Identifying Land-Use Related Potential Disaster Risk Drivers in the Ayeyarwady Delta (Myanmar) during the Last 50 Years (1974–2021) Using a Hybrid Ensemble Learning Model

Abstract

Land-use and land-cover change (LULCC) dynamics significantly impact deltas, which are among the world’s most valuable but also vulnerable habitats. Non-risk-oriented LULCCs can act as disaster risk drivers by increasing levels of exposure and vulnerability or by reducing capacity. Making thematically detailed long-term LULCC data available is crucial to improving understanding of those dynamics interlinked at different spatiotemporal scales. For the Ayeyarwady Delta, one of the least studied mega-deltas, such comprehensive information is still lacking. This study used 50 Landsat and Sentinel-1A images spanning five decades from 1974 to 2021 in five-year intervals. A hybrid ensemble model consisting of six machine-learning classifiers was employed to generate land-cover maps from the images, achieving accuracies of about 90%. The major identified potential risk-relevant LULCC dynamics include urban growth towards low-lying areas, mangrove deforestation, and the expansion of irrigated agricultural areas and cultivated aquatic surfaces. The novel area-wide LULCC products achieved through the analyses provide a basis to support future risk-sensitive development decisions and can be used for regionally adapted disaster risk management plans and models. Developed with freely available data and open-source software, they hold great potential to increase research activity in the Ayeyarwady Delta and will be shared upon request.

Keywords

• Ayeyarwady Delta
• Myanmar
• mega-delta
• land cover classification
• ensemble learning
• Landsat