Editing IPCC:AR6/WGI/Chapter-10 (section)

=== 10.3.3 Model Performance and Added Value in Simulating and Projecting Regional Climate ===

<div id="h2-17-siblings" class="h2-siblings"></div>

Assessing model performance is a prerequisite for building confidence in regional climate projections. This subsection assesses the performance of different model types at simulating regional climate and climate change. The subsection builds on the assessment of global model performance in Chapter 3, and complements the model assessment in Chapter 8, which focuses on the water cycle, and the Atlas.

While the ability of global models to simulate large-scale indicators of climate change has improved since AR5 (Chapter 3), the simulation of regional climate and climate change poses an additional challenge. Users demand regional climate projections for decision-making and have high expectations regarding accuracy and resolution ( [[#Rössler--2019a|Rössler et al., 2019a]] ), but some scientists consider such projections still a matter of basic research ( [[#Hewitson--2014a|Hewitson et al., 2014a]] ). For instance, large-scale circulation biases or the misrepresentation of regional topography as well as regional phenomena and feedbacks are very relevant ( [[#Hall--2014|Hall, 2014]] ; [[#Maraun--2018b|Maraun and Widmann, 2018b]] ). New global model ensembles such as CMIP6 ( [[#Eyring--2016a|Eyring et al., 2016a]] ), HighResMIP ( [[#Haarsma--2016|Haarsma et al., 2016]] ) or, at the regional scale, the convection permitting simulations from the CORDEX Flagship Pilot Study (FPS) on convective phenomena ( [[#Coppola--2020|Coppola et al., 2020]] ) have the potential to substantially improve the basis for generating regional climate information, yet uncertainties and (often unresolved) contradictions between model projections at the regional scale can be substantial ( [[#Fernández--2019|Fernández et al., 2019]] ).

Figure 10.6 shows the mean summer temperature and precipitation biases of several state-of-the-art climate model ensembles for the western Mediterranean. It additionally illustrates the role of observational uncertainty for model evaluation ( [[#10.2|Section 10.2]] ), where observations display differences that can be substantial. Model performance varies strongly from model to model, but also between ensembles. These biases are an expression of model error that leads to misrepresented phenomena and processes, and thus limit the confidence in future projections of regional climate. The focus of this subsection is therefore to evaluate the representation of relevant regional-scale phenomena for representing regional climate.

<div id="_idContainer026" class="Basic-Text-Frame"></div>

[[File:1474e7ebaea918e1e733ec2bc4fa8b01 IPCC_AR6_WGI_Figure_10_6.png]]

'''Figure''' '''10.6 |''' '''Illustration of some model biases in simulations performed with dynamical models. (a)''' Top row: Mean summer (June to August) near-surface air temperature (in °C) over the Mediterranean area in Berkeley Earth and respective mean bias for five multi-model historical experiments with global models (CMIP5, CMIP6 and HighResMIP) and regional climate models (CORDEX EUR-44 and EUR-11) averaged between 1986–2005. Bottom row: Box-and-whisker plot shows spread of the 20 annual mean summer surface air temperature averaged over land areas in the western Mediterranean region (33°N–45°N, 10°W–10°E, black quadrilateral in the first panel of the top row) for a set of references and single model runs of the five multi-model experiments (one simulation per model) between 1986–2005. Additional observation and reanalysis data included in the bottom row are CRU TS, HadCRUT4, HadCRUT5, E-OBS, WFDE5, ERA5, ERA-Interim, CERA-20C, JRA-25, JRA-55, CFSR, MERRA2, MERRA. Berkeley Earth is shown in the first box to the left. '''(b)''' As (a) but for precipitation rate (mm day <sup>–1</sup> ) and showing CRU TS in the first panel of the top row. Biases of the five multi-model experiments are shown with respect to CRU TS. Additional observation and reanalysis data included in the bottom row are GPCC, REGEN, E-OBS, GHCN, WFDE5, CFSR, ERA-Interim, ERA5, JRA-55, MERRA2, MERRA. CRU TS is shown in the first box to the left. All box-and-whisker plots show the median (line), and the interquartile range (IQR = Q3–Q1, box), with top whiskers extending to the last data less than Q3 + 1.5 × IQR and analogously for bottom whiskers. Data outside the whiskers range appear as flyers (circles). Further details on data sources and processing are available in the chapter data table (Table 10.SM.11).

<div id="10.3.3.1" class="h3-container"></div>

<span id="evaluation-diagnostics"></span>
==== 10.3.3.1 Evaluation Diagnostics ====

<div id="h3-24-siblings" class="h3-siblings"></div>

Since AR5, model evaluation has made use of a broad combination of diagnostics ( [[#Colette--2012|Colette et al., 2012]] ; [[#Kotlarski--2014|Kotlarski et al., 2014]] ; [[#Eyring--2016b|Eyring et al., 2016b]] ; [[#Gleckler--2016|Gleckler et al., 2016]] ; [[#Ivanov--2017|Ivanov et al., 2017]] , 2018; [[#Vautard--2021|Vautard et al., 2021]] ), ranging from long-term means to indices of extreme events (Zhang et al., 2011; [[#Sillmann--2013|Sillmann et al., 2013]] ) or a combination of these ( [[#Dittus--2016|Dittus et al., 2016]] ). This evaluation has shown that global models have pervasive biases in some aspects of their large-scale behaviour ( [[IPCC:Wg1:Chapter:Chapter-1#1.5.3.1|Section 1.5.3.1]] , Chapter 3). More complex diagnostics are used to characterize specific meteorological phenomena ( [[#Sprenger--2017|Sprenger et al., 2017]] ), such as feedbacks in the El Niño–Southern Oscillation (ENSO; [[#Bellenger--2014|Bellenger et al., 2014]] ), Madden-Julian Oscillation (MJO) characteristics (Benedict et al., 2014; [[#Jiang--2015|Jiang et al., 2015]] ; D. [[#Kim--2015|]] [[#Kim--2015|Kim et al., 2015]] ; [[#Ahn--2017|Ahn et al., 2017]] ), extratropical modes of variability ( [[#Lee--2019|Lee et al., 2019]] ), cyclone tracking ( [[#Neu--2013|Neu et al., 2013]] ; [[#Flaounas--2018|Flaounas et al., 2018]] ), front detection ( [[#Hope--2014|Hope et al., 2014]] ; [[#Schemm--2015|Schemm et al., 2015]] ), thunderstorm environment parameters ( [[#Bukovsky--2017|Bukovsky et al., 2017]] ), African easterly waves ( [[#McCrary--2014|McCrary et al., 2014]] ; [[#Martin--2015|Martin and Thorncroft, 2015]] ), land–atmosphere coupling ( [[#Spennemann--2015|Spennemann and Saulo, 2015]] ; [[#Santanello--2018|Santanello et al., 2018]] ), and sea–atmosphere coupling ( [[#Bellenger--2014|Bellenger et al., 2014]] ; [[#Mayer--2017|Mayer et al., 2017]] ).

New diagnostics for multivariate dependencies are needed to characterize compound events ( [[IPCC:Wg1:Chapter:Chapter-11#11.8|Section 11.8]] ; [[#Hobaek%20Haff--2015|Hobaek Haff et al., 2015]] ; [[#Wahl--2015|Wahl et al., 2015]] ; [[#Sippel--2016|Sippel et al., 2016]] , 2017; [[#Tencer--2016|Tencer et al., 2016]] ; [[#Bevacqua--2017|Bevacqua et al., 2017]] ; [[#Careto--2018|Careto et al., 2018]] ; [[#Zscheischler--2018|Zscheischler et al., 2018]] ). However, their success depends on the availability of adequate observational data ( [[#10.2.2|Section 10.2.2]] ). Multivariate dependencies discovered in compound events can also be used for designing and evaluating multivariate bias adjustment and statistical downscaling. Process-based diagnostics are useful for identifying the cause of model errors, although it is not always possible to associate a systematic error with a specific cause ( [[#Eyring--2019|Eyring et al., 2019]] ). AR5 discussed two approaches of process-based evaluation: (i) the isolation of physical components or parametrizations by dedicated experiments ( [[#10.3.2.4|Section 10.3.2.4]] ) and (ii) diagnostics conditioned on relevant regimes, usually synoptic-scale weather patterns. The regime-based approach has been used with both global models (e.g., [[#Barton--2012|Barton et al., 2012]] ; [[#Catto--2015|Catto et al., 2015]] ; [[#Taylor--2019|Taylor et al., 2019]] ) and RCMs ( [[#Endris--2016|Endris et al., 2016]] ; [[#Bukovsky--2017|Bukovsky et al., 2017]] ; [[#Whan--2017|Whan and Zwiers, 2017]] ; [[#Pinto--2018|Pinto et al., 2018]] ), but also with perfect prognosis and bias adjustment methods ( [[#Marteau--2015|Marteau et al., 2015]] ; [[#Addor--2016|Addor et al., 2016]] ; [[#Beranová--2016|Beranová and Kyselý, 2016]] ; [[#Soares--2018|Soares and Cardoso, 2018]] ; [[#Soares--2019b|Soares et al., 2019b]] ).

Recent studies highlight the importance of user-defined or user-relevant diagnostics for model evaluation ( [[#Maraun--2015|Maraun et al., 2015]] ; [[#Rhoades--2018|Rhoades et al., 2018]] ; [[#Rössler--2019b|Rössler et al., 2019b]] ; [[#Nissan--2020|Nissan et al., 2020]] ). Diagnostics have been used to assess the performance of climate models to produce useful input data for impact models as in the comparison between RCMs and convection-permitting models to capture flood-generating precipitation events in the Alps ( [[#Reszler--2018|Reszler et al., 2018]] ). Alternatively, the observed impact can be compared to that simulated by an impact model that uses input from both observations and climate models. This approach has been used to evaluate the influence of statistical downscaling and bias adjustment on hydrological ( [[#Rojas--2011|Rojas et al., 2011]] ; H. [[#Chen--2012|]] [[#Chen--2012|Chen et al., 2012]] ; [[#Gutiérrez--2019|Gutiérrez et al., 2019]] ; [[#Rössler--2019b|Rössler et al., 2019b]] ), agricultural ( [[#Ruiz-Ramos--2016|Ruiz-Ramos et al., 2016]] ; [[#Galmarini--2019|Galmarini et al., 2019]] ), forest and wildfire ( [[#Abatzoglou--2012|Abatzoglou and Brown, 2012]] ; [[#Migliavacca--2013|Migliavacca et al., 2013]] ) ( [[#Bedia--2013|Bedia et al., 2013]] ), snow depth ( [[#Verfaillie--2017|Verfaillie et al., 2017]] ), and regional ocean modelling (e.g., [[#Macias--2018|Macias et al., 2018]] ).

There is ''high confidence'' that to assess whether a climate model realistically simulates required aspects of present-day regional climate, and to increase confidence of future projections of these aspects, evaluation needs to be based on diagnostics taking into account multiple variables and process understanding.

<div id="10.3.3.2" class="h3-container"></div>

<span id="model-improvement-and-added-value"></span>
==== 10.3.3.2 Model Improvement and Added Value ====

<div id="h3-25-siblings" class="h3-siblings"></div>

Obtaining regional information from global simulations may involve a range of different methods ( [[#10.3.1|Section 10.3.1]] ). An approach with higher complexity or resolution is useful if it adds further, useful information to that of a reference model. [[#10.5|Section 10.5]] discusses the set of considerations that determine if the information is useful. This further useful information is often referred to as added value and is a function of variables, processes, and the temporal and spatial scales targeted taking into account the needs of specific users ( [[#Di%20Luca--2012|Di Luca et al., 2012]] ; [[#Ekström--2015|Ekström et al., 2015]] ; [[#Giorgi--2015|Giorgi and Gutowski, 2015]] ; [[#Torma--2015|Torma et al., 2015]] ; [[#Rummukainen--2016|Rummukainen, 2016]] ; [[#Falco--2019|Falco et al., 2019]] ). There is no common definition of added value, but here it is considered a characteristic that arises when one methodology gives further value to what another methodology yields.

