Urban Flood Mapping using Sentinel-1 SAR Data and Machine Learning: A Case of Maiduguri, Nigeria

Abstract

Flooding is one of the most devastating hydrological hazards, resulting in significant human and economic losses, particularly in rapidly urbanizing areas with poor urban planning. Recently, Maiduguri, Nigeria, faces recurrent floods, and traditional flood mapping methods relying on optical remote sensed data are often hindered by cloud cover. This study leverages Sentinel-1 Synthetic Aperture Radar (SAR) data and machine learning to overcome these limitations. The research employs threshold-based classification and change detection techniques to analyze pre-flood (January–August 2024) and post-flood (September–October 2024) SAR imagery. The methodology includes radiometric calibration, speckle filtering, and terrain correction to enhance flood detection accuracy. Flood extents were validated using ground reference points, achieving an overall accuracy of 88.6% and a Kappa coefficient of 0.82, confirming the reliability of SAR-derived flood maps. Findings reveal a 524.6 km² of inundated area, with severity concentrated along River Ngadda and low-lying areas. Zonal analysis highlights significant disparities, with the Maiduguri-Monguru Road experiencing 27.98% flooding, while elevated areas like Bornu Industrial Park (0.017%) remained minimally affected. The study also reveals that critical infrastructure are at risk. The research recommends improved urban drainage system, stricter land-use regulations and early warning systems to mitigate flood vulnerability.