Chapter 1 has extensively explored this topic in Section 1.4.1 and Cross-Chapter Box 1.2. A summary of the main points relevant to the Atlas chapter and the Interactive Atlas are provided here.

There is no standard baseline in the literature although the World Meteorological Organization (WMO) recommends a 30-year baseline approach such as the climate-normal period 1981–2010. However, it retains the 1961–1990 period as the historical baseline for the sake of supporting long-term climate change assessments (WMO, 2017). Using the WMO standards also provides sample sizes relevant to calculating changes in statistics other than the mean. The AR6 WGI has established the 1995–2014 period as the recent-past baseline period – for similar reasons to the 1986–2005 period used in AR5 WGI (IPCC, 2013b) – since 2014 (2005) is the final year of the historical simulations of the models (more details in Cross-Chapter Box 1.2).

The choice of a baseline can significantly influence the analysis results for future changes in mean climate (Cross-Chapter Box 1.2; Hawkins and Sutton, 2016) as well as its variability and extremes. Thus, assessing the sensitivity of results to the baseline period is important. The Interactive Atlas (Atlas.2) allows users to explore and investigate a wide range of different baseline periods when analysing changes for future time slices or global warming levels:

• 1995–2014 (AR6 20-year baseline);
• 1986–2005 (AR5 20-year baseline);
• 1981–2010 (WMO 30-year climate normal);
• 1961–1990 (WMO 30-year long-term climate normal);
• 1850–1900 (baseline used in the calculation of global warming levels).

This promotes cross-Working Group consistency and facilitates comparability with previous reports and across datasets. For instance, the AR5 and long-term WMO baselines facilitate the intercomparison of CMIP5, CORDEX and CMIP6 projections since all have historical simulations in these periods. Using more recent baselines introduces discontinuity for the CMIP5 and CORDEX models, since historical simulations end in 2005. A pragmatic approximation to deal with this issue is to use scenario data to fill the missing segment, for example for 2006–2014 use the first years of RCP8.5-driven transient projections in which the emissions are close to those observed. This approach is used in the Atlas chapter and Chapter 12.

When assessing changes over the recent past, many studies analyse datasets using a range of climatologically significant periods (i.e., 30 years or more) with precise start and end dates depending on data availability and the year of the study. To account for this, when generating assessments from this literature the term ‘recent decades’ is used to refer to a period of approximately 30 to 40 years which ends within the period 2010–2020. An equivalent approximate description using specific years would be ‘since the 1980s’.

Regarding the future reference periods, the Interactive Atlas presents projected global and regional climate changes at near-, mid- and long-term periods, respectively 2021–2040, 2041–2060 and 2081–2100, for a range of emissions scenarios (Atlas.1.4.3 and Cross-Chapter Box 1.4).

Notingthe approach taken in the recent IPCC Special Report on Global Warming of 1.5°C (SR1.5) above 1850–1900 levels (IPCC, 2018b), the Atlas also presents global and regional climate change information at different global warming levels (GWLs, see Cross-Chapter Box 11.1). In particular, to provide policy-relevant climate information and represent the range of outcomes from the emissions scenario and time periods considered, GWLs of 1.5°C, 2°C, 3°C and 4°C are considered. The information is computed from all available scenarios (e.g., only 1.5°C and 2°C GWL information can be computed from projections under the SSP1-2.6 scenario). The Interactive Atlas allows comparison of timings for global warming across the different scenarios and of spatial patterns of change, for example information at 2°C GWL is calculated from SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5 projections (Section 4.2.4).

To calculate GWL information for the datasets used in the Atlas (CMIP6 and CMIP5; see Atlas.1.4), this chapter adopted the procedure used in Cross-Chapter Box 11.1. A model future climate simulation reaches the defined GWL of 1.5°C, 2°C, 3°C or 4°C when its global near-surface air temperature change averaged over successive 20-year periods first attains that level of warming relative to its simulation of the 1851–1900 climate (1851–1900 defines the pre-industrial baseline period for calculating the required global surface temperature baseline, Cross-Chapter Box 1.2). Note that this process is different from the one used in the SR1.5 report which used 30-year future periods. If a projection stabilizes before reaching the required threshold it is unable to simulate climate at that GWL and is thus discarded. For CORDEX simulations, the periods of the driving GCM are used, as in Nikulin et al. (2018). Detailed reproducible information on the GWLs used in the Atlas is provided in the Atlas repository (Iturbide et al., 2021).

Climateinformation at many temporal scales and over a wide range of temporal averaging periods is required for the assessment of climate change and its implications. These range from annual to multi-decadal averages required to characterize low-frequency variability and trends in climate to hourly or instantaneous maximum or minimum values of impactful climate variables. In between, information on, for example, seasonal rainfall is important and implies the need to include averaging periods whose relevance are geographically dependent. As a result, the Atlas chapter presents results over a wide range of time scales, from daily to decadal, and averaging periods with the Interactive Atlas allowing a choice of user-defined seasons and a range of predefined daily to multi-day climate indices.

Many factors influence the spatial scales and regions over which climate information is required and can be reliably generated. Despite all efforts in researching, analysing and understanding climate and climate change, a key factor in determining spatial scales at which analysis can be undertaken is the availability and reliability of data, both observational and from model simulations. In addition, information is required over a wide range of spatial domains defined either from a climatological or geographical perspective (e.g., a region affected by monsoon rainfall or a river basin) or from a socio-economic or political perspective (e.g., least-developed countries or nation states). Chapter 1 provides an overview of these topics (Section 1.5.2). This subsection discusses some relevant issues, summarizes recent advances in defining domains and spatial scales used by AR6 analyses and how these can be explored with the Interactive Atlas.

Recent IPCC reports – AR5 Chapter 14 (Christensen et al., 2013) and SR1.5 Chapter 3 (Hoegh-Guldberg et al., 2018) – have summarized information on projected future climate changes over sub-continental regions defined in the SREX report (Seneviratne et al., 2012) and later extended in AR5 from the 26 regions in SREX to include the polar, Caribbean, two Indian Ocean, and three Pacific Ocean regions (hereafter known as AR5 WGI reference regions; Figure Atlas.2a). In recent literature, new sub-regions have been used, for example for North and South America, Africa and Central America, together with the new definition of reference oceanic regions. Iturbide et al. (2020) describes an updated version of the reference regions which is used in this report (hereafter known as AR6 WGI reference regions) and is shown in Figure Atlas.2b. The goal of these subsequent revisions was to ensure that they represented sub-continental areas of greater climatic coherency.

Figure Atlas.2 | WGI reference regions used in the (a) AR5 and (b) AR6 reports (Iturbide et al., 2020). Asterisks indicate regions that extend across both sides of the map. The latter includes both land and ocean regions and it is used as the standard for the regional analysis of atmospheric variables in the Atlas chapter and the Interactive Atlas. The codes used in the Interactive Atlas are included in the figure. The full description of the regions (grouped by continents) is as follows. North America: NWN (North-Western North America), NEN (North-Eastern North America), WNA (Western North America), CNA (Central North America), ENA (Eastern North America); Central America: NCA (Northern Central America), SCA (Southern Central America), CAR (Caribbean); South America: NWS (North-Western South America), NSA (Northern South America), NES (North-Eastern South America), SAM (South American Monsoon), SWS (South-Western South America), SES (South-Eastern South America), SSA (Southern South America); Europe: GIC (Greenland/Iceland), NEU (Northern Europe), WCE (Western and Central Europe), EEU (Eastern Europe), MED (Mediterranean); Africa: MED (Mediterranean), SAH (Sahara), WAF (Western Africa), CAF (Central Africa), NEAF (North Eastern Africa), SEAF (South Eastern Africa), WSAF (West Southern Africa), ESAF (East Southern Africa), MDG (Madagascar); Asia: RAR (Russian Arctic), WSB (West Siberia), ESB (East Siberia), RFE (Russian Far East), WCA (West Central Asia), ECA (East Central Asia), TIB (Tibetan Plateau), EAS (East Asia), ARP (Arabian Peninsula), SAS (South Asia), SEA (South East Asia); Australasia: NAU (Northern Australia), CAU (Central Australia), EAU (Eastern Australia), SAU (Southern Australia), NZ (New Zealand); Antarctica: WAN (Western Antarctica), EAS (Eastern Antarctica). The definition of the regions and companion notebooks and scripts are available at the Atlas repository (Iturbide et al., 2021). Figure from Iturbide et al. (2020).

Open figure

The rationale followed for the definition of the reference regions was guided by two basic principles: 1) climatic consistency and better representation of regional climate features, and 2) representativeness of model results (i.e., sufficient number of model grid boxes). The finer resolution of CMIP6 models (as compared, on average, to CMIP5) yields better model representation of the reference regions allowing them to be revised for better climatic consistency (e.g., dividing heterogeneous regions) while preserving the model representation. Figure Atlas.3 illustrates this issue, displaying the number of grid boxes (over land for land regions) in the AR6 reference regions for two Interactive Atlas reference grids of horizontal resolutions of 1° and 2°, representative of the typical resolution of CMIP6 and CMIP5 models respectively. This figure shows that the new reference regions are well suited for the assessment of model results, with poorest model coverage for the New Zealand (NZ), Caribbean (CAR) and Madagascar (MAD) regions.

Figure Atlas.3 | Number of land grid boxes for each AR6 WGI reference region for the reference grids representative of (a) CMIP6 and (b) CMIP5, at 1° and 2° resolution respectively. Colour shading indicates regions with fewer than 250 grid boxes (the darkest shading is for regions with fewer than 20 grid boxes). The polygons show the AR6 WGI reference regions of Figure Atlas.2. Detailed information on the grids used is provided at the Atlas repository (Iturbide et al., 2021).

Open figure

The AR6 WGI (land and open ocean) reference regions are used in the Interactive Atlas as the default regionalization for atmospheric variables. However, these regions are not optimum for the analysis of oceanic variables since, for instance, the five upwelling regions (Canary, California, Peru, Benguela and Somali) are mostly included in ‘land’ regions. Therefore, the alternative set of oceanic regions defined by their biological activity (Figure Atlas.4) is used in the Interactive Atlas for the regional analysis of oceanic variables (see Fay and McKinley, 2014; Gregor et al., 2019). Due to the many potential definitions of the regions relevant for WGI and WGII, some additional typological and socio-economic regions have also been included in the Interactive Atlas.

Figure Atlas.4 | Typological and socio-economic regions used in the Interactive Atlas. (a) Eleven ocean regions defined by their biological activity and used for the regional analysis of oceanic variables; (b) ocean regions for Small Islands, including the Caribbean (CAR) and the north Indian Ocean (ARS and BOB); (c) land monsoon regions of North America, South America, Africa, Asia and Australasia; (d) major river basins; (e) mountain regions; (f) WGII continental regions. These regions can be used alternatively to the reference regions for the regional analysis of climatic variables in the Interactive Atlas. The definition of the regions and companion notebooks and scripts are available at the Atlas repository (Iturbide et al., 2021).

Open figure

In addition to contiguous spatial domains discussed in the previous section, some domains are defined by specific climatological, geographical, ecological or socio-economic properties where climate is an important determinant or influencer. In these cases the domains are subject to particular physical processes that are important for its climatology or that involve systems affected by the climate in a way that observations and climate model simulations can be used to understand. Many of these are the basis of the chapters and cross-chapter papers of the AR6 WGII report, namely river basins, biodiversity hotspots, tropical forests, cities, coastal settlements, deserts and semi-arid areas, the Mediterranean, mountains and polar regions. It is therefore important to generate climate information relevant to these typological domains and some examples of these used in the Interactive Atlas are shown in Figure Atlas.4.

There are various sources of observational information available for global and regional analysis. Observational uncertainty is a key factor when assessing and attributing historical trends, so assessment should build on integrated analyses from different datasets (disparity, inadequacy and contradictions in existing datasets are assessed in Section 10.2). The Atlas chapter can supplement and complement Chapter 10 by providing the opportunity to visualize and expand on its assessment. This includes displaying maps of density of stations’ observations (including those that are used in the different datasets) and assessing observational uncertainty by using multiple datasets.

Two of the most commonly used variables in climate studies are gridded surface air temperature and precipitation. There are many datasets available (Annex I) and Chapter 2 provides an assessment of key global datasets, including blended land-air and sea surface temperature datasets to assess global mean surface temperature (GMST). The Atlas separately analyses atmospheric and oceanic variables, and for the former a number of common global datasets supporting the assessment done in other chapters is used, including those selected in Chapter 2, but considering land-only information for the blended products. In particular, for air temperature the Atlas uses CRUTEM5 – the land component of the HadCRUT5 dataset – (Osborn et al., 2021), Berkeley Earth (Rohde and Hausfather, 2020) and the Climatic Research Unit CRU TS4 (version 4.04 used here; Harris et al., 2020). For precipitation the Atlas includes CRU TS4, the Global Precipitation Climatology Centre (GPCC, v2018 used here; Schneider et al., 2011), and Global Precipitation Climatology Project (GPCP; monthly version 2.3 used here; Adler et al., 2018). Although the ultimate source of these datasets is surface-station reported values (GPCP also includes satellite information), each has access to different numbers of stations and lengths of records and employs different ways of creating the gridded product and ensuring quality control. For oceanic variables, the most widely used sea surface temperature (SST) datasets are HadSST4 (Kennedy et al., 2019), which is the oceanic component of the HadCRUT5 dataset, ERSST (B. Huang et al., 2017 ), and KaplanSST (Kaplan et al., 1998).

Figure Atlas.5 shows the spatial coverage of the total number of observation stations for different periods (1901–1910, 1971–1980, and 2001–2010) for two illustrative datasets: the CRU TS4 dataset for precipitation and the SST data in HadSST4. The former illustrates spatially the declining trend of station observation data used in the precipitation datasets for certain regions (South America and Africa) after the 1990s. This demonstrates the regional inhomogeneity and temporal change in station density, which is in part a consequence of many stations not reporting to the WMO networks and their data being held domestically or regionally. During early years (before 1950) a limited number of observations are available. This information is used in the Interactive Atlas to blank out regions not constrained with observations in those datasets providing station density information.

Figure Atlas.5 | Number of stations per 0.5° × 0.5° gridcell reported over the periods of 1901–1910, 1971–1980, and 2001–2010 (rows 1–3), and the global total number of stations reported over the entire globe (bottom row) for precipitation in the CRU TS4 dataset (left) and the HadSST4 dataset (right). Further details on data sources and processing are available in the chapter data table (Table Atlas.SM.15).

Open figure

In addition to surface observations, satellites have been widely used to produce rainfall estimates. The advantage of satellite-based rainfall products is their global coverage including remote areas but there is significant uncertainty in these products over complex terrain (Rahmawati and Lubczynski, 2018; Satgé et al., 2019). Another recent development has been on gridded datasets for climate extremes based on surface stations, such as HadEX3 (Dunn et al., 2020), as described in Section 11.2.2.

There are some studies assessing observational datasets globally (Beck et al., 2017; Q. Sun et al., 2018 ) and regionally (Manzanas et al., 2014; Salio et al., 2015; Prakash, 2019), reporting large differences among them and stressing the importance of considering observational uncertainty in regional climate assessment studies. Uncertainty in observations is also a key limitation for the evaluation of climate models, particularly over regions with low station density (Kalognomou et al., 2013; Kotlarski et al., 2019). More detailed information on these issues is provided in Section 10.2.

For regional studies, observational datasets with global coverage are complemented by a range of regional observational analyses and gridded products, such as E-OBS (Cornes et al., 2018) over Europe, Daymet (Thornton et al., 2016) over North America, or APHRODITE (Yatagai et al., 2012) over Asia. These are highlighted in various other chapters and the Atlas expands on their treatment, complementing discussions on discrepancies/conflicts in observations presented in Chapter 10 and expanding on and replicating their results for other regions. In particular, the Interactive Atlas includes the global and regional observational products described here to assess observational uncertainty over the different regions analysed.

There are currently many atmospheric reanalysis datasets with different spatial resolution and assimilation algorithms (see Annex I and Section 1.5.2). There are also substantial differences among these datasets due to the types of observations assimilated into the reanalyses, the assimilation techniques that are used, and the resolution of the outputs, amongst other reasons. For example, 20CR (Slivinski et al., 2019) only assimilates surface pressure and sea surface temperature (SST) to achieve the longest record but at relatively low resolution, while ERA-20C (Poli et al., 2016) only assimilates surface pressure and surface marine winds. At the other extreme, very sophisticated assimilation systems using multiple surface, upper air and Earth observation data sources are employed, for example ERA5 (Hersbach et al., 2020) and JRA-55 (Harada et al., 2016), which also have much higher resolutions. Most reanalysis datasets cover the entire globe, but there are also high-resolution regional reanalysis datasets which provide further regional detail (Kaiser-Weiss et al., 2019).

The Atlas and Interactive Atlas use information from ERA5 and from the bias-adjusted version WFDE5 (Cucchi et al., 2020) which is combined with ERA5 information over the ocean and used as the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP) observational reference dataset W5E5 (Lange, 2019b). This reference is also used in the Atlas for model evaluation (Atlas.1.4.4) and for bias-adjusting model outputs (Atlas.1.4.5).

The Atlaschapter (and the Interactive Atlas) uses global model simulations from both CMIP5 and CMIP6, mainly historical and future projections performed under ScenarioMIP (O’Neill et al., 2016). This facilitates backwards comparability and thus the detection of new salient features and findings from recent science and the latest CMIP6 ensemble. The selection of the models is based on availability of scenario data for the variables assessed in the Atlas chapter and for those included in the Interactive Atlas (Atlas.2.2). In particular, in order to harmonize the results obtained from the different scenarios as much as possible, only models providing data for the historical scenario and at least two emissions scenarios, RCP2.6, RCP4.5 and/or RCP8.5 (for CMIP5), and SSP1-2.6, SSP2-4.5, SSP3-7.0 and/or SSP5-8.5 (for CMIP6), were chosen, resulting in 29 and 35 models, respectively (see Cross-Chapter Box 1.4 for a description of the scenarios). In the Atlas chapter (similarly to the regional Chapters 11 and 12) a single simulation is taken from each model (see Atlas.12 for limitations of this choice). Since the RCP and SSP emissions scenarios are not directly comparable due to different regional forcing (Section 4.2.2), the Atlas includes GWLs as an alternative dimension of analysis (Cross-Chapter Box 11.1), which allows intercomparison of results from different scenarios as an alternative to the standard analysis based on time slices for particular scenarios (Atlas.1.3.1). This dimension allows for enhanced comparability of CMIP5 and CMIP6, since it constrains the regional patterns to the same global warming level for both datasets.

Building on this information, the Interactive Atlas displays a number of (mean and extreme) indices and climatic impact-drivers (CIDs), considering both atmospheric and oceanic variables (Atlas.2.2). Some of these indices have been selected in coordination with Chapters 11 and 12, in order to support and extend the assessment performed in these chapters (see Annex VI for details on the indices). In order to harmonize this information, the indices have been computed for each individual model on the original model grids and the results have been interpolated to a common 2° (for CMIP5) and 1° (CMIP6) horizontal resolution grids. In addition, for the sake of comparability with CMIP6 results (in particular when using baselines going beyond 2005), the historical period of the CMIP5 and CORDEX datasets has been extended to 2006–2014 using the first years of RCP8.5-driven transient projections (Atlas.1.3.1). Tables listing the CMIP5 and CMIP6 models used in the Atlas and in the Interactive Atlas for different scenarios and variables are included as Supplementary Material (Tables Atlas.SM.1 and Atlas.SM.2, respectively); moreover, full inventories including details on the specific Earth System Grid Federation (ESGF) versions are given in the Atlas GitHub repository (Iturbide et al., 2021).

Chapter 3 and Flato et al. (2013) describe the evaluation of CMIP6 and CMIP5 models, respectively, assessing surface variables and large-scale indicators. Section 10.3.3 assesses the general capability of GCMs to produce climate output for regions.

Information from the existing CMIP5 and CMIP6 datasets is supplemented with downscaled regional climate simulations from CORDEX. This facilitates an assessment of the effects from higher resolution, including whether this modifies the projected climate change signals compared to global models and adds any value, especially in terms of high-resolution features and extremes.

Global model data,as generated by the CMIP ensembles, although available globally, have spatial resolutions that are limited for reproducing certain processes and phenomena relevant for regional analysis (around 2° and 1° for CMIP5 and CMIP6, respectively). The Coordinated Regional Climate Downscaling Experiment (CORDEX; Gutowski Jr. et al., 2016) facilitates worldwide application of Regional Climate Models (RCMs, see Section 10.3.1.2), focusing on a number of regions (Figure Atlas.6) with a typical resolution of 0.44° (but also at 0.22° and 0.11° over some domains, such as Europe). However, only a few simulations are available for some domains (Annex II, Table AII.1), thus limiting the level of analysis and assessment that can be performed using CORDEX data in some regions. Moreover, there are regions where several domains overlap, thus providing additional lines of evidence. The use of multi-domain grand ensembles to work globally with CORDEX data have recently been proposed (Legasa et al., 2020; Spinoni et al., 2020). Ongoing efforts, such as the multi-domain CORDEX-CORE simulations are promoting more homogeneous coverage and thus more systematic treatment of CORDEX domains (Box Atlas.1).

Figure Atlas.6 | CORDEX domains showing the curvilinear domainboundaries resulting from the original rotated domains. The topography corresponding to the standard CORDEX 0.44° resolution is shown to illustrate the orographic gradients over the different regions.

Open figure

A lot of progress has been made by the regional climate modelling community since AR5 (Table AII.1) to produce and make available evaluation (reanalysis-driven) simulations over the different CORDEX domains along with downscaled CMIP5 historical and future climate projection information under a range of emissions scenarios, mainly RCP2.6, RCP4.5 and RCP8.5 (Tables AII.3 and AII.4). However, these ensembles cover only a fraction of the uncertainty range spanned by the full CMIP5 ensemble in the different domains (e.g., Figures Atlas.16, Atlas.17, Atlas.21, Atlas.22, Atlas.24, Atlas.26, Atlas.28 and Atlas.29; Ito et al., 2020b). Therefore, comparison of CMIP5 and CORDEX results should be performed carefully, providing results not only for the full CMIP5 ensemble but also for the sub-ensemble formed by the driving models since results can diverge (Fernández et al., 2019; Iles et al., 2020).

