A Global Approach for Mapping Multi-Hazard Susceptibility Using Deep Learning: A Case Study in Japan

Abstract Disaster preparedness is a key determinant in reducing multi-hazard risk by enabling timely and effective responses. A crucial, yet often missing, component of disaster preparedness is the development of a multi-hazard susceptibility map, which can be interpreted as the spatially distributed probability of occurrences of hazards. Here, we provide a globally applicable approach for multi-hazard susceptibility mapping based on deep learning, and demonstrate its use on a range of geophysical, atmospheric and hydrological hazards using Japan as case study. We show that single hazard susceptibility levels have a varied spatial distribution throughout Japan and are influenced by attributes such as terrain, sub-surface, atmospheric, and lithological factors. The multi-hazard susceptibility map shows a diverse range of values, with southern regions in Japan exhibiting elevated multi-hazard susceptibility, primarily driven by heatwave and earthquake hazards. The insights from our multi-hazard susceptibility map can help prioritise resources to the most vulnerable areas and support targeted resilience-building efforts in communities facing multiple hazards.