Tropical cyclone integrated kinetic energy in an ensemble of HighResMIP simulations

1. Introduction Tropical cyclones (TCs) are one of the costliest natural catastrophes (A. Benfield,2018; Emanuel,2005; Landsea,2000) as they can bring a number of hazards such as high winds, heavy precipitation, storm surges, inland and coastal flooding, tornadoes, and landslides. Hurricane Katrina (2005) is currently the costliest TC on record with US$160 billion in damage (National Hurricane Center,2018), but other recent examples such as Hurricane Dorian (2019) and Cyclone Harold (2020), which caused widespread damage across the Bahamas and the South Pacific respectively, have shown that TCs also have the potential to threaten some of the most vulnerable countries to devastating effect. Traditionally, the destructive potential of TCs has been estimated using the near surface 1-min maximum sustained wind speed and their corresponding classification on the Saffir-Simpson Scale (Simpson,1971). However, such classification is not always representative of the damage a storm can cause because it neglects an important component of the storm: its size. A number of studies have emphasized the benefits of considering the size and structure of the storm when estimating TC-related damage (Kantha,2006; Mahendran,1998; Wang & Toumi,2018a; Zhai & Jiang,2014). In fact, the past 2decades have also provided Abstract This study investigates tropical cyclone integrated kinetic energy, a measure which takes into account the intensity and the size of the storms and which is closely associated with their damage potential, in three different global climate models integrated following the HighResMIP protocol. In particular, the impact of horizontal resolution and of the ocean coupling are assessed. We find that, while the increase in resolution results in smaller and more intense storms, the integrated kinetic energy of individual cyclones remains relatively similar between the two configurations. On the other hand, atmosphere-ocean coupling tends to reduce the size and the intensity of the storms, resulting in lower integrated kinetic energy in that configuration. Comparing cyclone integrated kinetic energy between a present and a future scenario did not reveal significant differences between the two periods. Plain Language Summary The damage potential of tropical cyclones is often described by their maximum wind speed. However, maximum wind speed is not particularly well correlated with tropical cyclone losses because intensity defined in such a way does not take into account the size of the storm, which is an important factor driving tropical cyclone related losses. Tropical cyclone integrated kinetic energy on the other hand is a measure which takes into account both the size and the intensity of the cyclones and is more representative of its destructiveness. Here, we investigate integrated kinetic energy in three different global climate models following a common protocol. We find that an increase in horizontal model resolution results in smaller and more intense storms, but that the range of integrated kinetic energy produced by the models remains similar in both configurations. On the other hand, allowing the atmosphere and ocean to interact with each other in the models tends to reduce the size and the intensity of the storms, resulting in lower integrated kinetic energy. Comparing cyclone integrated kinetic energy between present conditions and a projected future climate scenario did not suggest notable changes between the two periods. KREUSSLER ET AL. © 2021. American Geophysical Union. All Rights Reserved. Tropical Cyclone Integrated Kinetic Energy in an Ensemble of HighResMIP Simulations Philip Kreussler1,2 , Louis-Philippe Caron1 , Simon Wild1 , Saskia Loosveldt Tomas1, Fabrice Chauvin3 , Marie-Pierre Moine4, Malcolm J. Roberts5, Yohan Ruprich-Robert1, Jon Seddon5 , Sophie Valcke4 , Benoît Vannière6, and Pier Luigi Vidale6 1Barcelona Supercomputing Center (BSC), Barcelona, Spain, 2GEOMAR – Helmholtz Centre for Ocean Research Kiel, Kiel, Germany, 3Centre National de Recherches Météorologiques, Université de Toulouse, CNRS, Météo-France, Toulouse, France, 4CECI, Université de Toulouse, CNRS, CERFACS, Toulouse, France, 5Met Office, Exeter, UK, 6National Centre for Atmospheric Science (NCAS), University of Reading, Reading, UK Key Points: • Increasing horizontal resolution leads to smaller and more intense tropical cyclones but relatively similar integrated kinetic energy • Coupling atmosphere and ocean tends to reduce the size and intensity of cyclones, generally resulting in lower integrated kinetic energy • Comparing integrated kinetic energy between present and projected future conditions does not reveal significant differences between the two Supporting Information: • Supporting Information S1 Correspondence to: P. Kreussler, philip.kr[email protected] Citation: Kreussler, P., Caron, L.