Downscaling is expected to improve the representation of a region’s climate compared to the driving global model ( [[#Di%20Luca--2015|Di Luca et al., 2015]] ). Arguably, there should be a clear physical reason for the improvement, which is applicable to the evaluation of added value in downscaled projections ( [[#Giorgi--2016|Giorgi et al., 2016]] ). The added value depends on the region, season, and governing physical processes ( [[#Lenz--2017|Lenz et al., 2017]] ; [[#Schaaf--2018|Schaaf and Feser, 2018]] ). Thus, added value of downscaling global model simulations is most likely where regional- and local-scale processes play an important role in a region’s climate, for example in complex or heterogeneous terrain such as mountains ( [[#Lee--2014|Lee and Hong, 2014]] ; [[#Prein--2016b|Prein et al., 2016b]] ), urban areas ( [[#Argüeso--2014|Argüeso et al., 2014]] ), along coastlines ( [[#Feser--2011|Feser et al., 2011]] ; [[#Herrmann--2011|Herrmann et al., 2011]] ; [[#Bozkurt--2019|Bozkurt et al., 2019]] ), or where convective processes are important ( [[#Prein--2015|Prein et al., 2015]] ). Examples of model improvements and added value are given in the following subsections and the Atlas.

A first step in determining added value in downscaling is to analyse whether the downscaling procedure gives detail on spatial or temporal scales not well-resolved by a global model, thus potentially representing climatic features missing in the GCM. This added detail, referred to as potential added value (PAV; [[#Di%20Luca--2012|Di Luca et al., 2012]] ), is insufficient for demonstrating added value in downscaling ( [[#Takayabu--2016|Takayabu et al., 2016]] ), but lack of PAV indicates that the downscaling method lacks usefulness. Added value is not guaranteed simply by producing model output at finer resolution. It depends on several factors, such as the simulation setup and the specific climatic variables analysed ( [[#Di%20Luca--2012|Di Luca et al., 2012]] ; [[#Hong--2014|Hong and Kanamitsu, 2014]] ; [[#Xue--2014|Xue et al., 2014]] ). A variety of performance measures are needed to assess added value ( [[#10.3.3.1|Section 10.3.3.1]] ; [[#Di%20Luca--2016|Di Luca et al., 2016]] ; [[#Wilks--2016|Wilks, 2016]] ; [[#Ivanov--2017|Ivanov et al., 2017]] , 2018; [[#Soares--2018|Soares and Cardoso, 2018]] ).

A further challenge, especially at increasingly higher resolutions, is that adequate observational data may not be available to assess added value ( [[#10.2|Section 10.2]] , e.g., [[#Di%20Luca--2016|Di Luca et al., 2016]] ; [[#Zittis--2017|Zittis et al., 2017]] ; [[#Bozkurt--2019|Bozkurt et al., 2019]] ). This implies a need for additional efforts to obtain, catalogue and quality-control higher resolution observational (or observation-based) datasets ( [[#Thorne--2017|Thorne et al., 2017]] ; [[#10.2|Section 10.2]] ). Univariate demonstration of added value is necessary, but may be insufficient, as better agreement with observations in the downscaled variable may be a consequence of compensating errors that are not guaranteed to compensate similarly as climate changes. Multivariate analysis of added value is better able to demonstrate physical consistency between observed and simulated behaviour ( [[#Prein--2013a|Prein et al., 2013a]] ; [[#Meredith--2015a|Meredith et al., 2015a]] ; [[#Reboita--2018|Reboita et al., 2018]] ).

<div id="10.3.3.3" class="h3-container"></div>

<span id="performance-at-simulating-large-scale-phenomena-and-teleconnections-relevant-for-regional-climate"></span>
==== 10.3.3.3 Performance at Simulating Large-scale Phenomena and Teleconnections Relevant for Regional Climate ====

<div id="h3-26-siblings" class="h3-siblings"></div>

Regional climate is often controlled by large-scale weather phenomena, modes of variability and teleconnections (e.g., Sections 2.3 and 2.4, Annex IV). In particular, extreme events are often caused by specific, in some cases persistent, circulation patterns (Sections 11.3–11.7). It is therefore important for climate models to reasonably represent not only continental, but also regional climate and its variability for such extremes. As explained in [[IPCC:Wg1:Chapter:Chapter-3#3.3.3|Section 3.3.3]] , standard resolution global models can suffer biases in the location, occurrence frequency or intensity of large-scale phenomena, such that statements about a specific regional climate and its change can be highly uncertain ( [[#Hall--2014|Hall, 2014]] ). RCMs have difficulties improving especially large-scale circulation biases, although some successful examples exist. But due to their enhanced representation of complex topography and coastlines, RCMs may add value to simulating the regional expression of teleconnections. Bias adjustment cannot mitigate fundamental misrepresentations of the large-scale atmospheric circulation ( [[#Maraun--2017|Maraun et al., 2017]] , Cross-Chapter Box 10.2). This subsection illustrates the relevance of large-scale circulation biases for regional climate assessments with selected examples from the mid- to high latitudes and tropics.

<div id="10.3.3.3.1" class="h4-container"></div>

<span id="mid--to-high-latitude-atmospheric-variability-phenomena-blocking-and-extratropical-cyclones"></span>
===== 10.3.3.3.1 Mid- to high-latitude atmospheric variability phenomena: Blocking and extratropical cyclones =====

<div id="h4-6-siblings" class="h4-siblings"></div>

Major large-scale meteorological phenomena for mid- to high latitude mean and extreme climate include atmospheric blocking and extratropical cyclones ( [[IPCC:Wg1:Chapter:Chapter-2#2.3.1.4|Section 2.3.1.4]] ). Atmospheric blocking is characterized by a quasi-stationary, long-lasting, high-pressure system that blocks and diverts the movement of synoptic cyclones ( [[#Woollings--2018|Woollings et al., 2018]] ). In regions where blocking occurs, it is known to lead to cold conditions in winter and warmth and drought during summer, defining the seasonal regional climate in certain years ( [[#Sousa--2017|Sousa et al., 2017]] , 2018b). Extratropical cyclones are storm systems that propagate preferentially in confined storm-track regions, characterized by large eddy kinetic energy, heat and momentum transports that shape regional weather at mid- to high latitudes ( [[#Shaw--2016|Shaw et al., 2016]] ). Given their importance in shaping mean and extreme regional climate (Sections 3.3.3.3, 11.3 and 11.4), an accurate representation of blocking and extratropical cyclones in global and regional climate models is needed to better understand regional climate variability and extremes as well as to project future changes ( [[IPCC:Wg1:Chapter:Chapter-11#11.7.2|Section 11.7.2]] ; [[#Grotjahn--2016|Grotjahn et al., 2016]] ; [[#Mitchell--2017|Mitchell et al., 2017]] ; [[#Rohrer--2018|Rohrer et al., 2018]] ; [[#Huguenin--2020|Huguenin et al., 2020]] ). An overview of CMIP5 and CMIP6 model performance in simulating blocking and extratropical cyclones is given in [[IPCC:Wg1:Chapter:Chapter-3#3.3.3.3|Section 3.3.3.3]] . CMIP6 models still suffer from long-standing blocking biases identified in previous generations of models. However, blocking location has improved compared to CMIP5, while comparable performance is seen for blocking frequency and persistence (Figure 10.7). Increasing horizontal model resolution to about 20 km in the HighResMIP experiments improves the representation of blocking frequency and its spatial pattern in most models, but no clear effect could be shown for blocking persistence. Biases associated with these two phenomena are highly region- and season-dependent and their amplitudes vary among CMIP models ( [[#Drouard--2018|Drouard and Woollings, 2018]] ; [[#Schaller--2018|Schaller et al., 2018]] ; [[#Woollings--2018|Woollings et al., 2018]] ; [[#Harvey--2020|Harvey et al., 2020]] ; [[#Schiemann--2020|Schiemann et al., 2020]] ).

<div id="_idContainer029" class="Basic-Text-Frame"></div>

[[File:055ebbc14be5907922f77eee0f954f09 IPCC_AR6_WGI_Figure_10_7.png]]

'''Figure 10.7''' '''|''' '''Northern Hemisphere blocking performance in historical coupled simulations for different multi-model ensembles.''' Coupled Model Intercomparison Project Phases 5 and 6 (CMIP5/6): CMIP5 and CMIP6 Diagnostic, Evaluation and Characterization of Klima (DECK) historical simulations, 1950–2005, LC/HC: Low- (LC)/high- (HC) resolution coupled simulations from the PRIMAVERA project, 1950–2014 following the hist-1950 experiment of the CMIP6 HighResMIP Protocol ( [[#Haarsma--2016|Haarsma et al., 2016]] ). (Top) blocking frequency, for example, fraction of blocked days; (middle) root-mean-squared error in blocking frequency; (bottom) 90th percentile of blocking persistence, aggregated over an Atlantic domain (left, ATL: 90°W–90°E, 50°–75°N) and a Pacific domain (right, PAC: 90°E–270°E, 50°–75°N). Results are for boreal winter (December–January–February, DJF) and summer (June–July–August, JJA). Box-and-whisker plots for CMIP5/6 follow the methodology used in Figure 10.6 and show median (line), mean (triangle), and interquartile range (box) across 29 models for each ensemble. The reference estimate (ERA, asterisk) is from a 50-year reanalysis dataset that merged ERA-40 (1962–1978) and ERA-Interim (1979–2011) reanalyses. An estimate of internal variability for each metric (IV) is shown as a box-and-whisker plot over the asterisk and is obtained from a single-model ensemble (ECMWF-IFS high-resolution hist-1950 experiment, 6 × 65 years). For details on the methodology see ( [[#Schiemann--2020|Schiemann et al., 2020]] ). Further details on data sources and processing are available in the chapter data table (Table 10.SM.11).

RCMs have a very limited ability to reduce large-scale circulation errors of the driving GCM ( [[#Hall--2014|Hall, 2014]] ). In a study of five ERA-Interim-driven RCMs, [[#Jury--2018|Jury et al. (2018)]] showed that RCMs typically simulate fewer blocking events over Europe than are present in the driving data, irrespective of the RCM horizontal resolution. Based on a simple blocking bias-decomposition method, they suggest that blocking frequency biases can contribute to the RCM mean surface biases. Over some large domains, reanalysis-driven RCMs can significantly improve the representation of storm characteristics compared to the driving reanalysis near regions with complex orography and/or large water masses ( [[#Poan--2018|Poan et al., 2018]] ). However, this is not necessarily true if the domain is large enough because the RCM and its biases will then control the circulation leading to a biased performance with regard to storm characteristics ( [[#Pontoppidan--2019|Pontoppidan et al., 2019]] ). An ensemble of 12 RCMs with and without air-sea coupling reasonably reproduced the climatology of Mediterranean cyclones, and air-sea coupling had a rather weak impact ( [[#Flaounas--2018|Flaounas et al., 2018]] ). Over the Gulf Stream, however, air-sea coupling played an important role in representing cyclone development ( [[#Vries--2019|Vries et al., 2019]] ). [[#Sanchez-Gomez--2018|Sanchez-Gomez and Somot (2018)]] showed that the effect of RCM internal variability on density of cyclone tracks is very significant and larger than for other variables such as precipitation. It is larger in summer than in winter, in particular over the Iberian Peninsula, northern Africa and the eastern Mediterranean, which are regions of enhanced cyclogenesis during the warm season.

Biases in the representation of large-scale atmospheric circulation can result in biased representation of regional climate. While, in principle, the connection between large-scale and regional biases is obvious, given the strong control of regional climate by large-scale phenomena, research on this connection is still limited. [[#Munday--2018|Munday and Washington (2018)]] relate CMIP5 model rainfall biases over South Africa to anomalous low-level moisture transport across high topography due to upstream wind biases and inaccurate representation of unresolved orographic drag effects. [[#Addor--2016|Addor et al. (2016)]] show that the overestimated frequency of westerly synoptic situations was a significant contributor to the wet bias in several RCMs in winter over Switzerland. Pepler et al. (2014, 2016) suggest that better capturing westerly-driven synoptic systems such as cold fronts and cut-off lows in climate models could be key in simulating the observed pattern correlation between rainfall and zonal wind in southern south-east Australia. [[#Cannon--2020|Cannon (2020)]] shows global improvement in performance going from CMIP5 to CMIP6 for both frequency and persistence of circulation types.

