennium Ensemble (thick blue line = ensemble mean) (Otto-Bliesner et al., 2016) for the same regions. The shaded area represents the range (10th to 90th percentile) of historical and future (RCP8.5) PDSI (Penman–Monteith) simulations from 15 CMIP5 models and 34 ensemble members for the same regions (1900–2100; Cook et al., 2014). On the right (b, d, f, h, j), the distribution of annual PDSI values from the past and present (850 to 2005 CE) (black) is compared to the future distribution (2006 to 2100 CE) (blue). The distributions show each of the four ensemble members from the CESM LME simulations. The future component of the CESM LME follows the RCP8.5 scenario. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

In summary, there is high confidence that soil moisture will decline in semi-arid, winter-rainfall dominated areas including the Mediterranean, southern Africa, south-western North America, south-western South America, and south-western Australia, as well as in Central America and the Amazonian basin. In general, these regions are expected to become drier both due to reduced precipitation (medium confidence) and increases in evaporative demand (high confidence). These same regions are likely to experience increases in drought duration and/or severity (high confidence). The magnitude of expected change scales with emissions scenarios (high confidence) but even under low-emissions trajectories, large changes in drought and aridity are expected to occur (high confidence) with consequences for regional water availability. In the Mediterranean, Central Chile, and western North America, future aridification will far exceed the magnitude of change seen over the last millennium (high confidence). Freshwater Reservoirs Glaciers

Previous assessments have concluded that recent warming has led to a reduction in low-elevation snow cover (high confidence) (SROCC), permafrost (high confidence) (SROCC), and glacier mass (high to very high confidence) (AR5; SROCC). The SROCC noted that these declines are projected to continue almost everywhere over the 21st century (high confidence), with complete glacier loss expected in regions with only small glaciers (very high confidence). The SROCC supported the AR5 finding that glacier recession would continue even without further changes in climate. The SROCC concluded that cryosphere changes had already altered the seasonal timing and volume of runoff (very high confidence), which in turn had affected water resources and agriculture (medium confidence), and projected peak water runoff had already been reached before 2019 in some of the glacier regions considered.

(Chapter 9 provides detailed assessment of glacier observations and projections (Figures 9.20 and 9.21, and Section 9.5.1). Here, a summary of their key findings is presented. Since SROCC, the coordinated Glacier Model Intercomparison Project (GlacierMIP; Box 9.3; Marzeion et al., 2020) has advanced modelling efforts. Global glacier volumes will substantially decline in coming decades regardless of emissions scenario; under a high-emissions scenario some areas will lose nearly all of their glacier mass (Section The projected global glacier mass loss over 20152100 is 29,000 ± 20,000 Gt for SSP1-2.6 to 58,000 ± 30,000 Gt for SSP5-8.5 (Section 9.5.1). Because of their lagged response to warming, glaciers will continue to lose mass for decades even if global temperature is stabilized (very high confidence) (Section 9.5.1).

Global glacier mass loss projections show a scenario-dependent geographic partitioning of when peak in runoff occurs (Marzeion et al., 2020), consistent with previous studies (Radić et al., 2014; Huss and Hock, 2018; Hock et al., 2019b). Under a low-emissions scenario (Marzeion et al., 2020) all regions exhibit runoff in the decades prior to 2050. Under a high-emissions scenario however, low- and mid-latitude regions show peak runoff before approximately 2060, whereas Arctic regions peak in later decades around 20702090. Antarctic glacier losses will not have peaked by the end of the century in the high-emissions scenario. Globally, peak runoff of 2.5 to 3 mm yr–1sea level equivalent occurs around 2090 (Marzeion et al., 2020). Regional projections are presented in detail in Section 9.5.1 and Figure 9.21, and briefly summarized below.

Himalaya and Central Asia: Glaciers in the Himalayas feed ten of the world’s most important river systems and are critical water sources for nearly two billion people (Wester et al., 2019). However, they are some of the most vulnerable ‘water towers’ (Immerzeel et al., 2020) that are projected to experience volume losses of approximately 30 to 100% by 2100 depending on global emissions scenarios (Marzeion et al., 2020). Under mid-range emissions scenarios glaciers in this region are projected to reach peak runoff during the period 2020 to 2040 (Marzeion et al., 2020).

Alaska, Yukon, British Columbia: Post-AR5 but pre-SROCC projections indicated a potential 70 ± 10% reduced volume of glacier ice in western Canada relative to 2005 (Clarke et al. , 2015), with few glaciers remaining in the Interior and Rockies regions and maritime glaciers in north-western British Columbia surviving only in a diminished state. Recent global projections support these earlier findings, showing that glacier mass in western Canada and the USA may reduce by 50% under low-emissions scenarios and be completely lost under the highest emissions and most sensitive glacier model combinations (Figure 9.21; Marzeion et al. , 2020). Arctic Canada and Alaskan glaciers are projected to experience more modest mass loss (0–60% depending on region, scenario, and model; Marzeion et al., 2020).

Andes: Huss and Hock (2018) concluded that peak glacier mass was reached prior to 2019 for 82–95% of the glacier area in the tropical Andes. This is consistent with more recent global model simulations that show mass loss rates from low latitude glaciers that universally decline from the start of simulations in 2015, regardless of emissions scenario (Marzeion et al., 2020). Peak runoff in low-latitude Andean glacier-fed rivers has therefore already passed (Frans et al., 2015; Polk et al., 2017) but in the Southern Andes may occur in the latter half of the century under high-emissions scenarios (Marzeion et al., 2020).

In summary, glaciers are projected to continue to lose mass under all emissions scenarios (very high confidence). Runoff from glaciers is projected to peak at different times in different places, with maximum rates of glacier mass loss in low latitude regions taking place in the next few decades in all scenarios (high confidence). While runoff from small glaciers will typically decrease because of glacier mass depletion, runoff from larger glaciers will increase with increasing global warming until glacier mass is similarly depleted, after which runoff peaks and then declines and which tends to occurs later in basins with larger glaciers and higher ice-cover fractions (high confidence). Glaciers in the Arctic and Antarctic will continue to lose mass through the latter half of the century and beyond (high confidence). Seasonal snow cover

The AR5 assessed as very likely that the amount and seasonal duration of Northern Hemisphere (NH) snow cover will reduce under global warming (AR5 Sections and Changes in the total amount of water in the snow cover (snow water equivalent) are less certain because of the competing influences of temperature and precipitation.

As snow cover is assessed in Chapter 9 (Section, only an overview of that assessment is provided here. Changes in seasonality of snow cover are assessed in Box 8.2. The continued consistency of reported results across all generations of model projections, along with improvements in process understanding, has increased confidence in snow cover projections since AR5.

In summary, based on the results of Chapter 9, it is nowvirtually certain that future NH snow cover extent and duration will continue to decrease with global warming. While most studies have focused on the NH, process understanding suggests with high confidence that these results apply to the Southern Hemisphere (SH) as well. There is high confidence in snowmelt occurring earlier in the year. Changes to the timing and amount of snowmelt will have a strong influence on all the other aspects of the water cycle in regions with seasonal snow, including run-off, soil moisture, and evapotranspiration. Wetlands and lakes

The AR5 did not include specific projections for wetlands and lakes. The SRCCL and SROCC provided some discussion of wetlands projections. For coastal wetlands, SRCCL noted the importance of sea level rise for increased saltwater intrusion, although projections of coastal wetland area with sea level rise are inconclusive. Some studies project substantial decreases (Spencer et al., 2016) while others indicate possible increases (Schuerch et al., 2018). SRCCL also noted the general expectation for decreases in water resources, including wetlands, in areas of decreased rainfall due to increased evaporation.

Local studies of inland wetlands project decreases in a range of environments including mountain (Lee et al., 2015), mid- to high latitude (D. Zhao et al., 2018), and prairie(Sofaeret al. , 2016) regions. In addition to affecting wetland extent and density, changes in flooding can also affect the connectivity between wetlands and rivers (Karimet al. , 2016). Despite a number of uncertainties underlying the general response of wetlands to climate change, there are multiple ways climate change may cause considerable stress on both inland and coastal wetlands (Junket al. , 2013; Moomawet al. , 2018).

