Land Surface Model influence on the simulated climatologies of temperature and precipitation extremes in the WRF v.3.9 model over North America

The representation and projection of extreme temperature and precipitation events in regional and global climate models are of major importance for the study of climate change impacts. However, state-of-the-art global and regional climate model simulations yield a broad inter-model range of intensity, duration and frequency of these extremes. Here, we present a modeling experiment using the Weather Research and Forecasting (WRF) model to determine the influence of the land surface model (LSM) component on uncertainties associated with extreme events. First, we evaluate land-atmosphere interactions 5 within four simulations performed by the WRF model using three different LSMs from 1980 to 2012 over North America. Results show LSM-dependent differences at regional scales in the frequency of occurrence of events when surface conditions are altered by atmospheric forcing or land processes. The inter-model range of extreme statistics across the WRF simulations is large, particularly for indices related to the intensity and duration of temperature and precipitation extremes. Areas showing large uncertainty in WRF simulated extreme events are also identified in a model ensemble from three different Regional 10 Climate Model (RCM) simulations participating in the Coordinated Regional Climate Downscaling Experiment (CORDEX) project, revealing the implications of these results for other model ensembles. This study illustrates the importance of the LSM choice in climate simulations, supporting the development of new modeling studies using different LSM components to understand inter-model differences in simulating temperature and precipitation extreme events, which in turn will help to reduce uncertainties in climate model projections. 15 1 https://doi.org/10.5194/gmd-2020-86 Preprint. Discussion started: 20 April 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
General Circulation Models (GCMs) and Regional Climate Models (RCMs) are currently the most useful tools for the study of processes affecting the frequency, duration and intensity of extreme temperature and precipitation events, as well as projecting their evolution under different emission scenarios at global, regional and local scales. Both observational data and climate 20 model simulations confirm all of these statistics respond to climate change Orlowsky and Seneviratne, 2012;Jeong et al., 2016). However, state-of-the-art global and regional climate models differ substantially in their interpretation of the climatology and response to warming of various indices of temperature and precipitation extremes (Sillmann et al., 2013a, b). Climate information provided by models is currently employed by public and private institutions dedicated to the evaluation and management of risks from extreme events and associated disasters (IPCC, 2013;Arneth, 2019). It is, therefore, 25 essential that climate models represent extreme events and their evolution as realistically as possible to aid in the design of appropriate policies to mitigate climate change and build resilience. In this study, we evaluated the representation of a set of extreme indices, previously included in international reports such as IPCC (2013) and Seneviratne et al. (2012), as simulated by the Weather Research and Forecasting (WRF) model with different land surface model (LSM) components.
Land-atmosphere interactions have been identified as a key factor in the simulation of extreme events (e.g. Lorenz et al.,30 2016; Vogel et al., 2017). Soil conditions affect and are affected by near-surface atmospheric phenomena, through energy and water exchanges at the ground surface. For example, previous observational studies have shown the impact of soil moisture deficits on hot extreme temperatures through changes in evapotranspiration over southeastern and western Europe and Russia (Hirschi et al., 2011;Miralles et al., 2012;Hauser et al., 2016). Additionally, soil moisture regimes have been found to alter the energy and water exchanges at the surface, influencing inter-annual summer temperature variability in central parts of North 35 America (Donat et al., 2016), and precipitation events in western North America (Diro et al., 2014). Land-Atmosphere interactions, and consequently near-surface conditions, are influenced by vegetation and snow covers (Stieglitz and Smerdon, 2007;Diro et al., 2018). For example, Diro et al. (2018) showed that interactions between snow cover and atmospheric processes influence extreme events, increasing the frequency of cold events over western North America and affecting the variability in warm events over northeast Canada and the Rocky mountains. 40 Metrics built on the representation of land-atmosphere interactions have been employed as a basis for evaluating extreme temperature and precipitation events in climate model simulations (Knist et al., 2016;Davin et al., 2016;Lorenz et al., 2016;Sippel et al., 2017;Gevaert et al., 2018;García-García et al., 2019). For example, Lorenz et al. (2016) evaluated outputs from six GCMs participating in the Global Land-Atmosphere Coupling Experiment of the Coupled Model Intercomparison Project, Phase 5 (GLACE-CMIP5) and concluded that ranges of intensity, frequency and duration of extreme events among The complexity and variety of these LSM components are limited in order to reduce computational costs, affecting the quality of the represented land surface processes. This has already been noted by the scientific community, and some have attempted to address the issue by incorporating updated versions of LSMs in new land reanalysis products though offline LSM simulations forced by observational data products (LDAS, MERRA-land, ERA-Iterim/Land, Rodell et al., 2004;Reichle et al., 2011;Balsamo et al., 2015). Although these new products can be useful for LSM development and provide data about the soil states and fluxes (Balsamo et al., 2015), the offline character of the new land products inhibits the representation of land-atmosphere feedbacks.
Here, we perform a set of modeling experiments to evaluate for the first time the influence of the LSM component on the simulation of key extreme indices and land-atmosphere interactions within land-atmosphere coupled climate simulations at continental scales. For this purpose, four regional simulations are performed over North America  using the WRF 70 model including three different LSM components widely employed in model simulations and reanalysis products, as described in Section 2. The methodology for the analysis of land-atmosphere interactions and the representation of extreme events is described in Section 3. Section 4 presents the evaluation of land-atmosphere interactions, the analysis of LSM differences in the representation of temperature and precipitation extremes, and the comparison between the WRF simulations and three Coordinated Regional Climate Downscaling Experiment (CORDEX) Evaluation simulations. A discussion about previous 75 results and the main conclusions and implications of this study are presented in Section 5 and 6, respectively.

