Gauss Process Based Approach for Application on Landslide Displacement Analysis and Prediction

International audience In this paper, the Gauss process is proposed for application on landslide displacement analysis and prediction with dynamic crossing validation. The prediction problem using noisy observations is first introduced. Then the Gauss process method is proposed for modeling non-stationary series of landslide displacements based on its ability to model noisy data. The monitoring displacement series of the New Wolong Temple Landslide is comparatively studied with other methods as an instance to implement the strategy of the Gauss process for predicting landslide displacement. The dynamic crossing validation method is adopted to manage the displacement series so as to give more precise predictions. Different covariance functions are illustrated to give predictive results which show that different covariance functions result in varying levels of prediction accuracy. Comparisons with other methods are also discussed in this study. The results show that the Gauss process can perform better than the RBF network and the SVM methods in this problem in view of the trends according to the original data. Finally, the landslide criterion is given for creep-typed slopes that landslide event would occur imminently if the cross angle at the intersection point of displacement curve changes more than 45 .