AI-Powered Flood Risk Assessment for Gilgit-Baltistan Using Multi-Source Satellite Data and Machine Learning

Flood disasters are intensifying worldwide due to climate change, with mountainous regions among the most vulnerable yet least studied. This paper presents an AI-powered flood risk assessment framework for Gilgit-Baltistan, Pakistan, a high-mountain region prone to flash floods and glacial lake outburst floods (GLOFs). Multi-source satellite datasets—including CHIRPS precipitation, JRC Global Surface Water occurrence, ERA5-Land soil moisture, SRTM elevation, and MODIS land surface temperature—were integrated within Google Earth Engine and analyzed using a Random Forest classifier. District-level risk classes (high, medium, low) were derived using historical flood records and validated through Leave-One-Out Cross-Validation, achieving 88.9% accuracy. Results consistently identified Astore, Diamer, and Nagar as high-risk districts, with precipitation and water occurrence as dominant predictors. Unlike many flood studies in lowland regions, elevation and slope were secondary yet important drivers in this mountainous context. The study demonstrates that even with limited ground data, satellite-driven AI models can deliver actionable insights for disaster management. The framework is scalable to other mountain regions globally and provides a step toward operational early warning and climate adaptation systems.