Description of the modeling experiment
We performed four regional simulations over North America (NA) using the version 3.9 of the Advanced Research WRF (ARW-WRF) model (Michalakes et al., 2001) including three different land surface models: the NOAH LSM (NOAH, Tewari et al., 2004), the NOAH LSM with multiparameterizations options (NOAH-MP, Niu et al., 2011), and the Community Land Model version 4 LSM (CLM4, Oleson et al., 2010). Vegetation cover was prescribed in these three simulations (NOAH, NOAH-MP and CLM4); an additional simulation was conducted with dynamic vegetation cover in the NOAH-MP LSM (NOAH-MP-DV), allowing for the evaluation of the influence of dynamic vegetation on extremes (NOAH-MP-DV). The use  Oleson et al. (2010) of different LSM components in a RCM permits the study of the influence of surface and soil processes on the simulated climate system in contrast to LSM offline simulations (Laguë et al., 2019).

85
The LSM components employed have been previously included in climate model studies or in reanalysis products. The CLM4 LSM component has been coupled to several GCMs participating in the CMIP5 project (Collins et al., 2006;Vertenstein et al., 2012). The NOAH LSM has been extensively used for reanalysis products, as well as for RCM simulations as those participating in the CORDEX project (Mesinger et al., 2006;Katragkou et al., 2015). The NOAH-MP LSM has been selected for current studies using WRF (e.g. Liu et al., 2017 improves the computation of energy, water and carbon fluxes at the surface; a separate scheme for computing energy fluxes over vegetated surfaces and bare soils; a new 3-layer snow model; a more permeable frozen soil; and an improved description of runoff and soil moisture. Although the NOAH-MP LSM is the updated version of the NOAH LSM and has been shown to improve the simulation of surface processes in comparison to the NOAH LSM (e.g. Niu et al., 2011;Yang et al., 2011), the NOAH-MP LSM has not yet been implemented in any reanalysis product. The CLM4 represents one of the most advanced 100 LSM components, incorporating a detailed description of biogeophysics, hydrology and biogeochemistry. The CLM4 classifies vegetation cover according to 4 different plant functional types, considering the physiology and structure of different plants.
The soil vertical structure is divided into a layer for the vegetation canopy, 5 layers for snow cover, and 10 soil layers, placing the zero-flux bottom boundary condition at approximately 4.32 m. The main characteristics of the employed LSM components are summarized in Table 1.