Widespread changes are also projected for lakes (Woolway et al., 2020), including changes in lake temperature (Fang and Stefan, 1999; Sahoo et al., 2016), ice (Sharma et al., 2019), evaporation (W. Wang et al., 2018), and stability and mixing (Woolway and Merchant, 2019). Note that lake ice is also considered in Chapter 12 of this Report. To date, CO2 -induced lake acidification, analogous to ocean acidification, has not been the focus of many studies but may occur with continued emissions (Phillips et al., 2015). While glacier lakes in general increase with melting glaciers (Linsbauer et al., 2016; Colonia et al., 2017; Magnin et al., 2020) no clear projections are currently available (see discussion in Chapter 9). Projections of lake level means and variability show substantial changes for individual lakes (Bucak et al., 2017; Li et al., 2021) but can be sensitive to methodology, due to the competing processes involved (Notaro et al., 2015). Projected changes to wetlands and lakes due to climate change will occur in the context of widespread and continuing human-caused conversion and degradation of wetlands (e.g, Davidson, 2014), and where water withdrawals have a large impact on lake levels (e.g., Micklin, 2016).

In summary, there is medium confidencethat inland wetland extent will decrease in regions of projected precipitation decrease and evaporation increase, and high confidencethat sea level rise will increase saltwater intrusion into coastal wetlands. However, there is low agreementon the influence of sea level rise on the extent of coastal wetlands. Regarding lakes, there is high confidence for temperature increases and ice decreases, based on both projections and physical expectations, and low confidence for non-homogeneous decreases in mixing, given there is currentlylimitedevidence. Groundwater

Groundwater projections were not assessed in AR5. Groundwater processes are not explicitly included in most current CMIP6 models and so must be calculated separately with hydrologic models (e.g., R.G. Taylor et al. , 2013; Cuthbert et al. , 2019a). A range of factors are important in assessing groundwater projections, including the mean difference between precipitation and evaporation, the intensity of precipitation (R.G. Taylor et al., 2013a), and in changes in snow (Tague and Grant, 2009), glaciers (Gremaud et al., 2009), and permafrost (Okkonen and Kløve, 2011). Climate impacts on groundwater are occurring in the context of severe and growing human-caused groundwater depletion (WGII; Konikow and Kendy, 2005; Rodell et al., 2018; Bierkens and Wada, 2019), and water scarcity issues (Mekonnen and Hoekstra, 2016). Climate-related changes to the water cycle can influence water demand (for example, precipitation decreases in an irrigated area), and anthropogenic groundwater depletion can influence the water cycle through interactions with surface energy fluxes, surface water, and vegetation (Cuthbert et al., 2019a), although uncertainties in estimates of future groundwater depletion are large (Smerdon, 2017; Bierkens and Wada, 2019). Some aspects of groundwater change will be irreversible, including the increase of saltwater intrusion into coastal aquifers with sea level rise (Werner and Simmons, 2009), and depletion of fossil aquifers and aquifers with very long recharge times (Bierkens and Wada, 2019).

Globally, two modelling studies have shown substantial decreases in groundwater in regions including the Mediterranean, north-eastern Brazil and south-western Africa, with less clarity for other regions (Döll, 2009; Portmann et al. , 2013). Recent regional-scale analyses of the impact of water cycle changes on groundwater recharge (e.g., Meixner et al. , 2016; Tillman et al. , 2017; Shrestha et al. , 2018) suggest changes in both seasonality and spatial distribution, which are amplified under a higher greenhouse-gas emissions scenario (i.e., RCP 8.5 compared to RCP4.5). Seasonality changes are linked to increases during wet winter periods and declines during dry summer periods. Changes in spatial distribution are linked with increases in more humid regions and declines in more arid locations. Uncertainty in projections of groundwater were found to be substantially influenced by the conceptual and numerical models employed to estimate groundwater recharge (Meixner et al., 2016; Hartmann et al., 2017). Accordingly, current research on estimating water cycles change on groundwater includes a focus on improving the numerical representation of groundwater systems (Bierkens et al., 2015; Döll et al., 2016).

In summary, based on known limitations in current modelling, no confident assessment of groundwater projections is made here, although important climate-related changes in groundwater recharge are expected. In many environments, such climate-related impacts are expected to occur in the context of substantial human groundwater withdrawals depleting groundwater storage.

8.4.2 Projected Changes in Large-scale Phenomena and Regional Variability

A weakening of the tropical circulation represents a balance between thermodynamic increases in low level water vapour (about 7% °C–1) and smaller increases in global precipitation(13%°C–1) that are influenced by rapid adjustments to radiative forcings as well as slow responses to warming (Section; Bony et al. , 2013; Chadwick et al. , 2013; Ma et al. , 2018). Since AR5, additional drivers of tropical circulation weakening have been identified, including mean SST warming and changes in spatial patterns of SST (He and Soden, 2015), and the direct CO2 radiative effect (Bony et al., 2013; He and Soden, 2015; Merlis, 2015). ITCZ and Tropical Rain Belts

CMIP5 projections show no consistent shift in the zonal mean position of the ITCZ (Donohoe et al. , 2013; Donohoe and Voigt, 2017; Byrne et al. , 2018). The ITCZ position is strongly connected to cross-equatorial energy transport (Kang et al. , 2008; Bischoff and Schneider, 2014), which also shows no consistent change in future projections (Donohoe et al., 2013). Since AR5 it has been reported that most CMIP5 models project a narrowing of the ITCZ in response to surface warming together with intensified ascent in the core region and weakened ascent on the ITCZ edges (Lau and Kim, 2015; Byrne et al., 2018), implying a narrowing of precipitation regions influenced by the ITCZ. Modelled changes in the width and intensity of the zonal mean ITCZ are strongly anti-correlated, for example, narrowing is associated with increased intensity while broadening with decreased intensity. Such changes are associated with changes in tropical high cloud fraction and outgoing longwave radiation (Su et al., 2017; Byrne et al., 2018).

Regional shifts in tropical convergence zones are much larger than their zonal mean, and associated regional changes in precipitation (Chadwick et al., 2013; Mamalakis et al., 2021) are characterized by considerable uncertainties across models (Kent et al., 2015; Oueslati et al., 2016). Over the tropical oceans, shifts in rain bands are strongly coupled with changes in SSTs (Xie et al. , 2010; Huang et al. , 2013). Over tropical land, factors including remote SST increases (Giannini, 2010), the direct CO2 effect (Biasutti, 2013) and land–atmosphere interactions (Chadwick et al., 2017; Kooperman et al., 2018) influence projections. CMIP6 models project a clear northward ITCZ shift over eastern Africa and the Indian Ocean as well as a southward shift over the eastern Pacific and Atlantic oceans, as a result of regionally-contrasting inter-hemispheric energy flows (Mamalakis et al., 2021). The northward movement of the ITCZ over Africa has been linked to an intensification of the Saharan heat low associated with greenhouse gas (GHG) warming (Dong and Sutton, 2015), causing the tropical rain belt to seasonally migrate farther northward and reside there longer (Cook and Vizy, 2012; Dunning et al., 2018). In southern Africa, the projected delay in the wet season onset (Dunning et al. , 2018) is also associated with a circulation-based northward shift in the tropical rain band (Lazenby et al., 2018).

In summary, consistent with the AR5, the overall weakening of the tropical circulation is projected in CMIP5 and CMIP6 simulations with high confidence. It is likely that the zonal mean of the ITCZ will narrow and strengthen in the core region with projected surface warming (high confidence). Distinct regional shifts in the ITCZ will be associated with regional changes in precipitation amount and seasonality (medium confidence). Hadley Circulation and Subtropical Belt

The AR5 found that the Hadley cells are likely to slow down and expand in response to radiative forcing, but with considerable internal variability. Given the complexities in forcing mechanisms, AR5 assigned low confidence to near-term changes in the structure of the Hadley circulation. The widening Hadley cells were expected to result in a poleward expansion of subtropical dry zones.

Model simulations since AR5 project a more noticeable and consistent weakening of the Northern Hemisphere (NH) winter Hadley cell than the Southern Hemisphere (SH) winter cell (Seo et al., 2014; Zhou et al., 2016), related to changes in meridional temperature gradient, static stability, and tropopause height (Seo et al., 2014; D’Agostino et al., 2017). Changes in SST patterns reduces the magnitude of Hadley cell weakening (Gastineau et al., 2009; Ma et al., 2012). There is considerable structure in Hadley circulation strength changes with longitude, associated with cloud-circulation interactions (Su et al., 2014). Subtropical anticyclones are projected to intensify over the north Atlantic and south Pacific but to weaken elsewhere (He et al., 2017).