The robust quantification of the influence of atmospheric circulation errors on regional climate remains a challenge as many parametrized processes such as cloud radiative effects and soil moisture or snow feedbacks can also contribute and interact with the circulation errors. Atmospheric nudging experiments where the simulated circulation is constrained to be close to that observed have been used to separate the circulation effect from other contributions to regional climate biases ( [[#Wehrli--2018|Wehrli et al., 2018]] ). The nudging approach requires detailed and careful implementation in order to limit detrimental effects due to the added tendency term in the model equations ( [[#Zhang--2014|Zhang et al., 2014]] ; [[#Lin--2016|Lin et al., 2016]] ). Based on single-model experiments, [[#Wehrli--2018|Wehrli et al. (2018)]] show that the circulation induced biases are often not the main contributors to mean and extreme temperature and precipitation biases for many regions and seasons.

There is ''high confidence'' that atmospheric circulation biases can deteriorate the model representation of regional land surface climate. Assessing the relative contributions of atmospheric circulation and other sources of bias remains a challenge due to the strong coupling between the atmosphere and other components of the climate system, including the land surface.

<div id="10.3.3.3.2" class="h4-container"></div>

<span id="tropical-phenomena-enso-teleconnections"></span>
===== 10.3.3.3.2 Tropical phenomena: ENSO teleconnections =====

<div id="h4-7-siblings" class="h4-siblings"></div>

Model performance in simulating ENSO characteristics, including ENSO spatial pattern, frequency, asymmetry between warm and cold events, and diversity, is assessed in [[IPCC:Wg1:Chapter:Chapter-3|Chapter 3]] ( [[IPCC:Wg1:Chapter:Chapter-3#3.7.3|Section 3.7.3]] ). The ability of the recent generation of GCMs and RCMs to adequately simulate ENSO-related teleconnections is reviewed here along with relevant methodological issues (see also Annex IV2.3.2, Figure 3.38 and [[IPCC:Wg1:Chapter:Chapter-3#3.7.3|Section 3.7.3]] ).

[[#Langenbrunner--2013|Langenbrunner and Neelin (2013)]] show that there is little improvement in CMIP5 relative to CMIP3 in amplitude and spatial patterns of the ENSO influence on boreal winter precipitation (spatial pattern correlations against observations are typically less than 0.5). However, the CMIP5 ensemble accurately represents the amplitude of the precipitation response in regions where observed teleconnections are strong. [[#Garcia-Villada--2020|Garcia-Villada et al. (2020)]] found a decline in performance of the representation of simulated ENSO teleconnection patterns for model experiments with fewer observational constraints. They also show that ENSO warm phase (El Niño) teleconnections are better represented than those for the cold phase (La Niña). Individual CMIP5 and CMIP6 models show a good ability to represent the observed teleconnections at aggregated spatial scales ( [[#Power--2018|Power and Delage, 2018]] ; [[IPCC:Wg1:Chapter:Chapter-3#3.7.3|Section 3.7.3]] and Figure 3.38). The evaluation of the atmospheric dynamical linkages is also an important part of the assessment. [[#Hurwitz--2014|Hurwitz et al. (2014)]] showed that CMIP5 models broadly simulate the expected (as seen in the MERRA reanalysis) upper-tropospheric responses to central equatorial Pacific or eastern equatorial Pacific ENSO events in boreal autumn and winter. CMIP5 models also simulate the correct sign of the Arctic stratospheric response, consisting of polar vortex weakening during eastern and central Pacific Niño events and vortex strengthening during both types of La Niña events. In contrast, most CMIP5 models do not capture the observed weakening of the Southern Hemisphere polar vortex in response to central Pacific ENSO events ( [[#Brown--2013|Brown et al., 2013]] ).

In RCMs, the effects of tropical large-scale modes and teleconnections are inherited through the boundary conditions and influenced by the size of the numerical domain. [[#Done--2015|Done et al. (2015)]] and [[#Erfanian--2018|Erfanian and Wang (2018)]] claim that large domains that include source oceanic regions are required to capture the remote influence of teleconnections, although, without spectral nudging, this can lead to biased synoptic-scale patterns ( [[#Prein--2019|Prein et al., 2019]] ). RCMs generally reproduce the regional precipitation responses to ENSO, and can sometimes even improve the representation of these teleconnections compared to the driving reanalysis ( [[#Endris--2013|Endris et al., 2013]] ; [[#Fita--2017|Fita et al., 2017]] ), but the overall performance may depend both on the driving reanalysis or GCM ( [[#Endris--2016|Endris et al., 2016]] ; [[#Chandrasa--2020|Chandrasa and Montenegro, 2020]] ) and on the chosen RCMs ( [[#Whan--2017|Whan and Zwiers, 2017]] ).

New studies since AR5 have shown that model performance assessment regarding ENSO teleconnections remains a difficult challenge due to the different types of ENSO and model errors in ENSO spatial patterns, as well as the strong influence of atmospheric internal variability at mid- to high latitudes ( [[#Coats--2013|Coats et al., 2013]] ; [[#Polade--2013|Polade et al., 2013]] ; [[#Capotondi--2015|Capotondi et al., 2015]] ; [[#Deser--2017c|Deser et al., 2017c]] ; [[#Tedeschi--2017|Tedeschi and Collins, 2017]] ; [[#Garcia-Villada--2020|Garcia-Villada et al., 2020]] ). Another difficulty comes from the non-stationary aspects of teleconnections in both observations and models, raising methodological questions on how best to compare a given model with another model or observations ( [[#Herein--2017|Herein et al., 2017]] ; [[#Perry--2017|Perry et al., 2017]] ; [[#O’Reilly--2018|O’Reilly, 2018]] ; [[#O’Reilly--2019|O’Reilly et al., 2019]] ; [[#Abram--2020|Abram et al., 2020]] ).

There is ''robust evidence'' that an accurate representation of both atmospheric circulation and sea surface temperature (SST) variability are key factors for the realistic representation of ENSO teleconnections in climate models. A robust and thorough evaluation of model performance regarding ENSO teleconnections is a challenging task with many methodological issues related to asymmetry between the warm and cold phases, non-stationarity and time-varying interaction between the Pacific and other ocean basins, signal-to-noise issues in the mid-latitudes and observational uncertainties, particularly for precipitation ( [[#10.2.2.3|Section 10.2.2.3]] ).

<div id="10.3.3.4" class="h3-container"></div>

<span id="performance-at-simulating-regional-phenomena-and-processes"></span>
==== 10.3.3.4 Performance at Simulating Regional Phenomena and Processes ====

<div id="h3-27-siblings" class="h3-siblings"></div>

Regional climate is shaped by a wide range of weather phenomena occurring at scales from about 2000 km to 2 km (Figure 10.3). These modulate the influence of large-scale atmospheric phenomena and create the characteristic and potentially severe weather conditions. The climate in different regions will be affected by different mesoscale phenomena, of which several may be relevant. A skilful representation of these phenomena is a necessary condition for providing credible and relevant climate information for a given region and application. Therefore, it is important to understand the strengths and weaknesses of different model types in simulating these phenomena. The performance of different dynamical climate model types to simulate a selection of relevant mesoscale weather phenomena is assessed here.

<div id="10.3.3.4.1" class="h4-container"></div>

<span id="convection-including-tropical-cyclones"></span>
===== 10.3.3.4.1 Convection including tropical cyclones =====

<div id="h4-8-siblings" class="h4-siblings"></div>

Convection is the process of vertical mixing due to atmospheric instability. Deep moist convection is associated with thunderstorms and severe weather such as heavy precipitation and strong wind gusts. Convection may occur in single locations, in spatially extended severe events such as supercells, and organized into larger mesoscale convective systems such as squall lines or tropical cyclones, and embedded in fronts (see below). Shallow and deep convection are not explicitly simulated but parametrized in standard global and regional models. In consequence, these models suffer from several biases. AR5 has stated that many CMIP3 and CMIP5 models simulate the peak in the diurnal cycle of precipitation too early, but increasing resolution and better parametrizations help to mitigate this problem ( [[#Flato--2014|Flato et al., 2014]] ). Similar issues arise for RCMs with parametrized deep convection ( [[#Prein--2015|Prein et al., 2015]] ), which also tend to overestimate high cloud cover ( [[#Langhans--2013|Langhans et al., 2013]] ; [[#Keller--2016|Keller et al., 2016]] ).

Non-hydrostatic RCMs at convection-permitting resolution (4 km and finer) improve features such as the initiation and diurnal cycle of convection ( [[#Zhu--2012|Zhu et al., 2012]] ; [[#Prein--2013a|Prein et al., 2013a]] , b; [[#Fosser--2015|Fosser et al., 2015]] ; [[#Stratton--2018|Stratton et al., 2018]] ; [[#Sugimoto--2018|Sugimoto et al., 2018]] ; [[#Finney--2019|Finney et al., 2019]] ; [[#Berthou--2020|Berthou et al., 2020]] ; [[#Ban--2021|Ban et al., 2021]] ; [[#Pichelli--2021|Pichelli et al., 2021]] ), the triggering of convection by orographic lifting ( [[#Langhans--2013|Langhans et al., 2013]] ; [[#Fosser--2015|Fosser et al., 2015]] ), and maximum vertical wind speeds in convective cells ( [[#Meredith--2015a|Meredith et al., 2015a]] ). Also spatial patterns of precipitation ( [[#Prein--2013a|Prein et al., 2013a]] , b; [[#Stratton--2018|Stratton et al., 2018]] ), precipitation intensities ( [[#Prein--2015|Prein et al., 2015]] ; [[#Fumière--2020|Fumière et al., 2020]] ; [[#Ban--2021|Ban et al., 2021]] ; [[#Pichelli--2021|Pichelli et al., 2021]] ), the scaling of precipitation with temperature ( [[#Ban--2014|Ban et al., 2014]] ), cloud cover ( [[#Böhme--2011|Böhme et al., 2011]] ; [[#Langhans--2013|Langhans et al., 2013]] ) and its resultant radiative effects ( [[#Stratton--2018|Stratton et al., 2018]] ), as well as the annual cycle of tropical convection ( [[#Hart--2018|Hart et al., 2018]] ) are improved. Phenomena such as supercells, mesoscale convective systems, or the local weather associated with squall lines are not captured by global models and standard RCMs. Convection-permitting RCM simulations, however, have been shown to realistically simulate supercells ( [[#Trapp--2011|Trapp et al., 2011]] ), mesoscale convective systems, their life cycle and motion ( [[#Prein--2017|Prein et al., 2017]] ; [[#Crook--2019|Crook et al., 2019]] ), and heavy precipitation associated with a squall line ( [[#Kendon--2014|Kendon et al., 2014]] ). There is ''high confidence'' that simulations at convection-permitting resolution add value to the representation of deep convection and related phenomena.

Convection is the key ingredient of tropical cyclones. An intercomparison of high-resolution AGCM simulations ( [[#Shaevitz--2014|Shaevitz et al., 2014]] ) showed that tropical cyclone intensities appeared to be better represented with increasing model resolution. [[#Takayabu--2015|Takayabu et al. (2015)]] have compared simulations of typhoon Haiyan at different resolutions ranging from 20 km to 1 km (Figure 10.8). While the eyewall structure in the precipitation pattern was strongly smoothed in the coarse resolution simulations, it was well-resolved at the highest resolution. [[#Gentry--2010|Gentry and Lackmann (2010)]] found similar improvements in simulating hurricane Ivan for horizontal resolutions between 8 km and 1 km. High-resolution coupled ocean–atmosphere simulations improve the representation of the radial structure of core convection and thereby the rapid intensification of the cyclone ( [[#Kanada--2017b|Kanada et al., 2017b]] ). There is ''high confidence'' that convection-permitting resolution is required to realistically simulate the three-dimensional structure of tropical cyclones.