105
Beyond the structural differences among LSM components, the remaining options and parameters are identical for the four WRF simulations. Boundary conditions for the WRF experiments are provided by the North American Regional Reanalysis (NARR) product, which is formed by the NCEP Eta atmospheric model, the NOAH LSM and the Regional Data Assimilation System (RDAS); (Mesinger et al., 2006). NARR data are provided with a 32 km grid and three-hourly temporal resolution,  Barlage et al., 2005). The four WRF simulations start in January 1st 1979, which is the first year of the NARR product, and end in December 31st 2012, using a time-step of 300 seconds for the model integrations. We use the first year of each simulation as spin-up and the other 33 years for the analysis. The employed physics parameterizations include the WSM 6-class graupel 115 scheme for the microphysics (Hong and Lim, 2006), the Grell-Freitas ensemble scheme for cumulus description (Grell and Freitas, 2014), the Yonsei University scheme as planetary boundary layer scheme (YSU, , the revised MM5 monin-Obukhov scheme for the surface layer (Jiménez et al., 2012), and the CAM scheme for the integration of radiation physics each 20 min intervals (Collins et al., 2004).
The gap in resolution from the employed boundary conditions (32 km) to the final simulations (50 km) can be counter-  (Zscheischler et al., 2015;Gevaert et al., 2018;Sippel et al., 2017;Philip et al., 2018). The VAC index is segregated in four categories based on the simultaneous occurrence of some given extreme percentile rages of Surface Air Temperature (SAT) and latent heat flux (LH, Philip et al., 2018): Extremes of SAT and LH are defined as values exceeding (below) the 70th (30th) percentile, relative to a 20-year period  (Eq. 1). The 30th and 70th percentile thresholds were also employed in previous studies based on monthly data 5 https://doi.org/10.5194/gmd-2020-86 Preprint. Discussion started: 20 April 2020 c Author(s) 2020. CC BY 4.0 License. (Sippel et al., 2017). The VAC index classifies areas depending on the soil moisture regime into energy-limited areas, where atmospheric forcing controls processes at the land surface (VAC a and VAC b ), and transitional areas, where land surface pro-140 cesses are driven by soil moisture deficits (VAC c and VAC d ). As explained in Zscheischler et al. (2015), the VAC a category is associated with low SAT caused by the presence of clouds and precipitation, which leads to low vegetation activity likely rising soil moisture. The VAC b category is frequent in wet areas with high SAT, usually related to clear sky and high radiation, which is associated with the increase in vegetation activity inducing the depletion of soil moisture. During VAC c episodes, the combination of high SAT and soil moisture deficits leads to diminished vegetation activity, followed by low precipitation and 145 consequently reduced soil moisture and high SAT, promoting heat waves and droughts. The VAC d category is associated with high precipitation over dry soils which stimulates vegetation activity, increases soil moisture and decreases SAT.
We calculate the frequency of occurrence for each VAC category using deseasonalized and detrended monthly SAT and LH series following the typical methodology (Sippel et al., 2017) at each grid cell from 1980 to 2012, hereafter the analysis period. The frequency of occurrence for each VAC category is calculated by counting the VAC events for the analysis period After the evaluation of land-atmosphere interactions in our set of simulations, we assess the representation of extreme events across the WRF simulations coupled to different LSM components. There are several definitions of indices related to temperature and precipitation extremes, mainly using thresholds based on absolute values or statistical percentiles (e.g. Sillmann et al., 2013a). The evaluation of model simulations in representing indices based on absolute values could include 160 model-specific biases, that can be corrected by bias removal techniques. However, the advantage of applying bias removal techniques techniques is not clear for the study of future climate trends and climate variability, since they have been proven to modify the spatiotemporal consistency of climate models as well as internal feedback mechanisms and conservation terms (Ehret et al., 2012;Cannon et al., 2015). Additionally, the simulation of absolute temperatures are of central importance for temperature dependent processes that may have important consequences for society and ecosystems, such as soil carbon  Table 2). Since we are interested in the climatology of extreme events, temporal averages 170 of each annual index are computed for the analysis period at each grid cell for each WRF experiment. Then, we compute the inter-model range of each index across the WRF simulations (i.e., the difference between the maximum and minimum values at The LSM effect on the WRF simulation of extreme temperature and precipitation events was also compared with the repre-175 sentation of extreme events by three different RCMs participating in the North America CORDEX (NA CORDEX) program, using daily data from three Evaluation simulations (Table S1). These CORDEX simulations were performed by the WRF model (Skamarock et al., 2008), the RCA4 model (Samuelsson et al., 2011), and the CRCM-UQAM model (Martynov et al., 2013), using boundary conditions from the ERA-Interim reanalysis (Dee et al., 2011). The spatial domain and resolution of the NA CORDEX simulations are similar to that of the WRF simulations, as indicated in Section 2. Refer to Table S2 for 180 information about the availability of the data employed in this work.