A consistent poleward expansion of the edges of the Hadley cells is projected (Nguyen et al., 2015; Grise and Davis, 2020), particularly in the SH, consistent with observed trends (Figure 8.21 and Section; Nguyen et al. , 2015). The main driver of future expansion appears to be greenhouse gas forcing (Grise et al., 2019), with uncertainty in magnitude due to internal variability (Kang et al., 2013). Proposed mechanisms for poleward expansion include increased dry static stability (Frierson et al. , 2007; Lu et al. , 2007), increased tropopause height (Chen and Held, 2007; Chen et al. , 2008), stratospheric influences (Kidston et al., 2015) and radiative effects of clouds and water vapour (Shaw and Voigt, 2016; see also Section Hadley cell expansion is thought to be associated with the precipitation declines projected in many subtropical regions (Shaw and Voigt, 2016), but more recent work suggests that these reductions are mainly due to the direct radiative effect of CO2 forcing (He and Soden, 2015), land sea contrasts in the response to forcing (Shaw and Voigt, 2016; Brogli et al. , 2019) and SST changes (Sniderman et al., 2019). In semi-arid, winter rainfall-dominated regions (such as the Mediterranean), thermodynamic processes associated with the land sea thermal contrast and lapse rate changes dominate the projected precipitation decline in summer, whereas circulation changes are of greater importance in winter (Brogli et al. , 2019). The hydroclimates in these regions are projected to evolve with time due to changing contributions from rapid atmospheric circulation changes and their associated SST responses, as well as slower SST responses to anthropogenic forcing (Zappa et al., 2020).

In summary, CMIP5 and CMIP6 models project a weakening of the Hadley cells, with high confidence for the NH in boreal winter and low confidence for the SH in austral winter. The Hadley cells are projected to expand polewards with global warming, most notably in the SH (high confidence). There is currentlylow confidence in the impacts on regional precipitation in subtropical regions. Walker Circulation

The AR5 determined that the Pacific Walker circulation was likely to slow down over the 21st century, which would lead to decreased precipitation over the western tropical Pacific and increases over the central and eastern Pacific. Recent studies show consistency with AR5 conclusions but also show an eastward shift over the Pacific, mostly due to a shift towards more ‘El Niño-like’ conditions under global warming (Bayr et al., 2014). Other studies suggest that the weakening of the Walker circulation is related to the response of the western North Pacific monsoon and to changing land–sea temperature contrasts, while a positive ocean–atmosphere feedback amplifies the weakening of both east–west SST gradient and trade winds in the tropical Pacific (Zhang and Li, 2017).

Since AR5, the paradox between the projected weakening and the observed strengthening of the Walker circulation since the 1990s (Section has triggered debate about the drivers of these changes (England et al. , 2014; McGregor et al. , 2014; Kociuba and Power, 2015; Vilasa et al. , 2017; Chung et al. , 2019). Projected changes in equatorial SST gradients are not entirely consistent with observed trends (Coats and Karnauskas, 2017; Seager et al., 2019a), and one CMIP5 model that projects a future strengthening of the Walker circulation is more consistent with observations than other models (Kohyama et al., 2017). Other studies suggest that these differences arise from the dominant influence of internal climate variability to the observed trends (Chung et al., 2019), or as a consequence of a systematic cold bias of most CMIP5 models in their Equatorial Pacific cold tongues (Seager et al., 2019a). However, the latter hypothesis is based on a simplified model of tropical Pacific dynamics and is not consistent with the current physical understanding of the tropical circulation response to increasing CO2 levels (Section or with independent paleoclimate evidence suggesting a weaker Walker circulation under warmer climates (Tierney et al., 2019; McClymont et al., 2020). Different time scales of the tropical Pacific responses to global warming have been highlighted by numerical experiments with both comprehensive and simplified models. Results suggest a transient strengthening of the Walker circulation related to Indian Ocean warming (L. Zhang et al., 2018), followed by a slower weakening linked to a strengthened eastern Pacific cold tongue warming emerging after 50100 years (Section; Heede et al., 2020).

CMIP6 projections provide further evidence of a significant long-term weakening of the Walker circulation (Figure 8.21). For instance, a pronounced weakening of the upper-level tropical easterly jet is projected both over the Indian Ocean and tropical eastern Pacific, where declines are projected to exceed 70% by 2100 in the high-emissions SSP5-8.5 scenario (S. Huang et al., 2020). CMIP6 models agree on a future decrease of the equatorial zonal temperature gradient (Fredriksen et al., 2020), which can lead to weaker trade winds over the tropical Pacific. However, CMIP6 models show a diversity of SST warming patterns in the tropical Pacific (Freund et al., 2020), which contributes to uncertainties in the response of both Walker circulation and ENSO to continued warming.

Figure 8.21 | Schematic depicting large-scale circulation changes and impacts on the regional water cycle. The central figures show precipitation minus evaporation (P–E) changes at 3°C or global warming relative to an 1850–1900 base period (mean of 23 CMIP6 SSP5-8.5 simulations). Annual mean changes (large map) include contours (ocean only) depicting control climate P–E = 0 mm day1lines with the solid contour enclosing the tropical rain belt region and dashed lines representing the edges of subtropical regions. Confidence levels assess understanding of how large-scale circulation change affect the regional water.

In summary, there is high confidence that the Pacific Walker circulation will weaken by the end of the 21st century, and will be associated with decreased precipitation over the western tropical Pacific and increases farther east. Discrepancies between observed and simulated changes in SSTs in the tropics indicate that a temporary strengthening of the Walker Circulation can arise from a transient response to GHG radiative forcing (low confidence) and from internal variability (medium confidence). Monsoons

In AR5, monsoon precipitation over land was projected to intensify by the end of the 21st century, due to thermodynamic increases in moisture convergence despite weakening of the tropical circulation (see Section Following the definition of regional monsoons in Annex V and Figure 8.11, and the assessment of the observed changes (Section, here we provide an assessment of projected changes in regional monsoons. Assessment is provided either in terms of SSP and RCP scenarios and global warming levels available since AR5, or from the newly available CMIP6 projections (Figure 8.22 and Table 8.2). Table 8.2 provides projected changes across the five SSPs used in this Report for precipitation (mm day–1), P–E (mm day–1) and runoff (mm day–1) over the regional monsoons for the mid (20412060) and long term (20812100).

Table 8.2 | Monsoon mean water cycle projections in the mid-term (2041–2060) and long term (2081–2100) relative to present day (1995–2014), showing present-day mean and 90% confidence range across CMIP6 models (historical experiment) and projected mean changes and the 90% confidence range across the same set of models and a range of Shared Socio-economic Pathway scenarios. All statistics are in units of mm day–1. Further details on data sources and processing are available in the chapter data table (Table 8.SM.1).

Mid-term: 2041–2061 Minus Reference Period

Long Term: 2081–2100 Minus Reference Period

19952014 Reference Period











South and South East Asian Monsoon (June–July–August–September, JJAS)


8.42 [6.66 to 10.14]

0.44 [0.08 to 0.74]

0.47 [0.1 to 0.96]

0.42 [0.03 to 0.81]

0.32 [–0.08 to +0.94]

0.54 [0.11–1.18]

0.46 [0.16 to 0.7]

0.52 [0.13 to 1.09]

0.66 [0.16 to 1.1]

0.94 [0.3 to 1.78]

1.46 [0.66 to 2.49]


3.75 [1.8 to 5.71]

0.23 [0.1 to 0.38]

0.29 [0.02 to 0.65]

0.29 [–0.0 to +0.66]

0.24 [–0.04 to +0.52]

0.38 [0.07–0.78]

0.19 [–0.02 to +0.35]

0.29 [–0.04 to +0.65]

0.42 [0.04 to 0.83]

0.7 [0.12 to 1.2]

1.14 [0.36 to 2.05]


5.19 [3.68 to 6.5]

0.28 [0.03 to 0.52]

0.36 [–0.0 to +0.76]

0.36 [0.02 to 0.69]

0.3 [–0.04 to +0.85]

0.45 [0.06–0.95]

0.27 [0.06 to 0.38]

0.38 [0.11 to 0.76]

0.51 [0.02 to 0.83]

0.81 [0.24 to 1.56]

1.15 [0.45 to 1.84]

East Asian Monsoon (June–July–August, JJA)


5.59 [4.47 to 6.86]

0.37 [–0.09 to +0.93]

0.37 [–0.09 to +0.87]

0.34 [0.05 to 0.76]

0.22 [–0.16 to +0.88]

0.43 [0.03 to 1.1]

0.43 [0.07 to 1.02]

