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

==== 11.7.1.3 Model Evaluation ====

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

Accurate projections of future TC activity have two principal requirements: accurate representation of changes in the relevant environmental factors (e.g., SSTs) that can affect TC activity, and accurate representation of actual TC activity in given environmental conditions.In particular, models’ capacity to reproduce historical trends or interannual variabilities of TC activity is relevant to the confidence in future projections. One test of the models is to evaluate their ability to reproduce the dependency of the TC statistics in the different basins in the real world, in addition to their capability of reproducing atmospheric and ocean environmental conditions. For the evaluation of projections of TC-relevant environmental variables, AR5 confidence statements were based on global surface temperature and moisture, but not on the detailed regional structure of SST and atmospheric circulation changes such as steering flows and vertical shear, which affect characteristics of TCs (genesis, intensity, tracks, etc.). Various aspects of TC metrics are used to evaluate how capable models are of simulating present-day TC climatologies and variability (e.g., TC frequency, wind intensity, precipitation, size, tracks, and their seasonal and interannual changes) ( [[#Walsh--2015|Walsh et al., 2015]] ; [[#Camargo--2016|Camargo and Wing, 2016]] ; [[#Knutson--2019|Knutson et al., 2019]] , 2020). Other examples of TC climatology/variability metrics are spatial distributions of TC occurrence and genesis ( [[#Walsh--2015|Walsh et al., 2015]] ), seasonal cycles and interannual variability of basin-wide activity ( [[#Zhao--2009|Zhao et al., 2009]] ; [[#Shaevitz--2014|Shaevitz et al., 2014]] ; [[#Kodama--2015|Kodama et al., 2015]] ; [[#Murakami--2015|Murakami et al., 2015]] ; [[#Yamada--2017|Yamada et al., 2017]] ) or landfalling activity ( [[#Lok--2018|Lok and Chan, 2018]] ), as well as newly developed process-diagnostics designed specifically for TCs in climate models (D. [[#Kim--2018|]] [[#Kim--2018|]] [[#Kim--2018|Kim et al., 2018]] ; [[#Wing--2019|Wing et al., 2019]] ; [[#Moon--2020|Moon et al., 2020]] ).

Confidence in the projection of intense TCs, such as those of Category 4–5, generally becomes higher as the resolution of the models becomes higher. The Coupled Model Intercomparison Project Phases 5 and 6 (CMIP5/6) class climate models (around 100–200 km grid spacing) cannot simulate TCs of Category 4–5 intensity. They do simulate storms of relatively high vorticity that are at best described as ‘TC-like’, but metrics such as storm counts are highly dependent on tracking algorithms ( [[#Camargo--2013|Camargo, 2013]] ; [[#Wehner--2015|Wehner et al., 2015]] ; [[#Zarzycki--2017|Zarzycki and Ullrich, 2017]] ; [[#Roberts--2020a|Roberts et al., 2020a]] ). High-resolution GCMs (around 10–60 km grid spacing), as used in HighResMIP ( [[#Haarsma--2016|Haarsma et al., 2016]] ; [[#Roberts--2020a|Roberts et al., 2020a]] ), begin to capture some structures of TCs more realistically, as well as produce intense TCs of Category 4–5 despite the effects of parametrized deep cumulus convection processes ( [[#Murakami--2015|Murakami et al., 2015]] ; [[#Wehner--2015|Wehner et al., 2015]] ; [[#Yamada--2017|Yamada et al., 2017]] ; [[#Roberts--2018|Roberts et al., 2018]] ; [[#Moon--2020|Moon et al., 2020]] ). Convection-permitting models (around 1–10 km grid-spacing), such as used in some dynamical downscaling studies, provide further realism with capturing TC eye-wall structures ( [[#Tsuboki--2015|Tsuboki et al., 2015]] ). Model characteristics besides resolution, especially details of convective parametrization, can influence a model’s ability to simulate intense TCs ( [[#Reed--2011|Reed and Jablonowski, 2011]] ; [[#Zhao--2012|Zhao et al., 2012]] ; [[#He--2015|He and Posselt, 2015]] ; D. [[#Kim--2018|]] [[#Kim--2018|]] [[#Kim--2018|Kim et al., 2018]] ; [[#Zhang--2018|Zhang and Wang, 2018]] ; [[#Camargo--2020|Camargo et al., 2020]] ). However, models’ dynamical cores and other physics also affect simulated TC properties ( [[#Reed--2015|Reed et al., 2015]] ; [[#Vidale--2021|Vidale et al., 2021]] ). Both wide-area regional and global convection-permitting models without the need for parameterized convection are becoming more useful for TC regional model projection studies ( [[#Tsuboki--2015|Tsuboki et al., 2015]] ; [[#Kanada--2017a|Kanada et al., 2017a]] ; [[#Gutmann--2018|Gutmann et al., 2018]] ) and global model projection studies ( [[#Satoh--2015|Satoh et al., 2015]] , 2017; [[#Yamada--2017|Yamada et al., 2017]] ), as they capture more realistic TC eye wall structures ( [[#Kinter%20III--2013|Kinter III et al., 2013]] ) and are becoming more useful for investigating changes in TC structures ( [[#Kanada--2013|Kanada et al., 2013]] ; [[#Yamada--2017|Yamada et al., 2017]] ). Large ensemble simulations of GCMs with 60 km grid spacing provide TC statistics that allow more reliable detection of changes in the projections, which are not well captured in any single experiment ( [[#Yoshida--2017|Yoshida et al., 2017]] ; [[#Yamaguchi--2020|Yamaguchi et al., 2020]] ). Variable resolution global models offer an alternative to regional models for individual TC or basin-wide simulations ( [[#Yanase--2012|Yanase et al., 2012]] ; [[#Zarzycki--2014|Zarzycki et al., 2014]] ; [[#Harris--2016|Harris et al., 2016]] ; [[#Reed--2020|Reed et al., 2020]] ; [[#Stansfield--2020|Stansfield et al., 2020]] ). Computationally less intense than equivalent uniform resolution global models, they also do not require lateral boundary conditions, thus reducing this source of error ( [[#Hashimoto--2016|Hashimoto et al., 2016]] ). Confidence in the projection of TC statistics and properties is increased by the use of higher-resolution models with more realistic simulations.