<div id="_idContainer031" class="Basic-Text-Frame"></div>

[[File:1b451650234bba076f5e7d74e3147fc1 IPCC_AR6_WGI_Figure_10_8.png]]

'''Figure 10.8''' '''|''' '''Hourly accumulated precipitation profiles (mm hour''' –1 ''') around the eye of Typhoon Haiyan.''' Represented by '''(a)''' Global Satellite Mapping of Precipitation (GSMaP) data (multi-satellite observation), '''(b)''' Guiuan radar (PAGASA), '''(c)''' Weekly Ensemble Prediction System (WEPS) data (JMA; 60 km), '''(d)''' NHRCM (20 km), '''(e)''' NHRCM (5 km), and '''(f)''' WRF (1 km) models. Panels (b), (d–f) are adapted from [[#Takayabu--2015|Takayabu et al. (2015)]] , CCBY3.0 https://creativecommons.org/licenses/by/3.0 . Further details on data sources and processing are available in the chapter data table (Table 10.SM.11).

Initial studies with convection-permitting global models suggests that improvements in representing convection, as described for RCMs above, have a positive impact on the tropical and extratropical atmospheric circulation and, thus, regional climate ( [[#Satoh--2019|Satoh et al., 2019]] ; [[#Stevens--2019|Stevens et al., 2019]] ; see also [[IPCC:Wg1:Chapter:Chapter-8#8.5.1.2|Section 8.5.1.2]] and Chapter 7). Computational constraints currently limit these simulations to a length of few months only, such that they cannot yet be used for routine climate change studies.

<div id="10.3.3.4.2" class="h4-container"></div>

<span id="mountain-wind-systems"></span>
===== 10.3.3.4.2 Mountain wind systems =====

<div id="h4-9-siblings" class="h4-siblings"></div>

Mountain slope and valley winds are localized thermally generated diurnal circulations that have a strong influence on temperature and precipitation patterns in mountain regions. During the day, heating of mountain slopes induces upslope winds; during the night this circulation reverses. This phenomenon is not realistically represented by global models and coarse-resolution RCMs. RCM simulations at 4 km resolution showed good skill in simulating the diurnal cycle of temperature and wind on days of weak synoptic forcing in the Rocky Mountains ( [[#Letcher--2017|Letcher and Minder, 2017]] ) as well as in simulating the mountain-plain wind circulation over the Tianshan mountains in central Asia ( [[#Cai--2019|Cai et al., 2019]] ), while in the Alps, a 1 km resolution has been required ( [[#Zängl--2004|Zängl, 2004]] ).

Föhn winds are synoptically-driven winds across a mountain range that are warm and dry due to adiabatic warming in the downwind side. In an RCM study for the Japanese Alps, [[#Ishizaki--2009|Ishizaki and Takayabu (2009)]] found that at least 10 km resolution was required to realistically simulate the basic characteristics of Föhn events.

Synoptically-forced winds may be channelled and accelerated in long valleys. For instance, the Tramontana, Mistral and Bora are northerly winds blowing down-valley from central France and the Balkans into the Mediterranean ( [[#Flaounas--2013|Flaounas et al., 2013]] ). In winter, these winds may cause severe cold air outbreaks along the coast. [[#Flaounas--2013|Flaounas et al. (2013)]] have shown that a GCM with a horizontal resolution of roughly 3.75° longitude/1.875° latitude (roughly 400 km × 200 km depending on latitude) is unable to reproduce these winds because of the coarse representation of orography. Fifty-kilometre RCM simulations did not realistically represent the Mistral ( [[#Obermann--2018|Obermann et al., 2018]] ) and Bora winds ( [[#Belušić--2018|Belušić et al., 2018]] ), but simulations at 12 km added substantial value. Similarly, [[#Cholette--2015|Cholette et al. (2015)]] found that a 30 km RCM resolution was not sufficient to adequately simulate the channelling of winds in the St Lawrence River Valley in eastern Canada, whereas a 10 km resolution was.

There is ''high confidence'' that climate models with resolutions of around 10 km or finer are necessary for realistically simulating mountain wind systems such as slope and valley winds and the channelling of winds in valleys.

<div id="10.3.3.4.3" class="h4-container"></div>

<span id="coastal-winds-and-lake-effects"></span>
===== 10.3.3.4.3 Coastal winds and lake effects =====

<div id="h4-10-siblings" class="h4-siblings"></div>

Simulating coastal climates and the influence of big lakes are a modelling challenge, due to the complex coastlines, the different heat capacities of land and water, the resulting wind system, and differential evaporation. The AR5 concluded that RCMs can add value to the simulation of coastal climates.

Summer coastal low-level jets off the mid-latitude western continental coasts are forced by the semi-permanent subtropical anticyclones, inland thermal lows, strong across-shore temperature contrasts in upwelling regions, and high coastal topography. They are important factors in shaping regional climate by, for instance, preventing onshore advection of humidity and thereby causing aridity in the Iberian Peninsula ( [[#Soares--2014|Soares et al., 2014]] ), or by transporting moisture towards precipitating regions as in the North American monsoon ( [[#Bukovsky--2013|Bukovsky et al., 2013]] ).

Reanalyses and most global models do not well resolve the details of coastal low-level jets ( [[#Bukovsky--2013|Bukovsky et al., 2013]] ; [[#Soares--2014|Soares et al., 2014]] ), but they are still able to represent annual and diurnal cycles and interannual variability ( [[#Cardoso--2016|Cardoso et al., 2016]] ; [[#Lima--2019|Lima et al., 2019]] ). [[#Bukovsky--2013|Bukovsky et al. (2013)]] found RCM simulations at a 50 km resolution to improve the representation of the coastal low-level jet in the Gulf of California and the associated precipitation pattern compared to the driving global models. [[#Lucas-Picher--2017|Lucas-Picher et al. (2017)]] find indirect evidence via precipitation patterns that 12 km simulations further improve the representation. [[#Soares--2014|Soares et al. (2014)]] demonstrated that an 8 km resolution RCM simulated a realistic three-dimensional structure of the Iberian coastal low-level jet, and the surface winds compare well with observations. [[#Lucas-Picher--2017|Lucas-Picher et al. (2017)]] showed that a 0.44° resolution RCM underestimated winds along the Canadian east coast, whereas a 0.11° resolution version simulated more realistic 10 metre wind speed. Also, the Etesian winds in the Aegean Sea were realistically simulated by 12 km resolution RCMs ( [[#Dafka--2018|Dafka et al., 2018]] ).

A particularly relevant coastal phenomenon is the sea breeze, which is caused by the differential heating of water and land during the diurnal cycle and typically reaches several tens of kilometres inland. Reanalyses and global models have too coarse a resolution to realistically represent this phenomenon, such that they typically underestimate precipitation over islands and misrepresent its diurnal cycle ( [[#Lucas-Picher--2017|Lucas-Picher et al., 2017]] ). RCMs improve the representation of sea breezes and thereby precipitation in coastal areas and islands. Over Cuba and Florida only a 12 km-resolution RCM is able to realistically simulate the inland propagation of precipitation during the course of the day ( [[#Lucas-Picher--2017|Lucas-Picher et al., 2017]] ). RCM simulations at 20 km horizontal resolution realistically represented the sea breeze circulation in the Mediterranean Gulf of Lions including the intensity, direction and inward propagation ( [[#Drobinski--2018|Drobinski et al., 2018]] ). Even though a coupled ocean–atmosphere simulation improved the representation of diurnal SST variations, the sea breeze representation itself was not improved.

Big lakes modify the downwind climate. In particular during winter they are relatively warm compared to the surrounding land, provide moisture, destabilize the passing air column and produce convective systems. The increase in friction when moving air reaches land causes convergence and uplift, and may trigger precipitation. [[#Gula--2012|Gula and Peltier (2012)]] found that a state-of-the-art GCM does not realistically simulate these effects over the North American Great Lakes, but a 10 km RCM better represents them and thereby simulates realistic downwind precipitation patterns, in particular enhanced snowfall during the winter season. Similar results were found by [[#Wright--2013|Wright et al. (2013)]] , [[#Notaro--2015|Notaro et al. (2015)]] and [[#Lucas-Picher--2017|Lucas-Picher et al. (2017)]] . In a convection permitting simulation of the Lake Victoria region, a too strong nocturnal land breeze resulted in unrealistically high precipitation ( [[#Finney--2019|Finney et al., 2019]] ).

There is ''high confidence'' that climate models with sufficiently high resolution are necessary for realistically simulating lake and coastal weather including coastal low-level jets, lake and sea breezes, as well as lake effects on rainfall and snow.

In regions like Fenno-Scandinavia or central-eastern Canada, very large fractions of land are covered by small and medium sized lakes. Other regions have fewer but larger lakes, such as central-eastern Africa, the eastern border between the USA and Canada, and central Asia. In these regions it has been considered essential to include a lake model in an RCM to realistically represent regional temperatures ( [[#Samuelsson--2010|Samuelsson et al., 2010]] ; [[#Deng--2013|Deng et al., 2013]] ; [[#Mallard--2014|Mallard et al., 2014]] ; [[#Thiery--2015|Thiery et al., 2015]] ; [[#Pietikäinen--2018|Pietikäinen et al., 2018]] ), as well as remote effects ( [[#Spero--2016|Spero et al., 2016]] ). The most common approach in RCMs is the two-layer lake model, including a lake-ice model, with parametrized vertical temperature profiles ( [[#Mironov--2010|Mironov et al., 2010]] ; [[#Golosov--2018|Golosov et al., 2018]] ). For the Caspian Sea, it is found that a three-dimensional ocean model simulated the SST fields better than a one-dimensional lake model when coupled to the same RCM ( [[#Turuncoglu--2013|Turuncoglu et al., 2013]] ).

There is ''medium evidence'' and ''high agreement'' that it is important to include interactive lake models in RCMs to improve the simulation of regional temperature, in particular in seasonally ice-covered areas with large fractions of lakes. There is ''medium evidence'' of the local influence of lakes on snow and rainfall as well as the importance of including lakes in regional climate simulations.

<div id="10.3.3.4.4" class="h4-container"></div>

<span id="fronts"></span>
===== 10.3.3.4.4 Fronts =====

<div id="h4-11-siblings" class="h4-siblings"></div>

Weather fronts are two-dimensional surfaces separating air masses of different characteristics and are a key element of mid-latitude cyclones. In particular cold fronts are regions of relatively strong uplift and hence often associated with severe weather (e.g., [[#Schemm--2016|Schemm et al., 2016]] ). Stationary or slowly moving fronts may cause extended heavy precipitation. The evaluation of how climate models represent fronts, however, remains limited. [[#Catto--2014|Catto et al. (2014)]] found in both ERA-Interim and CMIP5 models that frontal frequency and strength were realistically simulated, albeit with some biases in the location. Follow-up investigations, for boreal and austral winter ( [[#Catto--2015|Catto et al., 2015]] ) found frontal precipitation frequency to be too high and the intensity too low, but these compensating biases resulted in only a small total precipitation bias. [[#Blázquez--2018|Blázquez and Solman (2018)]] found similar results for Southern Hemisphere (SH) winter, and also showed that CMIP5 models typically overestimate the fraction of frontal precipitation compared to total precipitation. As for the reference, the ERA-Interim reanalysis misrepresents conditional symmetric instability associated with fronts, and the corresponding precipitation ( [[#Glinton--2017|Glinton et al., 2017]] ). Only a few studies evaluating fronts in RCMs have been conducted. [[#Kawazoe--2013|Kawazoe and Gutowski (2013)]] diagnosed strong temperature gradients associated with extreme winter precipitation in the North American Regional Climate Change Assessment Program (NARCCAP) RCM ensemble ( [[#Mearns--2012|Mearns et al., 2012]] ) and found the models agreed well with gradients in a reanalysis. De Jesus et al. (2016) diagnosed the representations of cold fronts over southern Brazil in two RCMs, finding that they were only underestimated by about 5% across the year, but in one RCM, summer cold fronts were underestimated by 17%. An RCM-based reanalysis suggests that high-resolution RCM simulations improve the representation of orographic influences on fronts ( [[#Jenkner--2009|Jenkner et al., 2009]] ).

