O.M. Ibrahim, Z. Yang, T.J. Mulrooney
North Carolina Central University,
United States
Keywords: Synthetic Aperture Radar, Support Vector Machine, Random Forest Classification, Freeman-Durden Decomposition
Summary:
Accurate and timely flood inundation mapping is essential for disaster preparedness and environmental resilience. This study presents an automated geospatial framework integrating NASA UAVSAR fully polarimetric L-band data with machine-learning models to map flood extent in southeastern North Carolina following Hurricane Florence. A Python-based polarimetric decomposition workflow was used to generate flood-sensitive composites, which served as inputs for Random Forest, Support Vector Machine, K-Nearest Neighbor, and Maximum Likelihood classification models. Comparative results show Random Forest achieved the highest accuracy (87.37%), demonstrating strong capability for detecting flood extent in vegetated and complex hydrological environments. The study highlights the potential of combining SAR remote sensing, machine learning, and automated geospatial processing to support rapid flood assessment, emergency response, and resilience planning. This scalable framework provides practical applications for agencies seeking data-driven tools for disaster management and infrastructure protection.