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

=== 11.6.3 Model Evaluation ===

<div id="h2-41-siblings" class="h2-siblings"></div>

<div id="11.6.3.1" class="h3-container"></div>

<span id="precipitation-deficits-2"></span>
==== 11.6.3.1 Precipitation Deficits ====

<div id="h3-13-siblings" class="h3-siblings"></div>

ESMs generally show limited performance and large spread in identifying precipitation deficits and associated long-term trends in comparison with observations ( [[#Nasrollahi--2015|Nasrollahi et al., 2015]] ). Meteorological drought trends in the CMIP5 ensemble showed substantial disagreements compared with observations ( [[#Orlowsky--2013|Orlowsky and Seneviratne, 2013]] ; [[#Knutson--2018|Knutson and Zeng, 2018]] ) including a tendency to overestimate drying, in particular in mid- to high latitudes ( [[#Knutson--2018|Knutson and Zeng, 2018]] ). The CMIP6 models display a better performance in reproducing long-term precipitation trends or seasonal dynamics in some studies in Southern South America ( [[#Rivera--2020|Rivera and Arnould, 2020]] ), East Asia ( [[#Xin--2020|Xin et al., 2020]] ), southern Asia ( [[#Gusain--2020|Gusain et al., 2020]] ), and south-western Europe ( [[#Peña-Angulo--2020b|Peña-Angulo et al., 2020b]] ), but there is still too ''limited evidence'' to allow for an assessment of possible differences in performance between CMIP5 and CMIP6. Furthermore, ESMs are generally found to underestimate the severity of precipitation deficits and the dry day frequencies in comparison to observations ( [[#Fantini--2018|Fantini et al., 2018]] ; [[#Ukkola--2018|Ukkola et al., 2018]] ). This is probably related to shortcomings in the simulation of persistent weather events in the mid-latitudes ( [[IPCC:Wg1:Chapter:Chapter-10#10.3.3.3|Section 10.3.3.3]] ). ESMs also show a tendency to underestimate precipitation-based drought persistence at monthly to decadal time scales ( [[#Ault--2014|Ault et al., 2014]] ; [[#Moon--2018|Moon et al., 2018]] ). The overall inter-model spread in the projected frequency of precipitation deficits is also substantial ( [[#Touma--2015|Touma et al., 2015]] ; [[#Zhao--2016|Zhao et al., 2016]] ; [[#Engström--2018|Engström and Keellings, 2018]] ). Moreover, there are spatial differences in the spread, which is higher in the regions where enhanced drought conditions are projected and under high-emissions scenarios ( [[#Orlowsky--2013|Orlowsky and Seneviratne, 2013]] ). Nonetheless, some event attribution studies have concluded that droughts at regional scales can be adequately simulated by some climate models ( [[#Schaller--2016|Schaller et al., 2016]] ; [[#Otto--2018c|Otto et al., 2018c]] ).

<div id="11.6.3.2" class="h3-container"></div>

<span id="atmospheric-evaporative-demand-2"></span>
==== 11.6.3.2 Atmospheric Evaporative Demand ====

<div id="h3-14-siblings" class="h3-siblings"></div>

There is only ''limited evidence'' on the evaluation of AED in state-of-the-art ESMs, which is performed on externally computed AED, based on model output ( [[#Scheff--2015|Scheff and Frierson, 2015]] ; [[#Liu--2016|Liu and Sun, 2016]] , 2017). An evaluation of average AED in 17 CMIP5 ESMs for 1981–1999 based on potential evaporation show that the models’ spatial patterns resemble the observations, but the magnitude of potential evaporation displays strong divergence among models globally and regionally ( [[#Scheff--2015|Scheff and Frierson, 2015]] ). The evaluation of AED in 12 CMIP5 ESMs with pan evaporation observations in East Asia for 1961–2000 ( [[#Liu--2016|Liu and Sun, 2016]] , 2017) show that the ESMs capture seasonal cycles well, but that regional AED averages are underestimated due to biases in the meteorological variables controlling the aerodynamic and radiative components of AED. The CMIP5 ESMs also show a strong underestimation of atmospheric drying trends compared to reanalysis data ( [[#Douville--2017|Douville and Plazzotta, 2017]] ).

<div id="11.6.3.3" class="h3-container"></div>

<span id="soil-moisture-deficits-2"></span>
==== 11.6.3.3 Soil Moisture Deficits ====

