Editing IPCC:AR6/WGI/Chapter-11 (section)

=== 11.3.3 Model Evaluation ===

<div id="h2-26-siblings" class="h2-siblings"></div>

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

Initial analyses of CMIP6 simulations (H. [[#Chen--2020|]] [[#Chen--2020|Chen et al., 2020]] ; [[#Di%20Luca--2020a|Di Luca et al., 2020a]] ; [[#Kim--2020|Kim et al., 2020]] ; [[#Thorarinsdottir--2020|Thorarinsdottir et al., 2020]] ; [[#Wehner--2020|Wehner et al., 2020]] ; [[#Li--2021|Li et al., 2021]] ) indicate that the CMIP6 models perform similarly to the CMIP5 models regarding biases in hot and cold extremes. In general, CMIP5 and CMIP6 historical simulations are similar in their performance in simulating the observed climatology of extreme temperatures ( ''high confidence'' ) ''.'' The general warm bias in hot extremes and cold bias in cold extremes reported for CMIP5 models ( [[#Kharin--2013|Kharin et al., 2013]] ; [[#Sillmann--2013a|Sillmann et al., 2013a]] ) remain in CMIP6 models ( [[#Di%20Luca--2020a|Di Luca et al., 2020a]] ). However, there is some evidence that CMIP6 models better represent some of the underlying processes leading to extreme temperatures, such as seasonal and diurnal variability and synoptic-scale variability ( [[#Di%20Luca--2020a|Di Luca et al., 2020a]] ). Whether these improvements are sufficient to enhance our understanding of past changes, or to reduce uncertainties in future projections, remains unclear. The relative error estimates in the simulation of various indices of temperature extremes in the available CMIP6 models show that no single model performs the best on all indices, and the multi-model ensemble seems to outperform any individual model due to its reduction in systematic bias ( [[#Kim--2020|Kim et al., 2020]] ). Figure 11.10 show errors in the 1979–2014 average annual TXx and annual TNn simulated by available CMIP6 models in comparison with HadEX3 and ERA5 ( [[#Kim--2020|Kim et al., 2020]] ; [[#Wehner--2020|Wehner et al., 2020]] ; [[#Li--2021|Li et al., 2021]] ). While the magnitude of the model error depends on the reference dataset, the model evaluations drawn from different reference datasets are quite similar. In general, models reproduce the spatial patterns and magnitudes of both cold and hot temperature extremes quite well. There are also systematic biases. Hot extremes tend to be too cool in mountainous and high-latitude regions, but too warm in the eastern USA and South America. For cold extremes, CMIP6 models are too cool, except in north-eastern Eurasia and the southern mid-latitudes. Errors in seasonal mean temperatures are uncorrelated with errors in extreme temperatures and are often of opposite sign ( [[#Wehner--2020|Wehner et al., 2020]] ).

<div id="_idContainer049" class="Basic-Text-Frame"></div>

[[File:77e630c252e7fc746ad65fc1d186f932 IPCC_AR6_WGI_Figure_11_10.png]]

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

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

Several regional climate models (RCMs) have also been evaluated in terms of their performance in simulating the climatology of extremes in various regions of the Coordinated Regional Downscaling Experiment (CORDEX) ( [[#Giorgi--2009|Giorgi et al., 2009]] ), especially in East Asia ( [[#Ji--2015|Ji and Kang, 2015]] ; [[#Yu--2015|Yu et al., 2015]] ; [[#Park--2016|Park et al., 2016]] ; [[#Bucchignani--2017|Bucchignani et al., 2017]] ; [[#Gao--2017a|Gao et al., 2017a]] ; [[#Niu--2018|Niu et al., 2018]] ; Y. [[#Sun--2018b|]] [[#Sun--2018|Sun et al., 2018]] b ; [[#Wang--2019|Wang et al., 2019]] ), Europe ( [[#Vautard--2013|Vautard et al., 2013]] , 2021; [[#Smiatek--2016|Smiatek et al., 2016]] ; [[#Gaertner--2018|Gaertner et al., 2018]] ; [[#Cardoso--2019|Cardoso et al., 2019]] ; [[#Lorenz--2019|Lorenz et al., 2019]] ; [[#Jacob--2020|Jacob et al., 2020]] ; [[#Kim--2020|Kim et al., 2020]] ), and Africa (J. [[#Kim--2014|]] [[#Kim--2014|Kim et al., 2014]] ; [[#Diallo--2015|Diallo et al., 2015]] ; [[#Dosio--2017|Dosio, 2017]] ; [[#Samouly--2018|Samouly et al., 2018]] ; [[#Mostafa--2019|Mostafa et al., 2019]] ). Compared to GCMs, RCM simulations show an added value in simulating temperature-related extremes, though this depends on topographical complexity and the parameters employed (see [[IPCC:Wg1:Chapter:Chapter-10#10.3.3|Section 10.3.3]] ). The improvement with resolution is noted in East Asia ( [[#Park--2016|Park et al., 2016]] ; W. [[#Zhou--2016|]] [[#Zhou--2016|Zhou et al., 2016]] ; [[#Shi--2017|Shi et al., 2017]] ; [[#Hui--2018|Hui et al., 2018]] ). However, in the European CORDEX ensemble, different aerosol climatologies with various degrees of complexity were used in projections ( [[#Bartók--2017|Bartók et al., 2017]] ; [[#Lorenz--2019|Lorenz et al., 2019]] ) and the land surface models used in the RCMs do not account for physiological CO <sub>2</sub> effects on photosynthesis leading to enhanced water-use efficiency and decreased evapotranspiration ( [[#Schwingshackl--2019|Schwingshackl et al., 2019]] ), which could lead to biases in the representation of temperature extremes in these projections ( [[#Boé--2020|Boé et al., 2020]] ). In addition, there are key cold biases in temperature extremes over areas with complex topography ( [[#Niu--2018|Niu et al., 2018]] ). Over North America, 12 RCMs were evaluated over the ARCTIC-CORDEX region ( [[#Diaconescu--2018|Diaconescu et al., 2018]] ). Models performed well at simulating climate indices related to mean air temperature and hot extremes over most of the Canadian Arctic, with the exception of the Yukon region where models displayed the largest biases related to topographic effects. Two RCMs were evaluated against observed extremes indices over North America over the period 1989–2009, with a cool bias in minimum temperature extremes shown in both RCMs ( [[#Whan--2016|Whan and Zwiers, 2016]] ). The most significant biases are found in TXx and TNn, with fewer differences in the simulation of annual minimum daily maximum temperature (TXn) and annual maximum daily minimum temperature (TNx) in Central and Western North America. Over Central and South America, maximum temperatures from the Eta RCM are generally underestimated, although hot days, warm nights, and heatwaves are increasing in the period 1961–1990, in agreement with observations ( [[#Chou--2014b|Chou et al., 2014b]] ; [[#Tencer--2016|Tencer et al., 2016]] ; [[#Bozkurt--2019|Bozkurt et al., 2019]] ).

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

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

<div id="11.3.4" class="h2-container"></div>

<span id="detection-and-attribution-event-attribution"></span>