PRISI: A novel piecewise radar impervious surface index for urban monitoring using Sentinel-1 data

Abstract

• We propose a novel piecewise radar impervious surface index (PRISI). • High backscatter and strong coherence of impervious surfaces are considered jointly. • We design an adaptive impervious surface extraction strategy using finite mixture models. • The proposed PRISI owns robust regional universality and seasonal suitability. The impervious surface (IS) is an important symbol of the urban ecological environment and urbanization process. An accurate, simple, and effective IS extraction method is crucial to environmental and social-economic research. The widely used optical data are frequently disturbed by cloud cover in low latitude and humid areas, bringing great difficulties to remote sensing monitoring of IS. Thus, it is necessary to fully explore the capability of IS extraction relying on SAR data sources alone. However, due to the diversity of IS materials and low spectral consistency, IS extraction in an accurate manner using the backscatter characteristics remains to be challenging. Therefore, we innovatively analyze the complementary effect between backscatter and interferometric coherence from the perspective of IS heterogeneity. Further, we propose a Sentinel-1-based Piecewise Radar Impervious Surface Index (PRISI) that takes into account both the high scattering characteristics and the physical stability of IS. We demonstrate that the proposed adaptive index PRISI-A based on the Gaussian mixture model (GMM), Log-Gaussian mixture model (LGMM) and Expectation maximum (EM) algorithm leads to superior performance. Moreover, we provide a non-adaptive index PRISI-D, which adopts empirical parameter settings from trial and error. The computational-friendly PRISI-D yields comparable results compared to PRISI-A. The experimental results show that the IS extraction overall accuracy of PRISI-D and PRISI-A achieve 87.50 % and 88.11 %, which consistently outperform the previous optical indexes by 6.2–11.7 % and the existing radar indexes by 5.2–21.4 % on average. The results reveal that PRISI owns robust regional universality and seasonal suitability, which significantly improves the capability of IS extraction relying on SAR data sources alone.