The Atlas chapter and the Interactive Atlas use CORDEX information for the following 11 individual CORDEX domains (out of the 14 domains shown in Figure Atlas.6): North, Central and South America; Europe; Africa; South, East and South East Asia; Australasia; Arctic and Antarctica; in addition, oceanic information has been used from the Mediterranean domain, which provides simulations from coupled atmosphere–ocean regional climate models (RCMs). In order to harmonize the information across domains and to maximize the size of the resulting ensembles, all the available simulations for each individual CORDEX domain (including the standard 0.44° CORDEX and the 0.22° CORDEX-CORE) have been interpolated to a common regular 0.5°-resolution grid to provide a grand ensemble covering the historical and future emissions scenarios RCP2.6, RCP4.5 and RCP8.5, and also the reanalysis-driven simulations for evaluation purposes. In the case of the European domain, the dataset considered is the 0.11° simulations (CORDEX EUR-11, the same dataset as used in Chapter 12), which has been interpolated to a regular 0.25° resolution grid (the same used for the regional observations). In the case of the Mediterranean domain, oceanic information (sea surface temperature, SST) is interpolated to a regular 0.11° grid. In all cases, the indices are computed on the original grids and the interpolation process is applied to the resulting indices. Moreover, for the sake of comparability with CMIP6 results (in particular when using baseline periods beyond 2005), the historical period of the CORDEX datasets has been extended to 2006–2014 using the first years of RCP8.5-driven transient projections in which the emissions are close to those observed (see Atlas.1.3.1); note that this procedure is also applied to CMIP5 simulations.

For the different CORDEX domains, the full ensembles of models (GCM-RCM matrix) used in the Atlas for the different scenarios and variables are described in the Supplementary Material (Tables Atlas.SM.3–Atlas.SM.14) and in the Atlas repository (Iturbide et al., 2021), including full metadata relative to ESGF versions used and the periods with data available for the different simulations. In particular, the historical scenario information is only available from 1970 onwards for some models and therefore the common period 1970–2005 is used for historical CORDEX data in the Atlas. As a result, the WMO baseline period 1961–1990 is not available in the Interactive Atlas for CORDEX data.

Sections Atlas.4 to Atlas.11 assess research on CORDEX simulations over different regions, analysing past and present climate as well as future climate projections. They also focus on regional model evaluation in order to extend and complement the validation of global models in Chapter 3, considering the specific regional climate and relevant large-scale and regional phenomena, drivers and feedbacks (Section 10.3.3). Besides the literature assessment, some simple evaluation diagnostics have been computed for the simulations used in the Atlas chapter to provide some basic information on model performance across regions. In particular, biases for mean temperature and precipitation have been calculated for the 11 CORDEX domains analysed.

Figure Atlas.7 shows mean temperature and precipitation biases over the North American domain in RCM simulations driven by reanalysis and historical GCM simulations (Section 10.3.2.5). Annual and seasonal (December–January–February (DJF) and June–July–August (JJA)) biases are computed for both the RCMs and driving GCMs. Biases in the reanalysis-driven RCMs result from intrinsic model errors, with the results displayed being spatially aggregated for each reference region. This same analysis is performed for the GCM-driven RCM simulations over the historical period 1986–2005. This allows comparison of the intrinsic bias of the RCMs with the biases resulting when driven by the different GCMs and patterns of behaviour in the RCMs, for example intrinsic warm and dry biases in ENA and WNA respectively or reduced RCM warm biases compared to the CCCma GCM in NEN and ENA. Similar results for the other CORDEX domains are included as Supplementary Material (Figures Atlas.SM.1–Atlas.SM.10).

Figure Atlas.7 | Evaluation of annual and seasonal air temperature and precipitation for the six North America sub-regions, NWN, NEN, WNA, CNA, ENA and NCA (land only) for CORDEX-NAM RCM simulations driven by reanalysis or historical GCMs. Seasons are June–July–August (JJA) and December–January–February (DJF). Rows represent sub-regions and columns correspond to the models. Magenta text indicates the driving historical CMIP5 GCMs (including ERA-Interim in the first set of slightly separated columns) and the black text to the right of the magenta text represents the driven RCMs. The colour matrices show the mean spatial biases; all biases have been computed for the period 1985–2005 relative to the observational reference (E5W5, see Atlas.1.4.2). Further details on data sources and processing are available in the chapter data table (Table Atlas.SM.15).

Open figure

Bias adjustment is often applied to data from climate model simulations to improve their applicability for assessing climate impacts and risk (e.g., in the Inter-Sectoral Impact Model Intercomparison Project, ISIMIP; Rosenzweig et al., 2017). Bias-adjustment approaches (Section 10.3.1.3) are particularly beneficial when threshold-based indices are used, but they can introduce other biases, in particular when applied directly to coarse-resolution GCMs (Cross-Chapter Box 10.2). Bias-adjustment techniques should be chosen carefully for a specific application. In the Atlas, bias adjustment is not applied systematically (in particular, it is not applied for the variables assessed in the Atlas chapter), and only some threshold-dependent extreme indices and climatic impact-drivers (CIDs) included in the Interactive Atlas are bias adjusted (in particular TX35 and TX40 in coordination with Chapter 12). To facilitate integration with WGII, the Atlas uses the same bias-adjustment method as in ISIMIP3 (Lange, 2019a) and the same observational reference (W5E5, see Atlas.1.4.2), upscaled to the same resolution as the model to avoid downscaling artefacts (Cross-Chapter Box 10.2). The ISIMIP3 bias-adjustment method is a trend-preserving approach that is recommended for general applications, as it reduces biases while preserving the original climate change signal (Casanueva et al., 2020). Following the recommendations given in Chapter 10, results in the Interactive Atlas are displayed for both the adjusted and the raw model output.

The primary purpose of the IPCC is to provide a policy-relevant, non-prescriptive assessment of the state of knowledge on climate change and its impacts. This purpose is different from the provision of information targeted to implement climate policies, which is the focus of climate services and national climate change assessment communities. IPCC assessments are based on quantitative observational and model-generated data that are also used in many activities supporting the development of climate policies. However, the functionality of the Interactive Atlas is primarily aimed at supporting the knowledge assessment.

Much of the assessment in this report is based on multiple lines of evidence (Cross-Chapter Box 10.3). The Interactive Atlas facilitates combining multiple observational and model-generated datasets and spatial and temporal analyses that combine to support statements on the characteristics of the climate system. The use of predefined spatial and temporal aggregations imposes constraints on the ability to make specific or tailored assessments but does provide essential background and uncertainty information to generate broad findings and provide confidence statements on these. Also, the inclusion of a selection of extremes and climatic impact-drivers (CIDs) is a new element in the Interactive Atlas and facilitates broader application, including the handshake with WGII. Below, some guidelines on the use, interpretation and limitations of the Interactive Atlas are given.

Africa has many varied climates which can be categorized as dry regime in the Saharan region, tropical humid regime in West and East Africa except for parts of the Greater Horn of Africa (alpine) and the Sahel (semi‐arid), and a dry/wet season regime in the northern and southern African region including the Namib and Kalahari deserts; each climate region has its local variations resulting in very high spatial and temporal variations (Peel et al., 2007). Based on the varied climates, nine sub-regions are defined for Africa (Figure Atlas.1 6): the Mediterranean region (MED) including North Africa, Sahara including parts of the Sahel (SAH), West Africa (WAF), Central Africa (CAF), North Eastern Africa (NEAF), South Eastern Africa (SEAF), West Southern Africa (WSAF), East Southern Africa (ESAF) and Madagascar (MDG).

The climatic features that characterize the intra-seasonal and interannual variability of Africa are mainly the Madden–Julian Oscillation (MJO), which is confined to the deep tropics during boreal winter, Pacific Decadal Variability (PDV), and the shift of the Atlantic Inter-tropical Convergence Zone in response to changes in the meridional SST gradient. A positive phase of PDV weakens African monsoons (Figure AIV.8d; Meehl and Hu, 2006), and MJO phase 4 suppresses convection over equatorial Africa (Figure AIV.10a; see Annex IV). Other features influence specific sub-regions. For instance, El Niño events increase precipitation in eastern Africa and decrease precipitation in southern Africa. Over southern Africa there is a strong link between ENSO and droughts (Meque and Abiodun, 2015). The positive phase of the Indian Ocean Dipole (IOD) increases rainfall in eastern tropical Africa in boreal autumn to early winter (Figure AIV.5d), while the negative phase induces the reduction in rainfall. The West African Monsoon is influenced by Atlantic Zonal Mode (AZM) with decreased rainfall over the Sahel and increased rainfall over Guinea (Losada et al., 2010). Positive Atlantic Multi-decadal Variability (AMV) influences positive anomalies all year round over a broad Mediterranean region, including North Africa.

The most recent IPCC reports, AR5 and SR1.5 (Christensen et al., 2013; Hoegh-Guldberg et al., 2018), state that over most parts of Africa, minimum temperatures have warmed more rapidly than maximum temperatures during the last 50 to 100 years (medium confidence). In the same period, minimum and maximum temperatures have increased by more than 0.5°C relative to 1850–1900 (high confidence). While the quality of ground observational temperature measurements tends to be high compared to that of measurements for other climate variables, Africa remains an under-represented region as reported in SR1.5 (Hoegh-Guldberg et al., 2018; IPCC, 2018c). Based on the Coupled Model Intercomparison Project Phase 5 (CMIP5) ensemble and reported in IPCC AR5 and SR1.5, surface air temperatures in Africa are projected to rise faster than the global average increase and are likely to increase by more than 2°C and up to 6°C by the end of the century, relative to the late 20th century, if global warming reaches 2°C (Bindoff et al., 2013; Niang et al., 2014; Hoegh-Guldberg et al., 2018). The higher temperature magnitudes are projected during boreal summer. Southern Africa is likely to exceed the global mean land surface temperature increase in all seasons by the end of the century. Temperature projections for East Africa indicate considerable warming under RCP8.5 where average warming across all models is approximately 4°C by the end of the century. According to SROCC, eastern Africa like other regions with smaller glaciers is projected to lose more than 80% of its glaciers by 2100 under RCP8.5 (medium confidence) (Hock et al., 2019b).

West Africa has also experienced an overall reduction of rainfall over the 20th century, with a recovery towards the last 20 years of the century (Christensen et al., 2013). Over the last three decades rainfall has decreased over East Africa, especially between March and May/June. Projected rainfall changes over Africa in the mid- and late 21st century is uncertain. In regions of high or complex topography such as the Ethiopian Highlands, downscaled projections indicate likely increases in rainfall and extreme rainfall by the end of the 21st century. However, North Africa and the south-western parts of South Africa are likely to have a reduction in precipitation.

The consequence of increased temperature and evapotranspiration, and decreased precipitation amount, in interaction with climate variability and human activities, have contributed to desertification in dryland areas in sub-Saharan Africa (medium confidence) as reported in SRCCL (Mirzabaev et al., 2019).

Summer (June–August) mean temperature in eastern China has increased by 0.82°C since reliable observations were established in the 1950s (Sun et al., 2014). Based on historical meteorological observations, the best estimate of the linear trend of annual mean surface air temperature (SAT) for China with 95% uncertainty ranges is 0.38°C ± 0.05°C per decade for 1979–2015 (Li et al., 2017). From 1960 to 2010, theincreasing trend of temperature was about 0.34°C per decade in the arid region of north-west China, higher than the average over China (B. Li et al., 2012 ; Xu et al., 2015). Over South Korea, warming is 1.4–2.6 times larger than global trends. The increase is 1.90°C during 1912–2014 and 0.99°C during 1973–2014 (Park et al., 2017) with a 25–45% urbanization contribution. The annual temperature increased in large cities at a rate of 0.29°C ± 0.08°C per decade compared with 0.11°C ± 0.08°C per decade in other stations in South Korea from 1960 to 2010 (H.-S. Kim et al., 2016 ). A relatively high increase in annual mean temperature at the rate of 3.0°C per century was detected in the Tokyo metropolitan area for the period 1901–2015 (Matsumoto et al., 2017). Trends of annual temperature for the period of 1961–2015 are shown in Figure Atlas.11. Most areas of East Asia have significant warming trends exceeding 0.1°C per decade, and the strongestwarming (0.3°C–0.4°C per decade) occurs in northern China.

Observational studies indicated significant decadal variations in the EAWM (Wang and Lu, 2016; He et al., 2017). It weakened significantly around the late 1980s, being relatively strong during 1976–1987 and weaker during 1988–2001. The EAWM has recovered in intensity after 2004 and caused frequent and prevalent severe cold spells, as well as a number of unusually harsh cold winters in many parts of East Asia during the period 2004–2012 (Wang and Chen, 2014; Kug et al., 2015; Ge et al., 2016; Gong et al., 2018). Negative zonal mean winter SAT anomalies were observed over the whole of East Asia from 1980 to 1988, with positive anomalies observed over high and low latitudes from 1988 to 2010 (Miao and Wang, 2020).

Precipitation trends over East Asia show considerable regional differences (medium confidence). Mean precipitation has shown negligible sensitivity to the warming trend with consequently limited overall trends in China though summer rainfall daily frequency and intensity show respectively decreasing and increasing trends from 1961 to 2014 (Zhou and Wang, 2017). The summer precipitation trends over eastern China display a dipole pattern, characterized by positive anomalies in central-eastern China along the Yangtze River Valley and negative anomalies in north China since the 1950s (Section 8.3.2.4.2). This pattern has changed with the enhanced rainfall in the Huaihe River Valley and decreased in the regions south of the middle and lower reaches of the Yangtze River Valley since the 2000s (Liu et al., 2012; Zhao et al., 2015). The climate in north-west China changed from ‘warm–dry’ to ‘warm–wet’ condition in the mid-1980s (Peng and Zhou, 2017; Wang et al., 2020), with an increased rate of annual precipitation of about 3.7% per decade from 1961 to 2015 (P. Wu et al., 2019 ) and 11.2 mm per decade between 1960 and 2011 in northern Xinjiang (Xu et al., 2015). Mean rainfall and the number of rainy days during the Meiyu-Baiu-Changma period from June to September have increased during 1973–2015 in Korea (Lee et al., 2017). The precipitation trend has caused a large increase in summer precipitation at a rate of 40.6 ± 4.3 mm per decade, resulting in an increase of annual precipitation of 27.7 ± 5.5 mm per decade in South Korea from 1960 to 2010 (H.-S. Kim et al., 2016 ). Precipitation amounts exhibited a slight decrease at both the annual and seasonal scales in Japan for the period 1901–2012 (Duan et al., 2015).

Agriculture intensification through oasis expansion in Xinjiang region has increased summer precipitation in the Tian Shan mountains (high confidence from medium evidence with high agreement) (Zhang et al., 2009, 2019b; Deng et al., 2015; Guo and Li, 2015; Yao et al., 2016; Xu et al., 2018; Cai et al., 2019). However, there is very low confidence of the effect of oasis expansion on the temperature warming trend (Han and Yang, 2013; Li et al., 2013; Yuan et al., 2017).

In the context of climate warming, intense snowfalls have hit China frequently in recent winters and have caused severe damages to the sustainability of society (Sun et al., 2019). Observations generally show a decrease in the frequency and an increase in the mean intensity of snowfalls in north-western, north-eastern and south-eastern China and the eastern Tibetan Plateau since the 1960s (Zhou et al., 2018), but the results may depend on the objective criteria for identifying winter snowfall (J. Luo et al., 2020 ).

Current climate models perform poorly insimulating the mean precipitation in East Asia, including the phase of the northward progression of the seasonal rainband (M. Zhang et al., 2018 ). Although there has been an improvement in the simulation of mean states, interannual variability and past climate changes in the progression from CMIP3 to CMIP5, some previously documented biases (such as the ridge position of the western North Pacific Subtropical High and the associated rainfall bias) are still evident in CMIP5 models (Sperber et al., 2013; Zhou et al., 2017). Most models capture the main characteristics of the winter mean circulation over East Asia reasonably well, but they still suffer from difficulty in predicting the interannual variability of the EAWM (Shin and Moon, 2018). Models have improved from CMIP5 to CMIP6 for climatological temperature and EAWM (D. Jiang et al., 2020 ). Some CMIP6 models also show improvements in simulating the annual mean and interannual variation of precipitation (Sellar et al., 2019; Tatebe et al., 2019; T. Wu et al., 2019 ). The performance of models is sensitive to cumulus convection schemes and horizontal resolution (Haarsma et al., 2016; Wu et al., 2017; Kusunoki, 2018b). High-resolution atmospheric global climate models (AGCM) successfully reproduce the intensity and the spatial pattern of the EASM rainfall (Li et al., 2015; Yao et al., 2017; Ito et al., 2020a) and improve the simulation of the diurnal cycle of precipitation rates and the probability density distributions of daily precipitation over Korea, Japan and northern China (Lin et al., 2019), but increasing horizontal resolution (at the typical scales used in GCMs) is not always a panacea for solving model biases (Roberts et al., 2018).

Recent studies using CORDEX-EA models with resolution of about 12–25 km showed that the RCMs produce relatively more detailed regional features of the temperature distribution compared with the driving GCMs (Tang et al., 2016). Over China, RCMs provide more spatial details and in general reduce the biases of their driving GCMs, in particular in DJF (December–January–February) and over areas with complex topography (Wu and Gao, 2020). However, RCMs also show biases in simulating East Asian precipitation and its variability (Park et al., 2016; Zhou et al., 2016; Zou and Zhou, 2016), and do not always show added value compared to the driving GCMs (Li et al., 2018b). For example, by comparing inter-GCM and inter-RCM differences around the Japan archipelago, it was found that RCM generate relatively large differences in precipitation (Suzuki-Parker et al., 2018). The RCM multi-model ensemble produces superior simulation compared to that of a single model (Jin et al., 2016; D.-L. Guo et al., 2018 ). A comparative study of RCMs at different spatial resolutions showed that with coarse resolution they present some limitations and high-resolution RCMs offer added value for several evaluation metrics (Park et al., 2020).

The development of climate models provides a solid basis for projection of future monsoon changes under different global warming scenarios. Coupled model simulations indicate that East Asia and the Tibetan Plateau will likely experience higher warming than the global mean conditions across all global warming levels (Figure Atlas.1 7) and with the projected warming greater in ECA and TIB than EAS. Also, in the CMIP6 ensemble, the multi-model mean and 90th percentile warming for a given period and emissions scenario are consistently greater than in the CMIP5 ensemble. Larger warming magnitudes are projected to occur in the southern, north-western, and north-eastern regions of China, parts of Mongolia, the Korean Peninsula, and Japan than in other regions (Li et al., 2018a). Projections indicate winter increases in SAT over the East Asian continent and in precipitation over the northern East Asian continent with 1.5°C and 2.0°C global warming under the RCP4.5 and RCP8.5 scenarios (Miao et al., 2020).

Projected annual precipitation changes in the CMIP5 and CMIP6 ensembles are positive for all warming levels in ECA and TIB and for the higher warming levels in EAS. Changes in precipitation per degree Celsius global warming are larger in DJF than in JJA in ECA but show smaller seasonal difference in EAS (Figure Atlas.1 7). The EASM precipitation is projected to increase but with a complex spatial structure (Kitoh, 2017; Moon and Ha, 2017). Simulations from CMIP5 models show that compared with the current summer climate, both SAT and precipitation increase significantly over the East Asian continent during the 1.5°C warming period (L. Chen et al., 2019 ) and that the main mode of EASM precipitation changes from tripolar to dipolar (Wang et al., 2018). The increase in precipitable water in the wet EASM region is only slightly greater than the global average but the increase in precipitation is much greater (Z. Li et al., 2019 ). The monsoon circulation in the lower troposphere is projected to strengthen due to the enhanced thermal forcing by the Tibetan Plateau (He et al., 2019; He and Zhou, 2020), which causes the increased summer precipitation over the East Asian continent. Precipitation over eastern China increases for almost all months under global warming in projections from GCMs with different horizontal resolutions (Kusunoki, 2018a). Also, under RCP scenarios, in the 21st century, mean precipitation is projected to increase (Kim et al., 2020), especially in the late afternoons (Oh and Suh, 2018), over the Korean Peninsula due to global warming and associated changes in EASM. Increase in JJA mean precipitation is projected in northern East Asia consistently among the CMIP models, while northward migration of early summer East Asian rainbands such as the Meiyu-Baiu-Changma is delayed along with that of the mid-latitude westerly jet in the future (Horinouchi et al., 2019). However, the geographical distribution of precipitation change tends to depend more on the cumulus convection scheme (Ose, 2017) and horizontal resolution of models rather than on SST distributions. Under the RCP4.5 and the RCP8.5 scenarios, the interannual variability in EASM rainfall is projected by the multi-model ensemble mean to increase in the 21st century (Ren et al., 2017). Further studies show a projected increase in heavy rainfall together with increases in rainfall intensity (Endo et al., 2017). Multi-model intercomparison indicates significant uncertainties in future projections of climate change in East Asia, although precipitation increases consistently across models (Zhou et al., 2017). Simulations under the RCP4.5 scenario project that the number of snow days will be reduced by the end of the 21st century relative to 1986–2005, primarily owing to the decline of light snowfall events. The total amount is projected to increase in north-western China but decrease in the other sub-regions (Zhou et al., 2018).

The increasing temperature trends under RCP scenarios were consistently reproduced in projections using CORDEX-EA models (Y. Kim et al., 2016 ) as reported in AR5 using GCMs. However, changes in annual and seasonal mean precipitation exhibit significant inter-RCM differences with larger magnitudes and variability than in the GCMs (Ham et al., 2016; Ozturk et al., 2017; H. Sun et al., 2018 ; D. Zhang et al., 2018 ). RCM simulations project that the Meiyu-Baiu-Changma heavy rainfall will significantly increase in northern Japan at the end of the 21st century under the RCP8.5 scenario (Osakada and Nakakita, 2018), but projected precipitation amount and the number of precipitation days in summer around and over Japan differ as a result of RCM uncertainty (Suzuki-Parker et al., 2018). Annual total snowfall is projected to decrease in most parts of Japan except for Japan’s northern island under RCP2.6 (Kawase et al., 2021).