<div id="10.3.3.5" class="h3-container"></div>

<span id="performance-at-simulating-regional-feedbacks"></span>
==== 10.3.3.5 Performance at Simulating Regional Feedbacks ====

<div id="h3-28-siblings" class="h3-siblings"></div>

Both SRCCL ( [[#Jia--2019|Jia et al., 2019]] ) and SROCC ( [[#Hock--2019|Hock et al., 2019]] ) highlight the weaknesses of climate models at simulating atmosphere–surface feedbacks. The performance at simulating some of these feedbacks is assessed below (climate feedbacks in urban areas are discussed in Box 10.3).

The snow-albedo feedback contributes to enhanced warming at high elevations ( [[IPCC:Wg1:Chapter:Chapter-8#8.5|Section 8.5]] ; [[#Pepin--2015|Pepin et al., 2015]] ). Global models often do not simulate it realistically due to their misrepresentation of orography in complex terrain ( [[#Hall--2014|Hall, 2014]] ; [[#Walton--2015|Walton et al., 2015]] ). The elevation dependence of historical warming, which is partly caused by the snow-albedo effect, is realistically represented across Europe by the ENSEMBLES RCMs ( [[#Kotlarski--2015|Kotlarski et al., 2015]] ). Some EURO-CORDEX RCMs simulate a spring snow–albedo feedback close to that observed, whereas others considerably overestimate it ( [[#Winter--2017|Winter et al., 2017]] ). In a multi-physics ensemble RCM experiment, the cold bias in north-eastern Europe is amplified by the albedo feedback ( [[#García-Díez--2015|García-Díez et al., 2015]] ). For the Rocky Mountains, RCM simulations generally reproduce the observed spatial and seasonal variability in snow cover, but strongly overestimate the snow albedo ( [[#Minder--2016|Minder et al., 2016]] ). There is ''high confidence'' ( ''medium evidence'' and ''high agreement'' ) that RCMs considerably improve the representation of the snow-albedo effect in complex terrain.

Soil-moisture feedbacks influence changes in both temperature and precipitation. More than 30% of CMIP5 models overestimate the influence of preceding precipitation (a proxy for soil moisture) on temperature extremes in Europe and the USA ( [[#Donat--2018|Donat et al., 2018]] ), and many CMIP5 models simulate an unrealistic influence of evaporation on temperature extremes for wet regions in Europe and the US ( [[#Ukkola--2018|Ukkola et al., 2018]] ). RCMs were found to realistically simulate the correlation between latent and sensible heat fluxes and temperature (coupling strength) over Africa ( [[#Knist--2017|Knist et al., 2017]] ; [[#Careto--2018|Careto et al., 2018]] ) and in northern and southern Europe, but to overestimate it in central Europe ( [[#Knist--2017|Knist et al., 2017]] ). Land surface models driven by global reanalysis agreed relatively well with observations. However, the coupling strength varied strongly across models at the regional scale, and a realistic partitioning of the incoming radiation into latent and sensible heat fluxes did not necessarily result in a realistic soil moisture-temperature coupling ( [[#Gevaert--2018|Gevaert et al., 2018]] ; [[#Boé--2020a|Boé et al., 2020a]] ).

Evaluating the representation of soil-moisture–precipitation feedbacks in climate models is challenging as different processes may induce feedbacks including moisture recycling, boundary-layer dynamics and mesoscale circulation. Moreover, the effects of soil moisture on precipitation may be region and scale dependent and may even change sign depending on the strength of the background flow ( [[#Taylor--2013|Taylor et al., 2013]] ; [[#Froidevaux--2014|Froidevaux et al., 2014]] ; [[#Guillod--2015|Guillod et al., 2015]] ; [[#Larsen--2016|Larsen et al., 2016]] ; [[#Tuttle--2016|Tuttle and Salvucci, 2016]] ). On seasonal-to-interannual time scales, CMIP5 models showed a stronger soil-moisture–precipitation feedback than estimated by satellite data ( [[#Levine--2016|Levine et al., 2016]] ). [[#Taylor--2013|Taylor et al. (2013)]] found that convection-permitting RCMs perform well at simulating surface-induced mesoscale circulations in daytime convection and the observed negative soil moisture feedback, whereas an RCM with parametrized convection, even when run at the same resolution, simulated an unrealistic positive feedback. There is ''medium evidence'' and ''high agreement'' that simulations at convection-permitting resolution are required to realistically represent soil-moisture–precipitation feedbacks.

Ocean–atmosphere RCMs have successfully been used to understand and simulate phenomena involving strong regional feedbacks like tropical cyclones in the Indian Ocean ( [[#Samson--2014|Samson et al., 2014]] ), Indian summer monsoon ( [[#Samanta--2018|Samanta et al., 2018]] ), East Asian summer monsoon ( [[#Zou--2016|Zou et al., 2016]] ), near coastline intense precipitation in the Mediterranean ( [[#Berthou--2015|Berthou et al., 2015]] , 2018), air-sea fluxes influencing heat and humidity advection over land ( [[#Sevault--2014|Sevault et al., 2014]] ; [[#Lebeaupin%20Brossier--2015|Lebeaupin Brossier et al., 2015]] ; [[#Akhtar--2018|Akhtar et al., 2018]] ) or snow bands in the Baltic region ( [[#Pham--2017|Pham et al., 2017]] ). The positive impact of ocean-coupling on the simulation of strongly convective phenomena such as Medicanes, a class of severe cyclones in the Mediterranean, can only be diagnosed when using relatively fine atmospheric resolution of about 10 km ( [[#Akhtar--2014|Akhtar et al., 2014]] ; [[#Flaounas--2018|Flaounas et al., 2018]] ; [[#Gaertner--2018|Gaertner et al., 2018]] ). A positive impact of ocean coupling has been quantified in marginal sea regions with reduced large-scale influence (e.g., in the Baltic Sea area during weak phases of the NAO and thus weak influence of Atlantic westerlies ( [[#Kjellström--2005|Kjellström et al., 2005]] ; [[#Pham--2018|Pham et al., 2018]] ). There is some evidence that coupled ocean components also positively impact RCM simulations of inland climates such as precipitation extremes in central Europe ( [[#Ho-Hagemann--2017|Ho-Hagemann et al., 2017]] ; [[#Akhtar--2019|Akhtar et al., 2019]] ). There ''is high confidence'' that coupled ocean–atmosphere RCMs improve the representation of ocean–atmosphere feedbacks and related phenomena.

The influence of ice-sheet mass balance on regional climate, explored with global and regional models by ( [[#Noël--2018|Noël et al., 2018]] ; [[#Fettweis--2020|Fettweis et al., 2020]] ), is discussed in [[IPCC:Wg1:Chapter:Chapter-9#9.4|Section 9.4]] .

<div id="10.3.3.6" class="h3-container"></div>

<span id="performance-at-simulating-regional-drivers-of-climate-and-climate-change"></span>
==== 10.3.3.6 Performance at Simulating Regional Drivers of Climate and Climate Change ====

<div id="h3-29-siblings" class="h3-siblings"></div>

Dust, with its regional character in both emissions and climatic influences, has traditionally been specified in climate simulations with a climatological estimate. In CMIP5 models, the influence of vegetation changes on mineral dust is largely underestimated while the influence of surface wind and precipitation are overestimated, resulting in a low bias of dust load ( [[#Pu--2018|Pu and Ginoux, 2018]] ). Interactive dust emission modules that simulate the dust optical depth in most of the key emission regions have only been recently introduced ( [[#Pu--2018|Pu and Ginoux, 2018]] ). However, coarse dust is underestimated in global models ( [[#Adebiyi--2020|Adebiyi and Kok, 2020]] ). Simulations of future changes in dust are hindered by the uncertainties in future regional wind and precipitation as the climate warms ( [[#Evan--2016|Evan et al., 2016]] ), in the effect of CO <sub>2</sub> fertilization on source extent ( [[#Huang--2017|Huang et al., 2017]] ), in the dust feedbacks ( [[#Evans--2019|Evans et al., 2019]] ), and in the effect of human activities that change land use and disturb the soil, including cropping and livestock grazing, recreation and urbanization, and water diversion for irrigation ( [[#Ginoux--2012|Ginoux et al., 2012]] ).

Volcanoes also provide forcings with a marked regional impact (Cross-Chapter Box 4.1). This implies that models are expected to capture these effects ( [[#Bethke--2017|Bethke et al., 2017]] ). Both proxy analyses and simulations have demonstrated reduced Asian monsoon precipitation after tropical and Northern Hemisphere (NH) volcanic eruptions due to reduced humidity and divergent circulation ( [[#Man--2014|Man and Zhou, 2014]] ; [[#Zhuo--2014|Zhuo et al., 2014]] ; F. [[#Liu--2016|]] [[#Liu--2016|Liu et al., 2016]] ; [[#Stevenson--2016|Stevenson et al., 2016]] ). Global model experiments ( [[#Zanchettin--2013|Zanchettin et al., 2013]] ; [[#Ortega--2015|Ortega et al., 2015]] ; [[#Sjolte--2018|Sjolte et al., 2018]] ; [[#Michel--2020|Michel et al., 2020]] ) have suggested that tropical volcanic eruptions (larger than the one from Mount Pinatubo in 1991) may lead to a positive phase of the winter NAO in the following few years (with an uncertainty on the exact years affected), but this influence is not well-reproduced in climate models and requires very large ensembles ( [[#Driscoll--2012|Driscoll et al., 2012]] ; [[#Toohey--2014|Toohey et al., 2014]] ; [[#Swingedouw--2017|Swingedouw et al., 2017]] ; [[#Ménégoz--2018b|Ménégoz et al., 2018b]] ). The ability to simulate the effect of volcanic aerosol in global models is evaluated in VolMIP ( [[#Zanchettin--2016|Zanchettin et al., 2016]] ). Given the relevance of volcanic aerosol, a good knowledge of the initial conditions is important because the response has proven to be sensitive to them ( [[#Ménégoz--2018a|Ménégoz et al., 2018a]] ; [[#Zanchettin--2019|Zanchettin et al., 2019]] ). A few decadal prediction systems have illustrated that current systems can predict some aspects of regional climate a few years in advance ( [[#Swingedouw--2017|Swingedouw et al., 2017]] ; [[#Illing--2018|Illing et al., 2018]] ; [[#Ménégoz--2018a|Ménégoz et al., 2018a]] ; [[#Hermanson--2020|Hermanson et al., 2020]] ). However, a better performance requires information about volcanic location ( [[#Haywood--2013|Haywood et al., 2013]] ; [[#Pausata--2015|Pausata et al., 2015]] ; [[#Stevenson--2016|Stevenson et al., 2016]] ; F. [[#Liu--2018a|]] [[#Liu--2018|]] [[#Liu--2018|Liu et al., 2018]] a ), strength ( [[#Emile-Geay--2008|Emile-Geay et al., 2008]] ; H.-G. [[#Lim--2016|]] [[#Lim--2016|Lim et al., 2016]] ; F. [[#Liu--2018b|]] [[#Liu--2018|]] [[#Liu--2018|Liu et al., 2018]] b ), and seasonality ( [[#Stevenson--2017|Stevenson et al., 2017]] ; [[#Sun--2019a|Sun et al., 2019a]] , b).

Some recent regional climate changes can only be simulated by climate models if anthropogenic aerosols are correctly included (Sections 10.4.2.1, 10.6.3 and 10.6.4; Chapters 6 and 8). Examples of the importance of correctly representing anthropogenic aerosols are the recent enhanced warming over Europe ( [[#Nabat--2014|Nabat et al., 2014]] ; [[#Dong--2017|Dong et al., 2017]] ), the cooling over the East Asian monsoon region, leading to a weakening of the monsoon ( [[IPCC:Wg1:Chapter:Chapter-8#8.3.2.4|Section 8.3.2.4]] ; [[#Song--2014|Song et al., 2014]] ; Q. [[#Wang--2017|]] [[#Wang--2017|Wang et al., 2017]] ), as well as changes in the monsoons of West Africa (Sections 8.3.2.4 and 10.4.2.1) and South Asia (Sections 8.3.2.4 and 10.6.3; [[#Undorf--2018|Undorf et al., 2018]] ). The relevance of appropriately representing anthropogenic aerosols has been widely studied in regional models ( [[#Boé--2020a|Boé et al., 2020a]] ; [[#Gutiérrez--2020|Gutiérrez et al., 2020]] ), with an advantage for models with interactive aerosol schemes ( [[#Drugé--2019|Drugé et al., 2019]] ; [[#Nabat--2020|Nabat et al., 2020]] ). Without a fully coupled chemistry module, radiative forcing can be simulated by including simple models of sulphate chemistry or specifying the optical properties from observations and prescribing the effect of aerosols on the cloud-droplet number ( [[#Fiedler--2017|Fiedler et al., 2017]] , 2019; [[#Stevens--2017|Stevens et al., 2017]] ). In all cases, the specification of the aerosol load limits the trustworthiness of the simulations at the regional scale when enough detail is not provided ( [[#Samset--2019|Samset et al., 2019]] ; [[#Shonk--2020|Shonk et al., 2020]] ; Z. [[#Wang--2021|]] [[#Wang--2021|Wang et al., 2021]] ).

