Study On Ortho-rectification Methodology Of Space-borne Synthetic Aperture Radar Imagery

The advent of space-borne Synthetic Aperture Radar (SAR) with SEASAT mission in 1978 and the follow-on Shuttle Imaging Radar (SIR) series, the ERS SAR-1/2 SAR,JERS-1 SAR, RADARSAT-1 SAR and the newly launched ENVISAT ASAR in the new century, has generated widespread use of SAR data products across such diverse scientific disciplines as forestry, agriculture, ecology, geology, oceanography, hydrology and cartography etc.. However, the application of SAR imagery to the measurement of geophysical and biophysical parameters is constrained by geometric and radiometric effects caused by variable terrain. Such terrain can introduce large displacements in the SAR image geometry that inhibits the collocation of SAR-derived quantities with geographically referenced information acquired from other sources. In order to improve the usefulness of SAR images, their inherent geometric distortions must be eliminated by generating a radar ortho-imagery that corresponds to a well defined map projection. In this paper, the ortho-rectification methodologies of space-borne SAR imagery were studied in detail with objectives to solve key technique problems raised by developing an efficient and operational SAR imagery processing software system. One key test site located in Zengchen district of Guandong Province was established for the validation of the algorithms developed in this paper. The collected ground true data includes topography map with scale of 1:50 000 and the corresponding Digital Elevation Model (DEM). A series of ERS-2 SAR data were acquired from 1996 to 1998 for this test site. One scene of ERS-2 SLC imagery was selected as major test data and Ground Control Points (GCP) were manually collected. The ENVISAT ASAR, ERS-2 SAR,SIR-C/X SAR, JERS-1 SAR and RADARSAT-1 SAR imagery products acquired in other places such as Zhejiang Province, Fujian Province, Northeast of China were also used for this study. The ortho-rectification schemes developed in this paper are based on Range Doppler (RD) geo-location model. The methods to construct RD geo-location model and the algorithms to solve the model directly and indirectly were explored in detail. As supports for RD geo-location model construction, several satellite orbit modeling methods were compared and validated and a general frame to derive RD geo-location model parameters from meta data of different SAR imagery product was finally defined. No significant geo-location accuracy differences have been found for different orbit modeling methods if the input state vectors were provided with high accuracy in both between time interval and satellite position coordinates. Two new direct geo-location algorithms based on RD model, SCTGM and AIRGM, were developed in this paper. Validation results show that the two new algorithms can achieve higher geo-location accuracy compared with pure analysis geo-location method, while the time is much less consumed than the pure numerical geo-location method built upon many iterating routes. Based on these established geo-location methods, two methods to generate geocoded terrain corrected SAR image were investigated. The first is just a nonlinear Least Square Error model parameter refinement method with manually collected GCPs and a DEM. Instead of refining the model parameters directly, the second scheme establishes polynomial functions to relate the inversely estimated SAR image coordinates and the real image coordinates of GCPs, where the geo-location model is just constructed with the initial parameters derived from the SAR imagery product meta data. The technique details to implement these two schemes were presented and their validity and geo-location accuracy were analyzed using numerical analysis method based on one simulated zero-error geo-location model and the GCPs system. It is proved that both of the two ortho-rectification methods are valid and of high geolocation accuracy. It is time consuming and inefficient to perform the map to image transformation from every map pixel with the refined geo-location model. Therefore, one...