Estimating the travel distance of channelized rock avalanches using genetic programming and support vector machine

Abstract

Channelized rock avalanche travel distance (CRATD) is one of key parameters in disaster risk analysis. Although traditional regression analysis methods is widely used in estimating CRATD, there is lack of studies on whether there is a room for further improvement. In this study, 34 channelized rock avalanche events triggered by Wenchuan earthquake in Fujiang River Basin were assembled to develop a robust model for estimating CRATD using two machine learning methods (Genetic Programming (GP) and Support Vector Machine (SVM)) and a widely accepted traditional regression analysis method (Power Form model (PFM)). It was found that GP model performed best among the three methods when the influence of source area, height difference between the head scarp crown and the base of the collapsed slope, average inclination angle of the source zone, and average slope angle of the travel path on the travel distance were considered in GP model. The proposed GP model was verified and compared against six previous models using 15 channelized rock avalanche events induced by Wenchuan earthquake in Tuojiang River Basin. The proposed GP model shows significant improvement in estimating CRATD. In view of the limited number of channelized rock avalanche events, the application range of the proposed GP model is suggested. In conclusion, the proposed GP model could play a beneficial role in related disaster prevention and land management.