The inclusion of irrigation in global and regional models over the South Asian monsoon region ( [[#10.6.3|Section 10.6.3]] ) has been found to be important to represent the monsoon circulation and rainfall correctly ( [[#Lucas-Picher--2011|Lucas-Picher et al., 2011]] ; [[#Guimberteau--2012|Guimberteau et al., 2012]] ; [[#Shukla--2014|Shukla et al., 2014]] ; [[#Tuinenburg--2014|Tuinenburg et al., 2014]] ; [[#Cook--2015a|Cook et al., 2015a]] ; [[#Devanand--2019|Devanand et al., 2019]] ). Similarly, the inclusion of irrigation over northern India and western Pakistan could be important for the correct simulation of precipitation over the Upper Indus Basin in northern Pakistan ( [[#Saeed--2013|Saeed et al., 2013]] ). Irrigation in the East African Sahel inhibits rainfall over the irrigated region and instead enhances rainfall to the east, coherent with both observations and theoretical understanding of the local circulation anomalies induced by the lower surface air temperatures over the irrigated region ( [[#Alter--2015|Alter et al., 2015]] ). Although several studies show how modelled irrigation reduces daytime temperature extremes, few compare modelled results with observations. Global model studies have found improvements in simulated surface temperature when including irrigation ( [[#Thiery--2017|Thiery et al., 2017]] ), in particular in areas where the model used has a strong land-atmosphere coupling ( [[#Chen--2019|Chen and Dirmeyer, 2019]] ). An RCM study over the North China Plain showed that the inclusion of irrigation led to a better representation of the observed nighttime warming ( [[#Chen--2018|Chen and Jeong, 2018]] ).

There is ''medium confidence'' that representing irrigation is important for a realistic simulation of South Asian monsoon precipitation. There is ''limited evidence'' that including irrigation in climate models improves the simulation of maximum and minimum daily temperatures as well as precipitation for other regions.

Regional land-radiation management, including modifying the albedo through, for instance, no-tillage practices, has been suggested as a measure to decrease regional maximum daily temperatures (see review in [[#Seneviratne--2018|Seneviratne et al., 2018]] ), but although modelled results and theoretical understanding are coherent, few studies have verified the results with observations. [[#Hirsch--2018|Hirsch et al. (2018)]] is an exception, showing that implementing minimal tillage, crop residue management and crop rotation in a global model over regions where it is practiced, improves the simulation of surface heat fluxes.

<div id="10.3.3.7" class="h3-container"></div>

<span id="statistical-downscaling-bias-adjustment-and-weather-generators"></span>
==== 10.3.3.7 Statistical Downscaling, Bias Adjustment and Weather Generators ====

<div id="h3-30-siblings" class="h3-siblings"></div>

The performance of statistical downscaling models, bias adjustment and weather generators is determined by the chosen model structure (e.g., to represent variability and extremes or spatial dependence) and, if applicable, the predictors selected ( [[#Maraun--2019a|Maraun et al., 2019a]] ). The VALUE initiative has assessed a range of such methods in a perfect-predictor experiment where the predictors are taken from reanalysis data ( [[#Maraun--2015|Maraun et al., 2015]] , 2019a; [[#Gutiérrez--2019|Gutiérrez et al., 2019]] ). Table 10.2 shows an overview comprising performance results from VALUE and other studies. These results isolate the performance of the statistical method in the present climate. The overall performance in a climate change application also depends on the performance of the driving climate model (Sections 10.3.3.3–10.3.3.6) and the fitness of both the driving model and the statistical method for projecting the climatic aspects of interest ( [[#10.3.3.9|Section 10.3.3.9]] ).

<div id="_idContainer024" class="Basic-Text-Frame"></div>

'''Table''' '''10.2 |''' '''Performance of different statistical method types in representing local weather at daily resolution.''' Individual state-of-the-art implementations may perform better. ‘+’: should work reasonably well based on empirical evidence and/or expert judgement; ‘o’: problems may arise depending on the specific context; ‘–’: weak performance either by construction or inferred from empirical evidence; ‘?’: not studied. The categorisation assumes that predictors are provided by a well-performing dynamical model. Statements about extremes refer to moderate events occurring at least once every 20 years. Adopted and extended from Maraun and Widmann (2018b).

[[File:8cd8d3e180d27cffffe2f61936399f18 IPCC_AR6_WGI_Chapter_10_Table_10_2.png]]
<div id="10.3.3.7.1" class="h4-container"></div>

<span id="performance-of-perfect-prognosis-methods"></span>
===== 10.3.3.7.1 Performance of perfect prognosis methods =====

<div id="h4-12-siblings" class="h4-siblings"></div>

Perfect prognosis methods can perform well when the synoptic forcing (i.e., the explanatory power of large-scale predictors) is strong ( [[#Schoof--2013|Schoof, 2013]] ). Using this approach, downscaling of precipitation is particularly skilful in the presence of strong orographic forcing. The representation of daily variability and extremes requires analogue methods or stochastic regression models, although the former typically do not extrapolate to unobserved values ( [[#Gutiérrez--2019|Gutiérrez et al., 2019]] ; [[#Hertig--2019|Hertig et al., 2019]] ). Temporal precipitation variability is well-represented by analogue methods and stochastic regression, but analogue methods typically underestimate temporal dependence of temperature ( [[#Maraun--2019b|Maraun et al., 2019b]] ). Spatial dependence of both temperature and precipitation is only well-represented by analogue methods, for which analogues are defined jointly across locations, and by stochastic regression methods explicitly representing spatial dependence ( [[#Widmann--2019|Widmann et al., 2019]] ). Overall, there is ''high confidence'' that analogue methods and stochastic regression are able to represent many aspects of daily temperature and variability, but the analogue method is inherently limited in representing climate change ( [[#Gutiérrez--2013|Gutiérrez et al., 2013]] ).

<div id="10.3.3.7.2" class="h4-container"></div>

<span id="performance-of-bias-adjustment-methods"></span>
===== 10.3.3.7.2 Performance of bias adjustment methods =====

<div id="h4-13-siblings" class="h4-siblings"></div>

This subsection assesses the performance of bias adjustment in a perfect predictor context. In practice, climate model imperfections may cause substantial additional issues in the application of bias adjustment. These are assessed separately in Cross-Chapter Box 10.2.

Bias adjustment methods, if driven by reanalysis predictors, in principle adjust well all the aspects that they intend to address ( [[#Maraun--2018b|Maraun and Widmann, 2018b]] ). For temperature, all univariate methods are good for adjusting means, variance, and high quantiles ( [[#Gutiérrez--2019|Gutiérrez et al., 2019]] ; [[#Hertig--2019|Hertig et al., 2019]] ). For precipitation, means, intensities, wet-day frequencies, and wet–dry and dry–wet transitions are well-adjusted ( [[#Gutiérrez--2019|Gutiérrez et al., 2019]] ; [[#Maraun--2019b|Maraun et al., 2019b]] ). The representation of high quantiles depends on the chosen method, although flexible quantile mapping performs best ( [[#Hertig--2019|Hertig et al., 2019]] ). Empirical (non-parametric) methods perform better than parametric methods over the observed range, but it is unclear how this translates into extrapolation to unobserved values (IPCC, 2015; [[#Hertig--2019|Hertig et al., 2019]] ). Many quantile mapping methods overestimate interannual variability ( [[#Maraun--2019b|Maraun et al., 2019b]] ). Temporal and spatial dependence are usually not adjusted and thus inherited from the driving model ( [[#Maraun--2019b|Maraun et al., 2019b]] ; [[#Widmann--2019|Widmann et al., 2019]] ). Spatial fields are thus typically too smooth in space, even after bias adjustment ( [[#Widmann--2019|Widmann et al., 2019]] ).

Several studies show improved simulations of present-day impacts, when the impact model is fed with bias-adjusted climate model output, including the assessment of river discharge ( [[#Rojas--2011|Rojas et al., 2011]] ; [[#Muerth--2013|Muerth et al., 2013]] ; [[#Montroull--2018|Montroull et al., 2018]] ), forest fires ( [[#Migliavacca--2013|Migliavacca et al., 2013]] ), crop production ( [[#Ruiz-Ramos--2016|Ruiz-Ramos et al., 2016]] ), and regional ocean modelling ( [[#Macias--2018|Macias et al., 2018]] ).

There is ''high confidence'' that bias adjustment can improve the marginal distribution of simulated climate variables, if applied to a climate model that adequately represents the processes relevant for a given application (Cross-Chapter Box 10.2).

<div id="10.3.3.7.3" class="h4-container"></div>

<span id="performance-of-weather-generators"></span>
===== 10.3.3.7.3 Performance of weather generators =====

<div id="h4-14-siblings" class="h4-siblings"></div>

Weather generators represent well most aspects that are explicitly calibrated. This typically includes mean, variance, high quantiles (for precipitation, if explicitly modelled), and short-term temporal variability for both temperature and precipitation, whereas interannual variability is strongly underestimated ( [[#Frost--2011|Frost et al., 2011]] ; [[#Hu--2013a|Hu et al., 2013a]] ; [[#Keller--2015|Keller et al., 2015]] ; [[#Dubrovsky--2019|Dubrovsky et al., 2019]] ; [[#Gutiérrez--2019|Gutiérrez et al., 2019]] ; [[#Hertig--2019|Hertig et al., 2019]] ; [[#Maraun--2019b|Maraun et al., 2019b]] ; [[#Widmann--2019|Widmann et al., 2019]] ). There is growing evidence that some spatial weather generators fairly realistically capture the spatial dependence of temperature and precipitation ( [[#Frost--2011|Frost et al., 2011]] ; [[#Hu--2013a|Hu et al., 2013a]] ; [[#Keller--2015|Keller et al., 2015]] ; [[#Evin--2018|Evin et al., 2018]] ; [[#Dubrovsky--2019|Dubrovsky et al., 2019]] ). There is ''high confidence'' that weather generators can realistically simulate a wide range of local weather characteristics at single locations, but there is ''limited evidence'' and ''low agreement'' of the ability of weather generators to realistically simulate the spatial dependence of atmospheric variables across multiple sites.

<div id="10.3.3.8" class="h3-container"></div>

<span id="performance-at-simulating-historical-regional-climate-changes"></span>
==== 10.3.3.8 Performance at Simulating Historical Regional Climate Changes ====

<div id="h3-31-siblings" class="h3-siblings"></div>

This section assesses how well climate models perform at realistically simulating historical regional climatic trends. Current global model ensembles reproduce global to continental-scale surface temperature trends at multi-decadal to centennial time scales (CMIP5, CMIP6), but underestimate precipitation trends (CMIP5) (Sections 3.3.1 and 3.3.2). For regional trends, AR5 concluded that the CMIP5 ensemble cannot be taken as a reliable representation of reality and that the true uncertainty can be larger than the simulated model spread ( [[#Kirtman--2014|Kirtman et al., 2014]] ). Case studies of regional trend simulations by global models can be found in Sections 10.4.1 and 10.6, and region-by-region assessments in the Atlas. A key limitation for assessing the representation of regional observed trends by single transient simulations of global models (or downscaled versions thereof) is the strong amplitude of internal variability compared to the forced signal at the regional scale ( [[#10.3.4.3|Section 10.3.4.3]] ). Even on multi-decadal time scales, an agreement between observed and individual simulated trends would be expected to occur only by chance (Laprise, 2014).

