Accuracy Evaluation and Analysis of a Localized Geometric Transformation Approach for Two Dimensional Landslide Displacement Measurements Using IRS P5 Satellite Images

Determination of the displacement vectors due to the landslide phenomenon is the initial stage for generating landslide inventory and susceptibility maps. The main objective of this paper is to evaluate the achievable accuracy of a simple two-dimensional localized geometric transformation method for the determination of the landslide displacement vectors. For this purpose, two IRS P5 backward images taken in two different revisit epochs over the Ardabil Province are used. The main objective of this paper is to compare the acquired image based displacement vectors with the ground measurements. The 2D localized transformation approach may be utilized to produce a small and medium scale country-wide landslide inventory maps and for continuous monitoring of the areas susceptible to large landslides. With regard to the fact that the coverage of the landslide zones in satellite imageries are usually small as compared with the entire image frame, the achieved accuracy figures suggest that the localized transformation approach may fulfill the required demands. Onsite displacement measurements performed by the GPS receivers, conducted by the ISTA SANJ DAGHIGH Co, were utilized to assess the accuracy of the localized transformation method. To compare the displacement vectors generated from the images with the ground observations, spatial registration between the points measured on the image and the points measured on the ground was conducted using the satellite supplied RPCs through which the ground coordinates for the measured image points are calculated. However, due to the fact that the supplied RPCs have a systematic shift error, to generate the ground coordinates for the measured image points, it is necessary to eliminate the systematic shift of the RPCs. This is achieved by identifying fixed feature points (outside the landslide zone) on the stereo P5 images for which ground observation were available through the large scale map of the area. The RPC shift error is then calculated by comparing the ground coordinates of the fixed points generated by the RPCs and their corresponding ground coordinates extracted from the large scale map. On the other hand, due to the lack of coincidence between the dates of the image acquisition and the ground observations, a temporal interpolation was also necessary. The temporal registration was performed using a linear interpolation approach assuming a linear land displacement. The average landslide speed on the ground was adopted as the amount of linear displacement trend for the period over which the assessment was conducted. The outline of the evaluation approach and a detailed description of the test results and the final accuracy figures are presented in this paper.