Exploring the World of Multi-Hazard Susceptibility Mapping With Deep Learning

Abstract

A crucial component of disaster preparedness is the development of a multi-hazard susceptibility map, which plays a vital role in comprehensive risk assessment, resource allocation, land use planning, emergency management, community preparedness, and decision-making. Recently deep learning methods have been showing potential to map susceptibility at a finer resolution. While prior research has predominantly focused on advanced single-hazard or simplified multi-hazard susceptibility mapping, an approach to explore multi-hazard susceptibility mapping using deep learning methods and explainable AI’s remains lacking to date. Addressing this gap, our research employs an ensemble Convolutional Neural Networks, to develop a multi-hazard susceptibility map. Leveraging diverse datasets and the MYRIAD-HESA framework, our analysis considers a range of hazards and their interactions, offering a more integrated view of the complex risk landscape faced by communities. Using Japan as a case study, the resulting susceptibility map serves as a valuable tool for informing land use and urban planning, resilient infrastructure development, and identification of suitable locations for critical facilities. Furthermore, it supports emergency management by facilitating resource prioritization, coordination, evacuation planning, and community awareness. This research contributes to evidence-based decision-making, policy development, and global disaster preparedness efforts.