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

==== Atlas.7.2.3 Assessment of Model Performance ====

<div id="h3-48-siblings" class="h3-siblings"></div>

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--2009|Giorgi et al., 2009]] ; [[#Boulanger--2016|Boulanger et al., 2016]] ; [[#Ambrizzi--2019|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--2013b|Blázquez and Nuñez, 2013b]] ; [[#Gulizia--2013|Gulizia et al., 2013]] ; [[#Joetzjer--2013|Joetzjer et al., 2013]] ; [[#Jones--2013|Jones and Carvalho, 2013]] ; [[#Torres--2013|Torres and Marengo, 2013]] ; [[#Gulizia--2015|Gulizia and Camilloni, 2015]] ; [[#Zazulie--2017|Zazulie et al., 2017]] ; [[#Abadi--2018|Abadi et al., 2018]] ; [[#Barros--2018|Barros and Doyle, 2018]] ; [[#Solman--2019|Solman and Blázquez, 2019]] ; [[#Fan--2020|Fan et al., 2020]] ; [[#Rivera--2020|Rivera and Arnould, 2020]] ; [[#Teichmann--2021|Teichmann et al., 2021]] ). During the dry season, precipitation is underestimated in most models over Amazonia ( ''medium evidence, high agreement'' ) ( [[#Torres--2013|Torres and Marengo, 2013]] ; [[#Yin--2013|Yin et al., 2013]] ; [[#Solman--2019|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--2015|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--2017|Zazulie et al., 2017]] ; [[#Rivera--2020|Rivera and Arnould, 2020]] ; [[#Díaz--2021|Díaz et al., 2021]] ). Most models show a dry bias over SES ( [[#Díaz--2017|Díaz and Vera, 2017]] ; [[#Barros--2018|Barros and Doyle, 2018]] ; [[#Solman--2019|Solman and Blázquez, 2019]] ; [[#Díaz--2021|Díaz et al., 2021]] ) associated with an underestimation of the northern flow that brings water vapour into the region ( ''medium confidence'' ) ( [[#Gulizia--2013|Gulizia et al., 2013]] ; [[#Zazulie--2017|Zazulie et al., 2017]] ; [[#Barros--2018|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--2013|Joetzjer et al., 2013]] ; [[#Gulizia--2015|Gulizia and Camilloni, 2015]] ; [[#Díaz--2017|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--2016|Tedeschi and Collins, 2016]] ; [[#Perry--2020|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--2014|Llopart et al., 2014]] ; [[#Sánchez--2015|Sánchez et al., 2015]] ; [[#Falco--2019|Falco et al., 2019]] ; [[#Solman--2019|Solman and Blázquez, 2019]] ; [[#Ciarlo%60--2021|Ciarlo` et al., 2021]] ; [[#Teichmann--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--2013|Solman et al., 2013]] ; [[#Falco--2019|Falco et al., 2019]] ) and seasonal cycles over the different climate regimes, including the main SAmerM features ( ''high confidence'' ) ( [[#Jacob--2012|Jacob et al., 2012]] ; [[#Solman--2013|Solman, 2013]] ; [[#Llopart--2014|Llopart et al., 2014]] ; [[#Reboita--2014|Reboita et al., 2014]] ; [[#de%20Jesus--2016|de Jesus et al., 2016]] ; [[#Lyra--2018|Lyra et al., 2018]] ; [[#Bozkurt--2019|Bozkurt et al., 2019]] ; [[#Ashfaq--2021|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--2012|Jacob et al., 2012]] ; [[#Solman--2013|Solman et al., 2013]] ; [[#Bozkurt--2019|Bozkurt et al., 2019]] ; [[#Falco--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--2013|Solman et al., 2013]] ; [[#Falco--2019|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--2013|Solman et al., 2013]] ; [[#Reboita--2014|Reboita et al., 2014]] ; [[#Solman--2016|Solman, 2016]] ; [[#Falco--2019|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--2018|Lyra et al., 2018]] ; [[#Bozkurt--2019|Bozkurt et al., 2019]] ; [[#Bettolli--2021|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--2010|Souvignet et al., 2010]] ; [[#Mendes--2014|Mendes et al., 2014]] ; [[#Palomino-Lemus--2015|Palomino-Lemus et al., 2015]] , 2017, 2018; [[#Soares%20dos%20Santos--2016|Soares dos Santos et al., 2016]] ; [[#Troin--2016|Troin et al., 2016]] ; [[#Borges--2017|Borges et al., 2017]] ; [[#Bettolli--2018|Bettolli and Penalba, 2018]] ; [[#Araya-Osses--2020|Araya-Osses et al., 2020]] ; [[#Bettolli--2021|Bettolli et al., 2021]] ). Temperature simulations are fewer but show added value to GCM simulations ( ''medium evidence'' , ''high agreement'' ) ( [[#Souvignet--2010|Souvignet et al., 2010]] ; [[#Borges--2017|Borges et al., 2017]] ; [[#Bettolli--2018|Bettolli and Penalba, 2018]] ; [[#Araya-Osses--2020|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--2018|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'' ).

<div id="Atlas.7.2.4" class="h3-container"></div>

<span id="atlas.7.2.4-assessment-and-synthesis-of-projections"></span>