Storm and flood data (Deliverable D6.1)

Exploring Tipping Points and Their Impacts Using Earth System Models Storm and flood data Deliverable D6.1 D6.1 Storm and flood data About this document First submission date to the European Commission: 28 August 2025 (first submission) Revised on: 8 October 2025, with integrations in section 2.2.4 by the authors Dissemination Level: Public (PU) Work package: WP6 Climate-driven tipping of ecological and social systems Authors: Potsdam-Institut für Klimafolgenforschung e.V. (PIK), PP6, Jacob Schewe, [email protected]; Dánnell Quesada Chacón, Sandra Zimmermann Contributors: Potsdam-Institut für Klimafolgenforschung e.V. (PIK), PP6, Jan Volkholz, Inga Sauer Reviewer: Danish Meteorological Institute (DMI), PP1, Chiara Bearzotti [email protected] Disclaimer: Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Climate, Infrastructure and Environment Executive Agency (CINEA). Neither the European Union nor the granting authority can be held responsible for them. 2 D6.1 Storm and flood data TABLE OF CONTENTS 1. Abstract 4 2. Work Done 4 2.1 Storm data 4 2.1.1 Data sources 4 2.1.2 Wind field calculation 6 2.1.3 Sampling of storms 6 2.2 Flood data 7 2.2.1 Hydrological modelling 7 2.2.2 Hydrodynamic modelling 8 2.2.3 Accounting for flood protection 9 2.2.4 Flooded fraction 9 3. Results Achieved 9 4. Contribution to the TipESM objectives 10 5. References 10 3 D6.1 Storm and flood data 1. Abstract This deliverable describes the datasets of areas affected by strong tropical cyclone winds and flooded areas made accessible through the ISIMIP data repository. These datasets are derived from state-of-the-art climate, tropical cyclone, hydrological, and hydrodynamic model simulations, respectively, prepared in the context of CMIP6 and ISIMIP3. They describe the annual maximum storm intensity (in terms of wind speed) and flooded area fraction, respectively, in each grid cell globally, under historical and future climatic conditions according to different scenarios of greenhouse gas concentrations. The data can be used for assessing the risks associated with these two hazards, such as risks to life, livelihoods, health, or economic assets. In the context of TipESM, they support the assessment of potential tipping dynamics in household recovery from climate-induced shocks and related poverty risks, as well as flood-induced human displacement risk. 2. Work Done 2.1 Storm data The storm data were produced according to the methodology presented below. The methodology represents a reproducible, bias-adjusted workflow for generating tropical cyclone (TC) windfield datasets, based on a sampling approach. The aim is to ensure that the statistical properties of simulated TC windfields closely align with observations, while allowing extension across multiple climate models and scenarios. 2.1.1 Data sources Observed TC tracks are sourced from the IBTrACS archive, which provides storm center positions from 1851 to present-day records. Simulated TC tracks are derived from WindRiskTech's global synthetic datasets (Emanuel et al., 2008), utilising boundary conditions such as sea surface temperature, air temperature, and winds at height, from five global climate models (GCMs) and up to six experiments. Each available model-experiment combination includes 1500 synthetic tracks per year. However, not all combinations are 4 D6.1 Storm and flood data available for the full duration of the respective experiments, and coverage varies by model and scenario. See Tab. 1 for further details. Table 1: Available simulated tropical cyclone tracks per GCM, experiment, and period. The data is provided by WindRiskTech. GCM Experiment Period GFDL-ESM4 Historical 1850–2014 GFDL-ESM4 ssp585 2061–2100 IPSL-CM6A-LR Historical 1850–2014 IPSL-CM6A-LR piControl 1850–2014 IPSL-CM6A-LR ssp126 2061–2100 IPSL-CM6A-LR ssp245 2015–2100 IPSL-CM6A-LR ssp370 2015–2100 IPSL-CM6A-LR ssp585 2015–2100 MPI-ESM1-2-HR Historical 1850–2014 MPI-ESM1-2-HR piControl 1850–2014 MPI-ESM1-2-HR ssp126 2061–2100 MPI-ESM1-2-HR ssp245 2015–2100 MPI-ESM1-2-HR ssp370 2015–2100 MPI-ESM1-2-HR ssp585 2015–2100 MRI-ESM2-0 Historical 1950–2014 MRI-ESM2-0 piControl 1850–2014 MRI-ESM2-0 ssp126 2061–2100 MRI-ESM2-0 ssp245 2015–2100 MRI-ESM2-0 ssp370 2015–2100 MRI-ESM2-0 ssp585 2015–2100 UKESM1-0-LL Historical 