0.44 [–0.0 to +1.08]

0.51 [0.11 to 1.09]

0.59 [0.02 to 1.31]

0.84 [0.24 to 1.74]


2.24 [1.28 to 3.41]

0.11 [–0.16 to +0.4]

0.13 [–0.19 to +0.42]

0.13 [–0.15 to +0.4]

0.15 [–0.29 to +0.76]

0.2 [–0.11 to +0.72]

0.16 [–0.08 to +0.49]

0.16 [–0.13 to +0.58]

0.22 [–0.13 to +0.64]

0.36 [–0.05 to +0.87]

0.51 [0.06 to 1.24]


2.41 [1.51 to 3.31]

0.1 [–0.31 to +0.51]

0.13 [–0.2 to +0.48]

0.17 [–0.04 to +0.53]

0.17 [–0.2 to +0.75]

0.23 [–0.09 to +0.86]

0.16 [–0.07 to +0.57]

0.18 [–0.18 to +0.65]

0.24 [–0.1 to +0.76]

0.4 [–0.08 to +0.93]

0.5 [–0.13 to +1.34]

North American Monsoon (July–August–September, JAS)


3.05 [2.24 to 3.96]

0.13 [–0.08 to +0.43]

0.07 [–0.27 to +0.32]

0.02 [–0.32 to +0.41]

–0.03 [–0.37 to +0.38]

–0.03 [–0.43 to +0.52]

0.18 [–0.05 to +0.44]

0.04 [–0.35 to +0.39]

–0.1 [–0.51 to +0.37]

–0.19 [–0.76 to +0.44]

–0.15 [–0.96 to +0.57]


0.46 [0.09 to 0.87]

0.03 [–0.04 to +0.12]

0.03 [–0.07 to +0.16]

0.02 [–0.1 to +0.14]

–0.0 [–0.1 to +0.14]

–0.0 [–0.11 to +0.14]

0.04 [–0.03 to +0.15]

–0.0 [–0.19 to +0.15]

–0.03 [–0.22 to +0.14]

–0.05 [–0.23 to +0.19]

–0.06 [–0.29 to +0.23]


0.78 [–0.1 to +1.45]

0.06 [–0.1 to +0.2]

0.02 [–0.18 to +0.24]

0.0 [–0.22 to +0.23]

–0.03 [–0.24 to +0.2]

–0.04 [–0.31 to +0.27]

0.09 [–0.06 to +0.31]

0.01 [–0.22 to +0.25]

–0.08 [–0.28 to +0.25]

–0.17 [–0.68 to +0.25]

–0.18 [–0.72 to +0.38]

South American Monsoon (December–January–February, DJF)


8.44 [5.98 to 10.22]

0.09 [–0.2 to +0.3]

0.12 [–0.29 to +0.62]

0.09 [–0.47 to +0.62]

0.07 [–0.55 to +0.62]

0.07 [–0.5 to +0.71]

0.02 [–0.32 to +0.36]

0.09 [–0.33 to +0.58]

0.07 [–0.63 to +0.81]

0.05 [–1.17 to +0.82]

–0.0 [–1.22 to +1.19]


2.49 [1.11 to 4.38]

–0.02 [–0.23 to +0.26]

–0.01 [–0.43 to +0.53]

0.01 [–0.45 to +0.46]

–0.03 [–0.49 to +0.36]

–0.03 [–0.56 to +0.53]

–0.04 [–0.27 to +0.28]

–0.01 [–0.41 to +0.39]

–0.01 [–0.58 to +0.55]

–0.06 [–0.81 to +0.24]

–0.04 [–0.85 to +0.93]


4.5 [2.83 to 6.01]

0.04 [–0.23 to +0.25]

0.08 [–0.26 to +0.47]

0.04 [–0.43 to +0.53]

0.04 [–0.5 to +0.61]

0.02 [–0.45 to +0.58]

–0.01 [–0.32 to +0.29]

0.03 [–0.34 to +0.43]

–0.02 [–0.63 to +0.62]

–0.02 [–1.03 to +0.72]

–0.09 [–1.11 to +0.98]

Australian and Maritime Continent Monsoon (December–January–February, DJF)


8.63 [6.79 to 10.7]

0.26 [0.04 to 0.49]

0.22 [–0.23 to +0.53]

0.28 [–0.2 to +0.79]

0.25 [–0.14 to +0.73]

0.38 [0.0 to 0.84]

0.15 [–0.09 to +0.34]

0.24 [–0.36 to +0.74]

0.5 [–0.1 to +1.07]

0.65 [–0.08 to +1.33]

0.9 [0.09 to 1.76]


3.82 [1.78 to 7.25]

0.2 [–0.01 to +0.48]

0.23 [–0.11 to +0.48]

0.29 [–0.11 to +0.7]

0.24 [–0.13 to +0.56]

0.35 [–0.03 to +0.87]

0.12 [–0.06 to +0.39]

0.29 [–0.08 to +0.88]

0.49 [0.09 to 1.25]

0.61 [–0.09 to +1.05]

0.92 [0.14 to 1.83]


4.8 [3.19 to 6.63]

0.22 [0.03 to 0.47]

0.13 [–0.23 to +0.42]

0.2 [–0.16 to +0.7]

0.2 [–0.14 to +0.62]

0.27 [–0.09 to +0.61]

0.12 [–0.1 to +0.31]

0.16 [–0.31 to +0.54]

0.38 [–0.05 to +0.75]

0.54 [–0.08 to +1.13]

0.69 [0.09 to 1.27]

West African Monsoon (June–July–August–September, JJAS)


5.14 [3.62 to 7.18]

0.16 [–0.19 to +0.4]

0.14 [–0.22 to +0.56]

0.24 [–0.14 to +0.72]

0.3 [–0.1 to +0.85]

0.38 [–0.12 to +1.24]

0.06 [–0.25 to +0.52]

0.1 [–0.25 to +0.57]

0.25 [–0.32 to +0.91]

0.38 [–0.49 to +1.14]

0.49 [–0.55 to +1.56]


1.43 [0.34 to 2.57]

0.06 [–0.07 to +0.22]

0.05 [–0.18 to +0.27]

0.14 [–0.13 to +0.54]

0.2 [–0.05 to +0.7]

0.24 [–0.1 to +0.8]

–0.01 [–0.2 to +0.21]

0.03 [–0.25 to +0.35]

0.1 [–0.25 to +0.51]

0.25 [–0.28 to +0.85]

0.3 [–0.33 to +0.93]


2.41 [1.05 to 4.07]

0.08 [–0.2 to +0.35]

0.1 [–0.2 to +0.4]

0.2 [–0.11 to +0.63]

0.23 [–0.11 to +0.74]

0.36 [–0.06 to +1.11]

–0.01 [–0.27 to +0.35]

0.07 [–0.2 to +0.44]

0.18 [–0.21 to +0.6]

0.28 [–0.38 to +0.95]

0.46 [–0.44 to +1.4]

Figure 8.22 | Projected regional monsoons precipitation changes. Percentage change in projected seasonal mean precipitation over regional monsoon domains (as defined in Figure 8.11, Section and Annex V) for near term (2021–2040), mid-term (2041–2060), and long term (2081–2100) periods based on 24 CMIP6 models and three SSP scenarios (SSP1-2.6, SSP2-4.5 and SSP5-8.5). Further details on data sources and processing are available in the chapter data table (Table 8.SM.1). South and South East Asian Monsoon

In AR5, South and South East Asian monsoon (SAsiaM) precipitation was projected to increase by the end of the 21st century but with a weakening of the circulation, withhigh agreement across the CMIP5 models (Kitoh et al. , 2013; Menon et al. , 2013; Sharmila et al. , 2015; Sooraj et al. , 2015; Kitoh, 2017; Kulkarni et al. , 2020). Since AR5, most studies have confirmed projected increases in South Asian monsoon precipitation (high confidence), while one high-resolution model (35 km in latitude/longitude) projects monsoon precipitation decreases during the 21st century following the RCP4.5 scenario (Krishnan et al., 2016).