-P., Wild, S., Loosveldt Tomas, S., Chauvin, F., Moine, M.-P., et al. (2021). Tropical cyclone integrated kinetic energy in an ensemble of HighResMIP simulations. Geophysical Research Letters, 48, e2020GL090963. https://doi. org/10.1029/2020GL090963 Received 28 SEP 2020 Accepted 12 FEB 2021 10.1029/2020GL090963 RESEARCH LETTER 1 of 11 Geophysical Research Letters clear examples of this: Ivan (2004), Katrina (2005), and Sandy (2012) were not particularly high on the Saffir-Simpson scale at landfall, yet caused extensive damage. For Hurricane Ivan and Sandy in particular, the damage can be explained by the continuous expansion of the storm in the mature stage (Wang & Toumi,2018a). To address this shortcoming, the so-called Integrated Kinetic Energy (IKE; Powell & Reinhold,2007) metric was developed. This metric takes the size of the storm into account by integrating the energy over the entire wind field of the storm. With this metric, the aforementioned storms rate as significantly more dangerous than on the Saffir-Simpson scale owing to the large extent of their wind fields at landfall, which is more in line with the large devastation they brought (Kantha,2006; Kozar & Misra,2019; Powell & Reinhold,2007). With a typical diameter of a few hundred kms (Chavas & Emanuel,2010), TCs are (relatively) small-scale phenomena which cannot be entirely resolved by CMIP-type global climate models. In fact, TC activity is notoriously difficult to reproduce in certain basins and the importance of increasing horizontal model resolution to address this issue has long been recognized (Bengtsson etal.,2007; Camargo,2013; Camargo & Wing,2016; Moon etal.,2020; Caron etal.,2011; Roberts etal.,2015; Strachan etal.,2013; Wehner etal.,2015). It is thought that horizontal resolution will address other long standing issues as well, such as more realistic TC wind speeds and size. As such, many different technical solutions have been attempted, from dynamical downscaling (Caron & Jones,2012; Lavender & Walsh,2011; Patricola etal.,2018), to global-variable resolution models (Caron etal.,2013; Daloz etal.,2012; Zarzycki & Jablonowski,2014) to statistical-dynamical downscaling through random seeding (Baudouin etal.,2019; Emanuel,2013; Lee etal.,2018). Here, we investigate how IKE, which has been relatively unexplored in climate models compared to other TC characteristics, is represented in a series of global climate models and how it is impacted by the horizontal resolution and the coupling of the atmosphere model to the ocean model. The impact of the resolution is investigated using a brute-force approach, wherein the global climate models are all run at both a standard and a higher resolution following a strict protocol. The structure of this study is as follows: Section2 provides an overview of the experimental setup and the cyclone tracker, while the impacts of resolution, coupling and change in atmospheric forcings are investigated in Section3. A brief conclusion is provided in Section4. 2. Models, Data, and Experimental Setup This study is carried out using data produced within the European PRIMAVERA project (Roberts etal.,2018) as a part of the CMIP6 High Resolution Model Intercomparison Project (HighResMIP; Haarsma etal.,2016). HighResMIP provides a protocol for investigating the impact of horizontal resolutions on climate simulations and as such, requires all model configurations to be the same for both resolutions so that differences can directly be attributed to the change in resolution rather than to model physics or adjustments of parameterizations. The experiments evaluated here are performed at two different model resolutions, a standard/lower resolution (LR) and a higher resolution (HR), and each model configuration was run in both an atmosphere-only mode using prescribed ocean and sea ice conditions and in coupled mode. Each configuration covers a historical period (1950–2014) which uses observed atmospheric forcings (present) and a future projection (2015–2050) which uses forcings derived from the SSP5-8.5 scenario (future). In each simulation, we track the storms between 0° and 