1850–2014 UKESM1-0-LL piControl 1960–2100 UKESM1-0-LL ssp126 2061–2100 UKESM1-0-LL ssp245 2015–2100 UKESM1-0-LL ssp370 2015–2100 UKESM1-0-LL ssp585 2015–2100 5 D6.1 Storm and flood data 2.1.2 Wind field calculation Both observed and simulated tracks are converted into windfields using the Holland (2008) parametric model, as implemented in the climada Python package (Siguan et al., 2023). Windfield generation is restricted to in-land grid cells at a spatial resolution of 300 arcseconds (~0.0833°). A wind speed threshold of 17.5 m/s (34 knots), corresponding to the tropical storm classification on the Saffir-Simpson scale, is applied; i.e., grid cells where wind speed is estimated to be lower than this threshold are considered unaffected, and their wind speed is set to zero. All subsequent filtering and wind speed thresholds are applied directly to the windfields rather than the original tracks, ensuring consistency throughout the workflow. 2.1.3 Sampling of storms To produce regionally meaningful datasets, these global windfields are further processed on a subbasin basis. This step involves bias-adjusting the windfield statistics for each region to account for differences in climatology, observational coverage, and data quality. Subbasin-specific adjustment ensures that the final statistics accurately reproduce observed TC characteristics within each region during the historical period. The procedure is similar to that described in Geiger et al. (2021). Bias correction is applied to the timeseries of both event counts and mean intensities, using normalization periods tailored to each subbasin: West Pacific (WP, 1950-2015), North Atlantic (NA, 1950-2015), East Pacific (EP, 1950-2015), North Indian (NI, 1980-2015), South Indian (SI, 1980-2015), South Pacific (SP, 1980-2015), and South Atlantic (SA, 1980-2015). For each basin, the simulated event count is scaled by the ratio of the sums of observed to simulated events over the normalisation period, . Similarly, mean α=Σ 𝐸𝑣𝑒𝑛𝑡𝑠𝑜𝑏𝑠 Σ 𝐸𝑣𝑒𝑛𝑡𝑠𝑠𝑖𝑚 intensity (i.e., each storm’s maximum wind speed over land, averaged over all storms during the normalisation period) and the intensity’s standard deviation are scaled using the ratio of observed to simulated corresponding values, and respectively. This approach ensures β=𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑜𝑏𝑠 𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑠𝑖𝑚 γ=σ𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑜𝑏𝑠 ( ) σ𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑠𝑖𝑚 ( ) alignment with both the average frequency and intensity of observed TCs in each subbasin. For each year and basin, the number of events to be sampled is determined by first calculating the ratio of the calibrated expected annual event count to the total of 1500 simulated tracks per year, then 𝑓𝑟𝑒𝑞𝑦𝑒𝑎𝑟, applying the bias-adjustment factors described above, yielding a bias-corrected expected basin-wise annual 6 D6.1 Storm and flood data event count . To reflect the probabilistic nature of TC occurrence, 100 independent 𝑁=α·𝑓𝑟𝑒𝑞𝑦𝑒𝑎𝑟 1500 realisations are generated per year. In each realisation, the number of events is drawn from a Poisson 𝑁 distribution with the expected value set to the bias-adjusted mean event count . events are then 𝑁𝑁 randomly sampled from the pool of precomputed windfields corresponding to that basin and year. The mean intensity of this set of events should be between the range 𝑁, where 𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑚𝑒𝑎𝑛−𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑠𝑡𝑑, 𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑚𝑒𝑎𝑛+𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑠𝑡𝑑 ⎡⎢⎣⎤⎥⎦ and while the bias-adjusted 𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑚𝑒𝑎𝑛=β·𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑚𝑒𝑎𝑛 𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑠𝑡𝑑=γ·𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑠𝑡𝑑 standard deviation of the intensity should be between 0.5 and 10 m/s. If the set does not satisfy these conditions, it is discarded, and a new set is drawn. If no suitable event set is found after 10,000 iterations, the aforementioned acceptable intensity range is expanded by ±1 m/s. This process is repeated until 100 realisations have been generated. This basin-wise dataset was then aggregated to indicate the yearly maximum windspeed per pixel per realisation, and the basins were merged into a global grid. This final dataset is saved in NetCDF format. The workflow is modular and fully parameterised, supporting straightforward adaptation to additional GCMs, experiments, resolutions, or regions. 