Hierarchical fusion of optical and dual-polarized SAR on impervious surface mapping at city scale

Abstract

The combined use of the data from Sentinel-1 (S1) dual-polarized Synthetic Aperture Radar (SAR) and Sentinel-2 (S2) Multispectral Sensor Images (MSI) has been successfully applied in many remote sensing applications. However, when it comes to the mapping of impervious surface (IS), the non-negligible shadows on S2 images can severely degrade the mapping accuracy. How to effectively incorporate S1 and S2 data to precisely extract IS from cities has not been well investigated yet. Besides, it still remains challenging to extract appropriate polarimetric features of S1 to discriminate IS and supplement the drawbacks of S2. In this research, we explore more effective polarimetric features of S1 and propose a hierarchical framework for IS mapping at the city scale by synergetic fusion of dual polarized SAR and multispectral information. The new polasrimetric features of S1 is initially explored based on the backscattering mechanism of dual polarized SAR. Then the IS is identified from shadow-free and shadow areas in two consecutive steps with different feature sets. The result indicates that our new polarimetric features of S1, i.e., the diagonal elements of coherence matrix, are more suitable for IS mapping due to their associated physical meaning with different land covers. With suitable feature sets, our proposed methods performed well on the urban area with abundant shadows, with an Overall Accuracy (OA) of 93.9% and a Kappa coefficient of 0.92 achieved.