Projejctions based on statistical downscaling of 37 CMIP5 GCMs for Xinjiang, China, show pronounced temperature increases of 0.27°C to 0.51°C per decade from 2021 to 2060 while precipitation changes were projected to be between –1.7% to 6.8% per decade and varying seasonally and spatially (Luo et al., 2018). A decrease of precipitation was projected in the western region of Xinjiang during summer. More extreme rainfall events were projected to occur during summer and autumn.

In East Asia annual mean temperature has been increasing since the 1950s (high confidence). The linear trend of annual mean surface air temperature likely exceeded 0.1°C per decade over most of East Asia from 1961 to 2015. Trends of annual precipitation show considerable regional differences with areas of both increases and decreases (medium confidence), and with increases over north-west China and South Korea (high confidence). Agricultural intensification through oasis expansion in Xinjiang region has increased summer precipitation in the Tian Shan mountains (high confidence).

GCMs still show poor performance in simulating the mean rainfall and its variability over East Asia, especially over regions characterized by complex topography. The CMIP6 models have improved from CMIP5 for climatological temperature and winter monsoon but show little improvements for the summer monsoon. The RCMs produce relatively more detailed regional features, but do not always produce superior simulations compared with the driving GCMs.

The annual mean surface temperature over East Asia and the Tibetan Plateau will very likely increase under all emissions scenarios and GWLs. Larger warming magnitudes will likely occur in the northern part of EAS and in ECA and TIB. Precipitation is likely to increase over land in most of EAS at the end of the 21st century under higher-emissions scenarios (SSP3-7.0, RCP8.5 and SSP5-8.5) and global warming levels, and in ECA and TIB under all emissions scenarios and global warming levels. Summer precipitation increase is likely to occur in East Asia, corresponding to the strengthened summer monsoon circulation.

Increases in surface air temperature (SAT) have been observed since the mid-1970s over the whole of North Asia (Frolov et al., 2014), and particularly over the north-eastern part (Figure Atlas.11; Gruza et al., 2015). Trends of annual SAT in the northern part of the region during the last decades were very likely twice as strong as the global average (Figure Atlas.11; Frolov et al., 2014; Mokhov, 2015; Sherstyukov, 2016) with trends in RFE of 0.8°C–1.2°C per decade for the 1976–2014 period and more intense warming strengthening from south to north observed in spring in ESB (Frolov et al., 2014; Ippolitov et al., 2014; Kokorev and Sherstiukov, 2015).

Recent strong warming in polar regions (Section Atlas.11.2) was accompanied by cooling in winter in mid-latitude regions particularly in the southern part of WSB and ESB (Cohen et al., 2014; Ippolitov et al., 2014; Gruza et al., 2015; Kharyutkina et al., 2016; Overland et al., 2016; Perevedentsev et al., 2017; Wegmann et al., 2018). These temperature decreases were strongly correlated with significant warming over the Barents-Kara Sea (greater than 2.5°C per decade during 2003–2017) and sea ice loss, suggesting a causal link (Outten and Esau, 2012; Semenov et al., 2012; Overland et al., 2016; Semenov, 2016; Wegmann et al., 2018; Meleshko et al., 2019; Susskind et al., 2019), though recent studies (Blackport et al., 2019; Clark and Lee, 2019) have shown that both phenomena result from mid-latitude circulation variability (see also Cross-Chapter Box 10.1). In addition, significant warming in the last decade has halved the cooling trend in southern WSB from –0.6°C per decade during 1976–2012 to –0.3°C per decade during 1976–2018 (high confidence) (Frolov et al., 2014; Roshydromet, 2019).

Annual precipitation totals very likely increased over North Asia in the last half century along with more heavy and less light precipitation, more freezing rain and less freezing drizzle (Figure Atlas.11 and the Interactive Atlas; Wen et al., 2014; Groisman et al., 2016; Ye et al., 2017; Chernokulsky et al., 2019). The highest increase was observed over regions of Siberia and RFE with estimated trends of 10–25 mm per decade for the 1976–2014 period (Kokorev and Sherstiukov, 2015) or 5% per decade for the 1976–2018 period (Roshydromet, 2019). Increases over southern RFE are the largest (over 50 mm per decade) and are mostly due to positive changes in convective precipitation intensity in the region in the summer season (JJA) during 1966–2016 (medium confidence) (Chernokulsky et al., 2019). A decreasing trend was observed in central WSB, northern ESB, the Baikal and Transbaikal regions, the Amur River region, and Primorie territories of RFE (the Kamchatka and Chukchi peninsulas) with up to –20 mm per decade for the 1976–2014 period (Kokorev and Sherstiukov, 2015) or 15–20% per decade for the 1976–2018 period (Roshydromet, 2019). Overall, solid precipitation predominantly decreased in North Asia and very likely caused both less snow cover extent (SCE) and snow water equivalent (SWE), attributable to the anthropogenic influence with high confidence (Sections 2.3.2.2 and 3.4.2).

Snow characteristics depend on both temperature and precipitation, and observed trends over North Asia show large spatial heterogeneity and interannual variability (Figure Atlas.1 8) leading to medium confidence that maximum snow depth has increased over Siberia, the Okhotsk Sea coast and in southern RFE since the 1960s (Callaghan et al., 2011; Loginov et al., 2014), with trends during 1976–2016 of 1.8 cm (in WBS), 1.1 cm (in ESB), and 4.6 cm (in RFE) per decade (Bulygina et al., 2017). Snow cover duration increased in Yakutia, Sakhalin Island and some other coastal areas of the Pacific Ocean in RFE during 1980–2009 (Callaghan et al., 2011), and decreased in WSB and ESB (Bulygina et al., 2017; Roshydromet, 2019). However, Gorbatenko et al. (2019) reported that in south-eastern WSB maximal snow depth has increased by 5–20 cm and duration of steady snow cover by between 4 and 10 days during 1989–2016 (Figure Atlas.1 8).

Figure Atlas.18 | Linear trends for the 1980–2015 period based on station data from the World Data Centre of the Russian Institute for Hydrometeorological Information (RIHMI-WDC; Bulygina et al., 2014). (a) Snow-season duration from 1 July to 31 December (days per decade); (b) snow-season duration from 1 January to 30 June (days per decade); (c) maximum annual height of snow cover (mm per decade). Trends have been calculated using ordinary least squares regression and the crosses indicate non-significant trend values (at the 0.1 level) following the method of Santer et al. (2008) to account for serial correlation. Further details on data sources and processing are available in the chapter data table (Table Atlas.SM.15).

Open figure

Temperature trends and means derived from reanalysis datasets (JRA-25 and MERRA) correctly represented temperature variability shown in observational data over the Asian territory of Russia for the 1976–2010 period (Loginov et al., 2014). Assessment of CRU TS 3.22, CRUTEMP4, ERA-Interim and NCEP2 datasets against station data over North Asia for annual and seasonal air temperature has shown that the ERA-Interim reanalysis outperforms others for the 1981–2005 period (Kokorev and Sherstiukov, 2015). The latter reanalysis also underestimates summer precipitation and shows large wet biases over north-east Asia during spring and underestimates mean seasonal temperature over north-east Asia in spring (MAM), autumn (SON), and winter (DJF), but overestimates it in summer (JJA) compared with the CRU dataset (medium confidence) (Ozturk et al., 2017; Top et al., 2021).

GCMs capture the main synoptic processes affecting North Asia and the CMIP5 ensemble simulates the temporal evolution of the magnitude and position of the Siberian High (SH) over the period 1872–2005 (Fei and Yong-Qi, 2015). CMIP5 models simulate a weakened intensity of the winter SH and a strengthened interannual variability compared to observations (Fei and Yong-Qi, 2015). The characteristics of blocking events over the region (number, duration, intensity and frequency) were reasonably well reproduced by GCMs (Mokhov et al., 2014), and most overestimate the annual mean temperature over northern Eurasia (Interactive Atlas). Biases in simulated annual surface air temperature simulation primarily come from the winter (DJF) season and are relatively smaller in other seasons (Miao et al., 2014; Peng et al., 2019). Most GCMs capture the main decadal SAT trend (Miao et al., 2014), though CMIP5 GCMs fail to capture the decreasing temperature trend over East Siberia (Fei and Yong-Qi, 2015). Possible causes of GCMs’ inability to represent the recent slowdown of warming is further discussed in Cross-Chapter Box 3.1. For CMIP5, models with higher resolution do not always perform better than those with lower resolutions (medium confidence) (Miao et al., 2014).

Sixteen CMIP5 model simulations of SAT variability over Eurasia were evaluated against CRU observations for permafrost sub-regions (Peng et al., 2019), showing a warm bias in north-west Eurasia, capturing the climate warming over the 20th century and its acceleration during the late 20th century. CMIP5 GCMs generally underestimate daily temperature range compared with observations over north-eastern Russia (Sillmann et al., 2013; Lindvall and Svensson, 2015). Currently there is no literature on the CMIP6 ensemble over the region though a few single-model studies are available (Voldoire et al., 2019; T. Wu et al., 2019 ).

There is very limited use of RCMs for North Asia. CORDEX-CAS covers North Asia, except parts of RFE, and ARCTIC-CORDEX covers the northern regions (Figure Atlas.6). For CORDEX-CAS three RCMs (REMO, ALARO-0 and CLMcom) have been used and have warm biases for maximum temperatures, cold biases for minimum temperatures and a wet bias in the north during the winter (Top et al., 2021). Rain gauges, however, are known to have problems in terms of measuring properly solid precipitation (e.g., due to drifting snow) which can greatly affect the accuracy of precipitation observations over North Asia (Harris et al., 2014).

CMIP5 and CMIP6 projections are consistent in the direction and ranges of surface temperature change which are higher than the global average and with ensemble-mean warming of around 6°C for the 4°C GWL. Projected precipitation changes are also consistent with significant increases in winter, of up to 40% in the ensemble mean for the highest warming levels, and lower increases in summer except for WSB where changes are small and suggest drying at the 4°C GWL (Figure Atlas.1 7 and the Interactive Atlas).

The CMIP5 ensemble projects a warming of the annual mean SAT over northern Eurasia in the 21st century, likely in the range of 0.8°C–1.0°C (RCP2.6), 2.3°C–3.1°C (RCP4.5) and up to 7.2°C (RCP8.5) (Miao et al., 2014; Peng et al., 2019). Mid-latitude permafrost sub-regions in Eurasia are projected to warm more than the global mean and non-permafrost territories, with ensemble area-averaged changes of 1.7°C (RCP2.6), 3.2°C (RCP4.5) or 6.4°C (RCP8.5) in 2081–2100 relative to 1986–2005 (Peng et al., 2019).

Over the Central Asia CORDEX domain, RegCM4.3.5 simulations driven by two different CMIP5 GCMs (HadGEM2-ES and MPI-ESM-MR) project SAT warming for 2071–2100 relative to 1971–2000 of about 3°C–4°C during the summer for RCP4.5 to over 7°C for all seasons for RCP8.5. Projected warming is most evident on the large continental Siberian Plateau with boreal and sub-boreal climates and biomes (i.e., taiga forests and tundra) during the winter season (Ozturk et al., 2017). The Voeikov Main Geophysical Observatory (MGO) RCM, driven by five CMIP5 GCMs for the RCP8.5 scenario, projects a faster increase in annual minimum temperature as compared with maximum temperature over the whole territory of Russia (Kattsov et al., 2017), and the smallest change in growing season lengths (i.e., periods with daily temperatures over 5°C, 10°C and 15°C) in the area of northern taiga in WSB and ESB comparable with other territories of Russia during the 21st century (Torzhkov et al., 2019).

For precipitation, MGO RCM projects for the Arctic-CORDEX domain under the RCP8.5 scenario increases in annual totals for northern North Asia, a decrease in summer over ESB for 2006–2100 relative to 1951–2005 and significant increases in the upper limit of intense precipitation over most of the region in winter (Kattsov et al., 2017; Khlebnikova et al., 2018). Other RCM projections show that in most seasons and for all future periods, precipitation in Siberia is not projected to change with respect to the 1971–2000 period, except under the RCP8.5 scenario for the winter and autumn (Ozturk et al., 2017). This very limited and controversial evidence leads to low confidence in RCM precipitation projections for North Asia and since the projections of GCMs and ESMs are more physically consistent, assessment of future precipitation changes is based on CMIP5/CMIP6 presented in Figure Atlas.1 7 and the Interactive Atlas.

Annual surface air temperature and precipitation have very likely increased and maximum snow depth has likely increased over most of North Asia since the mid-1970s. The highest warming has been found in spring in ESB and RFE, strengthening from south to north with linear trends of 0.8°C–1.2°C per decade over the 1976–2014 period (high confidence). A temperature decrease was identified just in winter in the southern part of WSB and ESB as a result of natural variability, but halved from –0.6°C per decade in 1976–2012 to –0.3°C per decade for the longer 1976–2018 period due to recent warmer winters (high confidence). Over North Asia annual precipitation increases with estimated trends of 5–15 mm per decade in the 1976–2014 period have been recorded with an exception over the Kamchatka and the Chukchi peninsulas, where decreases of up to –20 mm per decade in the same period have been found (medium confidence). Snow cover duration has very likely decreased over Siberia and increases in maximum snow depths of 1.8 cm, 1.1 cm and 4.6 cm per decade have been observed for WSB, ESB and RFE respectively from 1976 to 2016 (limited evidence).

Most of the CMIP5 and some CMIP6 GCMs overestimate the annual mean air temperature and precipitation over the North Asia region (medium confidence). GCMs generally represent the observed decadal temperature trend (medium confidence) and biases primarily come from the winter (DJF) season (high confidence). Results of a very limited number of RCMs applied over the whole region show that they have warmer biases for maximum and colder biases for minimum temperatures (limited evidence, medium agreement). Sparsity of observational data particularly in the northern part of ESB and the whole of the RFE results in low confidence in the assessments of model performance in North Asia.

Surface air temperature and precipitation in North Asia are projected to increase further (high confidence) with warming higher than the global average and around 6°C at the 4°C GWL. Temperature change in 2080–2099 relative to 1981–2000 is likely in the range of 3°C in summer to 4.9°C in winter under the RCP4.5 scenario, and 5.6°C in summer to 9.7°C in winter under the RCP8.5 scenario. Precipitation is projected to increase with ensemble-mean changes of 9% in summer under both RCP4.5 and RCP8.5, and of 22% and 56% in winter respectively.

Recent studies show that annual mean land temperatures over Indiawarmed at a rate of around 0.6°C per century during 1901–2018, which was primarily contributed by a significant increase in annual maximum temperature of 1.0°C per century, while the annual minimum temperature showed a lesser increasing trend of 0.18°C per century during this period, with a significant rise only in the recent few decades (1981–2010) at a rate of 0.17°C per decade (Srivastava et al., 2017, 2019). The annual average of daily maximum and minimum temperatures has increased over almost all Pakistan with a faster increasing trend in the south (high confidence). Minimum temperatures have increased faster (0.17°C–0.37°C per decade) than maximum temperatures (0.17°C–0.29°C per decade) with the diurnal temperature range reduced (–0.15°C to –0.08°C per decade) in some regions (Khan et al., 2019).

There has been a noticeable declining trend in rainfall with monsoon deficits occurring with higher frequency in different regions in South Asia (see also Section 8.3.2.4 on the South Asian monsoon). Concurrently, the frequency of heavy precipitation events has increased over India, while the frequency of moderate rain events has decreased since 1950 (high confidence) (Goswami et al., 2006; Dash et al., 2009; Christensen et al., 2013; Krishnan et al., 2016; Kulkarni et al., 2017; Roxy et al., 2017). There is a considerable spread in the seasonal and annual mean precipitation climatology and interannual variability among the different observed precipitation datasets over India (Collins et al., 2013; Prakash et al., 2014; Kim et al., 2018; Ramarao et al., 2019). Yet, the regions of agreement among datasets lend high confidence that there has been a decrease in mean rainfall over most parts of the eastern and central north regions of India (Singh et al., 2014; Roxy et al., 2015; Juneng et al., 2016; Krishnan et al., 2016; Guhathakurta and Revadekar, 2017; Jin and Wang, 2017; Latif et al., 2017). A global modelling study with high resolution over South Asia (Sabin et al., 2013) indicated that a juxtaposition of regional land-use changes, anthropogenic-aerosol forcing and the rapid warming signal of the Equatorial Indian Ocean was crucial to simulate the observed Indian summer monsoon weakening in recent decades (medium confidence).

A dipole-like structure in summer monsoon rainfall trends is observed over the northern Indo-Pakistan area with significant increases over Pakistan and decreases over central north India resulting from strengthening (weakening) of vertically integrated meridional moisture transport over the Arabian Sea (Bay of Bengal) (low confidence) (Latif et al., 2017). Positive annual precipitation trends are observed in global and regional datasets (Figure Atlas.11 and the Interactive Atlas) during 1961–2015 and over arid provinces of Pakistan (for rabi and kharif cropping seasons) during 1951–2015 of 2.8–34.8 mm per decade (Khan et al., 2020) imply high confidence for increased precipitation in Pakistan. Observations located in the monsoon-dominated strip in Pakistan indicate that the mean monsoon onset became earlier during 1971–2010 (Ali et al., 2020).

Snow and glaciers are major water resources of all countries in South Asia. Glacier melting is mainly controlled by natural phenomena but anthropogenic emissions of black carbon (BC) are now making a significant contributing to total glacial melting in the Hindu Kush Himalaya (HKH) region (Menon, 2002; Ramanathan et al., 2007; Ramanathan and Carmichael, 2008). BC concentration is seven to 10 times higher in mid-altitudes (1000–4000 metres above sea level) than at high altitudes (>4000 metres above sea level). The concentration of BC sampled from the surface of snow/ice samples as well as ice-core records shows decreasing ice albedo and an acceleration in glacier melting (Cross-Chapter Box 10.4; Wester et al., 2019). Karakoram and western HKH snow cover is increasing, a phenomena known as the ‘Karakoram anomaly’, and partially attributed to an increase in the strength of westerly disturbances (Wester et al., 2019).

Significant glacier retreat has been observed since 1960 in TIB with lower rates in the interior of the region (Yao et al., 2007). A large inter-decadal variation in snow cover is also observed from 1960 to 2010. Observations and model simulations showed that the increasing temperature of frozen grounds is leading to thawing and reduced depth of permafrost, with further significant reductions projected under future global warming scenarios (medium confidence) (Yang et al., 2019).

Whilst simulations of Indian summer monsoon rainfall (ISMR) have improved in CMIP5 compared to CMIP3 in terms of northward propagation, time for peak monsoon and withdrawal (Sperber et al., 2013), they fail to simulate the trends in monsoon rainfall and the post-1950 weakening of monsoon circulation (Saha et al., 2014). This is partially attributed to the failure of coarse-resolution CMIP5 models to simulate fine-resolution processes such as orographic effects or land surface feedback, and problems in cloud parametrization result in an overestimation of convective precipitation fraction (M.S. Singh et al., 2017 ). In CMIP6, a significant improvement is found in capturing the monsoon spatio-temporal patterns over India, particularly in the Western Ghats and north-eastern Himalayan foothills (Gusain et al., 2020). Over Pakistan the CMIP6 models simulate surface temperature better in JJA than DJF (Karim et al., 2020). The CMIP6 ensemble underestimates annual mean temperature over all of South Asian with mixed results for precipitation (Almazroui et al., 2020b). The CMIP6 GCMs have a large cold bias in both mean annual maximum and minimum temperatures in the complex Karakorum and Himalayan mountain ranges but exhibit warm biases in mean annual minimum temperature in most of the rest of South Asia.

Regional climate model (RCM) downscaling of CMIP5 models as part of CORDEX South Asia uses higher resolution (50 km) and improved surface fields such as topography and coastlines to resolve better the complexities of the monsoon and other hydrological processes (Giorgi et al., 2009). The added value of their simulations, relative to the driving GCMs, presents a complex picture. CORDEX RCMs better represent spatial patterns of temperature (Sanjay et al., 2017), the spatial features of precipitation distribution associated with the Indian summer monsoon (Choudhary and Dimri, 2018), and the simulation of monsoon active- and break-phase composite precipitation (Karmacharya et al., 2017b). The RCMs follow the driving GCMs in underestimating seasonal mean surface air temperature and overestimating spatial variability in precipitation. They amplify CMIP5 cold biases over almost the entire region, including over the HKH region, Afghanistan and south-west Pakistan during winter (Iqbal et al., 2017), and substantial cold biases of 6°C–10°C are found over the Himalayan watersheds of the Indus basin (Nengker et al., 2018; Hasson et al., 2019). Neither RCMs nor their driving CMIP5 GCMs reproduce well the region’s precipitation climatology (Mishra, 2015). In addition, important characteristics of ISMR such as northward and eastward propagation, onset, seasonal rainfall patterns, intra-seasonal oscillations and patterns of extremes did not show consistent improvement (S. Singh et al., 2017 ). Also, these RCM simulations have not demonstrated added value in capturing the observed changes in ISMR characteristics over recent decades, though RegCM4 simulations at 25 km showed high accuracy in capturing monsoon precipitation characteristics and atmospheric dynamics in historical simulations (Ashfaq et al., 2021).

Evaluation of four global reanalysis products (ERA5 and ERA-Interim, JRA-55 and MERRA-2; Atlas.1.4.2) for snow depth and snow cover over TIB was performed against 33 in situ station observations, Interactive Multisensor Snow and Ice Mapping System (IMS) snow cover and a satellite microwave snow-depth dataset (Orsolini et al., 2019). Most of the reanalyses showed a systematic overestimation. Only ERA-Interim assimilated IMS snow cover at high altitudes, whereas ERA5 did not and the excessive snowfall, snow depth and snow cover in ERA5 was attributed to this difference. The analysis of annual maximum consecutive snow-covered days for the period 1980–2018 over TIB using JRA-55 and passive microwave satellite observations showed a decreasing trend in all time periods and in recent snow seasons for MERRA-2 (Bian et al., 2020). The uncertainty assessment of model physics in snow modelling over TIB using ground-based observations and high-resolution snow cover satellite products from the Moderate Resolution Imaging Spectroradiometer (MODIS) and FengYun-3B suggests that errors can be overcome by optimizing parametrizations of the snow cover fraction rather than optimizing physics-scheme options (Y. Jiang et al., 2020 ).

