Optimal estimation of tropospheric delay corrections to INSAR results from GPS observations based on SVM

Abstract

Interferometric synthetic aperture radar (InSAR) has been demonstrated useful for topographic mapping and surface deformation measurement. However, the atmospheric disturbance, especially the tropospheric heterogeneity, represents a major limitation to accuracy. It is usually difficult to accurately model and correct the atmospheric effects. Consequently, significant errors are often resulted in misinterpretation of InSAR results. The purpose of this paper is to seek to reduce the atmospheric effects on repeat-pass InSAR using independent datasets, viz. Global Positioning System (GPS). A between-site and between-epoch double-differencing algorithm for the generation of tropospheric corrections to InSAR results based on GPS observations is applied. In order to correct the radar results on a pixel-by-pixel basis, the Support Vector Machine (SVM) with adaptive parameters is introduced to regressively estimate tropospheric corrections over unknown points using the sparse GPS-derived corrections. The feasibility of applying SVM in troposphetic corrections estimation is examined by using data from the Southern California Integrated GPS Network (SCIGN). Cross-validation tests show that SVM method is more suitable than the conventional inverse distance weighted (IDW) method; it accounts for not only topography-dependent but also topography-independent atmospheric effects, so it seems optimal to estimate the tropospheric delay corrections of unknown pixels from GPS data.