Evaluation of land-atmosphere interactions in WRF simulations
All WRF simulations with different LSM components display similar spatial patterns for VAC categories, agreeing in the seasonality and broadly in the areas with high probability of episodes when atmospheric forcing or soil conditions control 185 processes at the land surface (Figures 1 and 2). Atmospheric forcing controls surface processes at middle and high latitudes in MAM, JJA and SON, moving southward in DJF ( Figure 1). Areas frequently driven by soil processes are displayed over the western Mexican coast in DJF, spreading across low and middle latitudes in MAM, JJA and SON ( Figure 2). Despite the broad agreement between LSM simulations in the spatial distribution of the VAC categories, there are regional differences in their representation of land-atmosphere coupling. These regional differences allow us to identify the NOAH LSM as the one categories to these episodes is broadly similar across LSMs, with slightly higher VAC a in all seasons; modest LSM-specific differences include a tendency for the NOAH simulation to show slightly higher VAC a probabilities across all seasons (but especially DJF) ( Figures S1 and S2). LSM differences in the representation of VAC a and VAC b probabilities suggest the LSM influence on the evolution of atmospheric conditions.  northwestern North America in DJF also indicated by the CLM4 simulation, but absent in the NOAH-MP and NOAH-MP-DV simulations (Figure 2). The probability of land control episodes over the western Mexican coast is higher in the CLM4 and NOAH-MP simulations than in the NOAH and NOAH-MP-DV simulations in DJF. In JJA, however, the NOAH-MP-DV simulation presents a stronger land control at low and middle latitudes than the NOAH-MP simulation (Figure 2). There are also regional differences between LSM simulations in SON, particularly over the southeastern US coast where the CLM4 210 shows the strongest land control, followed by the NOAH-MP simulation (Figure 2). Exploring the contribution of respective VAC c and VAC d separately, it is shown they present small differences; for example, the VAC c probability in DJF is slightly higher than the VAC d probability for all simulations, showing the opposite behavior in JJA for the NOAH-MP and the NOAH-MP-DV simulations ( Figures S3 and S4). The LSM differences shown in the representation of land control VAC categories likely imply LSM differences in the simulated statistic of extreme events because of the relationship between VAC c episodes 215 and heat waves and droughts (Zscheischler et al., 2015).

Climatologies of temperature and precipitation extremes in WRF simulations
The climatologies of temperature and precipitation extremes for the analysis period, represented by their means, show similar spatial patterns across all WRF simulations with different LSM component ( Figures S5, S6 and S7). Figure (Figures 3 and 5b).
Note that ranges of more than 2% in the number of hot days and nights correspond to differences of more than 7 Figure S8). However, the seasonal decomposition of the range for the duration index of cold events shows a region with large uncertainty over western NA in MAM, corresponding to an area with marked differences between LSM simulations in the MAM atmospheric control VAC categories (Figures S8c and 1). For the simulation 300 of warm extremes, large LSM differences in the intensity indices correspond to LSM differences in the JJA VAC categories associated with atmospheric control episodes (Figures 1 and 5). Areas with large uncertainty in the JJA simulation of warm frequency indices coincide with areas showing strong land control on surface processes as well as regional differences between LSM simulations (Figures S9ab and 2). The duration index of warm extremes also shows large inter-model range in JJA over regions under land control ( Figures S9c and 2). The range of the intensity index of precipitation extremes displays a large JJA 305 component over areas under land control at low latitudes and under atmospheric control at middle and high latitudes ( Figures   S10a, 1 and 2). The MAM and SON components of the range for the precipitation intensity index also show large values over 16 https://doi.org/10.5194/gmd-2020-86 Preprint. Discussion started: 20 April 2020 c Author(s) 2020. CC BY 4.0 License. areas with atmospheric control in MAM and over areas with land control in SON ( Figures S10a, 1 and 2). The frequency index of precipitation events presents large inter-model range over small regions in JJA, coinciding with areas under atmospheric control ( Figures S10b and 1). The inter-model range of dry periods coincides with land control areas at low latitudes in all 310 seasons and with atmospheric control areas at high latitudes in MAM ( Figures S10c 1 and 2). The inter-model range in the simulation of consecutive wet days is large in JJA over a small Mexican region classified under atmospheric control with different degree of coupling between simulations (Figures S10d and 1).
In order to address the LSM influence on the simulation of extreme events, we compute the ranges among WRF simulations using the 95th percentile of the analysis period for each extreme index. The uncertainty in the WRF simulations due to the 315 LSM component when using the 95th percentile for each extreme index leads to similar conclusions (Figures S11 and S12).
The LSM differences using the 95th percentile of the analysis period are larger for all extreme temperature and precipitation indices than using the period mean as expected, but the marked areas are analogous ( Figures 5, 6, S11 and S12). The agreement in the representation of areas with large uncertainty in extreme indices between results using mean and extreme climatologies suggests the LSM influence on extreme events at climatological and shorter time scales.