40°N in both the Northern Pacific and Northern Atlantic (for convenience referred to as Northern Hemisphere [NH]) basins from June to November. Here, we use three global climate models (GCMs), ranging in nominal atmospheric resolution from 250– 100km in LR to 50km in HR: CNRM-CM6-1 (Voldoire etal.,2019), EC-Earth3P (Haarsma etal.,2020), and HadGEM3-GC31 (Roberts etal.,2019). The 6-h model output used is available on the Earth System Grid Federation nodes under references: CNRM-CM6-1 (Voldoire,2019a, 2019b), EC-Earth3P (EC-Earth,2018; EC-Earth,2019), and HadGEM3-GC31 (Roberts,2017a, 2017b). A brief overview of the tracking algorithm and model resolutions used to carry out these experiments are provided in the supporting information (TextS1) and TableS1. KREUSSLER ET AL. 10.1029/2020GL090963 2 of 11 Geophysical Research Letters 2.1. Tropical Cyclone Tracker The cyclone tracker used to detect the formation and propagation of TCs is the Geophysical Fluid Dynamics Laboratory Vortex Tracker V3.5b (Biswas etal.,2018) which we implemented in ESMValTool (Eyring etal.,2019; Righi etal.,2020). Information on the variables required by the tracker and the tracking process can be found in TextS1. The advantage of using this cyclone tracker is the direct output of IKE by the tracker. Unlike other integrated measures like the Accumulated Cyclone Energy index (ACE; Bell etal.,2000), IKE takes into account the entire wind field at any given time step. IKE is defined as the volume integral of the kinetic energy   2 2 U KE per volume unit of the horizontal wind field of a storm over a uniform 1-m slice, centered at 10m height with a spatially uniform air density ρ=1 kgm−3:         2 22 1 22 VV A U IKE KE dV dV kg m U dA (1) All grid points with wind speeds larger than 18ms−1 (threshold for tropical storms) are used to compute the storms' IKE. Reducing the threshold to 10ms−1 increases IKE but did not significantly impact the results. Also, only grid points within a 2,000-km square around the storm center are considered for the computation of the IKE. An example of an area considered in the computation of IKE is provided in FigureS1. We also integrated the IKE of each storm over their entire lifetime and summed up each storm's total IKE during that season. Henceforth referred to as the seasonally accumulated IKE, this metric reflects integrated TC characteristics, such as numbers, duration, and time at maximum intensity. Finally, the lifetime maximum IKE of observed TCs is estimated using the λ model described in Wang and Toumi(2016) using data from IBTrACS (Knapp etal.,2018) for storms located between 0° and 40°N for which information on the size of the storm was available. 3. Results We first provide an overview of the GCMs' ability to represent TCs in the NH. Figure1 shows all TCs generated in the models over a 30-year period. For all three models, it is obvious that increasing resolution leads to more storms, which is consistent with previous studies (Caron etal.,2011; Manganello etal.,2012; Roberts etal.,2015, 2020a, 2020b). The Western North Pacific basin (WNP) is the basin that sees the largest increase in absolute numbers, but the Eastern North Pacific (ENP) and the North Atlantic (NA) basins see the largest increase in relative terms due to the fact that generally few storms form in these basins in LR. In particular, the ENP is biased extremely low in all LR experiments with respect to observations (Ramsay,2017) and almost all model configurations strongly underestimate the hurricane frequency in the main development region (MDR; Goldenberg & Shapiro,1996) over the Atlantic. In fact, none of the LR simulations display what could be referred to as a MDR over the tropical Atlantic. Historically, simulating realistic TC activity in both these basins has proven particularly challenging for climate models (Camargo,2013; Camargo etal.,2016). Overall, only the HadGEM3-GC31 HR configurations are able to produce mean annual storm counts close to observations (WNP: 25, ENP: 15, NA: 13; Ramsay,2017) while the remaining HadGEM3-GC31 configurations and all EC-Earth3P and CNRM-CM6-1 configurations are biased low with respect to observations. We note that using a resolution-dependent intensity threshold for TC detection as suggested by Walsh etal.