Over South Asia, the moisture-bearing monsoon low-level jet is projected to shift northward in CMIP3 and CMIP5 models (Sandeep and Ajayamohan, 2015). Greater warming over the Asian land region compared to the ocean contributes to intensification of the monsoon low-level south-westerly winds and precipitation (Endo et al., 2018), even though the combined effect of upper and lower tropospheric warming makes the Asian monsoon circulation response rather complicated. A high resolution model projection, based on the RCP8.5 scenario, indicates that a northward shift of the low-level jet and associated weakening of the large-scale monsoon circulation can induce a large reduction in the genesis of monsoon low pressure systems by the late 21st century (Sandeep et al., 2018). Experiments with constant forcing indicate that at 1.5°C and 2°C global warming levels, mean precipitation and monsoon extremes are projected to intensify in summer over India and South Asia (Chevuturi et al., 2018; D. Lee et al., 2018) and that a 0.5°C difference would imply a 3% increase of precipitation (Chevuturi et al., 2018). CMIP5 models project an increase in short intense active days and decrease in long active days, with no significant change in the number of break spells for India (Sudeepkumar et al., 2018).

Future monsoon projections from CMIP6 models show an increase of SAsiaM precipitation across all the scenarios and across all the time frames (Figure 8.22) with the maximum increase at the end of the 21st century in SSP5-8.5 (Almazroui et al. , 2020c; Z. Chen et al. , 2020b; Ha et al. , 2020; Wang et al. , 2021). Table 8.2 confirms that changes in runoff and P–E over SAsiaM region are positive and largest in the higher emissions scenarios considered, as in precipitation. On the other hand, changes in the ensemble mean for all the variables considered in the SSP1-1.9 scenario are negative for both mid and long-term periods (Table 8.2). This is also consistently reflected in the spatial map of future precipitation changes (Figure 8.15). Different near-term projections of the SAsiaM may result given the diversity in the future aerosol emissions pathways and policies for regulating air pollution (Wilcox et al., 2020). Additionally, near-term projections of SAsiaM precipitation are expected to be constrained by internal variability associated with the PDV (X. Huang et al., 2020a). CMIP6 models also indicate a lengthening of the summer monsoon over India by the end of the 21st century, at least in SSP2-4.5, with considerable inter-model spread in the projected late retreat (Ha et al., 2020).

In summary, consistent with AR5, there is high confidence that SAsiaM precipitation is projected to increase during the 21st century in response to continued global warming across the CMIP6 higher emissions scenarios, mostly in the mid- and long terms. East Asian Monsoon

In AR5, the East Asian monsoon (EAsiaM) was projected to intensify in terms of precipitation, with an earlier onset and longer duration of the summer season. Since AR5, there has been improved understanding of future projected changes in the EAsiaM.

CMIP5 projections indicated a possible intensification of the EAsiaM circulation during the 21st century, in addition to precipitation increase, although there is a lack of consensus on changes in the western North Pacific subtropical high, this is an important feature of the EAsiaM circulation (Kitoh, 2017). Furthermore, the EAsiaM precipitation enhancements in the CMIP5 projections are prominent over the southern part of the Baiu rainband by the late 21st century, with no significant changes in the Meiyu precipitation over central-eastern China (Horinouchi et al., 2019). It was also shown that the Baiu precipitation response in CMIP5 projections is accompanied by a southward retreat of the western North Pacific subtropical high and a southward shift of the East Asian subtropical jet (Horinouchi et al., 2019). According to the high-resolution MRI-AGCM global warming experiments, future summer precipitation could potentially increase on the southern side and decrease on the northern side of the present-day Baiu location in response to downward-motion tendencies which can offset the ‘wet-gets-wetter’ effect, but is subject to large model uncertainties (Ose, 2019). Future projections of land warming over the Eurasian continent (Endo et al., 2018) and intensified land sea thermal contrast (Z. Wang et al., 2016; Tian et al., 2019) can potentially intensify the EAsiaM circulation during the 21st century. However, there are large uncertainties in projected water cycle changes over the region (Endo et al., 2018), mostly in the near-term because of uncertainties in future aerosol emissions scenarios (Wilcox et al., 2020), as well as due to the interplay between internal variability and anthropogenic external forcing (Wang et al., 2021).

Inter-hemispheric mass exchange can act as a bridge connecting SH circulation with EAsiaM rainfall, however this inter-hemispheric link is projected to weaken in a future warmer climate as seen from a CCSM4 projection using the RCP8.5 scenario (Yu et al., 2018). A comparison of 1.5°C and 2°C global warming levels reveals how a 0.5°C difference could result in precipitation enhancement over large areas of East Asia (D. Lee et al. , 2018; J. Liu et al. , 2018; Chen et al. , 2019), with substantial increases in the frequency and intensity of extremes (Chevuturi et al., 2018; D. Li et al., 2019). Future monsoon projections from the CMIP6 models show increase of EAsiaM precipitation across all the scenarios (Z. Chen et al., 2020b), though with a large model spread mostly on the long-term and in the higher emissions scenarios (Figure 8.22). Considering all the five scenarios used across the report, changes in precipitation, runoff and P–E over the EAsiaM are positive and become larger for highest emissions scenarios and for the long-term mean, except for the mid-term SSP1-1.9 scenario where the changes are close to zero or even negative (Table 8.2). Additionally, CMIP6 models confirm a projected increased length of the EAsiaM season due to early onset and late retreat (Ha et al., 2020).

In summary, despite the uncertainties in the monsoon circulation response in CMIP5 and CMIP6 models, there is high confidence that summer monsoon precipitation over East Asia will increase in the 21st century and medium confidence that the monsoon season will be longer. West African Monsoon

The AR5 concluded that projections of West African monsoon (WAfriM) rainfall are highly uncertain in CMIP3 and CMIP5 models, but still suggest a small delay and intensification in late wet season rains. Studies published since AR5 are broadly consistent with this assessment. CMIP6 models agree on statistically significant projected increases in rainfall in eastern-central Sahel and a decrease in the west for the end of the 21st century (Roehrig et al., 2013; Biasutti, 2019; Monerie et al., 2020). However, the magnitude of WAfriM projected precipitation depends on the convective parametrization used (Hill et al., 2017), and large uncertainties remain in WAfriM projections because of large inter-model spread, particularly over the western Sahel (Roehrig et al., 2013; Biasutti, 2019; Monerie et al., 2020). CMIP6 models show a general increase of WAfriM precipitation across all future scenarios but with a substantial model spread for the SSP5-8.5 scenario (Figure 8.22). This sensitivity arises from the combined and contrasting influences of anthropogenic greenhouse gas and aerosol forcing that affect WAfriM precipitation (particularly over the Sahel) directly and also indirectly through subtropical North Atlantic SST changes (Giannini and Kaplan, 2019). The large model spread and associated uncertainties in projected precipitation changes is reflected also in runoff and P–E changes (Table 8.2). Regional climate models (RCMs) ensembles (e.g., Klutse et al., 2018) agree with CMIP5 projected rainfall trends but some individual models show rainfall declines (e.g., Sylla et al., 2015; Akinsanola et al., 2018), highlighting the existing large uncertainties in RCMs WAfriM rainfall projections.

Changes in seasonality (Box 8.2) are projected with a later monsoon onset (high confidence) over the Sahel and a late cessation (medium confidence), suggesting a delayed wet season as a regional response to global GHG forcing (Biasutti, 2013; Dunning et al., 2018; Akinsanola and Zhou, 2019). Rainfall distribution is projected to be highly variable with a decrease in the number of rainy days in the western Sahel, consistent with an increase in consecutive dry days and a reduction in the number of growing season days (Cook and Vizy, 2012; Diallo et al., 2016). A decrease in the frequency but an increase in the intensity of very wet events is projected to be more pronounced over the Sahel than over Guinean coast, and also under higher emissions scenarios (i.e., RCP8.5; Sylla et al., 2015; Akinsanola et al., 2018).

In summary, post-AR5 studies and newly available CMIP6 results indicate projected rainfall increases in the eastern-central WAfriM region but decreases in the west (high confidence), with a delayed wet season (medium confidence). Overall, WAfriM summer precipitation is projected to increase during the 21st century but with larger uncertainty noted under high-emissions scenarios (medium confidence). North American Monsoon

The AR5 concluded that the North American monsoon (NAmerM) will likely intensify in the future, even though there is low agreement among models. The AR5 reported medium confidence that precipitation associated with the NAmerM will arrive later in the annual cycle and persist longer.

