Global Flood Susceptibility Map (GFSM v1): A high resolution (30m) flood susceptibility dataset derived from multi-source Earth observation and geospatial data

The Global Flood Susceptibility Map (GFSM v1) is a globally harmonized, near-global flood susceptibility dataset at 30 m spatial resolution. It maps the relative propensity of land areas to flooding based on topographic, hydrological, meteorological, vegetation, and anthropogenic factors. GFSM is a susceptibility baseline, not a real-time flood forecast or an event-specific inundation model. The dataset and its validation are described in the open-access Scientific Data paper A high-resolution global flood susceptibility dataset derived from multi-source earth observation and geospatial data, published on 23 September 2026. Dataset overview A gradient-boosted tree model (XGBoost) was developed using approximately 30.45 million stratified, tile-balanced flood and non-flood samples. Training labels were derived from the Aqueduct Flood Hazard Maps v2 5-year baseline. The model used nine flood-conditioning factors: Elevation Slope Aspect Topographic Wetness Index (TWI) Height Above Nearest Drainage (HAND) Normalized Difference Vegetation Index (NDVI) Distance to water bodies Distance to roads Rainfall frequency The modeling framework was deployed across 192 country or country-climate units to generate 17,069 processed spatial tiles. The released five-class equal-interval (EI-5) product classifies the model output into globally consistent susceptibility categories. Dataset specifications Data product: GFSM v1 Zenodo record version: v2 Spatial resolution: 30 m Geographic coverage: Near-global land areas, approximately 80°N to 60°S Coordinate reference system: World Mercator (EPSG:3395) Format: Tiled GeoTIFF archives with accompanying metadata Output: Five-class equal-interval (EI-5) flood susceptibility raster NoData value: 0 (masked or unavailable pixels) Valid class values: 1–5 Spatial tiles: 17,069 Modeling units: 192 country or country-climate units Classification scheme The continuous model output was reclassified using equal-interval slicing: Value Class Underlying probability interval 0 NoData / masked Not applicable 1 Very Low 0.0–0.2 2 Low 0.2–0.4 3 Moderate 0.4–0.6 4 High 0.6–0.8 5 Very High 0.8–1.0 NoData pixels may represent open water, missing input coverage, or areas outside the modeled domain. Validation summary Internal validation: Median ROC-AUC of approximately 0.95 and median AUPRC of approximately 0.94 across the modeling units. Independent event-based plausibility assessment: 79% of Dartmouth Flood Observatory (DFO) flood-event locations corresponded with mapped susceptible areas within 5 km, increasing to 92% within 10 km. The internal metrics measure discrimination against model-derived labels. The DFO comparison provides event-based plausibility context and should not be interpreted as direct pixel-level inundation validation. Download and access The Zenodo archive is the canonical release of GFSM v1. Because of the dataset's size, the GeoTIFF tiles are grouped into regional compressed archives. Download GFSM_Tile_Index.zip, which contains the GeoPackage (.gpkg) tile index. Open the tile index in a GIS such as QGIS or ArcGIS and select the footprints intersecting your area of interest. Use the index to identify and download the required regional archive or archives, such as GFSM_N20W020.zip. Extract the GeoTIFF tiles and associated metadata JSON files for analysis. GFSM v1 is also available as a Google Earth Engine ImageCollection: projects/floodsus/assets/fsm_ei5 Additional access and usage resources: GFSM Explorer — interactive visualization GFSM GitHub repository — documentation, metadata, Earth Engine scripts, Python examples, and notebooks Published paper — methods, validation, and technical documentation Intended use and limitations GFSM v1 supports broad screening, regional comparison, exposure assessment, and geospatial flood-susceptibility analysis, particularly where consistent high-resolution information is limited. Flood risk also depends on exposure and vulnerability; GFSM does not represent these components. The training labels are model-derived and coarser than the 30 m output. The dataset does not directly model real-time conditions, individual flood events, future climate or land-use change, local flood defenses, fine-scale drainage infrastructure, or site-specific engineering conditions. Users should combine GFSM with local observations, regulatory flood maps, flood-defense and drainage information, hydrological expertise, and applicable planning standards. For rigorous area calculations, reproject the data to an appropriate equal-area coordinate reference system. Citation If you use GFSM v1, please cite both the published paper and the dataset release. Paper Waleed, M., Sajjad, M., Al-Ghamdi, S. G., & Gao, M. (2026). A high-resolution global flood susceptibility dataset derived from multi-source earth observation and geospatial data. Scientific Data. https://doi.org/10.1038/s41597-026-08338-1 Dataset Waleed, M. (2026). Global Flood Susceptibility Map (GFSM v1): A high resolution (30m) flood susceptibility dataset derived from multi-source Earth observation and geospatial data (Version v2) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.20568218 When reporting derived results, state the GFSM version, access route (Zenodo or Google Earth Engine), access date, and any preprocessing or reclassification applied. License The GFSM data, metadata, and documentation are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). Code examples in the GitHub repository are licensed under the MIT License. Funding and acknowledgments Mirza Waleed was supported by a postgraduate studentship from the Hong Kong Baptist University Research Grant Committee (PhD studentship, 2022–2026). Muhammad Sajjad was supported by the Research Grants Council of the Hong Kong Special Administrative Region, China (Grant No. RFS2223-2H02). The Hong Kong Baptist University Graduate School provided support for Mirza Waleed's research visit to King Abdullah University of Science and Technology (KAUST). Computational processing used the Shaheen supercomputing facilities at KAUST, and parts of the analysis used Google Earth Engine. The authors acknowledge the organizations and contributors who provided the open datasets used to produce GFSM v1.