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

==== Atlas.11.1.3 Assessment of Model Performance ====

<div id="h3-59-siblings" class="h3-siblings"></div>

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 [[IPCC:Wg1:Chapter:Chapter-9#9.4.2.2|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--2020a|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--2015|Agosta et al., 2015]] ; [[#Bracegirdle--2015|Bracegirdle et al., 2015]] ). A positive bias in near-surface temperature over the Antarctic plateau is seen in CMIP5 models ( [[#Lenaerts--2016|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--2017|Palerme et al., 2017]] ; [[#Roussel--2020|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--2020|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--2017|Palerme et al., 2017]] ; [[#Gorte--2020|Gorte et al., 2020]] ). GCM simulations of surface snow-melt processes are either of variable quality, with extremely simple representatons, or non-existent ( [[#Agosta--2015|Agosta et al., 2015]] ; [[#Trusel--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--2021|Kittel et al., 2021]] ). Since CMIP5, representation of snow ( [[#Lenaerts--2016|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--2020|Schneider et al., 2020]] ), though surface melting is still considerably overestimated compared to RCMs and satellite products ( [[#Trusel--2015|Trusel et al., 2015]] ; [[#Lenaerts--2016|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--2019|Retamales-Muñoz et al., 2019]] ; [[#Tetzner--2019|Tetzner et al., 2019]] ; [[#Dong--2020|Dong et al., 2020]] ; [[#Gorodetskaya--2020|Gorodetskaya et al., 2020]] ). The ability of the reanalyses to simulate precipitation and SMB is more variable; they generally overestimate the latter ( [[#Gossart--2019|Gossart et al., 2019]] ; [[#Roussel--2020|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%20Wessem--2018|van Wessem et al., 2018]] ), MAR ( [[#Agosta--2019|Agosta et al., 2019]] ; [[#Kittel--2021|Kittel et al., 2021]] ), COSMO-CLM2 ( [[#Souverijns--2019|Souverijns et al., 2019]] ), HIRHAM5 ( [[#Lucas-Picher--2012|Lucas-Picher et al., 2012]] ) and MetUM ( [[#Walters--2017|Walters et al., 2017]] ; [[#Mottram--2021|Mottram et al., 2021]] ) – and one stretched-grid GCM – ARPEGE ( [[#Beaumet--2019|Beaumet et al., 2019]] ). RCM simulations forced by ERA-Interim agree well with automatic weather station temperatures, with high correlation (R <sup>2</sup> &gt; 0.9) and low bias (&lt;1.5°C) except for high-resolution HIRHAM5 (–2.1°C) and MetUM (–3.4°C), which are not internally nudged models ( [[#Mottram--2021|Mottram et al., 2021]] ). RCMs generally underestimate the observed SMB but with biases lower than 20%, except for COSMO-CLM2 at lower elevations (&lt;1200 m) and HIRHAM5 and MetUM at higher elevations (&gt;2200 m) ( [[#Mottram--2021|Mottram et al., 2021]] ). These RCM simulations lead to estimates of the grounded Antarctic Ice Sheet SMB ranging from 2133 Gt yr <sup>–1</sup> to 2328 Gt yr <sup>–1</sup> when considering the four simulations compatible with the IMBIE2 Antarctic total mass budget (IMBIE team et al., 2018; [[#Mottram--2021|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--2019|Agosta et al., 2019]] ), cloud-precipitation microphysical processes ( [[#van%20Wessem--2018|van Wessem et al., 2018]] ) and snowpack modelling ( [[#Mottram--2021|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--2021|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%20Wessem--2018|van Wessem et al., 2018]] ; [[#Bozkurt--2020|Bozkurt et al., 2020]] ; [[#Donat-Magnin--2020|Donat-Magnin et al., 2020]] ; [[#Elvidge--2020|Elvidge et al., 2020]] ), and strength of katabatic winds ( [[#Bintanja--2014|Bintanja et al., 2014]] ; [[#Souverijns--2019|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--2019|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--2014|Grosvenor et al., 2014]] ; [[#Elvidge--2015|Elvidge et al., 2015]] , 2020; [[#Elvidge--2016|Elvidge and Renfrew, 2016]] ; [[#King--2017|King et al., 2017]] ; [[#Turton--2017|Turton et al., 2017]] ; [[#Bozkurt--2018b|Bozkurt et al., 2018b]] ; [[#Hines--2019|Hines et al., 2019]] ; Vignon et al., 2019; [[#Gilbert--2020|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--2020|Gorte et al., 2020]] ; [[#Kittel--2021|Kittel et al., 2021]] ). Run-time bias adjustment in atmospheric GCMs (Cross-Chapter Box 10.2; [[#Krinner--2019|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--2019|Krinner et al., 2019]] ).

<div id="Atlas.11.1.4" class="h3-container"></div>

<span id="atlas.11.1.4-assessment-and-synthesis-of-projections"></span>