(2007) also leads to higher TC numbers in HR compared to LR in all configurations (FigureS2), so this gap in TC activity between the two sets of simulations is not the product of the tracking algorithm but represents a genuine difference in TC formation in HR compared to LR. In general, the coupled experiments show fewer storms than the atmosphere-only experiments, although there are some notable exceptions such as the ENP in EC-Earth3P and CNRM-CM6-1 at high resolution. Furthermore, the coupled experiments show storms that are less intense than atmosphere-only experiments, likely due to the ocean feedback on the storm systems (Cione & Uhlhorn,2003; Emanuel,2001) and the CNRM-CM6-1 HR atmosphere-only experiment is the only integration which generates TCs of category 3 or higher on the Saffir-Simpson scale (Figure1c) and thus represents wind speeds closer to observations than the other integrations. This feature of the CNRM-CM6-1 model has been highlighted before (Roberts KREUSSLER ET AL. 10.1029/2020GL090963 3 of 11 Geophysical Research Letters etal.,2020a) and seems to be linked to a newly implemented turbulence scheme (Hersbach etal.,2020). Overall, these results are consistent with the results of Roberts etal.(2020a, 2020b) obtained using different tracking algorithms. Having established the general TC climatology of the different model configurations, we can now compare the IKE of the TCs in these simulations. Figure2 shows the distribution of lifetime maximum IKE in each model configuration (and for each basin individually in FiguresS3– S5). Most models tend to overestimate IKE, with EC-Earth3P HR and HadGEM3-GC31 HR in coupled mode producing IKE that are closest to the reference data set. The CNRM-CM6-1 model in particular produces storms with considerably higher IKE than the other two models, regardless of the configuration. We also notice that coupling tends to reduce IKE. Perhaps surprisingly, maximum IKE values for a given model show a comparable range in LR and HR in all cases, which stands in strong contrast to maximum surface winds and the intensity measured by the Saffir-Simpson scale (Figure1). However, it is obvious from Figure 3 that LR and HR achieve similar IKE values for different reasons: TCs in LR tend to be larger, but weaker, than TCs in HR. This is made clear by the different regression lines computed using only TCs in LR (full lines) or HR (dashed lines): These lines show that in all models and configurations, HR generally produces smaller storms for a given wind speed. The CNRM-CM6-1 model in particular produces storms with considerably higher IKE than the other two KREUSSLER ET AL. 10.1029/2020GL090963 4 of 11 Figure 1. Tropical cyclone tracks for the CNRM-CM6-1, EC-Earth3P, and HadGEM3-GC31 models, for both atmosphere-only and coupled configurations and at both standard and high resolution. The period considered is 1950–1980. The color represents the maximum intensity of the storms, as measured by the SaffirSimpson scale. Figure 2. Distribution of lifetime maximum IKE in the NH for each climate model configuration and for a reference observation-based model (black). The reference is based on the λ model developed by Wang and Toumi(2016) and uses data for which information on the size of TCs is available in IBTrACS (2001–2020). White dot indicates median of distribution. Note that the y-axis is on a log-scale. IKE, Integrated Kinetic Energy; NH, Northern Hemisphere; TC, Tropical cyclone. Geophysical Research Letters KREUSSLER ET AL. 10.1029/2020GL090963 5 of 11 Figure 3. Scatterplots showing the TC lifetime maximum IKE (color) as well as the corresponding maximum surface wind and the area contributing to the IKE for the CNRM-CM6-1 model (first row), EC-Earth3P (second row), and HadGEM3-GC31 (third row). Each circle (triangle) represents a storm at LR (HR). The atmosphere-only (coupled) experiments are in the left (right) column. The solid and dashed lines are regression lines for LR and HR, respectively. Linear regressions for storms in the NH are shown in black, WNP in blue, ENP in green, and NA in red. The period covered is 1950–2050. Observed historical storms are displayed with bigger markers and taken from Powell and Reinhold(2007). The size of the tropical storm wind field for observed storms in Powell and Reinhold(2007) is approximated by using the radius of the maximum wind speed Rmax and the radius of the outer threshold of tropical storm strength R18, assuming an axis-symmetric cylindrical storm and wind speeds below tropical storm strength at radii below Rmax. ENP, Eastern North Pacific; HR, higher resolution; IKE, Integrated Kinetic Energy; LR, lower resolution; NA, North Atlantic; NH, Northern Hemisphere; TC, Tropical cyclone; WNP, Western North Pacific. Geophysical