<![CDATA[Flood disasters are often triggered by deforestation and land cover changes, highlighting the need for precise spatial flood susceptibility modeling. This study aims to analyze the spatiotemporal dynamics of deforestation (2021–2025) in Langsa and compare the accuracy of three flood hazard prediction models: Multi-Criteria Decision Analysis (MCDA), Geomorphic Flood Index (GFI), and Machine Learning using Random Forest. The study employed a quantitative experimental approach through the cloud-computing architecture of Google Earth Engine (GEE) and offline simulation using QGIS to process multispectral Landsat 9 satellite imagery along with landscape physical data. The results indicate a reduction in vegetation cover area of 1,606.12 hectares (-11.48%), which directly contributed to the expansion of built-up areas and open land. Comparative evaluation demonstrated that the Random Forest algorithm achieved the highest flood modeling accuracy, with an Overall Accuracy of 91.17% and a Kappa Coefficient of 0.87, outperforming MCDA, which was prone to over-prediction bias, and GFI, which exhibited algorithmic blind spots in localized pluvial flood areas. Risk exposure analysis further revealed that 38.5% of current built-up infrastructure areas are located within high flood susceptibility zones due to the loss of ecological infiltration areas. The outputs of this modeling process were successfully integrated into an interactive Decision Support System (DSS) based on a WebGIS dashboard to facilitate spatial disaster mitigation dissemination for policymakers.]]>
Copyrights © 2026