Operational forecasting models also reproduce TCs, and their use for climate projection studies shows promise. However, there is limited application for future projections as they are specifically developed for operational purposes, and TC climatology is not necessarily well evaluated. Intercomparison of operational models indicates that enhancement of horizontal resolution can provide more credible projections of TCs ( [[#Nakano--2017|Nakano et al., 2017]] ). Likewise, high-resolution climate models show promise as TC forecast tools ( [[#Zarzycki--2015|Zarzycki and Jablonowski, 2015]] ; [[#Reed--2020|Reed et al., 2020]] ), further narrowing the continuum of weather and climate models, and increasing confidence in projections of future TC behaviour. However, higher horizontal resolution does not necessarily lead to an improved TC climatology ( [[#Camargo--2020|Camargo et al., 2020]] ).

Atmosphere–ocean interaction is an important process in TC evolution. Atmosphere–ocean coupled models are generally better than atmosphere-only models at capturing realistic processes related to TCs ( [[#Murakami--2015|Murakami et al., 2015]] ; [[#Ogata--2015|Ogata et al., 2015]] , 2016; [[#Zarzycki--2016|Zarzycki, 2016]] ; [[#Kanada--2017b|Kanada et al., 2017b]] ; [[#Scoccimarro--2017|Scoccimarro et al., 2017]] ). However, the basin-scale SST biases commonly found in atmosphere–ocean models can introduce substantial errors in the simulated TC number ( [[#Hsu--2019|Hsu et al., 2019]] ). Higher-resolution ocean models improve the simulation of TCs by reducing the SST climatology bias ( [[#Li--2018|Li and Sriver, 2018]] ; [[#Roberts--2020a|Roberts et al., 2020a]] ). Coarse resolution atmospheric models may degrade coupled model performance as well. For example, in a case study of Hurricane Harvey, [[#Trenberth--2018|Trenberth et al. (2018)]] suggested that the lack of realistic hurricane frequency and intensity within coupled climate models hampers the models’ ability to simulate SST and ocean heat content and their changes.

Even with higher-resolution atmosphere–ocean coupled models, TC projection studies still rely on assumptions in experimental design that introduce uncertainties. Computational constraints often limit the number of simulations, resulting in relatively small ensemble sizes and incomplete analyses of possible future SST magnitude and pattern changes ( [[#Zhao--2011|Zhao and Held, 2011]] ; [[#Knutson--2013|Knutson et al., 2013]] ). Uncertainties in aerosol forcing also are reflected in TC projection uncertainty ( [[#Wang--2014|Wang et al., 2014]] ).

Regional climate models (RCM) with grid spacing around 15–50 km can be used to study the projection of TCs. RCMs are run with lateral and surface boundary conditions, which are specified by the atmospheric state and SSTs simulated by GCMs. Various combinations of the lateral and surface boundary conditions can be chosen for RCM studies, and uncertainties in the projection can be further examined in general. They are used for studying changes in TC characteristics in a specific area, such as Vietnam ( [[#Redmond--2015|Redmond et al., 2015]] ) and the Philippines ( [[#Gallo--2019|Gallo et al., 2019]] ).

Less computationally expensive downscaling approaches that allow larger ensembles and long-term studies are also used in the projection of TCs ( [[#Emanuel--2006|Emanuel et al., 2006]] ; C.Y. [[#Lee--2018|]] [[#Lee--2018|]] [[#Lee--2018|Lee et al., 2018]] ). A statistical–dynamical TC downscaling method requires assumptions of the rate of seeding of random initial disturbances, which are generally assumed to not change with climate change ( [[#Emanuel--2008|Emanuel et al., 2008]] ; [[#Emanuel--2013|Emanuel, 2013]] ). The results with the downscaling approach might depend on the assumptions, which are required for the simplification of the methods.

In summary, various types of models are useful to study how TCs change in response to climate changes, and there is no unique solution for choosing a model type. However, higher-resolution models generally capture TC properties more realistically ( ''high confidence'' ). In particular, models with horizontal resolutions of 10–60 km are capable of reproducing strong TCs with Category 4–5 and those of 1–10 km are capable of the eye wall structure of TCs. Uncertainties in TC simulations come from details of the model configuration of both dynamical and physical processes. Models with realistic atmosphere–ocean interactions are generally better than atmosphere-only models at reproducing realistic TC evolutions ( ''hi'' ''gh confidence'' ).

<div id="11.7.1.4" class="h3-container"></div>

<span id="detection-and-attribution-event-attribution-4"></span>