Geometric Calibration of SAR Images to Eliminate Earth's Surface Topography Distortions

Geometric calibration and georeferencing are the most important processes of SAR raw images. Geometric distortions caused by platform instabilities, error in determining the relative height and displacements origin from topography. The prominent errors for the SAR imaging geometry, and target height changes are known as foreshortening and layover. Therefore, in this article our studies were focused on this problem. In order to correct these errors, an independent source of information was required such as imaging from another angle, topographic map or DEM. In this paper, a method for geometric calibration of SAR images is proposed. The method uses Range-Doppler (RD) equations and to implement the method used in this article, two SAR datasets are tested with RD modelling. These datasets are acquired by ALOS PALSAR spaceborne SAR sensor. Test areas covered by these datasets range from flat plains to mountainous areas, which the first dataset located in the border between United States and Mexico and the second one is in Iran. In this method, for the image georeferencing, the appropriate Digital Elevation Model (DEM) and also exact ephemeris data of the sensor is required. In the algorithm proposed in this paper, first digital elevation model transmit to range and azimuth direction. By applying this process, errors caused by topography such as foreshortening is removed in the transferred DEM. Then, the original image is registered to transfer DEM by transformation equations. The output is a georeferenced image without geometric distortions. The advantage of the method described in this article is in eliminating the requirement for any control point as well as the need for attitude and rotational parameters of the sensor. Furthermore, two experiments with different settings are designed and conducted to comprehensively evaluate the accuracy of the SAR georeferencing with RD model. Few experiments are done in this study for various purposes. The first one is to find the best transformation equation among the three types for registering images. In the first experiment the efficacy of three types of transformation equations on georeferencing of ALOS PALSAR images were evaluated with identified check points. To evaluate the accuracy of the georeferenced images, 25 check points in different parts of the image was selected. By comparing the obtained coordinates in georeferenced image and reference points in Google Earth, the RMSE was calculated for these points. In best situation, the planimetry accuracy were 20.11m for dataset A and 19.94m for dataset B and the altimetry accuracy were 30.28m for dataset A and 30.71m for dataset B. Since the ground resolution of multi-look image was 30 meters, the planimetry accuracy achieved in this research is acceptable. The other experiment is to compare the georeferenced SAR images generated from three DEMs to demonstrate the effectiveness of DEM spatial resolution on the accuracy of georeferencing SAR images. In addition we investigated the suitability of three typical DEM datasets for SAR georeferencing in RD model. The experimental results show that the best transferred DEM was obtained from the ASTER DEM of spatial resolution comparable to that of ALOS PALSAR images.