Research Letters models due to the fact that the storms are both larger and their surface winds are much stronger than either EC-Earth3P or HadGEM3-GC31, particularly at high resolution: The latter seem to reach their maximum intensity at approximately 42ms−1 while the former can reach values up to 76ms−1. While this impact of resolution on simulated TCs has been highlighted before, what is interesting to note here is that this has relatively little impact on the simulated IKE of the storms themselves. In other words, the increase in intensity with resolution tends to compensate for the decrease in size such that IKE remains relatively similar across resolutions. Figure3 also shows how simulated cyclones compare with five different notable Atlantic hurricanes at landfall (Powell & Reinhold,2007). Only the CNRM-CM6-1 model at high resolution manages to produce storms similar to four of the five storms selected here (larger storms like Hurricane Isabel [2003] and Hurricane Katrina [2005] are relatively common in that model), but the model generally overestimates IKE at both LR and HR (consistent with findings from Figure2) since the storms are much larger than what is typically observed. Smaller storms with similar intensity such as Hurricane Frances (2004) and Hurricane Ivan (2004) appear at the limit of what the model can produce while small intense storms like Hurricane Camille (1969) are beyond the capability of this model. And generally, the storms produced in EC-Earth3P and HadGEM3-GC31 bear little resemblance to observed storms, except possibly for the latter at HR in atmosphere-only mode. We also note differences between the individual basins. The overall TC distributions for the NH (black line) generally follow the WNP (blue line), due to the larger number of storms occurring in that basin compared to the other two. Atlantic hurricanes are generally smaller in size than typhoons, which aligns with observations (Chavas & Emanuel,2010). Finally, TCs simulated over the ENP are generally the smallest, especially in HadGEM3-GM31. This behavior is less evident in EC-Earth3P, however, possibly because of the low number of storms generated over that basin in that model. The coupled and atmosphere-only configurations generally have similar behavior: The difference between the two is mainly limited to the range of the maximum wind speed and maximum storm size. FigureS6 shows a shift in probability densities to smaller and less intense storms in the CNRM-CM6-1 and HadGEM3-GC31 models. This shift toward smaller and weaker storms is particularly obvious for the strongest cyclones, likely a result of the ocean feedback which limits the intensification of the cyclones (FigureS7). Having compared IKE in the different configurations, we will now try to understand some of the differences that were detected, and more broadly what modulates IKE in these configurations. Correlations between lifetime maximum IKE and (i) the size of the storm and (ii) the maximum wind speed near the surface clearly show that the storm size is the dominant factor modulating IKE in all configurations (triangles, FigureS8), which is consistent with results from Wang and Toumi(2018a). It is interesting to note that there is a tendency for the minimum in mean sea level pressure (MSLP) to be a better predictor of IKE than maximum surface wind speed (orange squares vs. orange circles along the y-axis) at HR. While some of the differences in correlation between IKE-MSLP and IKE-wind are small, they are present in all configurations, something that is not observed at standard resolution (same symbols, x-axis). To understand why that might be, we performed the same analysis by dividing our data into two samples along the median in IKE. The red symbols show the same correlations for storms in the upper part of the distribution. We note that the correlations with respect to MSLP and maximum wind speed are generally lower, but the differences between the two have increased, which suggest that this discrepancy is partly due to the inability of high-resolution climate models to capture the observed relationship between the minimum in MSLP and the maximum surface wind speeds. This is also supported by the fact that the correlations with wind speed are higher in the CNRM-CM6-1 model, for which the wind-pressure relationship follows more closely the observations (Roberts etal.,2020a). Part of the reason for this difference could also come from the fact that the central pressure deficit is more closely tied to the size of the storm than the maximum in surface wind (Chavas etal.,2017). 