CMIP5 and CMIP6 surface temperature projections for South Asia are consistent across the range of GWLs withincreases greater than the global average, more so over TIB (Figure Atlas.1 7). CMIP6 models show higher sensitivity to greenhouse gas emissions, projecting higher warming for a given emissions scenario. The north-western parts of South Asia, mainly covering the Karakorum and Himalayan mountain ranges, are projected to warm more (over 6°C under SSP5-8.5, with higher warming in winters than in summer; Interactive Atlas) and this will accelerate glacier melting in the region. The warming pattern of maximum and minimum temperatures are projected to intensify in higher latitudes compared with mid-latitudes of South Asia in CMIP5 simulations for all RCP scenarios (Ullah et al., 2020).

Seasonal precipitation projections show increased winter precipitation over the western Himalayas and decreased precipitation over the eastern Himalayas. On the other hand, summer precipitation projections show a robust increase over most of South Asia, with the largest over the arid region of southern Pakistan and in adjacent areas of India, under SSP5-8.5 (Almazroui et al., 2020b). Daily bias-adjusted projections from 13 CMIP6 GCMs using all emissions scenarios project a warmer (3°C–5°C) and wetter (13–30%) climate in South Asia in the 21st century (Mishra et al., 2020).

With continued global warming and anticipated reductions in anthropogenic aerosol emissions in the future, CMIP5 models project an increase in the mean and variability of summer monsoon precipitation over India by the end of the 21st century, together with substantial increases in daily precipitation extremes (medium confidence) (Krishnan et al., 2020), see also Section 8.4.2.4 on changes in the South Asian monsoon. The CMIP5 GCMs consistently project an increase in moisture transport over the Arabian Sea and Bay of Bengal towards the end of the 21st century, an increase in moisture convergence and consequent increases in monsoon rainfall over the Indo-Pakistan region which are higher under RCP8.5 than RCP4.5 (Srivastava and Delsole, 2014; Mei et al., 2015; Latif et al., 2018). Out of 20 CMIP5 GCMs, four showed an increase in magnitude and lengthening of the summer monsoon across India under RCP8.5. The intensity of both strong and weak monsoons is projected to increase during the period 2051–2099 (Srivastava and Delsole, 2014).

Summer precipitation changes in South Asia are consistent between CMIP3 and CMIP5 projections, but the model spread is large for winter precipitation changes. Changes in summer monsoon rainfall will dominate annual changes over South Asia (Woo et al., 2019). CMIP3 GCMs project a gradual increase in annual precipitation over monsoon-dominated areas of Pakistan throughout the 21st Century and increases in humid and semi-arid climate areas (Saeed and Athar, 2018).

Warming of 2.5°C–5°C is projected over northern Pakistan and India (Syed et al., 2014). CORDEX-South Asia projections over north-east India under RCP4.5 for the period 2011–2060, show increasing trends for both seasonal maximum and minimum temperature over north-east India (Interactive Atlas). The future projections of South Asian monsoon from the CORDEX-CORE exhibit a spatially robust delay in the monsoon onset, an increase in seasonality, and a reduction in the rainy season length over parts of South Asia at higher levels of radiative forcing (Ashfaq et al., 2021).

With TIB continuing to warm, snow cover and snow water equivalent are projected to decrease but with regional differences due to synoptic influences (Cross-Chapter Box 10.4; Wester et al., 2019). There is limited evidence on whether the ‘Karakoram Anomaly’ will persist in coming decades, but its long-term persistence is unlikely with continued projected warming (high confidence) (Section 9.5.1.1). It is projected that peak river flow at higher altitudes will commence earlier, due to warming influences on snow cover area and snow/glacier melt rates and with more precipitation falling as rain rather than snow, and the magnitude and seasonality of flow will change over South Asia (Charles et al., 2016).

Mean, minimum and maximum daily temperatures in South Asia are increasing and winters are getting warmer faster than summers (high confidence). The South Asian monsoon has shown contrasting behaviour over India and Pakistan. There is high confidence that there has been a decrease in mean rainfall over most parts of the eastern and central north regions of India and an increase in precipitation in Pakistan.

Global model performance over the region has improved from CMIP3 to CMIP5 to CMIP6 in the multi-model ensemble-mean simulation of the amplitude and phase of the seasonal cycles of temperature and precipitation. However, there was no appreciable improvement in regions with steep orography, and there has remained substantial inter-model spread in seasonal and annual mean temperatures over South Asia with generally cold biases which are largest in the complex Karakorum and Himalayan mountain ranges. CMIP6 GCMs also show a dry bias (15–20%) in mean annual precipitation in the majority of the South Asia region with a wet bias in Nepal, Pakistan and northern India.

It is likely that surface temperatures over South Asia will increase more than the global average and more so over TIB, with projected increases of 4.6°C (3.4°C–6.0°C) during 2081–2100 compared with 1995–2014 under SSP5-8.5 and 1.3°C (0.7°C– 2.0°C) under SSP1-2.6 (Interactive Atlas). Summer monsoon precipitation in South Asia is likely to increase by the end of the 21st century while winter monsoons are projected to be drier. Over the same time periods CMIP6 models project an increase in annual precipitation in the range 14–36% under SSP5-8.5 and 0.4–16% under SSP1-2.6 (medium confidence).

With continued warming, TIB snow cover and snow water equivalent are likely to decrease and with more precipitation falling as rain rather than snow in SAS. It is projected that the peak river flow at higher altitudes will commence earlier due to the effect of warming on snow cover and snow/glacier melt rates, causing changes in magnitude and seasonality of flow.

Within the last decade, there has been an increasing number of studies on climatic trends over South East Asia, carried out on a regional basis (Thirumalai et al., 2017; Cheong et al., 2018) or focused on specific countries (Cinco et al., 2014; Villafuerte et al., 2014; Mayowa et al., 2015; Villafuerte and Matsumoto, 2015; Guo et al., 2017a; Supari et al., 2017; Sa’adi et al., 2019; Tan et al., 2021). They document virtually certainsignificant increases in mean as well as extreme temperature. The minimum temperature extremes very likely warmed faster compared to the maximum temperature. Temperatures, including extremes, are strongly influenced by ENSO in the region (Cinco et al., 2014; Thirumalai et al., 2017; Cheong et al., 2018). Over much of the region, extreme high temperatures occurred mostly in April and almost all April extreme temperatures occur in El Niño years (Thirumalai et al., 2017). In most of South East Asia (except for the north-eastern areas), there was likely an increase in the number of warm nights with El Niño episodes within the period 1972–2010 (Cheong et al., 2018).

Changes in mean precipitation are less spatially coherent over South East Asia. Over Thailand, the average number of rain days has decreased by 1.3 to 5.9 days per decade while average daily rainfall intensity has increased by 0.24–0.73 mm day^–1per decade (Limsakul and Singhruck, 2016). Precipitation is also affected by ENSO events (Tangang et al., 2017; Supari et al., 2018). Over South East Asia, there has been a significant increase in the amount of precipitation and its extremes with La Niña episodes in the past decades, especially during the winter monsoon period (high confidence) (Villafuerte and Matsumoto, 2015; Limsakul and Singhruck, 2016; Cheong et al., 2018).

Figure Atlas.11 shows trends in mean temperature and precipitation during 1961–2015 for two global datasets, indicating a significant overall warming over South East Asia (high confidence), with higher rates of warming in Malaysia, Indonesia, and the southern areas of mainland South East Asia (low confidence). Annual mean precipitation trends (Atlas.1.4.1 and the Interactive Atlas, which includes the regional dataset Aphrodite) over the region are mostly not significant except for increases over parts of Malaysia, Vietnam and the southern Philippines (medium confidence).

It is important to note that the availability, quality, and temporal and spatial density of observation data may lead to uncertainties and varying results in South East Asia (Juneng et al., 2016). Some efforts have been made to produce better observationally-based gridded datasets for the region (e.g., Nguyen-Xuan et al., 2016; van den Besselaar et al., 2017; Yatagai et al., 2020).

Performance in simulating rainfall over South East Asia varies among CMIP5 GCMs (high confidence). Only some are capable of reasonably simulating the rainfall seasonal cycle and spatial pattern (Siew et al., 2013; Raghavan et al., 2018). Over mainland South East Asia, the performance of CMIP5 GCMs in simulating rainfall during the wet season was superior to that for annual and dry-season precipitation (J. Li et al., 2019 ).

RCMs have been intensively used over the region in recent years in a series of single or multi-model experiments and there is medium confidence that they reproduce reasonably well seasonal climate patterns of temperature, precipitation and large-scale circulation over the different sub-regions of South East Asia with added values compared to their host GCMs (Kwan et al., 2014; Ngo-Duc et al., 2014, 2017; Van Khiem et al., 2014; Juneng et al., 2016; Katzfey et al., 2016; Loh et al., 2016; Raghavan et al., 2016; Cruz et al., 2017; Ratna et al., 2017; Trinh-Tuan et al., 2018; Nguyen-Thuy et al., 2021). RCM ensemble means tend to outperform the individual models in representing the climatological mean state (Ngo-Duc et al., 2014; Trinh-Tuan et al., 2018; Nguyen-Thi et al., 2021). There is relatively high consistency among the simulations of historical climate over mainland South East Asia compared to those over the Maritime Continent for both seasonal and interannual variability (Ngo-Duc et al., 2017). The consistency in rainfall simulations was lower than for temperature simulations.

Some RCMs showed a systematic cold bias (Manomaiphiboon et al., 2013; Kwan et al., 2014; Ngo-Duc et al., 2014; Loh et al., 2016; Cruz and Sasaki, 2017; Cruz et al., 2017) that was mainly due to model physics (Manomaiphiboon et al., 2013; Kwan et al., 2014) and/or the biases in the SST forcing (Ngo-Duc et al., 2014). A few simulations revealed a warm bias over some areas such as in the Maritime Continent (Cruz et al., 2017) or Vietnam (Van Khiem et al., 2014). The biases for rainfall in GCMs and RCMs over South East Asia were found to be less systematic with wet or dry biases depending on the sub-regions (Manomaiphiboon et al., 2013; Kwan et al., 2014; Van Khiem et al., 2014; Juneng et al., 2016; Supari et al., 2020; Tangang et al., 2020; Nguyen-Thi et al., 2021), although wet biases were more pronounced in RCMs (Kwan et al., 2014; Van Khiem et al., 2014; Kirono et al., 2015; Juneng et al., 2016; Supari et al., 2020; Tangang et al., 2020). Some RCMs overestimated rainfall interannual variability (Juneng et al., 2016) while some others underestimated it (Kirono et al., 2015). Simulated rainfall amount is sensitive to the choice of convective scheme (Juneng et al., 2016; Ngo-Duc et al., 2017) and the choice of land surface scheme (Chung et al., 2018). Rainfall biases in current climate simulations can be greatly reduced if a bias adjustment method such as quantile mapping is applied (Trinh-Tuan et al., 2018). The pattern of tropical cyclone numbers in the region were reasonable represented by RCM outputs (Van Khiem et al., 2014; Kieu-Thi et al., 2016; Herrmann et al., 2020).

Mean temperature in South East Asia is projected to continue to rise through the 21st century (virtually certain, very high confidence). Projections by multi-model regional climate simulations of CORDEX-SEA showed a temperature increment over land under RCP8.5 to range from 3°C–5°C by the end of the 21st century relative to the pre-1986–2005 period (Tangang et al., 2018). For the same periods, the average mean temperature increase over land projected by CMIP5 (CMIP6) varies, with 10th–90th percentile ranges, from 0.7°C to 1.3°C (0.7°C to 1.8°C) under RCP2.6 (SSP1-2.6) to 2.8°C to 4.4°C (2.6°C to 4.8°C) under RCP8.5 (SSP5-8.5) (Interactive Atlas). For all GWLs the land region is projected to warm by a slightly smaller amount than the global average, with 10th–90th percentile ranges for CMIP5 (CMIP6) of 1.2°C–1.6°C (1.2°C–1.5°C) for the 1.5°C GWL and of 3.3°C–4.0°C (3.3°C–3.9°C) for the 4°C GWL relative to the 1850–1900 baseline (calculated from RCP8.5 (SSP5-8.5) projections). Changes for other warming levels, periods and emissions pathways are shown in Figure Atlas.1 7 and can be explored in the Interactive Atlas.

Projections of future rainfall changes are highly variable among sub-regions of South East Asia and among the models (high confidence). The CMIP5 and CMIP6 ensembles showed an increase in annual mean precipitation over most land areas by the mid- and late 21st century, although only with a strong model agreement for higher warming levels (Figure Atlas.1 7 and the Interactive Atlas), while CORDEX produces a general decrease in projected precipitation (Figure Atlas.1 7). Based on CORDEX South East Asia multi-model simulations, significant and robust increases of mean rainfall over Indochina and the Philippines were projected while there is a drying tendency over the Maritime Continent during DJF for the early, mid and end of the 21st century periods under both RCP4.5 and RCP8.5 (Figure Atlas.1 9; Tangang et al., 2020). At the end of the 21st century during DJF and under RCP8.5, an increase of 20% in mean rainfall is projected over Myanmar, northern central Thailand and northern Laos, and of 5–10% over the eastern Philippines and northern Vietnam. During JJA, significantly drier conditions are projected over almost the entire South East Asia region except over Myanmar and northern Borneo. Over the Indonesian region, especially Java, Sumatra and Kalimantan, as much as a 20–30% decrease in mean rainfall is projected during JJA by the end of the 21st century. The projected drier condition over Indonesia from CORDEX is consistent with that of Kusunoki (2017), Giorgi et al. (2019), Kang et al. (2019) and Supari et al. (2020) and is associated with enhanced subsidence over the region (Kang et al., 2019; Tangang et al., 2020).

Figure Atlas.19 | The RCM-projected changes in mean precipitation between the early (2011–2040), mid- (2041–2070) and late (2071–2099) 21st century and the historical period 1976–2005. Data are obtained from the CORDEX-SEA downscaling simulations. Diagonal lines indicate areas with low model agreement (less than 80%). Figure adapted from Tangang et al. (2020).

Open figure

It is virtually certain that annual mean temperature has been increasing in South East Asia in the past decades while changes in annual mean precipitation are less spatially coherent though with some increasing trends over parts of Malaysia, Vietnam and the southern Philippines (medium confidence).

Although various biases still exist, there is high confidence that the models can reproduce seasonal climate patterns well over the different sub-regions of South East Asia. There is medium confidence that the RCMs show added value compared to their host GCMs over the region.

Projections show continued warming over South East Asia, but likely by a slightly smaller amount than the global average. Projected changes in rainfall over South East Asia vary, depending on model, sub-region and season (high confidence), with consistent projections of increases in annual mean rainfall from CMIP5 and CMIP6 over most land areas (medium confidence) and decreases in summer rainfall from CORDEX projections over much of Indonesia (medium confidence).

Since AR5, there has been an increasing number of studies on past climate change in South West Asia though meteorological stations are sparsely scattered in the region. They are mainly located in the plains below 2 km of altitude, very scarce in mountainous areas and have declined in number in WCA since the end of the Soviet Union in 1991. This increases the uncertainty in both temperature and precipitation trends, particularly for elevated areas (high confidence) (Christensen et al., 2013; Huang et al., 2014). So researchers use other sources of climate data in the region, particularly freely available gridded data (Annex I).

Globally, drylands showed an enhanced warming over the past century of 1.2°C–1.3°C, significantly higher than the warming over humid lands (0.8°C–1.0°C) (J. Huang et al., 2017 ). A strong increase in annual surface air temperature of 0.27°C–0.47°C per decade has been found over WCA between 1960 and 2013 (very high confidence) (Han and Yang, 2013; Li et al., 2013; Hu et al., 2014, 2017; Huang et al., 2014; Deng and Chen, 2017; Zhang et al., 2017, 2019a; H. Guo et al., 2018 ; Haag et al., 2019; Yu et al., 2019). Warming is most prominent in the spring based on the CRU dataset with rates likely ranging from 0.64°C–0.82°C per decade (Hu et al., 2014). Analysis of seasonal temperature trends based on high-resolution 1 km × 1 km downscaled dataset CHELSA and 20 stations in Uzbekistan has confirmed the maximum significant trend in temperature from 0.6°C up to 1°C per decade in spring from 1979 to 2013 and no significant trend in winter (Khaydarov and Gerlitz, 2019). There is very high confidence (robust evidence, high agreement) that the shrinking of the Aral Sea has induced an increase in surface air temperature around the Aral Sea region in the range of 2°C–6°C (Baidya Roy et al., 2014; McDermid and Winter, 2017; Sharma et al., 2018). The plateau of Iran has experienced significant increases in the average monthly values of daily maximum and minimum temperatures with spatially varying rates of 0.1°C–0.3°C up to 0.3°C–0.4°C per decade and greater spatial variation in minimum temperatures (high confidence) (Mahmoudi et al., 2019; Fathian et al., 2020; Sharafi and Mir Karim, 2020).

Observed warming over northern ARP is higher than over the south, where minimum temperatures are increasing faster than maximum temperatures (Almazroui, 2020a). The rate of mean temperature increase is estimated at 0.10°C per decade over 1901–2010 (Attada et al., 2019), while it has reached 0.63°C (likely in the range of 0.24°C–0.81°C) per decade for the more recent period of 1978–2019 (Almazroui, 2020a).

An overall increasing trend of annual precipitation (0.66 mm per decade) was found over Central Asia based on GPCC v7 data for the period 1901–2013 (Hu et al., 2017), but annual trends were found not significant over the shorter period 1960–2013 (Figure Atlas.11 and Interactive Atlas). Winter precipitation saw a significant increase of 1.1 mm per decade (Song and Bai, 2016). These estimates have low to medium confidencesince the satellite precipitation products have large systematic and random errors in mountainous regions. Moreover CMORPH and TRMM products fail to capture the precipitation events in the ice/snow covered regions in winter and show a substantial false-alarm percentage in summer, but the gauge-corrected GSMAP performs better than other products (Song and Bai, 2016; Guo et al., 2017b; Hu et al., 2017; S. Chen et al., 2019 ). Over the elevated part of eastern WCA precipitation increases in the range of 1.3–4.8 mm per decade during 1960–2013 were observed (very high confidence) (Han and Yang, 2013; Li et al., 2013; Hu et al., 2014, 2017; Huang et al., 2014; Deng and Chen, 2017; Zhang et al., 2017, 2019a; H. Guo et al., 2018 ; Haag et al., 2019; Yu et al., 2019). Reductions in spring precipitation and increases in winter have been reported for Uzbekistan over the period 1979–2013 based on station data but these are not significant (Khaydarov and Gerlitz, 2019). There is very low confidence of the impact of the Aral Sea shrinking on precipitation (Chen et al., 2011; Jin et al., 2017).

A decreasing trend of precipitation is reported for ARP with the mean value of –6.3 mm per decade (range of –30 mm–16 mm) for the period 1978–2019 (low confidence) with large interannual variability over Saudi Arabia, which covers 80% of the region (AlSarmi and Washington, 2011; Almazroui et al., 2012; Donat et al., 2014). The same decreasing trend in precipitation totals and an increasing trend in the number of consecutive dry days are found for most of the Iranian Plateau (medium confidence) (Rahimi and Fatemi, 2019; Fathian et al., 2020; Sharafi and Mir Karim, 2020). January-to-March mean snow cover and depth over mountainous areas decreased between 2000 and 2019 (low to medium confidence due to limited evidence) (Safarianzengir et al., 2020).

There is limited evidence about the performance of GCMs and RCMs in representing the current climate of South West Asia due to very few studies evaluating models over this region, but literature is now emerging particularly on CMIP5/CMIP6 and CORDEX simulations.

Over ARP, surface temperature biases for 18 of 30 CMIP5 models are within one standard deviation of the observed variability (Almazroui et al., 2017). A warm bias in summer and a cold bias for other months along with an underestimation of wet-season precipitation and an overestimation in the dry season have been reported in 26 CMIP5 models (Lelieveld et al., 2016). Thirty CMIP6 GCMs have limited skill in simulating annual precipitation patterns, annual cycle statistics and long-term precipitation trends over Central Asia partially due to considerable wet biases of up to 100% in the southern Xinjiang and Hexi Corridor regions (Guo et al., 2021). Also, CMIP6 models display a wide range of performance in reproducing ENSO teleconnections that influence the region (Barlow et al., 2021).

RCM simulations using the CORDEX-MENA domain reproduce the main features of the mean surface climatology over ARP with moderate biases (high confidence). RegCM4 driven by five GCMs (HadGEM2, GFDL, CNRM, CanESM2 and ECHAM6) showed an ensemble-mean cold bias of about –0.7°C and a dry bias of –13% over ARP (Almazroui, 2016) with a cold (warm) bias over western (south-eastern) areas (Syed et al., 2019). Temperature biases in 30-year historical simulations with WRF using three different radiation parametrizations were within ±2°C and mostly caused by surface long-wave radiation errors which affected nighttime minimum temperatures over 70% of the domain (Zittis and Hadjinicolaou, 2017). Mean absolute errors in COSMO-CLM driven by ERA-Interim were about 1.2°C for temperature, 15 mm per month for precipitation and 9% for total cloud cover, and with new parametrizations of albedo and aerosols optimized for the region the RCM simulated the main climate features of this very complex area (Bucchignani et al., 2016). RegCM4.4 also simulated the main features of the observed climatology (especially for dry regions) with temperature biases within ±3.0°C. Annual precipitation was overestimated with winter and spring underestimated (Ozturk et al., 2018).

Four RCMs (REMO, RegCM4.3.5, ALARO-0, and COSMO-CLM5.0) driven by ERA-Interim, NCEP2 reanalyses and two different GCMs reproduced reasonably well the spatio-temporal patterns for temperature and precipitation though underestimated diurnal temperature range and had cold biases over mountainous and high plateau regions in all seasons. There is low confidence in this result because of low station density and a lack of high-elevation stations, and with biases dependent on the choice of the observational dataset. However, the performance of both GCMs and RCMs is better than reanalyses when compared to available observations (Mannig et al., 2013; Ozturk et al., 2017; Russo et al., 2019; Top et al., 2021).

Temperature and precipitation projections from CMIP5/CMIP6 and CORDEX for different GWLs, SSP and RCP scenarios, time periods and baselines are shown in Figure Atlas.1 7 and further details can be explored in the Interactive Atlas.

