Green Infrastructure Mapping in Urban Areas Using Sentinel-1 Imagery

High temporal resolution of synthetic aperture radar (SAR) imagery (e.g., Sentinel-1 (S1) imagery) creates new possibilities for monitoring green vegetation in urban areas and generating land-cover classification (LCC) maps. This research evaluates how different pre-processing steps SAR affect accuracy. Machine learning (ML) methods were applied three study areas: random forest (RF), support vector machine (SVM), extreme gradient boosting (XGB). Since the presence speckle noise is inevitable, adaptive filters examined. Using backscattering values S1 imagery, SVM classifier achieved a mean overall accuracy (OA) 63.14%, Kappa coefficient (Kappa) 0.50. with Lee filter window size 5×5 (Lee5) reduction, 73.86% 0.64 OA achieved, respectively. An additional increase LCC was obtained texture features calculated from grey-level co-occurrence matrix (GLCM). The highest extracted GLCM using classifier, Lee5 78.32% 0.69 values, improved an evaluation various radiometric confirmed ability to apply classifier. For supervised classification, method outperformed RF XGB methods, although computational time needed SVM, whereas performed fastest. These results suggest infrastructure mapping areas. Future should address use multitemporal data along ML algorithms described this research.