3.1. Projected Activity We conclude our analysis by evaluating whether there is a change in IKE between present and future conditions in each of our coupled simulations. Recent studies using climate models suggest that certain TC characteristics, including precipitation or the lifetime maximum intensity, are projected to increase (Knutson KREUSSLER ET AL. 10.1029/2020GL090963 6 of 11 Geophysical Research Letters etal.,2019). Projections of storm size are more scarce and uncertain, however, while some studies detected an increase in the size of TCs globally in a future climate (Emanuel,2020; Kim etal.,2014; Yamada etal.,2017), other studies did not (Gutmann etal.,2018; Knutson etal.,2015; Wang & Toumi,2018b). Figure4 shows the distribution of lifetime maximum IKE for the 1950–2050 period for each of the three models. While we detect some multi-annual variability, we do not detect a significant increase in the median IKE. Because most of the damage comes from the strongest storms and, as with changes in the surface wind, extremes are where we might expect to detect a change (Elsner etal.,2008; Elsner,2020). This is because the strongest storms are the most likely to have reached or be near their maximum potential intensity, KREUSSLER ET AL. 10.1029/2020GL090963 7 of 11 Figure 4. Boxplot of lifetime maximum IKE for storms in the NH for coupled integrations of CNRM-CM6-1 (top row), EC-Earth3P (second row) and HadGEM3-GC31 (bottom row) at LR (left column) and HR (right column). Each box represents a 5-year period. Trend lines (green) are shown for the median and the 95th percentile. The statistical significance of the trend in the 95th percentile is provided in TableS2. HR, higher resolution; IKE, Integrated Kinetic Energy; LR, lower resolution; NH, Northern Hemisphere. Geophysical Research Letters which is controlled by the thermodynamic of the ocean-atmosphere system, which itself is sensitive to the change in radiative forcings. Thus, we also evaluated the trends for storms in the 95th percentile of each 5-year period and again found no significant trends in any of the simulations (based on a Mann-Kendall test whose null hypothesis is no trend, at the 5% level). In order to evaluate whether the absence of trends in maximum IKE was due to compensating trends in intensity and size, we also evaluated the trends in the TC area used to compute the IKE. No such trend was detected (FigureS9). This result is consistent with Roberts etal.(2020b), who detected only a weak increase in the intensity of TCs in PRIMAVERA simulations. Comparing probability distributions in lifetime maximum IKE between the 1950–1980 and 2020–2050 periods does not reveal any significant differences between the two time periods either (not shown), thus indicating that anthropogenic climate change did not have a detectable impact on IKE in our simulations. We also evaluated the change in the mean seasonally accumulated IKE between the 1950–1980 and 2020– 2050 periods, globally and for each of the three basins individually. While the responses would tend to suggest an increase (FigureS10), we note that only HadGEM3-GC31 LR over the NA and EC-Earth3P HR over the WNP and the NH as a whole show a statistically significant increase over the two periods (TableS3). It is also worth noting that besides EC-Earth3P in the ENP and NA where TC activity is poorly simulated to begin with, only the CNRM-CM6-1 model suggests a decrease in mean seasonally accumulated IKE and only does so in one basin. Given that only TCs that come near land impact human activities, we repeated this latest analysis by considering only storms in the western part of the Pacific and Atlantic (i.e., storms that tracked west of 140°E in the Pacific and west of 60°W in the Atlantic). The results generally suggest an increase in TIKE (FigureS11), similar to what we obtained when we considered the entire basin, although most of the changes are not significant. The only simulations with a significant change in this case are HadGEM3-GC31 LR (increase) and CNRM-CM6-1 HR (decrease) over the North Atlantic. 4. Summary and Discussion This study has investigated the impact of an increase in horizontal resolution on the representation of tropical cyclones in different climate models, with a special focus on IKE, a measure closely associated with the damage potential of TCs. For all three models and in both configurations, we found that an enhancement of the resolution leads to an increase in wind speed (Figures1 and 3) and a decrease in storm size (Figure3). These changes combine such that the range in lifetime maximum IKE remains relatively similar across resolutions, except for the most intense storms produced by CNRM-CM6-1 in HR. This result may be somewhat unexpected but aligns with the study by Vannière etal.