Data and code for "Revealing unseen tail risks with a global flood catastrophe model"

Introduction This repository contains the data and code underpinning the under-review paper, Revealing unseen tail risks with a global flood catastrophe model by Wing et al. The README files within each data and scripts archive describes each file in detail. The data analyses, visualisation, and likely any subsequent research using this repository can be executed on a typical desktop machine (with typical runtimes being seconds, up to an hour for the most intensive scripts). Model data and access The global flood catastrophe model is made available via Fathom and can be run end-to-end using the open-source Oasis Loss Modelling Framework in a variety of ways, the easiest of which are with third-party platforms such as Verisk Model Exchange. Key datasets in this and other repositories include: Inundation maps from Wing et al. (2024). Exposure data using global_capital_stock.parquet and the Global Human Settlement Layer. Vulnerability functions from global_vulnerability.xlsx. For further information on data and model access, please contact Oliver Wing. Reproducing the main figures Fig. 1 Calculate a global exeedance-probability loss curve for flooding, including a breakdown by peril and occupancy. Run global_epc.m with global_losses.mat (global event loss data) and iso3_tiv.csv (exposure value per country). Fig. 2 Visualise the 250-year flood loss per country and calculate continental exceedance-probability loss curves. Use national_summaries.csv (loss metrics per country) to visualise the data in QGIS, and run regional_epc.m with global_region_losses.mat (global event loss data per region), iso3_tiv.csv (exposure value per country), and iso3_country_regions.csv (link between country, region, and continent). Fig. 3 Quantify observational data uncertainty for each continent's loss history and compare the spread to modelled outcomes. Run obs_v_mod_loss_all.m with emdat-flood-actual.csv (annual actual EM-DAT flood losses), emdat-flood-normalised.csv (annual normalised EM-DAT flood losses), emdat-flood-frequency.csv (annual EM-DAT flood loss event frequency), emdat-all-actual.csv (annual actual EM-DAT flood + TC losses), emdat-all-normalised.csv (annual normalised EM-DAT flood + TC losses), emdat-all-frequency.csv (annual EM-DAT flood + TC loss event frequency), event_losses.mat (event losses per region), and occurrence.csv (link between synthetic events and years). Fig. 4 Visualise each region's 250-year flood event in QGIS using /PML_event_continental_gpkg_/PML_event__.gpkg (flood return period per peril per catchment). Fig. 5 Analyse the natural variability in average annual losses for short model samples at various geographic and temporal scales. Run aal_sample_error.m with global_losses.mat (global event loss data), global_region_losses.mat (regional event loss data), DEU_losses.mat (Germany event loss data), bavaria_losses.mat (Bavaria event loss data). Fig. S2 Quantify observational data uncertainty for SE Asia's loss history and compare the spread to modelled outcomes. Run obs_v_mod_loss_sea.m with emdat-flood-actual.csv (annual actual EM-DAT flood losses), emdat-flood-normalised.csv (annual normalised EM-DAT flood losses), emdat-flood-frequency.csv (annual EM-DAT flood loss event frequency), emdat-all-actual.csv (annual actual EM-DAT flood + TC losses), emdat-all-normalised.csv (annual normalised EM-DAT flood + TC losses), emdat-all-frequency.csv (annual EM-DAT flood + TC loss event frequency), event_losses.mat (event losses per region), and occurrence.csv (link between synthetic events and years).