In the context of downscaling, the ability of downscaling methods to reproduce observed trends when driven with boundary conditions or predictors taken from reanalysis data (which reproduce the observed internal variability on long time scales) can be assessed. For temperature in the continental USA, reanalysis-driven RCMs skilfully simulated recent spring and winter trends, but did not reproduce summer and autumn trends, ( [[#Bukovsky--2012|Bukovsky, 2012]] ). Over Central America, observed warming trends were reproduced ( [[#Cavazos--2020|Cavazos et al., 2020]] ). In contrast, a reanalysis-driven coupled atmosphere–ocean RCM covering the Mediterranean could not reproduce the observed SST trend ( [[#Sevault--2014|Sevault et al., 2014]] ).

Similar studies have been carried out for statistical downscaling and bias adjustment using predictors from reanalyses (or in case of bias adjustment, dynamically downscaled reanalyses). For a range of different perfect prognosis methods, [[#Huth--2015|Huth et al. (2015)]] found that simulated temperature trends were too strong for winter and too weak for summer. The performance was similar for the different methods, indicating the importance of choosing informative predictors. Similarly, Maraun et al. (2019b) found that the performance of perfect prognosis methods depends mostly on the predictor and domain choice (for instance, temperature trends were only captured by those methods including surface temperature as predictor). Bias adjustment methods reproduced the trends of the driving reanalysis, apart from quantile mapping methods, which deteriorated these trends.

RCM experiments are often set up such that changes in forcing agents are included only via the boundary conditions, but not explicitly included inside the domain. [[#Jerez--2018|Jerez et al. (2018)]] demonstrated that not including time-varying GHG concentrations within the RCM domain may misrepresent temperature trends by 1–2°C per century. Including the past trend in anthropogenic sulphate aerosols in reanalysis-driven RCM simulations substantially improved the representation of recent brightening and warming trends in Europe ( [[#Nabat--2014|Nabat et al., 2014]] ; see Sections 10.3.3.6 and 10.6.4, and Atlas.8.4). Similarly, [[#Bukovsky--2012|Bukovsky (2012)]] argued that RCMs may not capture observed summer temperature trends in the USA because changes in land cover are not taken into account. [[#Barlage--2015|Barlage et al. (2015)]] have revealed that including the behaviour of groundwater in land schemes increases the performance of an RCM model to represent climate variability in the central USA. [[#Hamdi--2014|Hamdi et al. (2014)]] found that an RCM that did not incorporate the historical urbanization in the land-use, land-cover scheme is not able to reproduce the warming trend observed in urban stations, with a larger bias for the minimum temperature trend.

Overall, there is ''high confidence'' that including all relevant forcings is a prerequisite for reproducing historical trends.

<div id="10.3.3.9" class="h3-container"></div>

<span id="fitness-of-climate-models-for-projecting-regional-climate"></span>
==== 10.3.3.9 Fitness of Climate Models for Projecting Regional Climate ====

<div id="h3-32-siblings" class="h3-siblings"></div>

AR5 stated that confidence in climate model projections is based on the physical understanding of the climate system and its representation in climate models. A climate model’s credibility for future projections may be increased if the model is able to simulate past variations in climate (Sections 10.3.3.8, 10.4.1 and 10.6; [[#Flato--2014|Flato et al., 2014]] ). In particular, the credibility of downscaled information depends on the quality of both the downscaling method and of the global model providing the large-scale boundary conditions ( [[#Flato--2014|Flato et al., 2014]] ). Credibility is closely linked to the concept of adequacy or fitness-for-purpose ( [[IPCC:Wg1:Chapter:Chapter-1#1.5.4.1|Section 1.5.4.1]] ; [[#Parker--2009|Parker, 2009]] ). From a regional perspective, one may ask about the fitness of a climate model for simulating future changes of specific aspects of a specific regional climate. The required level of model fitness may depend on the user context ( [[#10.5|Section 10.5]] ). A key challenge is to link performance at representing present and past climate (Sections 10.3.3.3–10.3.3.8) to the confidence in future projections ( [[IPCC:Wg1:Chapter:Chapter-1#1.3.5|Section 1.3.5]] ; [[#Baumberger--2017|Baumberger et al., 2017]] ) and it is addressed in this subsection.

A general idea of model fitness for a given application may be obtained by checking whether relevant large- ( [[#10.3.3.4|Section 10.3.3.4]] ) and regional-scale (Sections 10.3.3.5 and 10.3.3.6) processes are explicitly resolved (Figure 10.3). The basis for confidence in climate projection is a solid process understanding ( [[#Flato--2014|Flato et al., 2014]] ; [[#Baumberger--2017|Baumberger et al., 2017]] ). Thus, the key to assessing the fitness-for-purpose of a model is the evaluation of how relevant processes controlling regional climate are represented ( [[#Collins--2018|Collins et al., 2018]] ). A process-based evaluation may be more appropriate than an evaluation of the variables of interest (e.g., temperature, precipitation), because biases in the latter may in principle be reduced if the underlying processes are realistically simulated (Cross-Chapter Box 10.2), while individual variables may appear as well-represented because of compensating errors ( [[#Flato--2014|Flato et al., 2014]] ; [[#Baumberger--2017|Baumberger et al., 2017]] ). Combining a process-based evaluation with a mechanistic explanation of projected changes further increases confidence in projections ( [[#Bukovsky--2017|Bukovsky et al., 2017]] ). Fitness-for-purpose can also be assessed by comparing the simulated response of a model with simulations of higher resolution models that better represent relevant processes ( [[#Baumberger--2017|Baumberger et al., 2017]] ). For instance, [[#Giorgi--2016|Giorgi et al. (2016)]] have corroborated their findings on precipitation changes comparing standard RCM simulations with convection-permitting simulations.

The evaluation of model performance at historical variability and long-term changes provides further relevant information ( [[#Flato--2014|Flato et al., 2014]] ). Trend evaluation may provide very useful insight, but has limitations, in particular at the regional scale, mainly due to multi-decadal internal climate variability ( [[#10.3.3.8|Section 10.3.3.8]] ), observational uncertainty (in both driving reanalysis and local trends; [[#10.2|Section 10.2]] ), and the fact that often not all regional forcings are known, and that past trends may be driven by forcings other than those driving future trends (Sections 10.4.1 and 10.6.3).

Increasing resolution ( [[#Haarsma--2016|Haarsma et al., 2016]] ) or performing downscaling may be particularly important when it modifies the climate change signal of a lower resolution model in a physically plausible way ( [[#Hall--2014|Hall, 2014]] ). Improvements may result from a better representation of regional processes, upscale effects, as well as the possibility of a region-specific model tuning ( [[#Sørland--2018|Sørland et al., 2018]] ). For instance, [[#Gula--2012|Gula and Peltier (2012)]] showed that a higher resolution allows for a more realistic simulation of lake-induced precipitation, resulting in a more credible projection of changes in the snow belts of the North American Great Lakes. Similarly, [[#Giorgi--2016|Giorgi et al. (2016)]] demonstrated that an ensemble of RCMs better represents high-elevation surface heating and in turn increased convective instability. As a result, the summer convective precipitation response was opposite to that simulated by the driving global models (Figure 10.9). Similarly, [[#Walton--2015|Walton et al. (2015)]] showed that a kilometre-scale RCM enables a more realistic representation of the snow-albedo feedback in mountainous terrain compared to standard resolution global models, leading to a more plausible simulation of elevation-dependent warming. [[#Bukovsky--2017|Bukovsky et al. (2017)]] argue that strong seasonal changes in warm-season precipitation in the Southern Great Plains of the USA, projected by RCMs, are more credible than the weaker global model changes because precipitation is better simulated in the RCMs.

<div id="_idContainer034" class="Basic-Text-Frame"></div>

[[File:922aad415db36de19ede6fb50098babc IPCC_AR6_WGI_Figure_10_9.png]]

'''Figure 10.''' '''9 |''' '''Projected changes in summer (June to August) precipitation (in percent with respect to the mean precipitation) over the Alps between the periods 2070–2099 and 1975–2004. (a)''' Mean of four global climate models (GCMs) regridded to a common 1.32° × 1.32° grid resolution; '''(b)''' mean of six regional climate models (RCMs) driven with these GCMs. The grey isolines show elevation at 200 m intervals of the underlying model data. Further details on data sources and processing are available in the chapter data table (Table 10.SM.11). Figure adapted from [[#Giorgi--2016|Giorgi et al. (2016)]] .

Including additional components, feedbacks and drivers can substantially modify the simulated future climate. For example, Kjellström et al. (2005) and [[#Somot--2008|Somot et al. (2008)]] have shown that a regional ESM can significantly modify the SST response to climate change of its driving global model with implications for the climate change signal over both the sea and land. In particular, coupled ocean–atmosphere RCMs may increase the credibility of projections in regions of strong air-sea coupling such as the East Asia–western North Pacific domain ( [[#Zou--2016b|Zou and Zhou, 2016b]] , 2017). Recent studies demonstrate the importance of including regional patterns of evolving aerosols in RCMs for simulating regional climate change ( [[#Boé--2020a|Boé et al., 2020a]] ; [[#Gutiérrez--2020|Gutiérrez et al., 2020]] ). RCMs not including the plant physiological response to increasing CO <sub>2</sub> concentrations have been shown to substantially underestimate projected increases in extreme temperatures across Europe compared to global models that explicitly model this effect ( [[#Schwingshackl--2019|Schwingshackl et al., 2019]] ).

A difference between the climate changes simulated by two models does not automatically imply the more complex or higher resolution model is superior (e.g., [[#Dosio--2019|Dosio et al., 2019]] ). Studies comparing convection-permitting RCM simulations to simulations of climate models with parametrized convection find, depending on the considered models, regions and seasons, either similar or qualitatively different projected changes in short duration extreme precipitation ( [[#Chan--2014a|Chan et al., 2014a]] , b, 2020; [[#Ban--2015|Ban et al., 2015]] ; [[#Tabari--2016|Tabari et al., 2016]] ; [[#Fosser--2017|Fosser et al., 2017]] ; [[#Kendon--2017|Kendon et al., 2017]] , 2019; [[#Vanden%20Broucke--2018|Vanden Broucke et al., 2018]] ). Process studies provide evidence that convection-permitting simulations better represent crucial local and mesoscale features of convective storms and thus simulate more plausible changes ( [[#Meredith--2015a|Meredith et al., 2015a]] ; [[#Prein--2017|Prein et al., 2017]] ; [[#Fitzpatrick--2020|Fitzpatrick et al., 2020]] ), but further research is required to confirm and reconcile the different findings.

Studies assessing the fitness of statistical approaches for regional climate projections are still very limited in number. For statistical downscaling, a key issue is to include predictors that control long-term changes in regional climate. Models differing only in the choice of predictors may perform similarly in the present climate, but may project opposite precipitation changes ( [[#Fu--2018|Fu et al., 2018]] ; [[#Manzanas--2020|Manzanas et al., 2020]] ). In addition to trend-evaluation studies ( [[#10.3.3.8|Section 10.3.3.8]] ), perfect-model experiments ( [[#10.3.2.5|Section 10.3.2.5]] ) have been used to assess whether a given model structure with a chosen set of predictors is capable of reproducing the simulated future climates ( [[#Gutiérrez--2013|Gutiérrez et al., 2013]] ; [[#Räty--2014|Räty et al., 2014]] ; [[#Dayon--2015|Dayon et al., 2015]] ; [[#Dixon--2016|Dixon et al., 2016]] ; [[#San-Martín--2017|San-Martín et al., 2017]] ). Importantly, it is found that standard analogue methods inherently underestimate future warming trends because of missing analogues for a warmer climate ( [[#Gutiérrez--2013|Gutiérrez et al., 2013]] ).