Since AR5, analyses of CMIP5 projections suggest little change in the overall amount of NAmerM precipitation in response to rising global surface temperature. However, significant declines are projected in the early monsoon season and increases in the late monsoon season, suggesting a shift in seasonality toward a delayed monsoon onset and demise (Cook et al., 2013). It is recognized that CMIP5 models are generally too coarsely-resolved to simulate the Gulf of California and the moisture surges associated with the NAmerM (Pascale et al., 2017). Under different RCPs, CMIP5 models tend to project a reduction in NAmerM precipitation but an increase in extreme precipitation events (Torres-Alavez et al. , 2014; Bukovsky et al. , 2015; Pascale et al. , 2019). The almost unchanged or slight decrease in NAmerM total precipitation amount under global warming projections is at odds with paleoclimate records that suggest increased monsoon precipitation under past warm conditions (D’Agostino et al., 2019; Seth et al., 2019). However, there is low agreement on how those changes and the mechanisms that drive them are affected under different RCPs since most simulations are model-dependant (Cook and Seager, 2013; Geil et al., 2013; Pascale et al., 2019). Projections from six CMIP6 models show a shortening of the NAmerM under the SSP5-8.5 scenario due to earlier demises (Moon and Ha, 2020). In addition, CMIP6 projections show a decrease in NAmerM precipitation under SSP2-4.5 and SSP5-8.5 scenarios by the end of the 21st century with large inter-model spread (Figure 8.22). This result is also supported by the analysis of 31 CMIP6 models under the SSTP5-8.5 scenario for the 20802099 period (Almazroui et al., 2021). Non-linearities and uncertainties in the NAmerM projected changes are valid for many water cycle variables, like precipitation, runoff and P–E (Table 8.2).

In summary, there is low agreement on a projected decrease of NAmerM precipitation, however there is high confidence in delayed onsets and demises of the summer monsoon. South American Monsoon

The AR5 reported medium confidence that the South American monsoon (SAmerM) overall precipitation will remain unchanged, and medium confidence in projections of extreme precipitation. The AR5 also stated high confidence in the spatial expansion of the SAmerM, resulting from increased temperature and humidity.

Since AR5, some studies indicate that the SAmerM would experience changes in its seasonal cycle, with delayed monsoon onsets under increasing GHG emissions associated to different RCPs (Fu et al. , 2013; Reboita et al. , 2014; Boisier et al. , 2015; Pascale et al. , 2016; Seth et al. , 2019; Sena and Magnusdottir, 2020). In contrast, other studies indicate projected earlier onsets and delayed retreats of the SAmerM under the RCP8.5 scenario based on six CMIP5 models (Jones and Carvalho, 2013). These differences have been linked to the methodology used to determine monsoon timing, and sensitivity to the monsoon domain considered (Section; Correa et al., 2021). Recent studies provide further evidence for the projection of delayed SAmerM onsets by the late 21st century (Sena and Magnusdottir, 2020). An analysis of six CMIP6 models under the SSP5-8.5 scenario confirm the projections of delayed SAmerM onsets by the end of the 21st century (Moon and Ha, 2020). In addition, projected changes in the intensity and length of the SAmerM season have been found to be model-dependent (Pascale et al., 2019). The analysis of CMIP5 projections of total monsoon rainfall indicate mixed signals in the Amazon and SAmerM regions (Jones and Carvalho, 2013; Marengo et al., 2014), with some studies suggesting increased summer precipitation in the core SAmerM region (Kitoh et al., 2013; Seth et al., 2013). Dynamical downscaling of CMIP5 projections under the RCP4.5 and RCP8.5 scenarios with the Eta RCM suggests reductions of austral summer precipitation over the SAmerM region throughout the 21st century (Chou et al., 2014). Further analysis using 15 different CMIP6 models for the SSP2-4.5 scenario suggest reductions in total SAmerM rainfall (B. Wang et al., 2020). However, other analyses of CMIP6 projections under different SSP scenarios do not report clear changes in the SAmerM precipitation throughout the 21st century (Figure 8.22; Z. Chen et al., 2020b; Jin et al., 2020). Similar uncertainties for all the SSP scenarios used across the report are found for other water cycle variables, including runoff and P–E (Table 8.2). Furthermore, there is disagreement in projected extreme precipitation in the region, with some CMIP5-based studies suggest reductions (Marengo et al., 2014), while others indicate increases based on CMIP5 and CMIP6 models (Kitoh et al., 2013; Sena and Magnusdottir, 2020).

In summary, there is high confidence that the SAmerM will experience delayed onsets in association with increases in GHG. However, there is low agreement on the projected changes in terms of total precipitation of the South American summer monsoon season. Australian and Maritime Continent Monsoon

The AR5 concluded that projected changes in Australian and Maritime Continent monsoon (AusMCM) rainfall and seasonality are uncertain in the CMIP5 models, with some projecting increases and others projecting decreases for the range of emissions scenarios. Models that perform better at simulating present day regional climate project little change or an increase in Australian monsoon rainfall (Jourdain et al., 2013; CSIRO and BoM, 2015; Brown et al., 2016b). CMIP6 models project increased AusMCM precipitation in the 21st century but with a more robust signal in SSP2-4.5 and SSP5-8.5 rather than in lower emissions scenarios (Figure 8.22). A reduced range of CMIP6 rainfall projections but continued disagreement on the sign of change is reported over Australia (Narsey et al., 2020).

The northern and eastern parts of the Maritime Continent have projected increases in rainfall in CMIP5 models (Siew et al., 2014), while there are projected decreases over Java, Sulawesi and southern parts of Borneo and Sumatra. Rainfall changes are correlated with the extent of warming in the western tropical Pacific in CMIP5 models (Brown et al., 2016b) but inter-model differences are also related to modelled large-scale zonal mean precipitation response in both CMIP5 and CMIP6 model ensembles (Narsey et al., 2020). Decomposition of projected rainfall changes indicates that the largest source of model uncertainty is associated with shifts in the spatial pattern of convection (Chadwick et al., 2013; Brown et al., 2016b). Uncertainties in capturing the spatial and temporal features of the Maritime Continent monsoon depend also on the horizontal resolution of coupled climate models (e.g., Jourdain et al., 2013).

The role of anthropogenic aerosol forcing in future projections of the Australian monsoon has been investigated for CMIP5 models (Dey et al., 2019a); decreases in anthropogenic aerosol concentrations over the 21st century are expected to produce relatively greater warming in the NH than SH, favouring a northward shift of the tropical rain belt (e.g., Rotstayn et al., 2015).

There are some clear projected changes in the rainfall variability and extremes of the Australian monsoon. Rainfall variability in the Australian monsoon domain increases on time scales from daily to decadal in CMIP5 models (Brown et al., 2017), indicating either more intense wet days or more dry days or both. There is also a projected increase in the intensity of extreme rainfall but a reduction in the frequency of heavy rainfall days for the Australian monsoon (Dey et al. , 2019a). This is consistent with Moise et al. (2020), who found an increase in Australian monsoon active phase or ‘burst’ rainfall intensity but a reduction in the number of burst days and events.

H. Zhang et al. (2013) examined changes in Australian monsoon onset and duration in CMIP3 models and found model agreement on a delay in onset and shortened duration to the north of Australia, but less agreement over the interior of the continent. An updated study of CMIP5 models found similar mean changes with delayed onset and shortened duration, but substantial model disagreement (H. Zhang et al., 2016).

In summary, CMIP6 projections show an increase of AusMCM precipitation across all emissions scenarios. There is strong model agreement on an increase in monsoon precipitation over the Maritime Continent while there is low agreement on the direction of change over northern Australia. There is a projected increase in rainfall variability over northern Australia, with increased intensity of rainfall during the active or ‘burst’ phase (medium confidence). Tropical Cyclones

Tropical cyclones (TCs) projections are primarily assessed in Section Here, we extend this analysis by assessing the implications of projected changes in tropical cyclones on the water cycle.

The AR5 concluded that TC rainfall rate was likely to increase through the 21st century. Section assesses that the average tropical cyclone rain-rate is projected to increase with warming (high confidence), and peak rain rates are projected to increase at greater than the Clausius–Clapeyron scaling rate of 7%°C–1 warming in some regions due to increased low-level moisture convergence (medium confidence). The increase in TC rainfall rate is explained by increased TC intensity resulting from increasing SSTs, and increased environmental water vapour (Chauvin et al., 2017; M. Liu et al., 2019).

