Building a natural disaster risk management system for resilient communities

Abstract

The human and economic cost of natural disasters have been increasing exponentially over time. According to UN experts, climate-related and geophysical disasters in the last 20 years have killed over 1.3 million people and left 4.4 billion injured, homeless, or in need of humanitarian assistance. This project aimed to design a new software architecture to integrate and improve existing Disaster Risk Management (DRM) technologies through better performing Artificial Intelligence models, engineering optimizations and scalable designs, so as to build a hazard-agnostic DRM software. There were several services and features developed and deployed to provide Disaster Management services, including Deep Learning pipelines on the cloud for Damage Localization and Damage Classification from Satellite imagery, Simulation tools for estimating damage losses and injuries, and a Progressive Web Application to provide a visual interface for using these features. The report documents the literature review, experiments and engineering procedures carried out for the development of the DRM features, as well as the challenges faced and potential areas of improvement in the future.