<div id="h3-15-siblings" class="h3-siblings"></div>

The performance of climate models for representing soil moisture deficits shows more uncertainty than for precipitation deficits since, in addition to the uncertainties related to cloud and precipitation processes, there is uncertainty related to the representation of complex soil hydrological and boundary-layer processes ( [[#van%20den%20Hurk--2011|van den Hurk et al., 2011]] ; [[#Lu--2019|Lu et al., 2019]] ; [[#Quintana-Seguí--2020|Quintana-Seguí et al., 2020]] ). Another limitation is the lack of observations, particularly for soil moisture, in most regions ( [[#11.6.2.3|Section 11.6.2.3]] ) and the paucity of land surface property data to parametrize land surface models, in particular soil types, soil properties and depth ( [[#Xia--2015|Xia et al., 2015]] ). The spatial resolution of models is an additional limitation since the representation of some land–atmosphere feedbacks and topographic effects requires detailed resolution ( [[#Nicolai-Shaw--2015|Nicolai-Shaw et al., 2015]] ; Van Der Linden et al., 2019). In addition to climate models, land surface and hydrological models are also used to derive historical and projected trends in soil moisture and related land water variables ( [[#Albergel--2013|Albergel et al., 2013]] ; [[#Cheng--2015|Cheng et al., 2015]] ; [[#Gu--2019b|Gu et al., 2019b]] ; [[#Padrón--2020|Padrón et al., 2020]] ; [[#Markonis--2021|Markonis et al., 2021]] ; [[#Pokhrel--2021|Pokhrel et al., 2021]] ).

Overall, there are contrasting results on the performance of land surface models and climate models in representing soil moisture. Some studies suggest that soil moisture anomalies are well captured by land surface models driven with observation-based forcing ( [[#Dirmeyer--2006|Dirmeyer et al., 2006]] ; [[#Albergel--2013|Albergel et al., 2013]] ; [[#Xia--2014|Xia et al., 2014]] ; [[#Balsamo--2015|Balsamo et al., 2015]] ; [[#Reichle--2017|Reichle et al., 2017]] ; [[#Spennemann--2020|Spennemann et al., 2020]] ), but other studies report limited agreement in the representation of interannual soil moisture variability ( [[#Stillman--2016|Stillman et al., 2016]] ; [[#Yuan--2017|Yuan and Quiring, 2017]] ; [[#Ford--2019|Ford and Quiring, 2019]] ) and noticeable seasonal differences in model skill in some regions ( [[#Xia--2014|Xia et al., 2014]] , 2015). Models with good skill can nonetheless display biases in absolute soil moisture ( [[#Xia--2014|Xia et al., 2014]] ; [[#Gu--2019a|Gu et al., 2019a]] ), but these are not necessarily of relevance for the simulation of surface water fluxes and drought anomalies ( [[#Koster--2009|Koster et al., 2009]] ). There is also substantial inter-model spread ( [[#Albergel--2013|Albergel et al., 2013]] ), particularly for the root-zone soil moisture ( [[#Berg--2017a|Berg et al., 2017a]] ).

Regarding the performance of regional and global climate models, an evaluation of an ensemble of RCM simulations for Europe ( [[#Stegehuis--2013|Stegehuis et al., 2013]] ) shows that these models display overly strong drying in early summer, resulting in an excessive decrease of latent heat fluxes, with potential implications for more severe droughts in dry environments ( [[#Teuling--2018|Teuling, 2018]] ; [[#van%20Der%20Linden--2019|van Der Linden et al., 2019]] ). Compared with a range of observational ET estimates, CMIP5 models show an overestimation of ET on annual scale, but an ET underestimation in boreal summer in many Northern Hemisphere mid-latitude regions, also suggesting a tendency towards excessive soil drying ( [[#Mueller--2014|Mueller and Seneviratne, 2014]] ), consistent with identified biases in soil-moisture–temperature coupling ( [[#Donat--2018|Donat et al., 2018]] ; [[#Vogel--2018|Vogel et al., 2018]] ; [[#Selten--2020|Selten et al., 2020]] ). Land surface models used in ESMs display a bias in their representation of the sensitivity of interannual land carbon uptake to soil moisture conditions, which appears related to a limited range of soil moisture variations compared to observations ( [[#Humphrey--2018|Humphrey et al., 2018]] ).

For future projections, the spread of soil moisture outputs among different ESMs is more important than internal variability and scenario uncertainty, and the bias is strongly related to the sign of the projected change ( [[#Ukkola--2018|Ukkola et al., 2018]] ; [[#Lu--2019|Lu et al., 2019]] ; [[#Selten--2020|Selten et al., 2020]] ). The CMIP5 ESMs that project more drying and warming in mid-latitude regions show a substantial bias in soil-moisture–temperature coupling ( [[#Donat--2018|Donat et al., 2018]] ; [[#Vogel--2018|Vogel et al., 2018]] ). Although CMIP6 and CMIP5 simulations for soil moisture changes are similar overall, some differences are found in projections in a few regions ( [[#11.9|Section 11.9]] ; [[#Cook--2020|Cook et al., 2020]] ). There is still ''limited evidence'' to assess whether there are substantial differences in model performance in the two ensembles, but improvements in modelling aspects relevant for soil moisture have been reported for precipitation ( [[#11.6.3.2|Section 11.6.3.2]] ), and a better performance has been found in CMIP6 for the representation of long-term trends in soil moisture in continental USA ( [[#Yuan--2021|Yuan et al., 2021]] ). Despite the mentioned model limitations, the representation of soil moisture processes in ESMs uses physical and biological understanding of the underlying processes, which can well represent the temporal anomalies associated with temporal variability and trends in climate. In summary, there is ''medium confidence'' in the representation of soil moisture deficits in ESMs and related land surface and hydrological models.

