Comparative Performance of Sentinel-1 SAR and Sentinel-2 MSI Imagery for Mapping Urban Infrastructure Footprints

Abstract

The choice of remote sensing data is critical for accurate urban mapping, particularly in regions with frequent cloud cover. This study evaluates and compares the capacity of Synthetic Aperture Radar (SAR) from Sentinel-1A and multispectral optical imagery from Sentinel-2 for mapping urban infrastructure footprints in Osogbo, Nigeria. Using the Google Earth Engine platform, we performed a supervised classification for the year 2023 on both datasets using a Random Forest classifier. Identical training and validation data were used for both classifications to ensure a robust comparison. The results demonstrated a stark contrast in performance. The classification based on Sentinel-2 optical imagery achieved an exceptionally high overall accuracy of 99.93% (Kappa = 0.999), effectively distinguishing between roads, buildings, vegetation, water, and bare ground. In contrast, the Sentinel-1A SAR-based classification achieved a moderate overall accuracy of 67.24% (Kappa = 0.568). The error matrix for the SAR classification revealed significant mixed classification, particularly between roads and buildings, and between certain urban features and vegetation. The study concludes that while Sentinel-1A offers all-weather capability, its utility for detailed urban infrastructure classification is limited when used independently due to its reliance on backscatter and texture, which lack the rich spectral information of optical sensors. For precise urban footprint mapping in studies of medium-sized cities, Sentinel-2 is vastly superior. However, the complementary all-weather capability of Sentinel-1A suggests that a synergistic multi-sensor fusion approach would be the most effective strategy for continuous urban monitoring in tropical regions.