Using machine learning, computer vision, and remote sensing to monitor earthquake recovery
Natural disasters have claimed thousands of lives, affected millions, and caused trillions in damage over the past two decades (Ritchie, Rosado, and Roser, 2024). While effective disaster management is crucial for mitigating damage and saving lives, the recovery stage remains the least researched (Ghaffarian, Rezaie, and Kerle, 2020). Despite significant funding from multilateral organizations, some regions still struggle with effective recovery. This dissertation addresses this gap by developing a tracking system using Machine Learning (ML) and Remote Sensing (RS).
Recovery lacks a unified definition, complicating tracking efforts. However, we recognize that both disasters and recovery impact the status quo of a region. Our focus is on tracking changes caused by disasters and subsequent recovery to gauge progress. Instead of tracking every change, our goal is to demonstrate that RS and ML can monitor changes with minimal human intervention, laying the groundwork for a monitoring tool that future research can build upon.
We select indicators in three essential domains of community recovery: the Built, the Economic, and the Natural Environments. These indicators and tracking methods are derived from the literature. Our innovation lies in tracking these domains collectively, adding more capabilities, and creating a replicable code adaptable to different regions.
Our chosen indicators are house reconstruction for the Built Environment, vegetation recovery for the Natural Environment, and GDP tracking via proxies for the Economic Environment. Additionally, we monitor Land Usage and Cover Change to capture ground-level transformations. We have developed four models to track these indicators and applied them to Antakya, Turkey, which was severely impacted by the February 6, 2023, earthquakes (UNDP, 2023). The models illustrate real-time recovery tracking, ensuring that improvements in one dimension do not compromise others and enabling timely interventions. They also provide insights for stakeholders and potential donors, encouraging continued funding and support for recovery efforts.
Our models reveal a 50% reduction in the Built Environment, consistent with government reports, while promised new construction is behind schedule (Michaelson, 2024). The Natural Environment model shows that vegetation recovered to pre-disaster levels; however, possible compromises are rising due to the construction of new zones. Analysis of Nighttime Lights (NTL) indicates many people still reside in temporary settlements, and NTL levels are still 10% less compared to pre-disaster levels.