Comparison between WRF simulations and three CORDEX Evaluation simulations
The climatologies of temperature and precipitation extreme statistics as simulated by the RCMs participating in the NA-CORDEX project (Table S1) show similar spatial patterns to those from the WRF ensemble ( Figures S5-S7 and S13-S15).
Although spatial patterns are similar in both ensembles, the WRF simulations yield colder minimum temperatures in DJF (TNn DJF) and less frequent cold nights (TX10p) than the CORDEX simulations ( Figures S5 and S13). The percentage of hot 325 days, however, is higher and warm spells are longer in the WRF simulations than in the CORDEX simulations, particularly over southwestern NA (Figures S6 and S14). The intensity of heavy precipitation extremes is generally higher within the WRF ensemble than in the CORDEX ensemble, while dry periods are longer in the CORDEX simulations ( Figures S7 and S15).
The uncertainties in the simulation of extreme statistics within the CORDEX ensemble show some similarities with the WRF uncertainties arising from the LSM component. For example, the simulated climatology of DJF coldest night (TNn 330 DJF) shows large uncertainties over the US for both ensembles, particularly over the eastern US (Figures 5a and 7a). The climatologies of DJF hottest day (TXx DJF) display large inter-model range within the WRF ensemble over areas where temperatures approximate to 0 o C, expanding southward for the CORDEX ensemble. The CORDEX inter-model ranges of the frequency indices for cold extremes do not show a clear spatial pattern in agreement with the WRF ensemble. There is, however, a region over the central US with slightly larger ranges among the CORDEX simulations than among the WRF 335 simulations (Figures 5b, 7b, and S16b). The duration of cold spells presents large uncertainties in the CORDEX ensemble over the eastern US/Mexican border and over western Canada, coinciding with a small region with large inter-model range among the WRF simulations (Figures 5c and 7c). For the simulation of warm temperature extremes, the uncertainties in the intensity indices among the CORDEX simulations show large ranges over the eastern US for the JJA hottest day (TXx JJA) in agreement with the WRF simulations, and at high latitudes for the coldest night (TNn JJA), including the eastern region 340 of Hudson Bay also marked by the WRF ensemble (Figures 5a and 7a). The frequency indices of warm events show large 19 https://doi.org/10.5194/gmd-2020-86 Preprint. Discussion started: 20 April 2020 c Author(s) 2020. CC BY 4.0 License.  (Figures 5b and 7b). The uncertainty in the duration of warm spells among the CORDEX simulations does not show large spatial differences, although the ranges are slightly larger at low latitudes coinciding with regions marked by the WRF ensemble and at very high latitudes (Figures 5c and 7c). The simulation of precipitation extreme statistics is generally more 345 uncertain across the CORDEX simulations than across the WRF simulations (Figures 6, 8, and S17). Interestingly, all regions with large uncertainties in the simulation of precipitation extremes among the WRF simulations are also identified as areas with large uncertainty across the CORDEX ensemble. There are, however, additional areas with large uncertainty in the CORDEX ensemble, particularly for the consecutive dry days index and the frequency index at middle and high latitudes (Figures 6 and   8). Thus, the comparison between the WRF and CORDEX ensembles suggests that results from this study may be applicable 350 to other model ensembles over some areas, particularly for the simulation of warm temperature and precipitation extremes.