2.2 Flood data The flood data were produced according to the following methodology: 2.2.1 Hydrological modelling The data is derived from river runoff simulations using multiple global hydrological models (GHMs), driven by atmospheric forcing from global climate models, in the framework of ISIMIP3b (Frieler et al., 2025). Daily runoff was simulated globally on a 0.5° horizontal grid for pre-industrial, historical, and projected future conditions, according to different Shared Socioeconomic Pathways (SSPs; Riahi et al., 2017). Table 2 lists the GHMs, GCMs, and SSPs used. Note that not all combinations of GHM, GCM, and SSP were simulated. 7 D6.1 Storm and flood data Table 2: Models and scenarios for which flood data are available. GHMs GCMs Scenarios CLASSIC gfdl-esm4, ukesm1-0-II historical, picontrol, ssp126, ssp370, ssp585 CWatM ukesm1-0-II, gfdl-esm4, ipsl-cm6a-lr, mpi-esm1-2-hr, mri-esm2-0, ukesm1-0-II historical, picontrol, ssp126, ssp370, ssp585 H08 ukesm1-0-II, gfdl-esm4, ipsl-cm6a-lr, mpi-esm1-2-hr, mri-esm2-0, ukesm1-0-II historical, picontrol, ssp126, ssp370, ssp585 JULES-W2 ukesm1-0-II, gfdl-esm4, ipsl-cm6a-lr, mpi-esm1-2-hr, mri-esm2-0, ukesm1-0-II historical, picontrol, ssp126, ssp370, ssp585 MIROC-INTEG-LAND ukesm1-0-II, gfdl-esm4, ipsl-cm6a-lr, mpi-esm1-2-hr, mri-esm2-0, ukesm1-0-II historical, picontrol, ssp126, ssp370, ssp585 WaterGAP2-2e ukesm1-0-II, gfdl-esm4, ipsl-cm6a-lr, mpi-esm1-2-hr, mri-esm2-0, ukesm1-0-II historical, picontrol, ssp126, ssp370, ssp585 WEB-DHM-SG ukesm1-0-II, gfdl-esm4, ipsl-cm6a-lr, mpi-esm1-2-hr, mri-esm2-0, ukesm1-0-II historical, picontrol, ssp126, ssp370, ssp585 2.2.2 Hydrodynamic modelling Daily runoff was interpolated horizontally from the 0.5° ISIMIP grid to CaMa-Flood’s unit catchments (using a map, inpmat-30min.bin, included with the model code) and then routed through the global hydrodynamic model CaMa-Flood (Yamazaki et al., 2011), version 4.0.0. The resulting daily river discharge at 0.25° resolution then served as a basis for the inundation modelling. The annual maximum flood depth at each grid cell was determined, resulting in a yearly estimate of maximum flood depth (above river channel), and associated flooded area fraction, at 0.25° resolution. 8 D6.1 Storm and flood data 2.2.3 Accounting for flood protection A generalised extreme value distribution (GEV; Willner et al., 2018) was fit to the pre-industrial distribution of daily discharge, separately at each grid cell. From this fit, the return period associated with any given discharge level was determined in both historical and future simulations. This allowed us to produce flood estimates that accounted for different levels of assumed flood protection by masking out flooding in grid cells and years where the annual maximum discharge was below the return period against which that grid cell was assumed to be protected. Table 2 describes the different flood protection assumptions. Table 3: Flood protection assumptions. File name specifier Explanation none no flood protection 2y protection against floods with a pre-industrial return period of 2 years or shorter 40y protection against floods with a pre-industrial return period of 40 years or shorter flopros protection standards according to the merged layer of FLOPROS (Scussolini et al., 2016) hanze protection standards according to HANZE within Europe (Paprotny et al., 2023) and FLOPROS for the rest of the world 2.2.4 Flooded fraction The final, usable result is the flooded area fraction at 0,25° horizontal resolution. This relates to the fraction of a unit catchment that is inundated. Each unit catchment corresponds to a 0.25° x 0.25° grid cell but is irregularly shaped, which means that the flood fraction does not precisely represent the inundated fraction of the grid cell; this should be kept in mind when analysing the data. 3. Results Achieved The datasets, produced according to the methodologies described above, are publicly available: 9