In WCA, projections for different GWLs are consistent not only in annual and seasonal warming but in the ranges of the projections. Under RCP8.5, annual mean temperature will likely exceed 2°C by mid-century (compared with 1995–2014) and reach up to 4.8°C–6°C by the end of the century (Yang et al., 2017), with faster warming projected by the CMIP6 ensemble under SSP5-8.5. In individual county-level studies on GCM future climate projections, temperatures increased by up to 7°C by the end of the century, depending on season and emissions scenario (Allaberdiyev, 2010; MENRPG, 2015; MNP, 2015; Gevorgyan et al., 2016; Osborn et al., 2016; Aalto et al., 2017; IDOE, 2017; Salman et al., 2017). Statistical downscaling of 18 CMIP5 GCMs projected an annual temperature increase of 0.37°C per decade (under RCP4.5) with the maximum in northern WCA and warming most conspicuous in summer (Luo et al., 2019). RCM downscaling of GCMs over Central Asia projected a larger increase of temperature under RCP8.5 for the 2071–2100 period, ranging from 5°C to 8°C (Ozturk et al., 2017).

In ARP, the projected change in ensemble mean annual temperature from 30 CMIP6 models is from 1.6°C (SSP1-2.6) to 5.3°C (SSP5-8.5) by 2070–2099 compared to 1981–2010 (Almazroui et al., 2020a). The projected warming is the highest in the north, reaching 5.9°C and lowest in the south (4.7°C). COSMO-CLM projections over the CORDEX-MENA domain show for ARP and WCA a strong warming with marked seasonality for the end of the 21st century, ranging from 2.5°C in winter under RCP4.5 to 8°C in summer under RCP8.5 and with large increases found over high-altitude areas in winter and spring (Bucchignani et al., 2018; Ozturk et al., 2018). The CMIP5 multi-model mean warming in boreal summer in 2070–2099, compared with 1951–1980, is projected to be about 2.5°C and 6.5°C at the 2°C and 4°C global warming levels respectively (Huang et al., 2014).

Future projections of precipitation in South West Asia have large uncertainties and thus low confidence. There are few significant changes, little consensus on the sign and with a tendency for reduction in CMIP5 being reversed in CMIP6 across all warming levels (Ozturk et al., 2018). Statistical downscaling of 18 CMIP5 GCMs under RCP4.5 projected an increase in precipitation of 4.6 mm per decade in South West Asia during 2021–2060 relative to 1965–2004 (Luo et al., 2019). CMIP5 simulations project a general decrease in precipitation over lowlands in Turkey, Iran, Afghanistan and Pakistan (Ozturk et al., 2017), and an increase over high-mountain regions (Aalto et al., 2017; Salman et al., 2018). At a 4°C global warming level, the multi-model mean annual precipitation for Turkmenistan and parts of Tajikistan and Uzbekistan is projected to decrease by 20%, with somewhat stronger relative decreases in summer (Reyer et al., 2017). Over northern WCA, the CMIP5 ensemble mean projects increases of over 3 mm per decade under RCP2.6 and over 6 mm per decade under RCP4.5 and RCP8.5 over the 21st century (Huang et al., 2014). Mean annual precipitation is projected to rise by 5.2% at the end of the 21st century (2070–2099) under RCP8.5, compared to 1976–2005, while mean annual snowfall is projected to decrease by 26.5% in Central Asia (Yang et al., 2017). However, regardless of the sign of the precipitation change in the high-mountain regions of Central Asia, the influence of the warming on the snowpack will very likely cause important changes in the timing and amount of the spring melt (Diffenbaugh et al., 2013).

In ARP, the projected change in ensemble mean annual precipitation from 30 CMIP6 models ranges from 3.8% (–2.6 to 28.8%) to 31.8% (12.0–106.5%) under SSP1-2.6 and SSP5-8.5 emissions for the period 2080–2100 compared with 1995–2014 (Almazroui et al., 2020a). North-west ARP precipitation is projected to decrease between –6 to –27% per decade and in the south precipitation to increase by up to 8.6% per decade. CMIP6 projections are in line with those from CMIP3 and CMIP5, however they are less variable in the central area in CMIP6. The uncertainty associated with precipitation over ARP is large because of very low annual amounts and high variability.

Increases in annual surface air temperature over South West Asia are very likely in the range of 0.24°C–0.81°C per decade over the last 50–60 years. Annual precipitation change over ARP since 1970 is estimated at –6.3 mm per decade (and in the range of –30 to 16 mm per decade) and over WCA is generally not significant except over the elevated part of eastern WCA where increases between 1.3 mm and 4.8 mm per decade during 1960–2013 have been observed (very high confidence). In mountainous areas, the scarcity and decline of the number of observation sites since the end of the former Soviet Union in 1991 increase the uncertainty of the long-term temperature and precipitation estimates (high confidence).

Mean temperature biases in RCMs are within ±3°C in South West Asia, and annual precipitation biases are positive in almost all parts of the region, except over the ARP where they are negative in the wet season (November to April) and over WCA in winter and spring (from December to May) (medium confidence). Since regional model evaluation literature has only recently emerged there is medium evidence about the performance of RCMs in South West Asia though with medium to high agreement on mean temperature and precipitation biases. RCMs simulate colder temperatures than observed over mountainous and high plateau regions (limited evidence, high agreement).

Further warming over South West Asia is projected in the 21st century to be greater than the global average, with rates varying from 0.25°C to 0.8°C per decade depending on the season and scenario, and the maximum rates found in the northern part of the region in summer (high confidence). The influence of the warming on the snowpack will very likely cause changes in the timing and amount of the spring melt. CMIP6 projected changes in annual precipitation totals are in the range of –3 to 29% (SSP1-2.6) and 12–107% (SSP5-8.5) in ARP (medium confidence). Strong spatio-temporal differences with overall precipitation decreases are projected in the central and northern parts of WCA in summer (JJA) with increases in winter (DJF) (medium confidence).

Australasia is divided into five regions for the Atlas (Figure Atlas.21), as follows: New Zealand (NZ), with a varied climate with diverse landscapes, mainly maritime temperate with four distinct seasons; Northern Australia (NAU), which is mainly tropical with monsoonal summer-dominated rainfall (monsoon season December to March, see Annex V), but with a hot, semi-arid climate in the south of the region; Central Australia (CAU) with a predominantly hot, dry desert climate; Eastern Australia (EAU) with a temperate oceanic climate at the coast to semi-arid inland; and Southern Australia (SAU), which ranges from Mediterranean and semi-arid in the west to mainly cool temperate maritime climate in the south-east. Various remote drivers have notable teleconnections to regions within Australasia, including an effect of the El Niño–Southern Oscillation (ENSO) and the Indian Ocean Dipole (Table Atlas.1 and Annex IV). Much of southern NZ and SAU are affected by systems within the westerly mid-latitude circulation, in turn affected by the Southern Annular Mode (SAM). The monsoon and the Madden–Julian Oscillation (MJO) affect rainfall variability in northern Australia.

Figure Atlas.21 | Regional changes over land in annual mean surface air temperature and precipitation relative to the 1995–2014 baseline for the reference regions in Australasia (warming since the 1850–1900 pre-industrial baseline is also provided as an offset). Bar plots in theleft panel of each region triplet show the median (dots) and 10th–90th percentile range (bars) across each model ensemble for annual mean temperature changes for four datasets (CMIP5 in intermediate colours; a subset of CMIP5 used to drive CORDEX in light colours; CORDEX overlying the CMIP5 subset with dashed bars; and CMIP6 in solid colours); the first six groups of bars represent the regional warming over two time periods (near-term 2021–2040 and long-term 2081–2100) for three scenarios (SSP1-2.6/RCP2.6, SSP2-4.5/RCP4.5 and SSP5-8.5/RCP8.5), and the remaining bars correspond to four global warming levels (GWLs: 1.5°C, 2°C, 3°C and 4°C). The scatter diagrams of temperature against precipitation changes display the median (dots) and 10th–90th percentile ranges for the above four warming levels for December–January–February (DJF; middle panel) and June–July–August (JJA; right panel), respectively; for the CMIP5 subset only the percentile range of temperature is shown, and only for 3°C and 4°C GWLs. Changes are absolute for temperature (in °C) and relative (as %) for precipitation. See Atlas.1.3 for more details on reference regions (Iturbide et al., 2020) and Atlas.1.4 for details on model data selection and processing. The script used to generate this figure is available online (Iturbide et al., 2021) and similar results can be generated in the Interactive Atlas for flexibly defined seasonal periods. Further details on data sources and processing are available in the chapter data table (Table Atlas.SM.15).

Open figure

The AR5 WGI and WGII reports (IPCC, 2013c; Stocker et al., 2013; Reisinger et al., 2014) give very high confidence that air and sea temperatures in the region have warmed; cool extremes have become rarer in Australia and New Zealand since 1950, while hot extremes have become more frequent and intense (e.g., it is very likely that the number of warm days and nights have increased). The AR5 reported that it is virtually certain that mean air and sea temperatures will continue to increase, with very high confidence that the greatest increase will be experienced by inland Australia and the smallest increase by coastal areas and New Zealand. The AR5 reported a range of different precipitation trends within the region. For example, while annual rainfall has been significantly increasing in north-western Australia since the 1950s (very high confidence), it has been decreasing in the north-east of the South Island of New Zealand over 1950–2004 (very high confidence) and over the south-west of the state of Western Australia. In line with these trends, AR5 reported it is likely that drought has decreased in north-west Australia. Future projections for precipitation extremes indicate an increase in most of Australia and New Zealand, in terms of rare daily rainfall extremes (i.e., current 20-year return period events) and of short duration (sub-daily) extremes (medium confidence). Likewise, however, there is a projected increase in the frequency of drought in southern Australia (medium confidence) and in many parts of New Zealand (medium confidence). Owing to hotter and drier conditions there is high confidence that the occurrence of fire weather will increase in most of southern Australia, and medium confidence that the fire danger index will increase in many parts of New Zealand.

The AR5 reported mean sea levels have also increased in Australia and New Zealand at average rates of relative sea level rise of 1.4 ± 0.6 mm yr^–1from 1900 to 2011, and 1.7 ± 0.1 mm yr^–1from 1900 to 2009, respectively (very high confidence). The assessment found that the volume of ice in New Zealand has declined by 36–61% from the mid- to late 1800s to the late 1900s (high confidence), while late-season significant snow depth has also declined in three out of four Snowy Mountain sites in Australia between 1957 and 2002 (high confidence). As mean sea level rise is projected to continue for at least several more centuries, there is very high confidence that this will lead to large increases in the frequency of extreme sea level events in Australia and New Zealand. On the other hand, the volume of winter snow and the number of days with low-elevation snow cover in New Zealand are projected to decrease in the future (very high confidence), while both snow depth and area are projected to decline in Australia (very high confidence).

The SROCC (Hock et al., 2019b) reports on the observed and projected decline in snow cover in Australasia, as well as the retreat of New Zealand glaciers following an advance in 1983–2008 due to enhanced snowfall. It also reports on the vulnerability of some Australian communities and ecosystems to sea level rise, increases in the intensity and duration of marine heatwaves driven by human influence (high confidence), the decrease in frequency of tropical cyclones’ landfall on eastern Australia since the late 1800s (low confidence in an anthropogenic signal), and presents a case study on the multiple hazards, compound risk and cascading impacts from climate extremes in Tasmania in 2015–2016 (including an attributable human influence on some events). The SRCCL (Mirzabaev et al., 2019) found widespread vegetation ‘greening’ has occurred in parts of Australia, and an increase in the desertification and drought risk in future in southern Australia.

Significant warming trends between 0.2°C and 0.3°C per decade have been observed in the three reference regions of Central America in the last 30 years (Planos Gutiérrez et al., 2012; P.D. Jones et al., 2016 a; Hidalgo et al., 2017), with the largest increases in the North American Monsoon region (high confidence) (Figure Atlas.11 and the Interactive Atlas; Cavazos et al., 2020). There is high confidence of increasing temperature over parts of NCA, reaching 0.5°C per decade in Mexico and southern Baja California, with a lower rate (0.2°C per decade) in the Yucatan Peninsula and the Guatemala Pacific coastal region (Cueto et al., 2010; García Cueto et al., 2013; Martínez-Austria et al., 2016; Martínez-Austria and Bandala, 2017; Navarro-Estupiñan et al., 2018; Cavazos et al., 2020) and CAR (McLean et al., 2015) over the last 30 to 40 years. Cooling trends have been detected in limited areas of Honduras and northern Panama (Hidalgo et al., 2017).

Changes in mean precipitation rates are less consistent and long-term trends are generally weak. Different databases show significant differences depending mainly on the type and resolution of data (Centella-Artola et al., 2020). Small positive trends were observed in the total annual precipitation (Stephenson et al., 2014). In SCA and CAR, trends in annual precipitation are generally non-significant, with the exception of small significant positive trends for sub-regions or limited periods (Planos Gutiérrez et al., 2012; Hidalgo et al., 2017), and the 1970–1999 trends in precipitation in SCA are generally non-significant (J.M. Jones et al., 2016; Hidalgo et al., 2017). Positive trends in the duration of the MSD have been found in this region over the past four decades (low confidence) (Anderson et al., 2019). For CAR see also Atlas.10 Small Islands.

The ability of climate models to simulate the climate in this region has improved in many key aspects (Karmalkar et al., 2013; Fuentes-Franco et al., 2014, Fuentes-Franco et al., 2015, Fuentes-Franco et al., 2017; Vichot-Llano et al., 2014; Vichot-Llano and Martínez-Castro, 2017; Martínez-Castro et al., 2018). Particularly relevant for this region are increased model resolution and a better representation of the land surface processes (high confidence).

Regional climate models (RCMs) forced with reanalyses and atmosphere-only global climate models provide simulations with a reasonably good performance over the core North American Monsoon region, mostly in NCA (high confidence) (Bukovsky et al., 2013; Cerezo-Mota et al., 2016). RCMs also reproduce the seasonal spatial patterns of temperature and the bimodal rainfall characteristics of the NCA, SCA and CAR (high confidence) (Karmalkar et al., 2013; Centella-Artola et al., 2015; Martínez-Castro et al., 2018; Cavazos et al., 2020; Vichot-Llano et al., 2021b), though in some sub-regions specific models overestimate and shift the month of the maxima. RCM simulations in the region do not necessarily improve with the size of the domain, as important features of the regional circulation and key rainfall climate features, such as the CLLJ and MSD, are well represented for a variety of domains of different sizes (high confidence) (Centella-Artola et al., 2015; Martínez-Castro et al., 2018; Cabos et al., 2019; Cavazos et al., 2020; Vichot-Llano et al., 2021b).

Figure Atlas.22 and the Interactive Atlas synthesize regional mean changes in annual mean surface air temperature and precipitation for the Central American reference regions for CMIP6, CMIP5 and CORDEX for different warming levels and time periods. At the 1.5°C GWL, it is very likely that average annual temperature in Central America over land surpasses 1.3°C (CAR), 1.7°C (NCA) and 1.6°C (SCA). For the 3°C GWL, the corresponding projected ensemble mean regional warming values are 2.7°C (CAR), 3.5°C (NCA) and 3.1°C (SCA). CAR average annual warming is below the level of global warming, while the two continental reference regions are close to the global warming level with CMIP6 and CMIP5 showing very consistent results (Figure Atlas.22). However, when focusing on time slices instead of warming levels, the CMIP6 projections show systematically higher median values than CMIP5. CORDEX results are also consistent with the previous findings, though the subset of driving models spans a smaller range of uncertainty, particularly over CAR. Results have also been reported for this region based on CMIP5, CMIP6 and downscaled simulations over the CORDEX CAM domain or similar smaller domains (Taylor et al., 2013b; Nakaegawa et al., 2014; Imbach et al., 2018; Vichot-Llano et al., 2019; Almazroui et al., 2021). Statistical downscaling methods have been also applied to CMIP5 projections to obtain bias-adjusted regional projections (Colorado-Ruiz et al., 2018; Taylor et al., 2018; Vichot-Llano et al., 2019).

Global and regional models consistently project warming in the whole region for the end of the century, under RCP4.5 and RCP8.5 for CMIP5 projections with greater warming for continental compared to insular territories, likely reaching values between 2°C and 4°C (high confidence) (Campbell et al., 2011; Karmalkar et al., 2011; Cavazos and Arriaga-Ramírez, 2012; Cantet et al., 2014; Chou et al., 2014; Coppola et al., 2014; Hidalgo et al., 2017; Colorado-Ruiz et al., 2018; Imbach et al., 2018). The greatest warming of 5.8°C for the end of the century was projected for northern Mexico under RCP8.5 (Colorado-Ruiz et al., 2018), using an ensemble of CMIP5 GCMs (Interactive Atlas).

Regarding precipitation, it is likely that the annual average precipitation changes for the 1.5°C GWL will be in the ranges of –11 to 0% in CAR, from –12 to 0% in SCA, and from –10 to +3% in NCA (Interactive Atlas). For the 3°C GWL, the corresponding annual average precipitation changes will be from –17 to –2% in CAR, from –16 to +2% in NCA, and from –23 to 0% in SCA. A clear drying tendency is observed for the 3°C GWL relative to the 1.5°C GWL. Maloney et al. (2014) examined 21st-century climate projections of North American climate in CMIP5 models under RCP8.5, including Central America and the Caribbean. Summer drying was projected in CAR and SCA for most of the models, with good agreement. The strongest drying is projected to occur during July and August which are the months when the MSD occurs in many sub-regions (Figure Atlas.22 and the Interactive Atlas). Intensification of the MSD in SCA was also projected by using the Rossby Centre Regional Climate Model (RCA4; Corrales-Suastegui et al., 2020), but with a future decrease in area and frequency (Cross-Chapter Box Atlas.2). They also found a projected intensification of CLLJ and drying for the future time slice of 2071–2095, relative to their baseline of 1981–2005. Decreased precipitation was also projected for SCA (Imbach et al., 2018) with the 8-km resolution Eta RCM during the rainy season, including an intensification of the MSD, although no significant change was projected for the CLLJ.

Colorado-Ruiz et al. (2018) assessed an ensemble of 14 GCMs from CMIP5 for a 1971–2000 baseline period, projecting precipitation decreases of between 5% and 10% by the end of the century for the RCP4.5 and RCP8.5 scenarios respectively. The greatest decrease in precipitation is projected during summer reaching 13%, especially in southern Mexico, Central America and the Caribbean. Dynamically downscaled simulations (Bukovsky et al., 2015) also projected a decrease of precipitation for the middle of the century (2041–2069) relative to 1971–1999 for the north of Mexico, though despite good agreement amongst the models, these results must be considered of low confidence, because of their poor simulation of important monsoon physical processes. Vichot-Llano et al. (2021a) used a multi-parameter ensemble of RegCM4, driven by the CMIP5 global model HagGEM2-ES projections to conclude that, relative to the 1975–2004 baseline, in the near (2020–2049) and more prominently in the far (2070–2099) future, drier conditions will prevail at over the eastern Caribbean. The projected future warming trend was statistically significant at the 95% confidence level over CAR and SCA. Almazroui et al. (2021) used an ensemble of 31 CMIP6 models to estimate climate change signals of temperature and precipitation in six reference regions in North and Central America and the Caribbean, finding a decrease in precipitation (10–30%) over Central America and the Caribbean under three scenarios with regional and seasonal variations.

There is high agreement and high confidence in the projected decrease of precipitation by the end of the century for most of the region, particularly for annual and summer precipitation, but there is low confidence on the magnitude of this decrease which varies between 5% and 50% for different projections and different sub-regions (see extended information in the Interactive Atlas).

The status of climateextreme trends and projections for the region has been assessed in Chapter 11 and the main findings are synthesized here. There is high confidence in the projections of significant heatwave events at the end of the century in SCA (Angeles-Malaspina et al., 2018) and an increase in warm days and warm nights over this region and CAR (Stennett-Brown et al., 2017). For CAR islands, using dynamically downscaled CMIP3 models, Karmalkar et al. (2013) projected an increase in drought severity at the end of the century, mainly due to a precipitation decrease during the early wet season. In SCA, projections suggest an increase in the MSD (Imbach et al., 2018) and an increase in consecutive dry days (Chou et al., 2014), consistent with the projections of Stennett-Brown et al. (2017).

Significant warming trends between 0.2°C and 0.3°C per decade have been observed in the three reference regions of Central America in the last 30 years, with the largest increases in the North American Monsoon region (high confidence). Changes in mean precipitation rates are less consistent and long-term trends are generally weak. Small positive trends were observed in the total annual precipitation in part of the region.

Warming in the continental part of the region is projected to increase in the range of the mean global values for GWL of 1.5°C and 3°C, but in the Caribbean regional warming will be lower. Precipitation is projected to decrease with increasing GWLs, especially for CAR and SCA.

Projected change in mean annual precipitation shows a large spatial variability across Central America and the Caribbean. Under moderate future emissions overall negative but non-significant precipitation trends are projected for the 21st century (low confidence). Under higher-emissions scenarios and at higher GWLs, average precipitation is likely to decrease in most of the region, particularly in the north-western and central Caribbean and part of continental Central America, especially in SCA.

Studies on climatic trends in South America indicate that mean temperature and extremely warm maximum and minimum temperatures have shown an increasing trend (high confidence), particularly for a large region in Northern South America and the south-western Andes (NSA, SAM, NES, SWS and the north of SES; Skansi et al., 2013; de Barros Soares et al., 2017). Also, the trend of the difference between the annual mean of the daily maximum temperature and the annual mean of the daily minimum temperature was positive – up to 1°C per decade – over the extratropics with the maximum temperature generally increasing faster than the minimum temperature, while a negative trend of up to –0.5°C per decade was observed over the tropics.

Regionally, analyses of temperatures point to an increased warming trend (high confidence) over Amazonia over the last 40 years, which reached approximately 0.6°C–0.7°C (Figure Atlas.11 and the Interactive Atlas) and with stronger warming during the dry season and over the south-east. The analyses also showed that 2016 was the warmest year since at least 1950 (Marengo et al., 2018b). Andean temperatures showed significant warming trends, especially at inland and higher-elevation sites, while trends are non-significant or negative at coastal sites (high confidence) (Vuille et al., 2015; Burger et al., 2018; Vicente-Serrano et al., 2018; Pabón-Caicedo et al., 2020). Over central Chile, positive trends are largely restricted to austral spring, summer and autumn seasons for mean, maximum and minimum temperatures (Burger et al., 2018; Vicente-Serrano et al., 2018). Over Peru, trends of maximum air temperature were mainly amplified during the austral summer, but trends of cold-season minimum air temperature showed an opposite pattern, with the strongest warming being recorded in the austral winter (Vicente-Serrano et al., 2018).

