Land-cover classification using multitemporal ERS-1/2 InSAR data

In this study the potential of ERS-1/2 Tandem InSAR data for land-cover classification was investigated at a 2500 km² study area around the Helsinki metropolitan area in Southern Finland. A time-series of 14 ERS-1/2 SAR Tandem image pairs was processed into 28 five-look intensity images, 14 Tandem coherence images and two coherence images with a longer temporal baseline (36 and 246 days). All image data was coregistered and orthorectified into map coordinates using an InSAR DEM. A two-stage hybrid classifier method was employed, where the water-class was classified separately in the first classifier stage, and the remaining classes were classified with an ISODATA classifier. Temporal averaging and Principal Components Transformation (PCT) were used to reduce the number of images fed into ISODATA. Classification accuracy was assessed using high-resolution aerial orthophotos, digital base maps and the Finnish National Forest Inventory (NFI). The overall accuracy for six classes (Field/Open Land, Dense Forest, Sparse Forest, Mixed Urban, Dense Urban, Water) was found to be 90% with kappa coefficient of 0.86. Interferometric coherence carries more land-cover related information than the backscattered intensity. This study confirms that the ERS-1/2 Tandem archives could be exploited for land-cover classification.