Debris Flow Forecasting of Northwest of Yunnan Province Based on LR, SVM, and RF Statistical Models

Abstract

Debris flow forecasting is of great significance for it could seriously endanger people's life and property safety. This paper considered the northwestern part of Yunnan Province as research area, and took the elevation, slope, rainfall, landform, evapotranspiration and NDVI (normalized difference vegetation index) as influential factors. Followed by two accuracy indicators TPR and FPR, best models of Logistic Regression (LR), Support Vector Machine (SVM), and Random Forest (RF) were established. Then a same test set were used to compare the effectiveness of them. The result shows that as a combined classifier, RF performed the best with TPR is 83.10% and FPR is 0.48%, SVM took second place with TPR is 74.99% and FPR is 1.98%, and LR is inclined to predict occurrence, causing its high FPR 22.71%. The LR, SVM and RF models built in this paper are quite effective and provide a theoretical base for prevention and reduction of debris flow. Additionally, 41 mud sensors data distributed in this region were collected, based on which the debris flow probability of these area were obtained by LR model to explore the effect of mud on debris flow. Experiments find that in some basins, mud has a positive impact on debris flow, and in the remain basins, mud may be slightly influenced by rainfall and thus cause a negative effect on debris flow.