Radargrammetry for Digital Elevation Model Generation Using Envisat Reprocessed Image and Simulation Image

Abstract

The digital elevation model (DEM) is one of the most important sources used for Earth surface analysis because of its various applications and its usability in subsequent studies. A DEM can be generated with interferometric synthetic aperture radar (InSAR) and radargrammetry techniques from synthetic aperture radar (SAR) images with orbital separation between them as the baseline. The on-board radar imaging system records both phase and intensity information of the backscattered signals. Radargrammetry is based on the disparity between two intensity images which is less affected by temporal and atmospheric decorrelation while InSAR uses the phase differences between two images. However, SAR intensity images may have low spatial resolution compared to optical images that are used in the photogrammetric DEM generation and the original SAR image is degraded by speckle noise. This leads to low accuracy of radargrammetric DEMs. It is necessary to develop methods focusing on original SAR images to improve the accuracy of radargrammetric DEM generation. In this letter, we propose that the single look complex (SLC) SAR image reprocessing method can be utilized to improve the image quality of radargrammetric DEMs. Simulated noise-free SAR images are used to generate radargrammetric DEMs for the purpose of comparison. Three SLC SAR images are selected for image quality improvement in each track. Other images in the same track are matched and processed to one reference image. The SAR simulation images are generated based on the reference image in each track. Therefore, the products in the same track have the same meta-information. Each DEM product is compared with conventional radargrammetric DEMs. This study shows that DEMs generated from re-processed SAR images are more accurate than radargrammetric DEMs from single-pair SAR images. Also, the accuracy of radargrammetric DEM produced with the proposed method is close to DEMs generated with noise-free simulated SAR images.