<div id="11.6.3.4" class="h3-container"></div>

<span id="hydrological-deficits-2"></span>
==== 11.6.3.4 Hydrological Deficits ====

<div id="h3-16-siblings" class="h3-siblings"></div>

Streamflow and groundwater are not directly simulated by ESMs, which only simulate runoff, but they are generally represented in hydrological models ( [[#Prudhomme--2014|Prudhomme et al., 2014]] ; [[#Giuntoli--2015|Giuntoli et al., 2015]] ), which are typically driven in a stand-alone manner by observed or simulated climate forcing. The simulation of hydrological deficits is much more problematic than the simulation of mean streamflow or peak flows ( [[#Fundel--2013|Fundel et al., 2013]] ; [[#Stoelzle--2013|Stoelzle et al., 2013]] ; [[#Velázquez--2013|Velázquez et al., 2013]] ; [[#Staudinger--2015|Staudinger et al., 2015]] ), since models tend to be too responsive to the climate forcing and do not satisfactorily capture low flows ( [[#Tallaksen--2014|Tallaksen and Stahl, 2014]] ). Simulations of hydrological drought metrics show uncertainties related to the contribution of both GCMs and hydrological models ( [[#Bosshard--2013|Bosshard et al., 2013]] ; [[#Giuntoli--2015|Giuntoli et al., 2015]] ; [[#Samaniego--2017|Samaniego et al., 2017]] ; [[#Vetter--2017|Vetter et al., 2017]] ), but hydrological models forced by the same climate input data also show a large spread ( [[#van%20Huijgevoort--2013|van Huijgevoort et al., 2013]] ; [[#Ukkola--2018|Ukkola et al., 2018]] ). At the catchment scale, the hydrological model uncertainty is higher than both GCM and downscaling uncertainty ( [[#Vidal--2016|Vidal et al., 2016]] ), and the hydrological models show issues in representing drought propagation throughout the hydrological cycle ( [[#Barella-Ortiz--2019|Barella-Ortiz and Quintana Seguí, 2019]] ). A study on the evaluation of streamflow droughts in seven global (hydrological and land surface) models compared with observations in near-natural catchments of Europe showed a substantial spread among models, an overestimation of the number of drought events, and an underestimation of drought duration and drought-affected area ( [[#Tallaksen--2014|Tallaksen and Stahl, 2014]] ).

<div id="11.6.3.5" class="h3-container"></div>

<span id="atmospheric-based-drought-indices-2"></span>
==== 11.6.3.5 Atmospheric-based Drought Indices ====

<div id="h3-17-siblings" class="h3-siblings"></div>

A number of studies have analysed the ability of models to capture drought severity and trends based on climatic drought indices. Given the limitations of ESMs in reproducing the dynamic of precipitation deficits and AED (11.6.3.1, 11.6.3.2), atmospheric-based drought indices derived from ESM data for these two variables are also affected by uncertainties and biases. A comparison of historical trends in PDSI-PM for 1950–2014 derived from CMIP3 and CMIP5, with respective estimates derived from observations ( [[#Dai--2017|Dai and Zhao, 2017]] ) show a similar behaviour at global scale (long-term decrease), but low spatial agreement in the trends except in a few regions (Mediterranean, South Asia, north-western USA). In future projections, there is an important spread in PDSI-PM and SPEI-PM among different models ( [[#Cook--2014a|Cook et al., 2014a]] ).

<div id="11.6.3.6" class="h3-container"></div>

<span id="synthesis-for-different-drought-types-1"></span>
==== 11.6.3.6 Synthesis for Different Drought Types ====

<div id="h3-18-siblings" class="h3-siblings"></div>

The performance of ESMs used to assessed changes in variables related to meteorological droughts, agricultural and ecological droughts, and hydrological droughts, shows the presence of biases and uncertainties compared to observations, but there is ''medium confidence'' in their overall performance for assessing drought projections given process understanding. Given the substantial inter-model spread documented for all related variables, the consideration of multi-model projections increases the confidence of model-based assessments, with only ''low confidence'' in assessments based on single models.

In summary, the evaluation of ESMs, land surface and hydrological models for the simulation of droughts is complex, due to the regional scale of drought trends, their overall low signal-to-noise ratio, and the lack of observations in several regions, in particular for soil moisture and streamflow. There is ''medium confidence'' in the ability of ESMs to simulate trends and anomalies in precipitation deficits and AED, and also ''medium confidence'' in the ability of ESMs and hydrological models to simulate trends and anomalies in soil moisture and streamflow deficits, on global and regional scales.

<div id="11.6.4" class="h2-container"></div>

<span id="detection-and-attribution-event-attribution-3"></span>