Bias adjustment assumes that model biases are time invariant (or more precisely, independent of the climate state), such that the adjustment made to present climate simulations is still applicable to future climate simulations. Many findings challenge the validity of this assumption, as already assessed in AR5 ( [[#Flato--2014|Flato et al., 2014]] ). Further research has addressed this issue by means of perfect model experiments ( [[#10.3.2.5|Section 10.3.2.5]] ) and process understanding. Perfect-model studies with GCMs found that circulation, energy, and water-cycle biases are roughly state-independent ( [[#Krinner--2018|Krinner and Flanner, 2018]] ), whereas temperature biases depend linearly on temperature ( [[#Kerkhoff--2014|Kerkhoff et al., 2014]] ). Others show that regional temperature biases may depend on soil moisture and albedo, and may thus be state-dependent ( [[#Maraun--2012|Maraun, 2012]] ; [[#Bellprat--2013|Bellprat et al., 2013]] ; [[#Maraun--2017|Maraun et al., 2017]] ; see Cross-Chapter Box 10.2 for further limitations of bias adjustment). The fitness of weather generators for future projections depends on whether they account for all relevant changes in their parameters, either by predictors or change factors ( [[#Maraun--2018b|Maraun and Widmann, 2018b]] ).

In any case, the fitness of regional climate projections based on dynamical downscaling or statistical approaches depends on the fitness of the driving models in projecting boundary conditions, predictors and change factors ( [[#Hall--2014|Hall, 2014]] ; [[#Maraun--2018b|Maraun and Widmann, 2018b]] ).

Overall, there is ''high confidence'' that an assessment of model fitness for projections applying process-based evaluation, process-based plausibility checks of projections and a comparison of different model types, increases the confidence in climate projections. There is ''high confidence'' that increasing model resolution, dynamical downscaling, statistical downscaling with well-simulated predictors controlling regional climate change, and adding relevant model components can increase the fitness for projecting some aspects of regional climate when accompanied by a process-understanding analysis.

<div id="10.3.3.10" class="h3-container"></div>

<span id="synthesis-of-model-performance-at-simulating-regional-climate-and-climate-change"></span>
==== 10.3.3.10 Synthesis of Model Performance at Simulating Regional Climate and Climate Change ====

<div id="h3-33-siblings" class="h3-siblings"></div>

Global models reproduce many of the features of observed climate and its variability at regional scales. However, global models can show a variety of biases in, for instance, precipitation and temperature at scales ranging from continental ( [[#Prasanna--2016|Prasanna, 2016]] ) to sub-continental scales ( [[#Lovino--2018|Lovino et al., 2018]] ), both in the mean and in higher order moments of the climatological distribution of the variable (Figure 10.6; [[#Ren--2019|Ren et al., 2019]] ; [[#Xin--2020|Xin et al., 2020]] ). Regional biases could occur even if all the relevant large-scale processes are correctly represented, but not their interaction with regional features such as orography or land–sea contrasts ( [[#10.3.3.4|Section 10.3.3.4]] ). These biases have been considered an important limiting factor in model usability, especially at the regional scale ( [[#Palmer--2016|Palmer, 2016]] ). In spite of this, global model simulations have been extensively used to create regional estimates of climate change (Chapters 11, 12 and Atlas), taking into account the result of a performance assessment (Chapter 11, Sections 10.3.3.3–10.3.3.8, and Atlas; [[#Jiang--2020|Jiang et al., 2020]] ). However, their application is limited in part by the effective resolution of these models ( [[#Klaver--2020|Klaver et al., 2020]] ).

Global model performance at the regional scale is assessed in terms of the time or spatial averages of key variables (see Atlas; [[#Brunner--2019|Brunner et al., 2019]] ), the ability to reproduce their seasonal cycle ( [[#Hasson--2013|Hasson et al., 2013]] ) or a set of extreme climate indicators (Chapter 11; [[#Luo--2020|Luo et al., 2020]] ) and the representation of regional processes and phenomena, feedbacks, drivers and forcing impacts (Sections 10.3.3.4–10.3.3.6). In many cases, the performance estimates have been used to select models for either an application or a more in-depth study ( [[#Lovino--2018|Lovino et al., 2018]] ), to select the models that provide boundary conditions to perform RCM simulations ( [[#McSweeney--2015|McSweeney et al., 2015]] ) or to weight the results of the global model simulations ( [[#Sanderson--2015|Sanderson et al., 2015]] ; [[#Brunner--2020|Brunner et al., 2020]] ). While some large-scale metrics are improved between the CMIP5 and CMIP6 experiments (Chapter 3; [[#Cannon--2020|Cannon, 2020]] ), there is not yet concluding evidence of a systematic improvement for surface variables at the regional scale.

The special class of high-resolution global models (Sections 1.5.3.1 and 10.3.3.1, Chapter 3; [[#Haarsma--2016|Haarsma et al., 2016]] ; [[#Prodhomme--2016|Prodhomme et al., 2016]] ) is expected to improve some of the regional processes that are not appropriately represented in standard global models ( [[#Roberts--2018|Roberts et al., 2018]] ). There is general consensus that increasing global model resolution improves some long-standing biases (Chapter 3, [[#10.3.3.3|Section 10.3.3.3]] , and Figures 10.6 and 10.7; [[#Demory--2014|Demory et al., 2014]] , 2020; [[#Schiemann--2014|Schiemann et al., 2014]] ; [[#Dawson--2015|Dawson and Palmer, 2015]] ; [[#van%20Haren--2015|van Haren et al., 2015]] ; [[#Feng--2017|Feng et al., 2017]] ; [[#Fabiano--2020|Fabiano et al., 2020]] ), although the resolution increase is not a guarantee of overall improvement ( [[IPCC:Wg1:Chapter:Chapter-8#8.5.1|Section 8.5.1]] ; [[#Fabiano--2020|Fabiano et al., 2020]] ; [[#Hertwig--2021|Hertwig et al., 2021]] ). For instance, increasing resolution in global models has been shown to improve Asian monsoon rainfall anchored to orography and the monsoon circulation ( [[#Johnson--2016|Johnson et al., 2016]] ), but fails to solve the major dry bias. It is also difficult to disentangle the role of resolution increase and model tuning on the performance of the GCM ( [[#Anand--2018|Anand et al., 2018]] ). Some efforts have been undertaken to complement the performance improvements of resolution by using stochastic parametrizations ( [[#Palmer--2019|Palmer, 2019]] ), which explicitly acknowledge the multi-scale nature of the climate system, in standard resolution global models with some success ( [[#Dawson--2015|Dawson and Palmer, 2015]] ; [[#MacLeod--2016|MacLeod et al., 2016]] ; [[#Zanna--2017|Zanna et al., 2017]] , 2019). The expectation is to achieve a similar performance to the increase in resolution at a reduced computational cost.

Despite their known errors that affect model performance, there is ''high confidence'' that global models provide useful information for the production of regional climate information. There is ''robust evidence'' and ''high agreement'' that the increase of global model resolution helps in reducing the biases limiting performance at the regional scale, although resolution per se does not automatically solve all performance limitations shown by global models. There is ''robust evidence'' that stochastic parametrizations can help to improve some aspects of the global model performance that are relevant to regional climate information.

Global models tend to have difficulties in simulating climate over regions where unresolved local scale processes, feedbacks and non-linear scale interactions result in a degradation of the model performance compared to models with higher resolution. In this case, RCMs and variable resolution global models can resolve part of these processes in the regions of interest at an acceptable computational cost ( [[#Rummukainen--2016|Rummukainen, 2016]] ; [[#Giorgi--2019|Giorgi, 2019]] ; [[#Gutowski%20Jr.--2020|Gutowski Jr. et al., 2020]] ).

The assessment of RCM performance needs to focus not only on mean climatology (Atlas), but also trends ( [[#10.3.3.8|Section 10.3.3.8]] ) and extremes (Chapter 11), and the RCM’s ability at correctly reproducing relevant processes, forcings and feedbacks including aerosols, plant responses to increasing CO <sub>2</sub> , and so on, ( [[#Schwingshackl--2019|Schwingshackl et al., 2019]] ; [[#Boé--2020a|Boé et al., 2020a]] ; Sections 11.2. and 10.3.3.3 to 10.3.3.8) to be fit for future projections ( [[#10.3.3.9|Section 10.3.3.9]] ).

When RCMs are driven by global models, part of the uncertainty in the RCM simulation is introduced by the global model biases ( [[#Kjellström--2018|Kjellström et al., 2018]] ; [[#Sørland--2018|Sørland et al., 2018]] ; [[#Christensen--2020|Christensen and Kjellström, 2020]] ). As RCMs are typically not able to mitigate global model biases in large-scale dynamical processes, if such biases are substantial, and if the corresponding large-scale processes are important drivers of regional climate, downscaling is questionable ( [[#10.3.3.3|Section 10.3.3.3]] ). However, when global models have weak circulation biases and regional climate change is controlled mainly by regional-scale processes and feedbacks, dynamical downscaling has the potential to add substantial value to global model simulations ( [[#10.3.3.4|Section 10.3.3.4]] and Atlas; [[#Hall--2014|Hall, 2014]] ; [[#Rummukainen--2016|Rummukainen, 2016]] ; [[#Giorgi--2019|Giorgi, 2019]] ; [[#Schwingshackl--2019|Schwingshackl et al., 2019]] ; [[#Boé--2020a|Boé et al., 2020a]] ; [[#Lloyd--2021|Lloyd et al., 2021]] ).

There is ''very high confidence'' ( ''robust evidence'' and ''high agreement'' ) that RCMs add value to global simulations in representing many regional weather and climate phenomena, especially over regions of complex orography or with heterogeneous surface characteristics and for local-scale phenomena. Realistically representing local-scale phenomena such as land–sea breezes requires simulations at a resolution of the order of 10 km ( ''high confidence'' ). Simulations at kilometre-scale resolution add value in particular to the representation of convection, sub-daily summer precipitation extremes ( ''high confidence'' ) and soil-moisture–precipitation feedbacks ( ''medium confidence'' ). Resolving regional processes may be required to correctly represent the sign of regional climate change ( ''medium confidence'' ). However, the performance of RCMs and their fitness for future projections depend on their representation of relevant processes, forcings and drivers in the specific context (Sections 10.3.3.4–10.3.3.8).

Statistical downscaling, bias adjustment and weather generators outperform uncorrected output of global and regional models for a range of statistical aspects at single locations due to their calibration ( [[#Casanueva--2016|Casanueva et al., 2016]] ), but RCMs are superior when spatial fields are relevant ( [[#Mehrotra--2014|Mehrotra et al., 2014]] ; [[#Vaittinada%20Ayar--2016|Vaittinada Ayar et al., 2016]] ; [[#Maraun--2019a|Maraun et al., 2019a]] ). Similarly, there is some evidence that bias adjustment is comparable in performance when applied to global models and dynamically downscaled global models only for single locations, but dynamical downscaling prior to bias adjustment clearly adds value once spatial dependence is relevant ( [[#Maraun--2019a|Maraun et al., 2019a]] ). These results may explain why dynamical downscaling does not add value to global model simulations for (single-site) agricultural modelling, when both global and regional models are bias adjusted ( [[#Glotter--2014|Glotter et al., 2014]] ), but dynamical downscaling adds value compared to bias-adjusted global model output for spatially distributed hydrological models ( [[#Qiao--2014|Qiao et al., 2014]] ).

Overall, statistical downscaling methods with carefully chosen predictors and an appropriate model structure for a given application realistically represent many statistical aspects of present-day daily temperature and precipitation ( ''high confidence'' , [[#10.3.3.7|Section 10.3.3.7]] ). Bias adjustment has proven beneficial as an interface between climate model projections and impact modelling in many different contexts ( ''high confidence'' ) ( [[#10.3.3.7|Section 10.3.3.7]] ). Weather generators realistically simulate many statistical aspects of present-day daily temperature and precipitation ( ''high confidence'' ) ( [[#10.3.3.7|Section 10.3.3.7]] ). The performance of these approaches and their fitness for future projections also depends on predictors and change factors taken from the driving dynamical models ( ''high confidence'' ) ( [[#10.3.3.9|Section 10.3.3.9]] ).

<div id="10.3.4" class="h2-container"></div>

<span id="managing-uncertainties-in-regional-climate-projections"></span>