An impervious surfaces extraction method based on optical, ascending and descending SAR remote sensing imagery in high-density urban core areas

• A new SAR shadowed index based on Ascending SAR and Descending SAR imagery (ADSI). • Hierarchical sampling (HS) in optical, Ascending SAR, and Descending SAR shadowed areas. • High accuracy impervious surface extraction in high-density urban core areas. Accurately monitoring impervious surface (IS) changes is vital for assessing the effects of urbanization on human activities and its broader environmental, economic, and social sustainability. However, optical building shadows pose significant challenges for IS extraction in high-density urban core areas. Even though combining optical and synthetic aperture radar (SAR) imagery makes it easier to identify impervious surfaces, buildings block SAR signals, causing them to mix with optical shadows. Ascending (AS) and descending (DE) SAR imagery offer potential solutions due to their AS/DE characteristics. This study proposes an IS extraction method utilizing hierarchical samples, optical and AS/DE SAR imagery (HS-OAD). Firstly, the AS/DE SAR-based shadowed index (ADSI) is proposed by analyzing the characteristics of building shadowed areas. Then, the OTSU method is adopted to threshold imagery into non-shadowed areas, optically shadowed areas, and SAR shadowed areas. Within each area type, samples representing four urban land covers—IS, vegetation, bare soil, and water bodies—are comprehensively labeled. Subsequently, a random forest classifier is employed to extract IS integrating optical, AS SAR, and DE SAR imagery. The results indicated that the integration of AS and DE SAR imagery significantly reduces the misclassification of IS and water bodies within optically shadowed areas. Furthermore, the hierarchical samples enhance the classification accuracy and reduce the misclassification between IS and vegetation in optically shadowed areas. Moreover, samples labeled in SAR-shadowed areas contribute to the reduction of misclassification between IS and bare soil. Our proposed method performed well in high-density urban core areas with extensive shadows, achieving an overall accuracy of 89.61% and a Kappa coefficient of 0.79.