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

=== Atlas.9.3 Assessment of Model Performance ===

<div id="h2-39-siblings" class="h2-siblings"></div>

CMIP6 models have been evaluated in the literature, although these studies have not included the full set of CMIP6 simulations. [[#Fan--2020|Fan et al. (2020)]] established on a continental basis for North America that temperature pattern correlations were quite accurate. [[#Thorarinsdottir--2020|Thorarinsdottir et al. (2020)]] compared maximum and minimum temperatures over Europe and North America with several observational datasets and found that the CMIP6 ensemble agreed better with ERA5 data than did CMIP5. [[#Srivastava--2020|Srivastava et al. (2020)]] evaluated historical CMIP6 simulations for precipitation, comparing them with several observational datasets over the continental US. Most models show a wet bias over the eastern half of the continental USA and the north-east region, while dry biases persist in the central part of the country ( [[#Akinsanola--2020a|Akinsanola et al., 2020a]] ; [[#Almazroui--2021|Almazroui et al., 2021]] ). The spatial structure of biases is similar in CMIP5 and CMIP6, but with lower magnitudes in CMIP6. [[#Agel--2020|Agel and Barlow (2020)]] examined 16 CMIP6 models over the north-eastern USA for precipitation and did not find a distinct improvement over CMIP5, although they did find the higher-resolution models tended to perform better. On the basis of the evidence so far, there is ''medium confidence'' that CMIP6 models are improved compared to CMIP5 in terms of biases in mean temperature and precipitation over North America.

North America has been extensively used as a test bed for regional climate model (RCM) experiments, such as the North American Regional Climate Change Assessment Program (NARCCAP; [[#Mearns--2009|Mearns et al., 2009]] ), the MultiRCM Ensemble Downscaling (MRED; [[#Yoon--2012|Yoon et al., 2012]] ), and NA-CORDEX ( [[#Bukovsky--2020|Bukovsky and Mearns, 2020]] ). Therefore, much performance evaluation has been conducted with a focus on specific climate features in North America. For the North American Monsoon region, multi-model performance evaluation ( [[#Bukovsky--2013|Bukovsky et al., 2013]] ; [[#Tripathi--2013|Tripathi and Dominguez, 2013]] ; [[#Cerezo-Mota--2016|Cerezo-Mota et al., 2016]] ) or a single-member performance ( [[#Lucas-Picher--2013|Lucas-Picher et al., 2013]] ; [[#Martynov--2013|Martynov et al., 2013]] ; [[#Šeparović--2013|Šeparović et al., 2013]] ) demonstrated the added value of RCMs, particularly more recent CORDEX simulations, through improved simulation of summer precipitation and the climatological winter storm tracks across the western USA. NA-CORDEX simulations were more successful at reproducing weather types compared to a single model-based large perturbed-physics ensemble ( [[#Prein--2019|Prein et al., 2019]] ). The application of a complex evaluation tool to the full suite of NA-CORDEX simulations found that the higher-resolution simulations (25 km compared with 50 km) of precipitation were improved, particularly for daily intensity ( [[#Gibson--2019|Gibson et al., 2019]] ).

However, deficiencies have also been reported. For example, excessive storm occurrence over the east coast of North America was found ( [[#Poan--2018|Poan et al., 2018]] ), and amplitude in the simulated annual cycle was generally excessive in NA-CORDEX simulations. RCMs tend to produce more (less) precipitation over mountains (the coastal plains; [[#Cerezo-Mota--2016|Cerezo-Mota et al., 2016]] ) and winter precipitation in the western USA had large positive biases in all RegCM simulations, regardless of the driving GCM ( [[#Mahoney--2021|Mahoney et al., 2021]] ).

Recently, convective-permitting RCMs have been used to simulate North American climate features and generated better simulations of precipitation. For example, summer precipitation over the south-western USA was improved due to better representation of organized mesoscale convective systems at the sub-daily scale ( [[#Castro--2012|Castro et al., 2012]] ; [[#Liu--2017|Liu et al., 2017]] ; [[#Prein--2017a|Prein et al., 2017a]] ; [[#Pal--2019|Pal et al., 2019]] ), the diurnal cycle of convection ( [[#Nesbitt--2008|Nesbitt et al., 2008]] ), and in terms of means (and extremes) for the north-eastern USA ( [[#Komurcu--2018|Komurcu et al., 2018]] ).

Recent studies have examined RCMs’ simulation of SWE, a quantity of primary importance notably for hydrological modelling, though its ground measurements are restricted by relatively high time and monetary costs ( [[#Smith--2017|Smith et al., 2017]] ; [[#Odry--2020|Odry et al., 2020]] ) which limit model assessment. Also, studies often emphasize that a false impression of model skill for SWE can be obtained by compensating temperature and precipitation biases. Assessment frameworks have dealt with these issues by considering observational uncertainty ( [[#Mccrary--2017|Mccrary et al., 2017]] ) and by decomposing SWE biases into their contributing processes ( [[#Rhoades--2018|Rhoades et al., 2018]] ; [[#Xu--2019|Xu et al., 2019]] ). SWE biases exceed observational uncertainty in several 50-km reanalysis-driven NARCCAP simulations over several regions, for all cold months ( [[#Mccrary--2017|Mccrary et al., 2017]] ). Analyses of NA-CORDEX simulations show that refining spatial resolution from 50 to 12 km improves certain (but not all) aspects of SWE, stemming from improved mean precipitation and topography-related temperature ( [[#Xu--2019|Xu et al., 2019]] ). Similarly an assessment of RCM simulations of freezing rain over eastern Canada found a mix of improved and deteriorated aspects from higher resolution ( [[#St-Pierre--2019|St-Pierre et al., 2019]] ).

<div id="Atlas.9.4" class="h2-container"></div>

<span id="atlas.9.4-assessment-and-synthesis-of-projections"></span>