Synergizing radar and optical data for accurate urban impervious surface mapping in cloud prone areas

Abstract

In the swiftly evolving urban environment, the rapid and nuanced developmental changes are gaining heightened attention. This emphasizes an urgent need for accurate, continuous, and timely urban monitoring, crucial for evaluating the impacts of urbanization and advancing sustainable development assessments. However, cloud interference distorts reflectance signals and hampers timely and accurate urban observations. While synthetic aperture radar (SAR) offers penetration capabilities, its effectiveness is impeded by imaging limitations, necessitating the integration of SAR and optical data. However, existing fusion techniques are primarily tailored for cloud-free scenarios, leaving a critical gap in addressing the challenge of information fusion under cloud-contaminated conditions. This study bridges this gap by introducing the SAR-Optical Dictionary Learning (SODL) method to synergize cloud-contaminated optical data and polarimetric SAR data to harness their complementary discriminative information, taking an example of estimating urban impervious surfaces (UIS). SODL avoids cloud interference by constructing an enhanced objective function to build a joint SAR-optical dictionary space, effectively maximizing the discriminative capabilities of both data sources. Experimental results across tropical and subtropical regions in China demonstrate the robust performance of SODL, with notable improvements in UIS estimation compared to both traditional and deep learning fusion methods, of up to 33.81% in overall accuracy (OA). Furthermore, SAR and optical fusion with SODL significantly outperforms the results using either SAR data or optical data alone, with increase in OA of up to 46.54%. These outcomes underscore the promising potential of SODL in accurate urban monitoring in cloudy regions.