Comparison of inter-model ranges across the WRF and CORDEX ensembles
In order to provide context for the applicability of these results to other model ensembles, we compared the inter-model range across the WRF simulations with the inter-model range across three CORDEX simulations in representing extreme events have showed that the spread of extreme events among ensemble members of an individual model is generally small compared to inter-model spreads (Kharin et al., 2007;Sillmann et al., 2013a).
Although CORDEX simulations were performed using boundary conditions from the ERA reanalysis product, the comparison with the WRF simulations is possible because we compute inter-model ranges across ensembles as a measure of the 365 uncertainty in each model ensemble. Thus, we compare model's uncertainty in both ensembles finding common areas with large inter-model ranges for the simulation of cold and warm temperature extremes and precipitation extremes, despite they used different products as boundary conditions. The similar uncertainties of extreme events in the CORDEX ensemble relative to the WRF simulations suggest that the LSM component may be an important source of uncertainty in the CORDEX ensemble.
That is, the LSM component employed in each CORDEX simulation (Table S1) Giorgi and Francisco (2000). These spatial averages allow identification of some regional differences between the WRF and the CMIP5 ensembles, for example over the eastern US coast (ENA region) where the WRF simulations yield warmer JJA maximum temperatures than the CMIP5 ensemble ( Figure 4 and Figure  in Sillmann et al. 2013a). Although this is a rough comparison between results presented in this article and in Sillmann et al. (2013a), this comparison suggests that our conclusions could be also applicable to the CMIP5 ensemble as it was the case for the CORDEX ensemble.

400
Increases in heat-related events have been directly and robustly associated with increases in mortality, for example in Europe during the heatwave of 2003 (Fischer et al., 2007) or in India (Mazdiyasni et al., 2017). Heavy precipitation events often lead to floods, which also are directly associated to economic loss and death toll (Hu et al., 2018). All climate change projections point out to a future increase in temperature and precipitation extreme events (Sillmann et al., 2013b), thus developing mitigation strategies will become necessary to preserve human health. Climate model simulations are our best source of information to 405 inform measure against climate change impacts. However, the results presented here indicate that the simulation of several extreme indices varies largely depending on the employed LSM component. This means that a climate model may simulate the climatology of heat extremes 5 o C warmer and 6 days longer depending on the employed LSM component, and similarly for cold extremes and heavy precipitation events. The accuracy of climate models in simulating extreme events will likely affect climate change policy, therefore having repercussions for society and environment.

410
The indices employed here to study the climatology of extreme temperature events were based on minimum and maximum temperature outputs. However, many studies have proven that the study of compound events using indices based on multiple variables, such as temperature and moisture outputs, are more representative of thermal stress in humans and ecosystems than standard indices (Zscheischler et al., 2018). The large LSM influence on the climatology of extreme temperature and precipitation events, suggests that the uncertainty arising from the LSM component could be higher on extreme indices based 415 on multiple variables. However, the analysis of the LSM influence on compound events is beyond the scope of this work, and constitutes an interesting line for future research.

Conclusions
WRF simulations coupled to different LSM components showed similar spatial patterns of land-atmosphere interactions, indicative of atmospheric control over surface conditions at middle and high latitudes and land surface control over lower lat-420 itudes, particularly in JJA. However, the simulation of land-atmosphere interactions differs at regional scales depending on the LSM choice in two directions; by altering land control on surface processes (VAC c and VAC d categories) and by altering atmospheric forcing and its influence on surface conditions (VAC a and VAC b categories). Thus, the NOAH LSM is associated with the weakest representation of land control on surface conditions, while the CLM4 LSM simulates one of the strongest land effect on surface conditions. The use of different LSM components leads to large ranges of represented extreme temperature 425 and precipitation events, affecting their simulation in intensity, frequency and duration. The CLM4 LSM yields the weakest cold events, the warmest hot days, and the heaviest precipitation events, while the NOAH simulation yields the weakest land control on surface conditions, the weakest warm temperature events and the weakest heavy precipitation events. This relationship between the degree of land control on surface conditions and the intensity of extreme events is in agreement with two case studies during the Russian 2010 heat wave and the Amazon 2010 drought (Zscheischler et al., 2015). Meanwhile, the NOAH-

430
MP LSM produces the driest simulation, yielding slightly wetter conditions when using dynamic vegetation at middle and low latitudes. Despite small differences between simulations with prescribed and dynamic vegetation, differences are much more marked among the WRF simulations due to different LSM components.
Previous studies using GCM simulations suggested a dependence of the simulated land-atmosphere interactions on the em- in regional and global climate models as well as in reanalysis products. The strong LSM dependency of climate model simulation of extremes is also of special importance for international reports focused on land, such as the IPCC Special Report on Climate Change, Desertification, Land Degradation, Sustainable Land Management, Food Security, and Greenhouse gas fluxes in Terrestrial Ecosystems (Arneth, 2019). Future sensitivity analyses to the LSM component using different regional and global climate models would be useful to understand models' differences in simulating temperature and precipitation extremes, 445 helping to narrow the inter-model range across reanalyses and climate model projections in simulating extreme events.
Code and data availability.