Measuring spatial urban recovery and expansion in tsunami-affected areas by satellite remote sensing and machine learning approach

This study proposes a novel method for monitoring long-term urban recovery and expansion by integrating freely available medium-resolution satellite data with machine learning. While traditional recovery assessments often rely on regional-based and timely sparse data, we demonstrated that two decades of building dynamics were assessed using time-series Landsat imagery. We applied this method to Banda Aceh, Indonesia, which was devastated by the 2004 Indian Ocean Tsunami. A random forest regression model was developed using spectral indices, trained on building counts from high-resolution imagery (2004, 2007, and 2024). The model achieved high estimation accuracy, with a coefficient of determination (R 2 ) of 0.79, showing strong agreement with visual interpretation and field investigations. Our analysis captured regional variations in recovery speed and significant recent expansion in both unaffected and inundated areas. These findings provide a scalable approach for disaster risk reduction and urban planning, offering a vital tool for policymakers to monitor reconstruction progress and manage future urban growth in disaster-prone regions.