Procedure of Data Processing for the Improvement of Failure Time Prediction of a Landslide Based on the Velocity and Acceleration of the Displacement

Abstract

AbstractTime prediction methods based on monitoring surface displacement (SD) are effective for early warning against shallow landslides. However, failure time prediction by Fukuzono’s original inverse-velocity (INV) method is less accurate due to variation in the inverse-velocity (1/v) caused by noise in the measured SD, which amplifies the fluctuation in the resultant 1/v. Therefore, the present study incorporates pre-analysis to acquire better prediction by reducing the effect of noise on the measured SD. The data extraction (DE) and moving average (MA) methods are used to filter the measured SD for better smoothing of 1/v. The root mean square error (RMSE) and determining factor (f) values are used to select the optimum SD interval (Δx) in the DE method. The RMSE and f values are used to evaluate the reproducibility of the measured data and the scattering in the relationship between velocity and acceleration in an orderly. The data, treated by the DE and MA methods, are utilized to predict the failure time based on the INV method and the relationship between velocity and acceleration on a logarithmic scale (VAA) method. Accordingly, Δx gives the smallest sum of the normalized RMSE and normalized (1-f), which offers a better prediction. When the SD at failure changes, Δx is changed. The best prediction is obtained by DE preprocessing with the VAA method because it minimizes the effect of the individual 1/v by reducing the scatter in the relationship between velocity and acceleration. However, the time prediction using data processed by the MA method shows poor prediction due to some scattering of the inverse velocity. In some cases, the prediction by the VAA method using MA data provides better prediction than the results of the INV method.