In general, the spatial patterns of observed trends in temperature are more consistent than for precipitation across the whole of South America (medium confidence) (Interactive Atlas; de Barros Soares et al., 2017). In south-east Brazil there is a region of highly significant decrease of rainfall in both wet and dry seasons recorded in the period 1979–2011 (Interactive Atlas; Rao et al., 2016). The most consistent evidence of positive rainfall trend occurs in the southern part of the La Plata basin (high confidence) (southern Brazil, Uruguay, and north-eastern Argentina; de Barros Soares et al., 2017). By contrast, there is high confidence that annual rainfall has decreased over north-east Brazil during the last decades (Carvalho et al., 2020). Contrary to temperature changes, trends in annual precipitation exhibit different signs across sectors in the Andes. For instance, annual precipitation trends in the north tropical (north of 8°S) and south tropical (8°S–27°S) Andes do not show a homogeneous pattern. Over the subtropical Andes, central Chile shows a robust signal of declining precipitation since 1970 (high confidence) (Pabón-Caicedo et al., 2020).

Observational studies show that the dry-season length over southern Amazonia has increased significantly since 1979 (high confidence) (Fu et al., 2013; Alves, 2016). In the Peruvian Amazon-Andes basin, there is no trend in mean rainfall during the period 1965–2007 (Lavado Casimiro et al., 2012) though statistically significant decreases in total annual rainfall in the central and southern Peruvian Andes from 1966 to 2010 were found (Heidinger et al., 2018). Despite that, recent analyses of Amazon hydrological and precipitation data suggest an intensification of the hydrological cycle over the past few decades (Gloor et al., 2015). In general, these changes are attributed mainly to decadal climate fluctuations (high confidence), ENSO, the Atlantic SST north–south gradient, feedbacks between fire and land-use change mainly across southern south-eastern Amazon, and changes in the frequency of organized deep convection (Fernandes et al., 2015; Sánchez et al., 2015; Tan et al., 2015).

Since AR5, there has been limited attribution literature in the South America. Recent publications based on observational and modelling evidence assessed that anthropogenic forcing in CMIP5 models explains the overall warming (high confidence) over the entire South American continent, including the increase in the frequency of extreme temperature events (Hannart et al., 2015). It has a detectable influence in explaining positive and negative precipitation trends observed in regions such as SES and the southern Andes (Vera and Díaz, 2015; de Barros Soares et al., 2017; Boisier et al., 2018; de Abreu et al., 2019). Despite that, there is limited evidence that human-induced greenhouse gas emissions had an influence on the 2014/2015 water shortage in south-east Brazil (Otto et al., 2015). Extreme event attribution on sub-continental scales is assessed in Chapter 11 and continental-scale attribution in Chapter 3.

In summary, analyses of historical temperature time series point strongly to an increased warming trend (high confidence) across many South American regions, except for a cooling off the Chilean coast. Annual rainfall has increased over South-Eastern South America and decreased in most tropical land regions, particularly in central Chile (high confidence). The number and strength of extreme events, such as extreme temperatures, droughts and floods, have already increased (medium confidence) (Table 11.7).

It is noted that the major barrier to the study of climate change in many regions of South America is still the absence or insufficiency of long time series of observational data (Carvalho, 2020; Condom et al., 2020). Most national datasets were created in the 1970s and 1980s, preventing a more comprehensive long-term trend analysis. To fulfil the users’ demand for climatological and meteorological data products covering the whole region, several interpolation techniques have been used with reanalysis and gridded gauge-analysis products to add the necessary spatial detail to the climate analyses over land and for climate variability and trend studies, but these are subject to uncertainties (Skansi et al., 2013; Rozante et al., 2020).

Since AR5 the number of publications on climate model performance and their projections in South America has increased, particularly for regional climate modelling studies (Giorgi et al., 2009; Boulanger et al., 2016; Ambrizzi et al., 2019) and the understanding of their strengths and weaknesses (high confidence).

Most global and regional climate models can simulate reasonably well the current climatological features of South America, such as seasonal mean and annual cycles. However, significant biases persist mainly at regional scales (high confidence) (Blázquez and Nuñez, 2013b; Gulizia et al., 2013; Joetzjer et al., 2013; Jones and Carvalho, 2013; Torres and Marengo, 2013; Gulizia and Camilloni, 2015; Zazulie et al., 2017; Abadi et al., 2018; Barros and Doyle, 2018; Solman and Blázquez, 2019; Fan et al., 2020; Rivera and Arnould, 2020; Teichmann et al., 2021). During the dry season, precipitation is underestimated in most models over Amazonia (medium evidence, high agreement) (Torres and Marengo, 2013; Yin et al., 2013; Solman and Blázquez, 2019). Over regions with complex orography, such as the tropical Andes of NWS, CMIP5 models tend to underestimate precipitation which is associated with the misrepresentation of the Pacific ITCZ and local low-level jets (Sierra et al., 2015, 2018), whereas over the subtropical central Andes in SWS, the models are found to overestimate both mean temperature and precipitation values (limited evidence, high agreement) (Zazulie et al., 2017; Rivera and Arnould, 2020; Díaz et al., 2021). Most models show a dry bias over SES (Díaz and Vera, 2017; Barros and Doyle, 2018; Solman and Blázquez, 2019; Díaz et al., 2021) associated with an underestimation of the northern flow that brings water vapour into the region (medium confidence) (Gulizia et al., 2013; Zazulie et al., 2017; Barros and Doyle, 2018). The biases in seasonal precipitation, annual precipitation and climate extremes over several regions of South America were reduced, including the Amazon, central South America, Bolivia, eastern Argentina and Uruguay, in the CMIP5 models when compared to those of CMIP3 (medium confidence) (Joetzjer et al., 2013; Gulizia and Camilloni, 2015; Díaz and Vera, 2017). The evidence is still insufficient to determine whether CMIP6 biases are reduced when compared with CMIP5 simulations regarding precipitation and its variability in South America. The temperature and precipitation patterns of anomalies associated with ENSO in tropical South America (NWS, NSA and NES) are better captured by GCMs in tropical South America (NWS, NSA and NES) than in extratropical South America (SES), particularly during austral summer and autumn (limited evidence, high agreement) (Tedeschi and Collins, 2016; Perry et al., 2020).

Based on regional simulations, studies showed that some RCMs improve the quality of the simulated climate when compared with the driving GCM (medium evidence, high agreement) (Llopart et al., 2014; Sánchez et al., 2015; Falco et al., 2019; Solman and Blázquez, 2019; Ciarlo` et al., 2021; Teichmann et al., 2021). Regional climate model (RCM) simulations over South America can reproduce the main features of temperature and precipitation in terms of both spatial distributions (Solman et al., 2013; Falco et al., 2019) and seasonal cycles over the different climate regimes, including the main SAmerM features (high confidence) (Jacob et al., 2012; Solman, 2013; Llopart et al., 2014; Reboita et al., 2014; de Jesus et al., 2016; Lyra et al., 2018; Bozkurt et al., 2019; Ashfaq et al., 2021). However, RCMs showed systematic biases such as precipitation overestimations and temperature underestimations along the Andes throughout the year (high confidence), although these biases may be artificially amplified by the lack of a dense observational station network (Jacob et al., 2012; Solman et al., 2013; Bozkurt et al., 2019; Falco et al., 2019). RCMs tended to show dry biases over the Amazon and the northern part of the continent (SAM, NSA) during DJF and during the maximum precipitation associated with the ITCZ over NSA during JJA (medium evidence, high agreement) (Solman et al., 2013; Falco et al., 2019). Temperature overestimation and precipitation underestimation over La Plata basin (in SES) are also RCM common biases, with the warm bias amplified for austral summer and the dry bias amplified for the rainy season (high confidence) (Solman et al., 2013; Reboita et al., 2014; Solman, 2016; Falco et al., 2019). Despite their relevance, RCM simulations at very high resolution (less than 10 km) are still few in South America (high confidence) and are mainly designed for specific regions or purposes (Lyra et al., 2018; Bozkurt et al., 2019; Bettolli et al., 2021).

The evaluation of statistical downscaling models (ESD) in representing regional climate features in South America has increased since AR5, however there are still few ESD studies over the different sub-regions. Precipitation simulations based on ESD models are able to reproduce mean precipitation over tropical and subtropical South American regions, especially over maximum precipitation areas in western Colombia, south-eastern Peru, central Bolivia, Chile and the La Plata basin (medium confidence) (Souvignet et al., 2010; Mendes et al., 2014; Palomino-Lemus et al., 2015, 2017, 2018; Soares dos Santos et al., 2016; Troin et al., 2016; Borges et al., 2017; Bettolli and Penalba, 2018; Araya-Osses et al., 2020; Bettolli et al., 2021). Temperature simulations are fewer but show added value to GCM simulations (medium evidence, high agreement) (Souvignet et al., 2010; Borges et al., 2017; Bettolli and Penalba, 2018; Araya-Osses et al., 2020).

Overall, climate modelling has made some progress in the past decade but there is no model that performs well in simulating all aspects of the present climate over South America (high confidence). The performance of the models varies according to the region, time scale and variables analysed (Abadi et al., 2018). There is also a fairly narrow spread in the representation of temperature and precipitation over South America by the CMIP5 GCMs and also the RCMs, with biases that can be associated with the parametrizations and schemes of surface, boundary layer, microphysics and radiation used by the models. Finally, observational reference datasets, such as reanalysis products, used in the calibration and validation of climate models can also be quite uncertain and may explain part of the apparent biases present in climate models (high confidence).

It is very likely that annual mean temperature will increase over South America, with a wide range of projected changes of 1.0°C–6.0°C by the end of the 21st century (from RCP2.6/SSP1-2.6 to RCP8.5/SSP5-8.5 emissions, Figure Atlas.22). Overall, GCMs project higher temperature change than RCMs in austral summer and winter over all sub-regions and in winter mainly over the central part of the continent (Interactive Atlas; Coppola et al., 2014; Llopart et al., 2021; Teichmann et al., 2021). The largest warmings over the South American continent are projected for the Amazon basin (SAM and NSA) and the central Andes range (southern SAM, northern SWS and south-eastern NWS; Figure Atlas.22), especially during the dry and dry-to-wet transition seasons (austral winter and spring) (high confidence) (Blázquez and Nuñez, 2013a; Coppola et al., 2014; Pabón-Caicedo et al., 2020; Teichmann et al., 2021).

Using warming levels (Figure Atlas.22), the temperature is projected to increase at or above the level of global warming in all regions apart from SSA with additional warming (compared to a 1995–2014 baseline) of over 4°C for the 4°C warming level in NSA and SAM. Changes for other warming levels, sub-regions and emissions pathways are shown in Figure Atlas.22 and can be explored with the Interactive Atlas.

In general, models show a wide regional range in the direction and the magnitude of mean precipitation change in many South American regions, with large significant increases and decreases (Figure Atlas.22 and the Interactive Atlas). In the medium and long term, under the high-emissions scenario, the CMIP5 multi-model ensemble projected an increase in precipitation (generally greater than 10%) in SES and NWS, and a decrease (less than 10%) in NSA across seasons (high confidence, robust evidence) (Solman, 2013; Chou et al., 2014; Coppola et al., 2014; Llopart et al., 2014, 2021; Reboita et al., 2014, 2021; Sánchez et al., 2015; Menéndez et al., 2016; Ruscica et al., 2016; Bozkurt et al., 2018a; Zaninelli et al., 2019). Also, in parts of SWS, annual precipitation is projected to decrease (up to 30%) by the late 21st century (Souvignet et al., 2010; Palomino-Lemus et al., 2017, 2018; Bozkurt et al., 2018a). Under high RCPs, the CMIP5 ensemble projects that all Brazilian regions will experience more rainfall variability in the future, so drier dry periods and wetter wet periods on daily, weekly, monthly and seasonal time scales, despite the future changes in mean rainfall being currently uncertain (medium confidence) (Alves et al., 2021). Regarding the SAmerM, it is very likely that the monsoon will experience changes in its life cycle by the end of the 21st century for both RCP4.5 and RCP8.5 emissions and, in particular, delayed onset. However there is low agreement on the projected changes in terms of extreme and total precipitation of the monsoon season in South America (Llopart et al., 2014; Ashfaq et al., 2021). Changes in the SAmerM are assessed in Section 8.3.2.4.5.

Projected changes in seasonal precipitation and their uncertainties generally agree with the annual changes, particularly for the decreases in SWS (Figure Atlas.22). DJF precipitation changes in NSA and SAM are largely uncertain, with weak agreements in the projections, particularly for CMIP5 and CMIP6 ensembles, which project almost no change, and decreasing precipitation for NSA and a narrow range from slight increases to no change respectively for SAM.

In summary, it is virtually certain that the climate of South America has warmed. Studies on climate trends in South America indicate that mean temperature and maximum and minimum temperatures have increased over the last 40 years. Long-term observed precipitation trends show an increase over South-Eastern South America and decreases in most tropical land regions (high confidence).

Evaluation of global and regional climate model simulations have increased over South America in the past decade and shown improved performance. However, the results reveal that no model performs well in simulating all aspects of the present climate (very likely) . On the other hand, there is still a lack of high-quality and high-resolution observational data that may explain part of the important biases present in climate models (high confidence).

Climate model projections show a general increase in annual mean surface temperature over the coming century for all emissions scenarios (RCPs and SSPs) (high confidence), consistent with the observed warming, and with all regions except SSA warming faster than the global average. Unlike temperature, annual precipitation has patterns of decrease in North-Eastern South America (NES) and South-Western South America (SWS), and increase in Southern South America (SES) and North-Western South America (NWS) (high confidence), with small changes projected under a low-emissions scenario. However, there is low confidence in the magnitude because of the large spread among models, both GCMs and RCMs.

Westerly winds and the accompanying Atlantic storm track with cyclones and anticyclones travelling from the Atlantic towards inland Europe are the main climatic features that characterize daily to interannual variability in the European region. The Siberian High in winter determines cold weather in Eastern Europe and can affect other regions with cold outbreaks. Intra-seasonal and interannual variations are driven by modes of climate variability such as the North Atlantic Oscillation (NAO; Table Atlas.1 and Annex IV.2). Global warming can lead to systematic changes in regional climate variability via thermodynamic responses such as altered lapse rates (Kröner et al., 2017; Brogli et al., 2019) and land-atmosphere feedbacks (Zampieri and Lionello, 2011; Boé and Terray, 2014). Regional feedbacks involving the land-sea contrast, sea surface, land surface, clouds, aerosols, radiation and other processes modulate the regional response to enhanced warming.

Four climatic regions are defined for Europe (Figure Atlas.24). The Mediterranean region (MED) in the south is characterized by mild winters and hot and dry summers (Mediterranean climate; Section 10.6.4.2). It covers both Europe and Africa, and MED assessments in this section generally imply the entire MED domain unless stated otherwise. The Western and Central Europe region (WCE) has distinct summer and winter seasons with increasing continentality of climate eastwards. The Northern Europe region (NEU), close to the Atlantic Ocean, is characterized by high humidity and relatively mild winters, and strong exposure to the Atlantic storm track. Eastern Europe (EEU) covers the western part of Russia and neighbouring territories and has continental characteristics. Many regional datasets and model projections assessed here do not sufficiently cover the EEU region.

The AR5 WGII (Kovats et al., 2014) reports with high confidence that observed climate trends show regionally varying changes in temperature and rainfall in Europe. The average temperature in Europe has continued to increase, with seasonally different rates of warming being greatest in high latitudes in Northern Europe. Annual precipitation has increased in Northern Europe and decreased in parts of Southern Europe. The SROCC (Hock et al., 2019b) reports with high confidence that a reduction in snow cover at low elevation and glacier extent is observed in recent decades, with consequent changes in annual and seasonal runoff patterns. According to the SRCCL report (IPCC, 2019b) there is high agreement that observed vegetation greening and forestation in the last 30 years cools summer surface temperature and warms winter temperature due to decreased snow cover and increased snow shading in forested areas. It is very likely that aerosol column amounts have declined over Europe since the mid-1980s.

The AR5 (Collins et al., 2013) reports that the ability of models to simulate the climate in Europe has improved in many important aspects. Particularly relevant for this region are increased model resolution and a better representation of the land surface processes in many of the models that participated in CMIP5. The spread in climate model projections is still substantial, partly due to pronounced internal variability in this region (particularly NAO and AMO). In the winter half year, NEU and WCE are likely to have increased mean precipitation associated with increased atmospheric moisture and moisture convergence, and intensification in extratropical cyclone activity. No change or a moderate reduction is projected for MED. In the summer half year, it is likely that NEU and WCE mean precipitation will have only small changes with a notable reduction in MED. According to SR1.5 (Hoegh-Guldberg et al., 2018), these precipitation changes are more pronounced at 2°C than at 1.5°C of global warming. For a 2°C global warming level, an increase in runoff is projected for north-eastern Europe while decreases are projected in the Mediterranean region, where runoff differences between 1.5°C and 2°C global warming will be most prominent (medium confidence). According to SROCC (Hock et al., 2019b) the RCP8.5 projections lead to a loss of more than 80% of the ice mass from small glaciers by the end of century in Central Europe (high confidence). Snow cover and glaciers are projected to decrease throughout the 21st century.

The recent-past climate of North America is characterized by high spatial heterogeneity and by variability at diverse temporal scales. Considering the traditional Köppen-Geiger classification, North America covers all main climate types (see reference region descriptions below). Important geographical features influence local climates over various distances, like the Rocky Mountains through cyclogenesis (Grise et al., 2013) and the Great Lakes through lake-effect snowfall (Wright et al., 2013). The cryosphere is an important component of the climate system in North America, with fundamental roles for sea ice cover, snow cover and permafrost. The ocean surrounding the continent also influences its climate, with water temperatures strongly influencing hurricane activity which impacts the coasts of eastern Mexico and south-eastern USA (Walsh et al., 2010). Temporal variability is influenced by several large-scale atmospheric modes (Table Atlas.1 and Annex IV) with the North Atlantic Oscillation (NAO) affecting north-eastern USA and eastern Canada precipitation (Whan and Zwiers, 2017), and El Niño–Southern Oscillation (ENSO) affecting temperature and precipitation in California, although in a complex and not yet fully understood manner (Yoon et al., 2015; Yeh et al., 2018).

The reference regions defined for summarising North America climate change (Figure Atlas.26) include: North-Western North America (NWN), characterized by a sub-Arctic climate with cool summers and rainfall all year round; North-Eastern North America (NEN), which also has a sub-Arctic climate with sections of tundra climate in the far north (these two northern regions are also discussed in Section Atlas.11.2, Polar Arctic); Western North America (WNA), which has a complex but mainly cold semi-arid climate; Central North America (CNA) with a mainly continental climate in the northern part of the region and a humid subtropical climate in the southern portion; Eastern North America (ENA) with a humid continental climate in the northern half and a humid subtropical climate to the south; Northern Central America (northern Mexico; NCA), has a temperate climate to the north of the Tropic of Cancer, with marked differences between winter and summer, modulated by the North American Monsoon (Peel et al., 2007).

The IPCC AR5 (Bindoff et al., 2013; Hartmann et al., 2013) found that the climate of North America has changed due to anthropogenic causes (high confidence), in particular with primarily increasing annual precipitation and annual temperature (very high confidence). Assessment of CMIP5 ensemble projections concluded that mean annual temperature over North America and annual precipitation north of 45°N will very likely continue to increase in the future. Also, CMIP5 projects increases in winter precipitation over Canada and Alaska and decreases in winter precipitation over the south-western USA and much of Mexico.

The CMIP5 multi-model ensemble generally reproduces the observed spatial patterns but somewhat underestimates the extent and intensity of the North American Monsoon, and also underestimates wetting over Central North America over the period of 1950–2012 during the winter season according to AR5 (Flato et al., 2013). In the long term (2081–2100), the largest changes of precipitation over North America are projected to occur in the mid- and high latitudes and during winter (Kirtman et al., 2013).

The SR1.5 (Hoegh-Guldberg et al., 2018) reported a stronger warming compared to the global mean over Central and Eastern North America, and a weakening of storm activity over North America under 1.5°C of global warming. The SROCC (Hock et al., 2019b) reported that snow depth or mass is projected to decline by 25% mainly at lower elevations over the high mountains in Western North America. The SRCCL (Mirzabaev et al., 2019) observed vegetation greening in Central North America with high confidence.

Many small islands lie in tropical regions and their climate varies depending on a range of factors with location, extent and topography having major influences. In general, their climate is determined by that of the broader region in which they lie as they have little influence on the regional climate, although steep topography can induce higher rainfall totals locally. Temperature variability tends to be low due to the influence of the surrounding ocean, most marked in the tropics where oceanic temperature ranges are small. However, seasonal rainfall variability can often be significant, both through the annual cycle and also interannually through the influence of many modes of variability (Cross-Chapter Box Atlas.2:, Annex IV and Atlas.7.1 for the Caribbean). Many small islands are exposed to tropical cyclones and the associated hazards of high winds, storm surges and extreme rainfall, and many low-lying islands are exposed to regular flooding from natural high-tide and wave activity. In the Pacific, phases of the El Niño–Southern Oscillation result in periods of warmer or cooler than average temperatures following the upper ocean warming of El Niño events or cooling of La Niña events, and respectively weaker and stronger trade winds. El Niño conditions also lead to drought in Melanesian islands and increased tropical cyclones and storm surges in French Polynesia with La Niña conditions causing drought in Kiribati. Other islands experience increased rainfall during these periods.