(2020). TCs derive their kinetic energy from latent heat release during the ascent of moist air in the tropical cyclone and Vannière etal.(2020) showed that the distribution of precipitation averaged over the entire TC had little sensitivity to resolution for a given model. Taken in combination, these results suggest that the thermodynamic efficiency of TCs is not affected by resolution. Analyses have shown that the storm size is the leading factor modulating IKE for all models and configurations. Interestingly, minimum in MSLP seems to be a better predictor of IKE in HR than the maximum wind speed, which tends to support the use of central pressure deficit as a better proxy than maximum surface winds to estimate TC damage (Klotzbach etal.,2020), even in climate simulations. Furthermore, our simulations show no significant changes in lifetime maximum IKE between present climate conditions and a projected climate scenario, which stands in contrast with previous studies. While we do not show it here, this result also applies to atmosphere-only simulations. It's worth noting that Roberts etal.(2020b) found that surface wind speed of TCs in PRIMAVERA simulations (including the models used here) projected an increase in TC intensity much below that of other modeling studies, which is somewhat consistent with the results obtained here. It is not yet clear why TCs in PRIMAVERA models appear less sensitive to an increase in forcings compared to that of other models. Finally, it is important to point out that even though the IKE of individual storms is similar across resolutions, the mean seasonally accumulated IKE is much higher at HR simply because the number of storms is significantly larger in these simulations. And while IKE appears relatively insensitive to the range of resolution covered here, it would be interesting to check whether this holds true for models with even higher KREUSSLER ET AL. 10.1029/2020GL090963 8 of 11 Geophysical Research Letters resolution (say a few km range). We hope that integrating the tracker in the ESMValTool environment will provide the opportunity to other groups who have access to such simulations to investigate this issue, and, more generally, that this will facilitate the analysis of TCs in climate models. Data Availability Statement The data sets used in this work are cited with appropriate DOIs in publicly available archives. The 6-h model output used is available on the Earth System Grid Federation nodes (https://esgf-index1.ceda.ac.uk/ search/cmip6-ceda) under references Roberts(2017a, 2017b), EC-Earth(2018); EC-Earth(2019) and Voldoire(2019a, 2019b). The IBTrACS data is provided by Knapp etal.(2018) and available under https://doi. org/10.25921/82ty-9e16. References Baudouin, J. P., Caron, L. P., & Boudreault, M. (2019). Impact of reanalysis boundary conditions on downscaled Atlantic hurricane activity. Climate Dynamics, 52(5–6), 3709–3727. https://doi.org/10.1007/s00382-018-4352-7. Retrieved from http://dx.doi.org/10.1007/ s00382-018-4352-7 Bell, G. D., Halpert, M. S., Schnell, R. C., Higgins, R. W., Lawrimore, J., Kousky, V. E., etal. (2000). Climate assessment for 1999 Table of contents. Bulletin of the American Meteorological Society, 81(6), 1–50. Benfield, A. (2018). Weather, climate & catastrophe insight2017 annual report. Aon Benfield UCL hazard research center report, 2–5. Retrieved from http://thoughtleadership.aonbenfield.com/Documents/20180124-ab-if-annual-report-weather-climate-2017.pdf Bengtsson, L., Hodges, K. I., Esch, M., Keenlyside, N., Kornblueh, L., Luo, J. J., & Yamagata, T. (2007). How may tropical cyclones change in a warmer climate? Tellus, Series A: Dynamic Meteorology and Oceanography, 59 A(4), 539–561. https://doi. org/10.1111/j.1600-0870.2007.00251.x Biswas, M. K., Stark, D., & Carson, L. (2018). GFDL Vortex tracker users' guide V3.9a., p. 35. Camargo, S. J. (2013). Global and regional aspects of tropical cyclone activity in the CMIP5 models. Journal of Climate, 26(24), 9880–9902. https://doi.org/10.1175/JCLI-D-12-00549.1 Camargo, S. J., Sobel, A. H., Delgenio, A. D., Jonas, J. A., Kelley, M., Lu, Y., etal. (2016). Tropical cyclones in the GISS ModelE2. Tellus, Series