Consistent with the observed poleward migration of tropical cyclone activity (Kossin et al., 2014), in the SH a larger proportion of storms are projected to decay south of 25°S at the end of the 21st century but with negligible changes in genesis latitude and storm duration for the Australian region (CSIRO and BoM, 2015; Sharmila and Walsh, 2018). An analysis of projections for North Pacific islands indicate that the maximum intensity of storms will increase but the number of tropical cyclones will decrease in some places, such as Guam and Kwajalein Atoll in the tropical north-western Pacific, or remain the same in other regions like near Okinawa (Japan) or Oahu (Hawaii) (Widlansky et al., 2019). TC-induced storm tides affecting landfall in the Pearl River delta over South China are projected to increase by the end of the 21st century (J. Chen et al., 2020b)

In summary, there is high confidence that heavy precipitation associated with tropical cyclones is projected to increase, in response to well-understood processes related to increased low-level moisture convergence and environmental water vapour. Stationary Waves

The AR5 did not provide an assessment of stationary wave projections as distinct from other related aspects of circulation, such as blocking, modes of variability, and storm tracks. Here we provide a brief assessment of stationary wave projections from the water cycle perspective, with the related circulation aspects considered separately in the following sections.

Several studies based on CMIP5 projections show changes in NH winter stationary waves that increase precipitation over the west coast of North America and decrease it over the eastern Mediterranean and parts of south-western North America (Neelin et al. , 2013; Seager et al. , 2014a, b, 2019b; Simpson et al. , 2016; Wills et al. , 2019), although the underlying dynamics are not yet fully understood (Seager et al., 2019b; Wills et al., 2019). For the NH winter global teleconnection pattern, the majority of the models analyzed in (Sandler and Harnik, 2020) project the development of a preferred longitudinal phasing for the pattern, but with strong disagreement among models over the details of the phasing and therefore the associated regional hydrologic impacts.

While the potential role of increasing hydrologic extremes with quasi-resonant stationary waves during NH summer has received considerable attention (see Section, as yet there is no clear evidence in model projections that this variability will increase (Teng and Branstator, 2019). The influence of the Arctic on mid-latitude circulation is assessed in Cross-Chapter Box 10.1, which reports that there is low confidence in the dominant contribution of Arctic warming compared to other drivers in future projections. Potential changes to the stratospheric polar vortex in CMIP5 models have a substantial influence on tropospheric stationary waves and associated hydrologic impacts in both the NH (Zappa and Shepherd, 2017) and SH (Mindlin et al., 2020). CMIP5 models have some important limitations in their representation of stationary waves (Lee and Black, 2013; Simpson et al., 2016; Garfinkel et al., 2020) and this aspect of CMIP6 models has not yet been comprehensively evaluated.

In summary, future changes in stationary waves may have an important influence on both the mean state and variability of the water cycle. Limitations in model representation, dynamical understanding, and the number of targeted studies on the topic currently constrain the assessment of future changes in stationary waves. Based on current knowledge, there is low confidence that projected changes in stationary wave activity will contribute to decreases of cold season precipitation over the eastern Mediterranean and increases over the west coast of North America. Atmospheric Blocking

In AR5, the increased ability of models to simulate blocking and higher agreement on projections led to an assessment with medium confidence that the frequency of NH and SH blocking will not increase, but future changes in blocking intensity and persistence were deemed uncertain (AR5 Chapter 14, ES and Box 14.2). Blocking influences precipitation (e.g., Trigo et al., 2004), flooding (e.g., Yamada et al. , 2016), drought (e.g., Dong et al. , 2018b), snow (e.g., García-Herrera and Barriopedro, 2006), and glacier melt (e.g., Hanna et al., 2013), and so is of broad importance to the water cycle in areas of blocking activity.

Blocking projections are assessed in this Report in Chapter 4 (Section, and model performance in simulating blocking is also discussed in Chapter 3 (Section CMIP5 projections suggest a complex response in blocking frequencies with an eastward shift in NH winter blocking, mid-latitude decreases in boreal summer except in eastern Europe western Russia, and SH decreases in the Pacific sector during austral spring and summer. CMIP6 projections (Figure. 4.28) show a notable decrease in blocking activity over Greenland and the North Pacific for the SSP3-7.0 and SSP5-8.5 scenarios. However, the continued large differences among current models as well as the sensitivity to blocking detection methods limits confidence in projected regional changes in blocking (see also Section The influence of blocking on multiple elements of the water cycle means that the uncertainty in blocking projections adds a corresponding layer of uncertainty to water cycle projections.

In summary, and despite recent improvements in the simulation of blocking, there is limited evidence in model projections of future changes, except for boreal winter over Greenland and the North Pacific where there is high confidence that blocking events are not expected to increase in the SSP3-7.0 and SSP5-8.5 scenarios. As with stationary waves, this adds uncertainty to mid-latitude water cycle projections at the regional scale. Extratropical Cyclones, Storm Tracks and Atmospheric Rivers Extratropical cyclones and storm tracks

The AR5 found that extratropical storms were expected to decrease in the Northern Hemisphere (NH), but only by a few percent. Meanwhile, precipitation associated with extratropical storms was projected to increase due to thermodynamic increases in moisture but potentially also due to intensification from increased latent heat release. Latent heating is a strong influence on extratropical storms, so it is plausible that changes in precipitation and associated latent heating could affect extratropical storm intensity and thus precipitation (Z. Zhang et al., 2019).

There is increased evidence that precipitation associated with individual extratropical storms is projected to increase, following thermodynamic drivers with negligible dynamic change (Yettella and Kay, 2017). Comparisons with reanalyses also support the projected increase in thermodynamic precipitation with little dynamic response for precipitation associated with extratropical storms (Li et al., 2014). There is high confidence that projected increases inprecipitation associated with extratropical storms in the NH (Marciano et al. , 2015; Pepler et al. , 2016; Michaelis et al. , 2017; Yettella and Kay, 2017; Zhang and Colle, 2017; Hawcroft et al. , 2018; Kodama et al. , 2019). A projected decrease in the number of extratropical cyclones over the NH during the boreal summer in CMIP5 models was reported by Chang et al. (2016) who related this decrease with a decrease in cloudiness and thus accentuating increased maximum temperatures. However, model spread was quite large, especially over North America, thus there is onlylow confidence in this seasonal signal.

In AR5, the Southern Hemisphere (SH) storm track was deemed likely to shift poleward, the North Pacific storm track more likely than not to shift poleward, while the North Atlantic storm track was unlikely to display any discernible changes. There was low confidence in regional storm track changes and the associated surface climate impacts, although a weakening of the Mediterranean storm track was a robust response of the models. Since AR5, the SH mid-latitude storm track is projected to shift poleward and the westerlies are projected to strengthen over Australia (CSIRO and BoM, 2015). Although thermodynamic effects were considered to be the most important factor in overall projections of increased mid-latitude precipitation, the general poleward shift in cyclogenesis and an enhanced latitudinal displacement of individual cyclones may play a role (Tamarin-Brodsky and Kaspi, 2017).

In AR5, several factors were identified as relevant to the uncertainties in projections of cyclone intensity, frequency, location of storm tracks and precipitation associated with ETCs. These include horizontal resolution, resolution of the stratosphere, and how changes in the Atlantic meridional overturning circulation (AMOC) were simulated. Since AR5, projections of extratropical cyclones and storm tracks have been examined further, largely confirming previous assessments. In particular, extratropical cyclone precipitation scales with the product of cyclone intensity (as measured by near-surface wind speed) and atmospheric moisture content (Pfahl and Sprenger, 2016). Booth et al. (2018) showed that the fraction of rainfall generated by the convection scheme in simulated extratropical cyclones is highly model- and resolution-dependent, which may be a source of uncertainty regarding their precipitation response to anthropogenic forcings. Also, increased moisture availability may increase the maximum intensity of individual storms while reducing the overall frequency as poleward energy transport becomes more efficient.

The role of temperature trends in influencing storm tracks has been further investigated, both in terms of upper tropospheric tropical warming (Zappa and Shepherd, 2017) and lower tropospheric Arctic amplification (J. Wang et al., 2017), including the direct role of Arctic sea ice loss (Zappa et al., 2018), and the competition between their influences (Shaw et al. , 2016). Physical linkages between Arctic amplification and changes in the mid-latitudes are uncertain, as discussed in Chapter 10 (Cross-Chapter Box 10.1). The remote and local SST influence has been further examined by Ciasto et al. (2016), who confirmed sensitivity of the storm tracks to the SST trends generated by the models and suggested that the primary greenhouse gas influence on storm track changes was indirect, acting through the greenhouse gas influence on SSTs. The importance of the stratospheric polar vortex in storm track changes has received more attention (Zappa and Shepherd, 2017; Mindlin et al. , 2020) and the anticipated recovery of the ozone layer further complicates the role of the stratosphere (Shaw et al., 2016; Bracegirdle et al., 2020b).

