Modernizing Quezon City's Building Risk Data with AI and Earth Observation

Developing, maintaining, updating, and forecasting city-scale building risk data is costly yet essential for tracking progress toward the UN Sendai Framework for Disaster Risk Reduction 2015-2030. Although advances in information technology and big data have enabled a shift from low- to high-resolution mapping, current state-of-practice methods remain limited in detail and accuracy due to the challenges in coarse-to-fine-grained mapping across spatiotemporal scales. To address these challenges, we develop a novel spatiotemporal framework that integrates the rich information from time-series Earth Observation data and leverages recent advances in artificial intelligence, including graph deep learning, state-space modeling, and probabilistic inference. Through an academic research collaboration with Earthquakes and Megacities Initiative (EMI) and Quezon City Disaster Risk Reduction and Management Office (QCDRRMO), we present a flexible probabilistic framework and an open-access dataset comprising annual spatiotemporal 10-meter maps of building exposure and physical vulnerability for Quezon City, Philippines, from 2016 to 2030, along with associated uncertainty estimates. Beyond demonstrating the potential of AI and Earth Observation in modernizing building risk data, our work has offered a dynamic approach to disaster risk auditing by capturing spatiotemporal changes in exposure and vulnerability, thereby empowering local governments with data-driven insights for effective disaster risk reduction.