Landslide Risk Assessment along Roads by Using Radar-driven Land Deformation and Rainfall Data

Abstract

To prevent damage from landslide disasters, traffic regulation based on records is implemented before disasters occur in Japan. Logistics and accordingly economic activities are halted once the traffic regulation is implemented. There are problems that the operation of the traffic regulation tends to be redundant in terms of temporal duration and spatial coverage. In this paper, to consider the effect of topography and land deformation and resolve the problems of redundant traffic regulation, we attempted to predict the land deformation using spatio-temporal statistical models whose objective variable was deformation estimated PSInSAR and explanatory variables were accumulated rainfall and maximum gradient angle. Three statistical models: low-rank GP model, separable covariance model, and product-sum covariance model were used. According to the results of experiments, three spatio-temporal models showed similar predictions; relatively small deformations were well fitted while relatively large deformations were poorly fitted. Since land deformation due to landslides is relatively large, it should be considered the measures to improve the prediction of larger deformations.