The AR5 noted observed temperature increases of 0.1°C–0.2°C per decade in the Pacific Islands and that warming was very likely to continue across all Small Islands regions (Christensen et al., 2013; IPCC, 2013a). It also reported decreased rainfall over the Caribbean, increases over the Seychelles, streamflow reductions over the Hawaiian Islands and projections of reduced rainfall over the Caribbean and drier rainy season for many of the south-west Pacific Islands (Christensen et al., 2013; IPCC, 2013a; Nurse et al., 2014). The remaining findings are derived from the SROCC (IPCC, 2019a). Ocean warming rates have likely increased in recent decades with marine heatwaves increasing and very likely to have become longer-lasting, more intense and extensive as a result of anthropogenic warming. Open ocean oxygen levels have very likely decreased and oxygen minimum zones have likely increased in extent. There is very high confidence that global mean sea level rise has accelerated in recent decades which, combined with increases in tropical cyclone winds and rainfall and increases in extreme waves, has exacerbated extreme sea level events and coastal hazards (high confidence). It is virtually certain that during the 21st century, the ocean will transition to unprecedented conditions with further warming and acidification virtually certain, increased upper ocean stratification very likely and continued oxygen decline (medium confidence). There is very high confidence that marine heatwaves and medium confidence that extreme El Niño and La Niña events will become more frequent. It is very likely that these changes will be smaller under scenarios with low greenhouse gas emissions. Global mean sea level will continue to rise and there is high confidence that the consequent increases in extreme levels will result in local sea levels in most locations that historically occurred once per century occurring at least annually by the end of the century under all RCP scenarios (high confidence). In particular, many small islands are projected to experience historical centennial events at least annually by 2050 under RCP2.6 and higher emissions. The proportion of Category 4 and 5 tropical cyclones, and associated precipitation rates and storm surges, along with average tropical cyclone intensity are projected to increase with a 2°C global temperature rise, thereby exacerbating coastal hazards.

Figure Atlas.30 (Antarctic map inset) shows near-surface air temperature trends for 1957–2016 and 1979–2016 at the stations where observations are available for at least 50 years and the detected trends have statistical significance of at least 90% according to the most recent (after SROCC) studies (Jones et al., 2019; Turner et al., 2020). It is very likely that the western and northern AP has been warming significantly since the 1950s (0.49°C ± 0.28°C per decade during 1957–2016 and 0.46°C ± 0.15°C during 1951–2018 at Faraday-Vernadsky station; 0.29°C ± 0.16°C per decade during 1957–2016 at Esperanza station), with no significant trends reported in the eastern AP during the same period (Gonzalez and Fortuny, 2018; Jones et al., 2019; Turner et al., 2020). Short-term cooling trends, strongest during austral summer, have been reported at AP stations during 1999–2016, but the absence of warming and cooling at some stations during 1999–2016 is consistent with natural variability, and there is no evidence of a shift in the overall warming trend observed since the 1950s (Turner et al., 2016, 2020; Gonzalez and Fortuny, 2018; Jones et al., 2019; Bozkurt et al., 2020).

Figure Atlas.30 | (Upper panels) Time series of annual surface mass balance (SMB) rates (in Gt a–1) for the Greenland Ice Sheet and its regions (shown in the inset map) for the periods 1972–2018 (Mouginot et al., 2019) and 1980–2012 (Fettweis et al., 2020) using 13 different models. (Lower panels) Time series of annual SMB rates (in Gt a^–1) for the grounded Antarctic Ice Sheet (excluding ice shelves) and its regions (shown in the inset map) for the periods 1979–2019 (Rignot et al., 2019) and 1980–2016 (Mottram et al., 2021) using five Polar-CORDEX regional climate models. The Antarctic inset map also shows the location of the stations discussed in Atlas.11.1.2 where observations are available for at least 50 years. Colours indicate near-surface air temperature trends for 1957–2016 (circles) and 1979–2016 (diamonds) statistically significant at 90% (Jones et al. 2019; Turner et al. 2020). Stations with an asterisk (*) are where significance estimates disagree between the two publications. Further details on data sources and processing are available in the chapter data table (Table Atlas.SM.15).

Open figure

Significant warming at the Byrd station (0.29°C ± 0.19°C per decade during 1957–2016) confirms and extends earlier trend estimates (0.42°C ± 0.24°C per decade during 1958–2010) and is representative of the entire WAN warming (0.22°C ± 0.12°C per decade from 1958 to 2012 averaged over WAN excluding AP, medium confidence due to lack of observations) (Bromwich et al., 2013, 2014; Jones et al., 2019). WAN and AP show statistically significant warming in the HadCRUTv5 observational dataset (Figure 2.11b). There is high confidence in the long-term warming trend at the AP and WAN, and also at the century scale based on reconstructions (Zagorodnov et al., 2012; Stenni et al., 2017; Lyu et al., 2020), confirming the trends estimated by earlier studies assessed in the SROCC (Meredith et al., 2019). The century-scale warming trend in the AP is very likely an emerging signal compared to natural variability, while the WAN warming trend falls in the high end of century-scale trends over the last 2000 years (medium confidence) (Stenni et al., 2017).

In EAN, during 1957–2016, three stations showed significant warming (Scott 0.22°C ± 0.15°C, Novolazarevskaya 0.13°C ± 0.09°C, and Vostok 0.15°C ± 0.13°C per decade), while other stations with long-term observations indicated no statistically significant trends (Figure Atlas.3 0). During 1979–2016, three coastal stations showed cooling, while at the South Pole a warming trend was detected, increasing to 0.61°C ± 0.34°C per decade during 1989–2018 (Figure Atlas.3 0; Jones et al., 2019; Clem et al., 2020; Turner et al., 2020). The century-scale warming in Queen Maud Land coast based on ice-core reconstructions is within the range of centennial internal variability (Stenni et al., 2017).

While a trend towards a positive phase of the SAM since the 1970s likely explains a significant part of the warming at the northern AP, it had a cooling effect on continental WAN and EAN (particularly strong in DJF; Table Atlas.1). Warming in western AP and over WAN during 1957–2016 (Figure Atlas.3 0) and through to 2020 (Figure 2.11) is likely due to significant contribution of other factors, such as tropical Pacific forcing through PDV, ENSO, Amundsen Sea Low position/strength and also anthropogenic climate change (Jones et al., 2019; Scott et al., 2019; Wille et al., 2019; Donat-Magnin et al., 2020; Turner et al., 2020). Since SROCC, new studies confirmed the influence of foehn wind and cloud radiative forcing on Larsen C surface melt (Elvidge et al., 2020; Gilbert et al., 2020; Turton et al., 2020). In WAN, summer surface-melt occurrence over ice shelves may have increased since the late 2000s (Scott et al., 2019). It is likely that increased meltwater ponding and resulting hydrofracturing have been important mechanisms of the rapid disintegration of the Larsen B ice shelf (Banwell et al., 2013; MacAyeal and Sergienko, 2013; Robel and Banwell, 2019). Ice-shelf disintegration and relevant processes are discussed in Sections 9.4.2.1 and 9.4.2.3.

Direct observations of snowfall in Antarctica using traditional gauges are highly uncertain and records from precipitation radars (Gorodetskaya et al., 2015; Grazioli et al., 2017; Scarchilli et al., 2020) are not long enough to assess trends. Estimates of precipitation and SMB are largely model-based due to the paucity of in situ observations in Antarctica (Lenaerts et al., 2019; Hanna et al., 2020). Antarctic SMB is dominated by precipitation and removal by sublimation with very small amounts of melt mostly important only on the ice shelves. Climate models and satellite records (IMBIE team et al., 2018; Rignot et al., 2019; Mottram et al., 2021) suggest that strong interannual variability of Antarctic-wide SMB over the satellite period currently masks any existing trend (Figure Atlas.3 0) in spite of a possible ozone depletion-related precipitation increase over the 1991–2005 period (Lenaerts et al., 2018). No significant Antarctic-wide SMB trend is inferred since 1979 (IMBIE team et al., 2018; Medley and Thomas, 2019). While ice-core reconstructions show a significant increase in the western AP SMB since the 1950s (Thomas et al., 2017; Medley and Thomas, 2019; Wang et al., 2019), this trend is not reproduced by regional climate models or the reanalyses used to drive them (Figure Atlas.3 0; van Wessem et al., 2016; Wang et al., 2019).

According to the ice-core reconstructions, SMB over WAN (including AP) has likely increased during the 20th century with trends of 5.4 ± 2.9 Gt yr^–1per decade (1900–2010; Wang et al., 2019) mitigating global mean sea level rise by, respectively, 0.28 ± 0.17 mm per decade (WAN excluding AP, during 1901–2000) and 0.62 ± 0.17 mm per decade (AP, during 1979–2000; Medley and Thomas, 2019). Significant spatial heterogeneity in SMB trends has been observed over AP and WAN:

• Western AP has likely experienced a significant increase in SMB beginning around 1930 and accelerating during 1970–2010, which is outside of the natural variability range of the past 300 years (Thomas et al., 2017; Medley and Thomas, 2019; Wang et al., 2019);
• eastern AP has no significant SMB trends during the same period (low confidence, observations limited to one ice core and large interannual variability) (Thomas et al., 2017; Engel et al., 2018);
• overall WAN SMB (excluding AP) was stable during 1980–2009 but exhibited high regional variability (Medley et al., 2013): significant increases (5–15 mm per decade during 1957–2000) to the east of the West Antarctic Ice Sheet divide and a significant decrease (–1 to –5 mm per decade during 1901–1956, and –5 to –15 mm per decade during 1957–2000) to the west (Medley and Thomas, 2019; Wang et al., 2019).

The SMB of EAN increased during the 20th century which mitigated global mean sea level rise by 0.77 ± 0.40 mm per decade during 1901–2000 (medium confidence) (Medley and Thomas, 2019). EAN SMB has been increasing at a much lower rate since 1979 as shown by observations, while regional climate models show strong interannual variability masking any trend (low confidence due to limited observations) (Figure Atlas.3 0; Medley and Thomas, 2019; Rignot et al., 2019). EAN SMB changes during the 20th century and recent decades showed large spatial heterogeneity:

• With significant increases likely in Queen Maud Land (QML): 5.2 ± 3.7% per decade during 1920–2011 measured in ice cores near the Kohnen station (Medley et al., 2018), an increase on the plateau (Altnau et al., 2015), and stable conditions during 1993–2010 along the annual stake line from Syowa (coast) to Dome F (plateau) (Y. Wang et al., 2015 ); increases during 1911–2010 (Thomas et al., 2017) with anomalously high SMB observed in 2009 and 2011 (Boening et al., 2012; Lenaerts et al., 2013; Gorodetskaya et al., 2014);
• increases in Wilkes Land and Queen Mary Land during 1957–2000 (low confidence due to limited observations and strong spatial variability) (Thomas et al., 2017; Medley and Thomas, 2019);
• a likely stable SMB in the interior of the east Antarctic plateau during the 1901–2000 period and the last decades (Thomas et al., 2017; Medley and Thomas, 2019);
• stable in Adelie Land (annual stake line during 1971–2008) (low confidence due to limited evidence) (Agosta et al., 2012).

Regional trends during recent 50 year (1961–2010) and 100 year (1911–2010) periods are within the centennial variability of the past 1000 years, except for coastal QML (unusual 100-year increase in accumulation) and for coastal Victoria Land (unusual 100-year decrease in accumulation) (Thomas et al., 2017). Nevertheless, the current EAN SMB is not unusual compared to the past 800 years (Frezzotti et al., 2013).

The geographic pattern of accumulation changes since the 1950s bears a strong imprint of a trend towards a more positive phase of the SAM (e.g., Medley and Thomas, 2019), which could be linked to ozone depletion (Lenaerts et al., 2018) or large-scale atmospheric warming (Frieler et al., 2015; Medley and Thomas, 2019). More evidence has emerged showing the importance of the Pacific–South American pattern, ENSO and Pacific Ocean convection, and large-scale blocking causing warm-air intrusions and both extreme precipitation and melt events, responsible for large interannual SMB variability (high confidence) (Gorodetskaya et al., 2014; Bodart and Bingham, 2019; Scott et al., 2019; Turner et al., 2019; Wille et al., 2019; Adusumilli et al., 2021). This strengthens evidence for an important connection between Antarctic climate and tropical sea surface temperature stated by SROCC (Meredith et al., 2019). Section 3.4.3 and SROCC (Meredith et al., 2019) provide a discussion of attribution of Antarctic ice-sheet changes.

This section provides evaluation of atmospheric global and regional climate models, including reanalyses. Evaluation of the ice-sheet models and relevant processes, including selection of the atmospheric models used to drive ice-sheet models, is given in Section 9.4.2.2.

One of the major systematic biases in CMIP5 and earlier GCMs was an equatorward bias in the latitude of the Southern Hemisphere mid‐latitude westerly jet, which is significantly reduced in the CMIP6 ensemble (Bracegirdle et al., 2020a). GCM Southern Ocean sea ice biases are also of importance as they influence 21st-century temperature projections in Antarctica and simulations of present-day temperatures are highly sensitive to these biases (Agosta et al., 2015; Bracegirdle et al., 2015). A positive bias in near-surface temperature over the Antarctic plateau is seen in CMIP5 models (Lenaerts et al., 2016).

CMIP6 GCMs showed an improved representation of the Antarctic near-surface temperature compared to CMIP5 but little improvement (maintaining positive bias) in Antarctic precipitation estimates (Palerme et al., 2017; Roussel et al., 2020). An analysis of the 1850–2000 SMB mean, trends, and interannual and spatial variability suggests slightly worse agreement with ice-core-based reanalyses in CMIP6 than CMIP5 (Gorte et al., 2020). Comparison of CMIP5 models with CloudSat satellite products and an ice-core-based SMB reconstruction showed almost all the models overestimate current Antarctic precipitation, some by more than 100% (Palerme et al., 2017; Gorte et al., 2020). GCM simulations of surface snow-melt processes are either of variable quality, with extremely simple representatons, or non-existent (Agosta et al., 2015; Trusel et al., 2015). Though most meltwater refreezes in the snowpack in current climate simulations, this may be an issue in the future climate simulations under global warming as runoff is projected to increase (Kittel et al., 2021). Since CMIP5, representation of snow (Lenaerts et al., 2016) and stable surface boundary layers (Vignon et al., 2018) has improved in some atmospheric GCMs. In one example, the CMIP6 model CESM2 simulation of cloud and precipitation showed substantial improvements (Schneider et al., 2020), though surface melting is still considerably overestimated compared to RCMs and satellite products (Trusel et al., 2015; Lenaerts et al., 2016).

Assimilation of observations in reanalysis products yields realistic temperature patterns and seasonal variations, with the recent ERA5 reanalysis showing improved performance compared to others for mean and extreme temperature, wind and humidity, though a warm bias in near-surface air temperatures remains (Retamales-Muñoz et al., 2019; Tetzner et al., 2019; Dong et al., 2020; Gorodetskaya et al., 2020). The ability of the reanalyses to simulate precipitation and SMB is more variable; they generally overestimate the latter (Gossart et al., 2019; Roussel et al., 2020), but are well suited to provide atmospheric and sea surface boundary conditions to drive RCMs.

Recent higher-resolution simulations covering the entire Antarctic Ice Sheet with a grid spacing of 12 to 50 km include five Polar-CORDEX RCMs – RACMO2 (van Wessem et al., 2018), MAR (Agosta et al., 2019; Kittel et al., 2021), COSMO-CLM2 (Souverijns et al., 2019), HIRHAM5 (Lucas-Picher et al., 2012) and MetUM (Walters et al., 2017; Mottram et al., 2021) – and one stretched-grid GCM – ARPEGE (Beaumet et al., 2019). RCM simulations forced by ERA-Interim agree well with automatic weather station temperatures, with high correlation (R²> 0.9) and low bias (<1.5°C) except for high-resolution HIRHAM5 (–2.1°C) and MetUM (–3.4°C), which are not internally nudged models (Mottram et al., 2021). RCMs generally underestimate the observed SMB but with biases lower than 20%, except for COSMO-CLM2 at lower elevations (<1200 m) and HIRHAM5 and MetUM at higher elevations (>2200 m) (Mottram et al., 2021). These RCM simulations lead to estimates of the grounded Antarctic Ice Sheet SMB ranging from 2133 Gt yr^–1to 2328 Gt yr^–1 when considering the four simulations compatible with the IMBIE2 Antarctic total mass budget (IMBIE team et al., 2018; Mottram et al., 2021). However, the simulated spatial pattern of SMB differs widely between models, suggesting the importance of missing or under-represented processes in the models, such as drifting-snow transport and sublimation (Agosta et al., 2019), cloud-precipitation microphysical processes (van Wessem et al., 2018) and snowpack modelling (Mottram et al., 2021). Comparisons of integrated SMB estimates between models are also complicated by different resolutions and continental ice masks, with models showing large differences in the absolute SMB (Mottram et al., 2021) but better agreement for SMB annual rates (Figure Atlas.3 0).

Finer-resolution RCM studies demonstrate improved representation of precipitation and temperature gradients (van Wessem et al., 2018; Bozkurt et al., 2020; Donat-Magnin et al., 2020; Elvidge et al., 2020), and strength of katabatic winds (Bintanja et al., 2014; Souverijns et al., 2019) in coastal and mountainous regions. Adequate representation of some processes is still lacking, including drifting snow, sublimation of falling snow or the spectral dependency of snow albedo (Lenaerts et al., 2019). Non-hydrostatic regional models, for example Polar-WRF, MetUM or HARMONIE-AROME at spatial resolutions up to 2 km further improve regional RCM simulations, but are still often unable to resolve relevant feedbacks and foehn processes (Grosvenor et al., 2014; Elvidge et al., 2015, 2020; Elvidge and Renfrew, 2016; King et al., 2017; Turton et al., 2017; Bozkurt et al., 2018b; Hines et al., 2019; Vignon et al., 2019; Gilbert et al., 2020).

Existing uncertainties in the Antarctic climate representation by both GCMs and RCMs cause significant spread in the future Antarctic climate and SMB projections (Gorte et al., 2020; Kittel et al., 2021). Run-time bias adjustment in atmospheric GCMs (Cross-Chapter Box 10.2; Krinner et al., 2019, 2020) has been proposed to provide low-bias present and consistently corrected future RCM forcing (reducing the need for coupled model selection), which could be used directly for Antarctic climate projections (Krinner et al., 2019).

This section provides an assessment of projections in temperature, precipitation and SMB. See Section 9.4.2 for projected changes in the ice-sheet total mass balance and relevant processes, and see Section 4.3.1 (Table 4.2) and Section 4.5.1 for Antarctic temperature projections relative to other regions.

The Antarctic region is very likely to experience a significant increase in annual mean temperature and precipitation by the end of this century under all emissions scenarios used in CMIP5 and CMIP6 (Figure Atlas.29; Bracegirdle et al., 2015, 2020b; Frieler et al., 2015; Lenaerts et al., 2016; Previdi and Polvani, 2016; Palerme et al., 2017). Ensemble means (and 10th–90th percentile ranges) of end-of-century (2081–2100) projected Antarctic surface air temperature change from 35 CMIP6 models and relative to 1995–2014 are 1.2°C (0.5°C–2.0°C) for the SSP1-2.6 emissions scenarios, 2.3°C (1.3°C–3.4°C) for SSP2-4.5, 3.5°C (2°C–5°C) for SSP3-7.0, and 4.4°C (2.8°C–6.4°C) for SSP5-8.5 (Interactive Atlas). Both temperature and precipitation projections are characterized by a relatively large multi-model range (Figure Atlas.29 and the Interactive Atlas). A strong regional variability is present with the projected changes over coastal Antarctica not scaling linearly with global forcing. While continental mean temperatures are linearly related to global mean temperatures in CMIP6 models, the relative increase in coastal temperatures are higher for low-emissions scenarios due to stronger relative Southern Ocean warming and relatively stronger effects of ozone recovery (Bracegirdle et al., 2020b). A higher multi-model average increase in temperature is projected by CMIP6 models compared to CMIP5, with a 1.3°C higher mean Antarctic near-surface temperature at the end of the 21st century (Kittel et al., 2021). While similar median temperature changes are projected for WAN and EAN, the former shows larger spread and higher projected temperature range in both CMIP5 and CMIP6 models and for all scenarios (Figure Atlas.29). CORDEX-Antarctica simulations show a mean and range in the future temperature changes similar to the subset of CMIP5 models used to drive them for the RCP8.5 scenario and 1.5°C, 2°C and 3°C GWLs (Figure Atlas.29).

There is high confidence that projected future surface air temperature increase over Antarctica will be accompanied by precipitation increase (Figure Atlas.29). CMIP6 models show a similar or larger but more constrained increase in precipitation (more models agreeing with larger precipitation increase) for the same GWLs compared to CMIP5. For example, over WAN during JJA for 3°C GWL, CMIP6 and CMIP5 models project a median 15% increase in precipitation with a 10th–90th percentile range of 7–25% in CMIP6 models and of 3–24% in CMIP5. Average precipitation changes relative to 1995–2014 over WAN and EAN are largely similar; they show projected increases for SSP2-4.5 (SSP5-8.5) of around 5% (5%) for 2021–2040, 7% (10%) for 2041–2060, and 12% (25%) for 2081–2100 with smaller increases projected for SSP1-2.6 emissions, reaching around 5% in 2081–2100. Regionally, the largest relative precipitation increase is projected (under all scenarios) for the eastern part of WAN, the western AP, large parts of the EAN plateau and over coastal EAN within 0°E–90°E longitudinal sector (Interactive Atlas). The largest increase in absolute precipitation amount is projected along the coastal regions, with the largest increase over coastal WAN and the western AP, and is projected to be largely driven by the increase in maximum five-day precipitation (Interactive Atlas), which is in line with the dominant contribution of extreme snowfall events to the total annual precipitation in the present Antarctic climate (Boening et al., 2012; Gorodetskaya et al., 2014; Turner et al., 2020). Under all emissions scenarios, the coastal precipitation increase corresponds to the snowfall increase, except for the northern and central part of the western AP, where snowfall is projected to decrease and rainfall to increase (similarly to the tendency towards increased precipitation, decreased snowfall and increase in rainfall over the Southern Ocean; Interactive Atlas).