Biases remain in cyclone locations, intensities, cloud features, and precipitation (Catto, 2016, Chang et al., 2016). Uncertainties in projected precipitation changes in many mid-latitude regions can be explained to a large degree by uncertainties in projected storm track or ETC changes. Multiple studies (Chang et al., 2013; Zappa et al., 2015; Chang, 2018) have shown strong relationships between model-projected precipitation change in many regions and model-projected change in storm track activity near that regions. While front frequency is well represented, frontal precipitation frequency is too high and the intensity is too low (Catto et al., 2015). Some of the bias in storm tracks appears to be related to limitations in model realization of blocking (Zappa et al., 2014). The CMIP6 generation of models has improved representation of storm tracks in both hemispheres (Bracegirdle et al., 2020a; Harvey et al., 2020). Simulation of storm tracks and their associated precipitation generally improve with increasing resolution beyond that used in most current climate models (Jung et al. , 2006; Michaelis et al. , 2017; Barcikowska et al. , 2018). In terms of projections, the decreases in cyclone occurrence over the Mediterranean were replicated in a higher resolution model (Raible et al., 2018).

The projected changes in storm tracks and the associated mechanisms have several important implications for water cycle projections. P–E changes in the Mediterranean, California and Chile are directly linked to storm track changes (Zappa et al. , 2020). Where the storm tracks are robustly projected to shift (SH, North Pacific) or weaken (Mediterranean), understanding the physical causes of the related changes in precipitation helps increase confidence in the projections. Understanding the competing influences provides context for why other regions do not exhibit a consistent signal and cautions against regional projections based on individual models. However, model bias and the need for relatively high resolution to reproduce the relevant dynamics is an important overall limit on confidence in current CMIP6 projections.

In summary, there is the high confidence that precipitation associated with extratropical storms will increase with global warming in most regions. The SH storm track will likely shift poleward, the North Pacific storm track more likely than not will shift poleward, and the North Atlantic storm track is unlikely to have a simple poleward shift/ display any discernible changes. There is low confidence in regional storm track changes, although a weakening of the Mediterranean storm track is a robust response of the models. Atmospheric rivers

Atmospheric rivers were not assessed in AR5 but are important in the water cycle as they are linked to extreme rainfall, flooding, and changes in terrestrial water storage including melt and ablation of glaciers and snowpack (Sections 8.2.3). In a warming world, there is high confidence that thermodynamical increases in atmospheric water vapour ensure that atmospheric rivers will become wetter, hence stronger, and longer-lasting (Payne et al., 2020). This is clearly observed in several regional (Ralph and Dettinger, 2011; Lavers et al., 2013; Gao et al., 2015; Payne and Magnusdottir, 2015; Warner et al., 2015; Hagos et al., 2016; Gershunov et al., 2019) and in one global study (V. Espinoza et al., 2018) of atmospheric river activity in CMIP5 model projections. Lavers et al. (2015) indicate that integrated vapour transport under RCP 8.5 and 4.5 could increase, and consequently this thermodynamic response (O’Gorman, 2015) could affect mid-latitude regions where orographic precipitation is important (Gershunov et al., 2019).

Under continued global warming, more intense moisture transport within atmospheric river events is projected to increase the magnitude of heavy precipitation events on the west coast of the USA (Ralph and Dettinger, 2011; Lavers et al., 2015; Warner and Mass, 2017), in Western Europe (Lavers et al., 2015; Ralph et al., 2016; Ramos et al., 2016), and in East Asia (very likely) (Kamae et al., 2019). All CMIP5 models analysed agreed under a range of scenarios, except over the Iberian Peninsula (Ramos et al., 2016) where there is onlylow confidence in projected changes. Kamae et al. (2019) reported a 1% increase per °C warming in the frequency of atmospheric rivers affecting East Asia, but this is strongly affected by SST changes. Emerging evidence of possible regional changes due to dynamical factors are uncertain (Lavers et al., 2013; Gao et al., 2015; Payne and Magnusdottir, 2015). The frequency, magnitude and duration of atmospheric rivers making landfall along the North American west coast are projected to increase (Gershunov et al., 2019). In contrast, V. Espinoza et al. (2018) suggest that the number of atmospheric river events is projected to slightly decrease globally.

In semi-arid regions where atmospheric rivers have historically been important and precipitation is mainly confined to the cold season, the contribution of atmospheric rivers to annual total precipitation may be expected to grow disproportionately. For example, in California decreases in precipitation frequency are projected as a result of fewer non-atmospheric river storms, while the projected increase in heavy and extreme precipitation events are almost entirely a result of increased atmospheric river activity (Gershunov et al., 2019). Interannual variability in precipitation amounts is projected to increase because of the overall decrease in the frequency of storms but a stronger dependence on extremes (Polade et al., 2014), particularly due to atmospheric rivers (Gershunov et al., 2019), especially where interaction with topography are important (Polade et al. , 2014; Gershunov et al., 2019).

In summary, there is high confidence that the magnitude and duration of atmospheric rivers are projected to increase in future, leading to increased precipitation. This is projected to increase the intensity of heavy precipitation events on the west coast of the USA and in western Europe (high confidence). Modes of Climate Variability and Regional Teleconnections

Following on from the assessment of projected changes in modes of climate variability (MoVs) and regional teleconnections (Section 4.5.3), here we assess their consequences for projected water cycle changes. Tropical modes

CMIP6 projections indicate that the amplitude of ENSO (Annex IV.2.3) variability will not substantially change during the 21st century (high confidence) (Section However, rainfall variability related to ENSO is projected to increase significantly by the second half of the 21st century, regardless of ENSO amplitude (Section Regional precipitation variability associated with ENSO increases due to increases in atmospheric moisture, regardless of changes in ENSO variability itself (Pendergrass et al., 2017). In many regions, the magnitude of the projected changes related to ENSO is small compared with historical interannual variability (Bonfils et al., 2015; Power and Delage, 2018; Perry et al., 2020). Uncertainties in precipitation projections related to ENSO depend on internal variability associated with the mode (Section 8.5.2), hence the need to have relatively large ensembles (about 15 members) to adequately estimate uncertainty (Deser et al. , 2018; N. Maher et al. , 2018; C. Sun et al. , 2018; Zheng et al., 2018).

Even over regions with statistically significant simulated rainfall teleconnections during the historical period, CMIP5 models do not project clear changes (Perry et al., 2020). Nonetheless, CMIP5 models that realistically reproduce Indian summer monsoon rainfall indicate a strengthening of its relationship with ENSO in RCP8.5 projections, though the response is not consistent for different varieties of ENSO events (Roy et al., 2019). Inconsistent changes in the ENSO–Indian summer monsoon relationship in response to global warming in CMIP5 and CMIP6 models may be related to statistical issues rather than dynamical changes (Bódai et al., 2020; Haszpra et al., 2020). Over East Africa during the boreal spring and summer, ENSO teleconnections are projected to become stronger in the future (Endris et al., 2019). Meteorological drought consequences of each strong El Niño are projected to become more severe in the region (Rifai et al., 2019).

Indian Ocean Dipole (IOD, Annex IV.2.4) and Indian Ocean Basin (IOB, Annex IV.2.4) interactions with ENSO are expected to persist in the future (Section but projected changes in the frequency and intensity of events remain uncertain (Hui and Zheng, 2018; Endris et al., 2019; McKenna et al., 2020). Climate extremes such as those associated with the extreme positive IOD event of 2019 are expected to occur more frequently under continued global warming (Cai et al., 2021). Projected changes in IOD teleconnections are linked to model performance in representing the IOD and its remote influence in the present climate, apparently dominated by a positive IOD event-like mean state (G. Wang et al., 2017; Huang et al., 2019). Interactions between the IOD and the Indian Ocean mean state, via atmosphere–ocean feedbacks, can affect the behaviour of the IOD (Ng et al., 2018). In the eastern Horn of Africa, OND rainfall is projected to increase because of IOD-ENSO related SST changes in the Indo-Pacific region and associated Walker circulation changes (Endris et al., 2019).

Sensitivity studies generally project increases in Madden Julian Oscillation (MJO, Annex IV.2.8) precipitation amplitude in a warmer climate, with increases of up to 14%°C–1of warming (Arnold et al., 2013, 2015; Caballero and Huber, 2013; Liu and Allan, 2013; Maloney and Xie, 2013; Schubert et al., 2013; Subramanian et al., 2014; Carlson and Caballero, 2016; Pritchard and Yang, 2016; Adames et al., 2017a;