Global Flood Susceptibility Map (GFSM v1)

Objective. Existing global flood datasets either lack spatial resolution or rely on inconsistent regional methodologies, limiting their utility for localized risk planning.

Approach. Architected a scalable machine-learning framework integrating nine topographic, hydrological, meteorological, and anthropogenic conditioning factors. An XGBoost model was trained on approximately 30.45 million samples from 17,069 curated tiles and deployed across 192 geopolitical and climate units using Google Earth Engine and the KAUST Shaheen supercomputer.

Impact. Published as an open-access Data Descriptor in Scientific Data, GFSM v1 provides five susceptibility classes at 30-meter resolution from approximately 80°N to 60°S. Cross-validation produced a median ROC-AUC of approximately 0.95. The dataset, reproducible workflow, and interactive dashboard are openly available for regional screening, exposure assessment, and disaster risk reduction planning, particularly in data-sparse regions.