From 2000 to 2100, the grounded Antarctic SMB is projected to mitigate sea level rise for RCP4.5 (RCP8.5) by the following sea level equivalents (SLEs), 0.03 ± 0.02 m (0.08 ± 0.04 m SLE) from 30 CMIP5 models and for SSP2-4.5 (SSP5-8.5) by 0.03 ± 0.03 m SLE (0.07 ± 0.04 m SLE) from 24 CMIP6 models (Gorte et al., 2020). Subsets or downscaling of CMIP AOGCMs lead to 21st-century cumulative projections in the range of 0.05 ± 0.03 m SLE for CMIP5 RCP8.5 and 0.08 ± 0.04 m SLE for CMIP6 SSP5-8.5 (Gorte et al., 2020; Nowicki et al., 2020; Seroussi et al., 2020; Kittel et al., 2021). Use of model subsets reduces spread leading to either lower or higher climate sensitivity in the Antarctic depending on the selection method. For example, models selected by Gorte et al. (2020) based on SMB ice-core reconstruction from Medley and Thomas (2019) tend to underestimate strongly winter sea ice area (Agosta et al., 2015; Roach et al., 2020) and show reduced 21st-century increase in Antarctic SMB compared to the full ensembles (Agosta et al., 2015; Bracegirdle et al., 2015). A different subset of models is used for ISMIP6 (Section 9.4.2.3) which gives a lower increase in Antarctic SMB than the full ensemble for CMIP5 but a larger increase for CMIP6.

Polar-CORDEX RCMs show higher variability in precipitation projections compared to CMIP5 models with a similar spatial pattern of the areas with precipitation increase over continental Antarctica but with higher local magnitude, and also showing a larger increase over the Weddell Sea ice shelves (Interactive Atlas). CMIP5 and CMIP6 models, bias adjusted based on regional climate model simulations, showed that the projected warming is expected to result in increased surface melting over the Antarctic ice shelves, with meltwater runoff under RCP8.5 and SSP5-8.5 becoming larger than precipitation over ice shelves over the period 2045–2050, surpassing intensities that were linked with the collapse of Larsen A and B ice shelves (Trusel et al., 2015; Kittel et al., 2021). Given the existing uncertainty in the present precipitation and SMB simulations and the significant range in the projected precipitation increase under various emissions scenarios in CMIP5, CMIP6 and CORDEX models, there is medium confidence that the future Antarctic SMB will have a negative contribution to sea level during the 21st century under all emissions scenarios (see Section 9.4.2.3 for assessment of the drivers of future Antarctic ice-sheet change and Section 9.4.2.6 for longer time scales).

Observations show a very likely widespread, strong warming trend starting in the 1950s in the Antarctic Peninsula. Significant warming trends are observed in other West Antarctic regions and at selected stations in East Antarctica (medium confidence). Antarctic precipitation and SMB showed a significant positive trend over the 20th century according to the ice cores, while large interannual variability masks any existing trend over the satellite period since the end of the 1970s (medium confidence).

An assessment of model performance for the present day shows that high-resolution regional climate models with polar-optimized physics are important for estimating SMB and generating climate information, and show improved realizations compared to reanalyses and GCMs when evaluated against observations. At the same time, CMIP6 GCMs showed an improved representation of the Antarctic near-surface temperature compared to CMIP5, though still struggle with the representation of precipitation. There is therefore medium confidence in the capacity of climate models to simulate Antarctic climate and SMB changes.

Under all assessed emissions scenarios, both West and East Antarctica are very likely to have higher annual mean surface air temperatures and more precipitation, which will have a dominant influence on determining future changes in the SMB (high confidence). However, due to the challenges of model evaluation over the region and the possibility of increased meltwater runoff described above, there is onlymedium confidence that the future contribution of the Antarctic SMB to sea level this century will be negative under all greenhouse gas emissions scenarios.

The Arctic has warmedat more than twice the global rate over the past 50 years with the greatest warming during the cold season (high confidence) (Davy et al., 2018; Box et al., 2019; Przybylak and Wyszyński, 2020; Xiao et al., 2020). This is based on various Arctic amplification processes, in particular the combined effect of several related feedback processes, including between various radiation components and (a) the albedo of sea ice and snow, (b) water vapour, and (c) clouds, as well as poleward energy transports. The annual average Arctic surface air temperature increased by 2.7°C from 1971 to 2017, with a 3.1°C increase in the cold season (October–May) and a 1.8°C increase in the warm season (June–September) (AMAP, 2019). Satellite-based data estimate the rate of annual warming for 1981–2012 over sea ice covered regions to be 0.47°C per decade, whereas the trend was significantly higher at 0.77°C per decade over Greenland and amplified in the northern Barents and Kara seas (Comiso and Hall, 2014). The largest Arctic warming in 2003–2017 was reported over the Barents and Kara seas with trends larger than 2.5°C per decade (Susskind et al., 2019), and Arctic temperatures from 2014 to 2018 have exceeded all previous records since 1900 (Blunden and Arndt, 2019).

Over the ARO, long-term temperature records are available from Spitsbergen (Svalbard Airport). For the period 1898–2018, the annual mean warming was 0.32°C per decade, about 3.5 times the global mean temperature for the same period and since 1991, it was 1.7°C per decade or about seven times the global average for the same period (Nordli et al., 2020). There is a positive trend in the annual temperature for all stations across Svalbard (Gjelten et al., 2016; Hanssen-Bauer et al., 2019; Dahlke et al., 2020) of 0.64°C–1.01°C per decade for 1971–2017 (Hanssen-Bauer et al., 2019), co-varying with regional changes in sea ice conditions (Dahlke et al., 2020). The largest temperature trends very likely occur in winter, with Svalbard Airport warming at 0.43°C per decade during 1898–2018 and 3.19°C per decade during 1991–2018 (Nordli et al., 2020), and Isaksen et al. (2016) reporting on substantial warming in western Spitsbergen, particularly in winter, while the summer warming is moderate.

A multi-dataset analysis for NEN shows a consistent warming (Rapaić et al., 2015), with the largest annual temperature trend greater than 0.3°C per decade during 1981–2010 over eastern NEN and also significant warming over northern Quebec and most of the Canadian Arctic north of the treeline. For the longer 1950–2010 period, a consistent warming is found over central and western NEN, but no trend or no consensus is found over the Labrador coast. The latter is related with cooling of the North Atlantic region during the 1970s. For western Greenland, however, summer temperatures increased (2.2°C in June, 1.1°C in July) from 1994 to 2015 (Saros et al., 2019). For neighbouring Arctic regions of NEU, WSE and ESB, datasets show a consistent warming of annual mean temperature since the mid-1970s and 1980 (Atlas.8 and Atlas.5.2).

Along with the amplified warming, the Arctic has become moister (Rinke et al., 2019; Nygård et al., 2020). AMAP reported Arctic precipitation increases of 1.5–2.0% per decade, with the strongest increase in the cold season (October–May) (medium confidence) (AMAP, 2019). Also, for neighbouring Arctic regions for example NEU, EEU and North Asia, mean annual precipitation has increased since the early 20th century (Atlas.8 and Atlas.5.2). Estimated trends for precipitation and snowfall fraction are mixed for the Arctic, with increases and decreases for different regions and seasons (Vihma et al., 2016). However, annual precipitation trends derived from different reanalyses do not agree, differ in sign and have low significance (Lindsay et al., 2014; Boisvert et al., 2018). Direct precipitation measurements are difficult and include uncertainties (among others measuring frozen precipitation), therefore precipitation estimates in the Arctic rely on climate models and reanalyses.

An average of five reanalyses for 2000–2010 suggests around 40% of Arctic Ocean precipitation falls as snow, though there is large uncertainty in this estimate (Boisvert et al., 2018). Rainfall frequency is estimated to have increased over the Arctic by 2.7–5.4% over 2000–2016 (Boisvert et al., 2018) with more frequent rainfall events reported for NEU and ARO (Svalbard; Maturilli et al., 2015; AMAP, 2019), and winter rain totals and frequency have increased in Svalbard since 2000 (medium confidence) (Łupikasza et al., 2019). Rain-free winters have rarely occurred since 1998 (Peeters et al., 2019).

Observational records (1966–2010) for the RAR region show changing precipitation characteristics (Ye et al., 2016), with higher precipitation intensity but lower frequency and little change in annual precipitation total. Precipitation intensity is reported to have increased in all seasons, strongest in winter and spring, weakest in summer, and at a rate of about 1–3% per degree Celsius of air temperature increase.

Evaluating simulated temperature and precipitation is problematic in the Arctic due to sparse weather station observations. The lack of reliable observed precipitation datasets for the Arctic thus makes it very unlikely to be able to evaluate objectively the skill of models to reproduce precipitation patterns (Takhsha et al., 2018).

The CMIP5 models reproduce the observed Arctic warming over the past century (medium confidence) (Chylek et al., 2016; Hao et al., 2018; Huang et al., 2019). The simulated mean Arctic warming for 1900–2014 averaged over 40 CMIP5 models is 2.7°C compared to the observed values of 2.2°C (NASA GISS data smoothed using a 1200-km radius) or 1.7°C (using a 250-km smoothing radius) (Chylek et al., 2016). However, there are large inter-model differences in the simulated warming which ranges from 1.2°C to 5.0°C. Although the CMIP5 models reproduce the spatially averaged observed warming over the past 50 to 100 years, the pattern is different from that of observations and reanalysis (Xie et al., 2016; Franzke et al., 2017; Hao et al., 2018). Zonal mean temperature trends in the CMIP5 models overestimate the warming in the cold season over high latitudes in the Northern Hemisphere (Xie et al., 2016). Overall, the amplified Arctic warming in recent decades is overestimated by CMIP5 models (Huang et al., 2019). Possible reasons are modelled sea surface temperature biases and an overestimated temperature response to the Arctic sea ice decline. Furthermore, some models, which have a warm or weak bias in their Arctic temperature simulations, closely relate the Arctic warming to changes in the large-scale atmospheric circulation. In other models, which show large cold biases, the albedo feedback effect plays a more important role for the temperature trend magnitude. This implies that the dominant simulated Arctic warming mechanism and trend may be dependent on the bias of the model mean state (Franzke et al., 2017). Compared to CMIP5 models, Davy and Outten (2020) found lower biases in CMIP6 models’ representation of sea ice extent and volume with improved extents linked to a better seasonal cycle in the Barents Sea.

Rapid temperature changes, such as the pronounced increase of 2°C yr^–1during 2003–2012 over the Kara and Barents seas in March is well captured in Arctic CORDEX simulations (Kohnemann et al., 2017). The models show adequate skill in capturing the general temperature patterns (Koenigk et al., 2015; Matthes et al., 2015; Hamman et al., 2016; Cassano et al., 2017; Brunke et al., 2018; Diaconescu et al., 2018; Takhsha et al., 2018), but tend to show a cold temperature bias which is largest in winter and depends on the reference dataset. Cassano et al. (2017) showed a large sensitivity of the simulated surface climate to changes in atmospheric model physics. In particular, large changes in radiative flux biases, driven by changes in simulated clouds, lead to large differences in temperature and precipitation biases.

The CMIP5 models perform well in simulating 20th-century snowfall for the Northern Hemisphere, although there is a positive bias in the multi-model ensemble relative to the observed data in many regions (Krasting et al., 2013). Lack of sufficient spatial resolution in the model topography has a serious impact on the simulation of snowfall. The patterns of relative maxima and minima of snowfall, however, are captured reasonably well by the models.

Arctic CORDEX RCMs reproduce the dominant features of regional precipitation patterns and extremes (e.g., Glisan and Gutowski, 2014; Hamman et al., 2016). Due to their higher spatial resolution, RCMs simulates larger amounts of orographic precipitation compared to reanalyses. Overall, the simulated precipitation is within the reanalysis and global model ensemble spread, but the Arctic river basin precipitation is closer to observations (Brunke et al., 2018). However, Takhsha et al. (2018) show that the RCMs’ precipitation bias highly depends on the observational reference dataset used.

The annual mean precipitation pattern of ensemble global atmospheric simulations with a high horizontal resolution agrees well with the observations, with precipitation maxima over the Greenland and Norwegian seas (Kusunoki et al., 2015). However, the simulated Arctic average annual precipitation shows a positive bias with excessive precipitation over Alaska and the western Arctic (Kattsov et al., 2017).

Regarding the Greenland Ice Sheet (region GIC), modelled surface mass balance (SMB) has decreased since the end of the 1990s (Fettweis et al., 2020). A multi-model intercomparison study (Fettweis et al., 2020) emphasized a simulated positive mean annual SMB of 338 ± 68 Gt yr^–1between 1980 and 2012, with a decreasing average rate of 7.3 ± 2.0 Gt yr^–2, mainly driven by an increase in meltwater runoff. Mouginot et al. (2019) stated that SMB played a strong role in the ice-sheet mass loss, where SMB dominated in the last two decades. Mottram et al. (2019) found that SMB processes dominate the ice-sheet mass budget over most of the interior, highlighting that the ice sheet is a contributor to global mean sea level rise between 1991 and 2015. More specifically, SMB models have improved (Fettweis et al., 2020; Hanna et al., 2021) due to increased availability and quality of remotely sensed (Koenig et al., 2016; Overly et al., 2016) and in situ observations (Machguth et al., 2016; Fausto et al., 2018; Vandecrux et al., 2019, 2020). Fettweis et al. (2020) showed that the models’ ensemble mean provides the best estimate of the present-day SMB relative to observations. This is the case for the patterns in all seven regions (regional division after Mouginot et al., 2019) apart from the SE accumulation zone where large discrepancies in modelled snowfall accumulation occurred where the spread can reach 2-m water equivalent per year. Montgomery et al. (2020) confirmed this, highlighting that RCMs (MAR and RACMO) are underestimating accumulation in south-east Greenland and that models misrepresent spatial heterogeneity due to an orographically forced bias in snowfall near the coast. Further, for north-east Greenland, Karlsson et al. (2020) found RCMs underestimate snow accumulation rates by up to 35%. The regional time series show that SMB has been gradually decreasing in all seven regions (1979–2017), although the trend is less strong in central-eastern and south-eastern regions. In the south-west, north-east and north-west, SMB turns negative or close to zero after 2000 and remains above zero in other regions (medium confidence) (Figure Atlas.3 0).

Mean temperature in the Arctic is projected to continue to rise throughthe 21st century significantly higher than the global average (Figure Atlas.29 and the Interactive Atlas). For the regions NWN and NEN, see Atlas.9. The Arctic is projected to reach a 2°C annual mean warming above the 1981–2005 baseline about 25 to 50 years before the globe as a whole under RCP8.5 and RCP4.5 (Post et al., 2019). The Arctic warming may be as much as 4°C in the annual mean and 7°C in late autumn under 2°C global warming, regardless of which scenario is considered (high confidence) (Post et al., 2019).

Projections from 40 CMIP5 models of the 2014–2100 Arctic annual warming under RCP4.5 vary from 0.9°C to 6.7°C, with a multi-model mean of 3.7°C (Chylek et al., 2016). All models agree on a projected Arctic amplification (of at least 1.5 times), but they disagree on the magnitude and spatial patterns. Arctic warming trends projected by models that include a full direct and indirect aerosol effect (‘fully aerosol–cloud interactive’) are significantly higher than those projected by models without a full indirect aerosol effect (Chylek et al., 2016).

Projected Arctic warming exhibits a very pronounced seasonal cycle, with exceptionally strong warming in the winter. In projections from 30 CMIP5 models, winter warming over ARO varies regionally from 3°C to 5°C by mid-century and 5°C to 9°C by late-century under RCP4.5 (high confidence) (AMAP, 2017). Averaged over the Arctic and based on 36 CMIP5 models, winter warming is 5.8°C ± 1.5°C by mid-century and 7.1°C ± 2.3°C by 2100 under RCP4.5 (Overland et al., 2019), and an exceptionally strong warming of up to 14.1°C ± 2.9°C is projected in December under RCP8.5 (Bintanja and Krikken, 2016). Bintanja and Van Der Linden (2013) estimated the Arctic winter warming over the 21st century to exceed the summer warming by at least a factor of four, irrespective of the magnitude of the climate forcing.

Overland et al. (2014) highlighted the difference between the near-term ‘adaptation timescale’ and the long-term ‘mitigation timescale’ for the Arctic. Only in the latter half of the century do the projections under RCP4.5 and RCP8.5 noticeably separate. End-of-the-century warming is approximately twice as large under RCP8.5 demonstrating the impact of the lower emissions under RCP4.5 (high confidence) (AMAP, 2017). More specifically under the strong forcing scenario, annual mean surface air temperature in the Arctic is projected to increase by 8.5°C ± 2.1°C over the course of the 21st century (Bintanja and Andry, 2017), and emerges as a ‘new Arctic’ climate being significantly different from that of the mid-20th century (Landrum and Holland, 2020). The end-of-the-century warming (2080–2099 relative to 1980–1999, RCP8.5) can exceed 15°C in autumn and winter over the Arctic Ocean (Koenigk et al., 2015). Projections averaged over the four best-performing CMIP5 models show an Arctic annual warming of 4.1°C (RCP2.6), 5.7°C (RCP4.5) and 10.6°C (RCP8.5) by 2100 compared to 1951–1980 (Hao et al., 2018). Also, for neighbouring Arctic regions, for example NEU, WSB and ESB, temperature is projected to increase towards the end of the century under both RCP4.5 and RCP8.5 (Atlas.8 and Atlas.5.2).

The ensemble of CMIP6 shows likely greater warming compared to CMIP5 (Figure Atlas.29). There is weak agreement among the models in projected temperature change over the Arctic North Atlantic under SSPs until the mid-century, but a robust warming signal clearly emerges even there by 2100 (Interactive Atlas). Generally, the largest annual warming is simulated over the Arctic Ocean, particularly over the Barents Sea and the Kara Sea. Future warming in CORDEX RCMs and the CMIP5 models are similar (Spinoni et al., 2020). The RCM warming over the AO is smaller, while the warming over land is larger in winter and spring but smaller in summer, compared with CMIP5 (Koenigk et al., 2015).

Mean precipitation in ARO, GIC and RAR is projected to rise in a warming climate (Figure Atlas.29), with different rates for the different seasons and scenarios. For NWN and NEN, see Atlas.9. The CMIP5 multi-model mean projected precipitation increase in the Arctic is likely of the order of 50% under RCP8.5 by the end of the 21st century, which is among the highest globally (Bintanja and Selten, 2014). Over 70°N–90°N, the precipitation increase is likely 62 ± 20% and 56 ± 13% for RCP4.5 and RCP8.5 respectively. For ARO (Svalbard), the increase in annual precipitation by 2100 is estimated to be about 45% for RCP4.5 and 65% for RCP8.5 (CMIP5 ensemble median; Van der Bilt et al., 2019). However, importantly the simulated Arctic precipitation increase varies by a factor of three to four between models (Bintanja and Selten, 2014). The projected increase is strongest in late autumn and winter (Vihma et al., 2016). The interannual variability of Arctic precipitation will likely increase markedly (up to 40% over the 21st century), especially in summer (medium confidence) (Bintanja et al., 2020).

The CMIP6 projections confirm precipitation will likely increase almost everywhere in the Arctic (Interactive Atlas). The largest increase is simulated over the Barents Sea, Kara Sea and East Siberian Sea regions, and over north-east Greenland. A pronounced uncertainty in the projection exists over the Arctic North Atlantic and south Greenland. There, the precipitation signal is not significant even by the end of the 21st century and under high-emissions scenarios (RCP8.5, SSP5-8.5). Consistent with the generally higher warming in CMIP6, compared to CMIP5, the projected precipitation increase is also higher (high confidence) (Figure Atlas.29).

The Arctic mean annual precipitation sensitivity has been estimated at a 4.5% increase per degree Celsius of temperature rise, compared to a global average of 1.6–1.9% per degree Celsius of temperature rise (Bintanja and Selten, 2014) based on a set of 37 CMIP5 GCMs. Koenigk et al. (2015) stress the different precipitation sensitivity in winter (0.8 mm per month per degree Celsius of temperature rise) and summer (2 mm per month per degree Celsius of temperature rise). The pattern and amplitude of precipitation changes agree in CORDEX simulations with their driving CMIP5 models (high confidence) (Koenigk et al., 2015; Spinoni et al., 2020). However, more small-scale variations over land and coastlines, and significantly larger precipitation changes in summer are obvious in the downscaling.

Rain is projected to become the dominant form of precipitation in the Arctic region by the end of the 21st century (Bintanja, 2018). The CMIP5 models show a decrease in annual Arctic snowfall under both RCP4.5 and RCP8.5 (high confidence) (Krasting et al., 2013; Bintanja and Andry, 2017). In the central Arctic, the snowfall fraction barely remains larger than 50%, with only Greenland still having snowfall fractions larger than 80% (Bintanja and Andry, 2017). The most dramatic reductions in snowfall fraction are projected to occur over the North Atlantic and, especially, the Barents Sea.

With ongoing warming and polar amplification in the Arctic, the Greenland Ice Sheet SMB will inevitably continue to change (high confidence) (Lenaerts et al., 2019). For the ice sheet, despite large differences between model scenarios, future projections and regions agree that increasing temperatures will increase runoff which will in turn dominate the future decrease of SMB (Rae et al., 2012; van Angelen et al., 2014; Mottram et al., 2017; Hofer et al., 2020), confirming the high sensitivity of the SMB to atmospheric warming. Changes in SMB will continue to dominate future mass loss from the ice sheet, and likely even more when marine-terminating glaciers retreat onto land, and solid ice discharge is reduced (Vizcaino, 2014; Lenaerts et al., 2019).

It is very likely that the Arctic has warmed at more than twice the global rate over the past 50 years and likely that annual precipitation has increased with the highest increases during the cold season. This is based on various Arctic amplification processes, in particular, a combination of several feedback-related processes such as sea ice and snow-cover albedo, poleward energy transports, and water-vapour-cloud-radiation feedbacks. The frequency of rainfall increased over the Arctic by 2.7–5.4% over the 2000–2016 period with more frequent rainfall events being reported for northern Europe and Svalbard (medium confidence).

The CMIP5 models reproduce the observed Arctic warming over the past century but overestimate the amplified Arctic warming in the recent decades (medium confidence). Arctic CORDEX simulations show adequate skill in capturing regional temperature and precipitation patterns and precipitation extremes (high confidence). SMB models have improved due to increased availability and quality of remotely sensed and in situ observations, and an ensemble mean of SMB model simulations provides the best estimate of the present-day SMB (medium confidence).

It is very likely that the Arctic annual mean temperature and precipitation will continue to increase, reaching around 11.5°C ± 3.4°C and 49 ± 19% over the 2081–2100 period (with respect to a 1995–2014 baseline) under the SSP5-8.5 scenario or 4.0°C ± 2.5°C and 17 ± 11% under the SSP1-2.6 scenario. These CMIP6 results show likely higher Arctic annual mean temperatures compared to CMIP5 for a given time-period and emissions scenario, though the projections are consistent for global warming levels.