Development of a Multi-Hazard Index for India: Applying CNN U-net Deep Learning framework to major Hydro-Meteorological Extremes

India’s increasing exposure to hydro-meteorological hazards under climate change calls for integrated, large-scale risk assessment methods that surpass traditional single-hazard frameworks. This study introduces a pioneering composite multi-hazard index for India at pixel and administrative levels, utilizing a multi-hazard susceptibility mapping framework with a CNN U-Net-based deep learning architecture. This framework captures spatial vulnerability patterns for four critical hydro-meteorological hazards: floods, droughts, heatwaves, and cyclones. The framework exploits the capability of deep learning to extract complex spatial relationships from diverse geospatial datasets, including digital elevation models and their derivatives (e.g. slope, curvature), climatological variables (e.g. temperature, rainfall, solar radiation), hydrological parameters (e.g. groundwater storage, TWI, soil moisture), and land cover classifications, enabling a high-resolution (90 metres) pixel-level hazard susceptibility prediction across India’s diverse physiographic landscapes. The hazard inventory data used for model training and validation were systematically compiled from reliable global and national primary and secondary datasets available in the public domain. Susceptibility mapping exhibited strong predictive accuracy, surpassing 90% across various hazard types. The maps effectively identify high-risk zones within river basins, coastal regions, and interior areas, demonstrating the robustness of the deep learning framework for nationwide assessments. Flood susceptibility is prominent in the Indo-Gangetic Plain, the Western regions, Northeast, and parts of the Eastern Ghats, while heatwaves are concentrated in the Indo-Gangetic Plains and the central Indian plateau. Regions such as the Indo-Gangetic Plain, Eastern Ghats, and Eastern Coast are particularly susceptible to multiple hazards, underscoring their significance for multi-hazard disaster risk management and climate adaptation strategies. The methodology demonstrates the scalability and operational feasibility of using deep learning for multi-hazard assessment, effectively capturing spatial patterns. It offers a transferable framework adaptable to regions facing complex climate-driven hazards, which can heighten overall risk through combined impacts. The multi-hazard index enables comparison of hazard exposure and vulnerability across regions, supporting multi-risk spatial planning, disaster preparedness, and climate adaptation at various governmental levels.