A: Dynamic Meteorology and Oceanography, 68(1), 1–21. https://doi.org/10.3402/tellusa.v68.31494 Camargo, S. J., & Wing, A. A. (2016). Tropical cyclones in climate models. Wiley Interdisciplinary Reviews: Climate Change, 7(2), 211–237. https://doi.org/10.1002/wcc.373 Caron, L. P., & Jones, C. G. (2012). Understanding and simulating the link between African easterly waves and Atlantic tropical cyclones using a regional climate model: The role of domain size and lateral boundary conditions. Climate Dynamics, 39(1–2), 113–135. https:// doi.org/10.1007/s00382-011-1160-8 Caron, L. P., Jones, C. G., Vaillancourt, P. A., & Winger, K. (2013). On the relationship between cloud-radiation interaction, atmospheric stability and Atlantic tropical cyclones in a variable-resolution climate model. Climate Dynamics, 40(5–6), 1257–1269. https://doi. org/10.1007/s00382-012-1311-6 Caron, L. P., Jones, C. G., & Winger, K. (2011). Impact of resolution and downscaling technique in simulating recent Atlantic tropical cylone activity. Climate Dynamics, 37(5), 869–892. https://doi.org/10.1007/s00382-010-0846-7 Chavas, D. R., & Emanuel, K. A. (2010). A QuikSCAT climatology of tropical cyclone size. Geophysical Research Letters, 37(18), 10–13. https://doi.org/10.1029/2010GL044558 Chavas, D. R., Reed, K. A., & Knaff, J. A. (2017). Physical understanding of the tropical cyclone wind-pressure relationship. Nature Communications, 8(1). https://doi.org/10.1038/s41467-017-01546-9. Retrieved from http://dx.doi.org/10.1038/s41467-017-01546-9 Cione, J. J., & Uhlhorn, E. W. (2003). Sea surface temperature variability in hurricanes: Implications with respect to intensity change. Monthly Weather Review, 131(8PART 2), 1783–1796. https://doi.org/10.1175//2562.1 Daloz, A. S., Chauvin, F., & Roux, F. (2012). Impact of the configuration of stretching and ocean-atmosphere coupling on tropical cyclone activity in the variable-resolution GCM ARPEGE. Climate Dynamics, 39(9–10), 2343–2359. https://doi.org/10.1007/s00382-012-1561-3 EC-Earth. (2018). ECEarthConsortium ECEarth3PHR model output prepared for CMIP6 HighResMIP. Retrieved from https://doi. org/10.22033/ESGF/CMIP6.2323 EC-Earth. (2019). ECEarthConsortium ECEarth3P model output prepared for CMIP6 HighResMIP. Retrieved from https://doi.org/10.22033/ ESGF/CMIP6.2322 Elsner, J. B. (2020). Continued increases in the intensity of strong tropical cyclones. Bulletin of the American Meteorological Society, 101(8), E1301–E1303. https://doi.org/10.1175/bams-d-19-0338.1 Elsner, J. B., Kossin, J. P., & Jagger, T. H. (2008). The increasing intensity of the strongest tropical cyclones. Nature, 455(7209), 92–95. https://doi.org/10.1038/nature07234 Emanuel, K. (2001). Contribution of tropical cyclones to meridional heat transport by the oceans. Journal of Geophysical Research, 106(D14), 14771–14781. https://doi.org/10.1029/2000JD900641 Emanuel, K. (2005). Increasing destructiveness of tropical cyclones over the past 30 years. Nature, 436(7051), 686–688. https://doi. org/10.1038/nature03906 Emanuel, K. (2013). Downscaling CMIP5 climate models shows increased tropical cyclone activity over the 21st century. Proceedings of the National Academy of Sciences of the United States of America, 110(30), 12219–12224. https://doi.org/10.1073/pnas.1301293110 Emanuel, K. (2020). Response of global tropical cyclone activity to increasing CO 2: Results from downscaling CMIP6 models. Journal of Climate, 1–54. https://doi.org/10.1175/JCLI-D-20-0367.1 Eyring, V., Bock, L., Lauer, A., Righi, M., Schlund, M., Andela, B., etal. (2019). ESMValTool v2.0 Extended set of large-scale diagnostics for quasi-operational and comprehensive evaluation of Earth system models in CMIP. Geoscientific Model Development Discussions, 1–81. https://doi.org/10.5194/gmd-2019-291 KREUSSLER ET AL. 10.1029/2020GL090963 9 of 11 Acknowledgments This research has been supported by the Horizon 2020 programme (PRIMAVERA, GA #641727). S. Wild received funding from the European Union Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement 2020-MSCA-COFUND-2016-754433 and financial support from the Spanish Agencia Estatal de Investigación (FJC2019041186-I/AEI/10.13039/501100011033). M. J. Roberts acknowledges the support from the UK-China Research and Innovation Partnership Fund through the Met Office Climate Science for Service Partnership (CSSP) China as part of the Newton Fund. Finally, the authors are most grateful to three anonymous reviewers for their helpful comments in improving a previous version of this manuscript.