Spatial Multi-scale Probability Prediction Of Typhoon Rainfall Landslides

The hills and mountains in the southeast of Fujian Province have been affected by typhoons and their consequent heavy rainfall for a long time,leading to frequent geological disasters such as typhoon rainfall landslides.Due to the fragile geological environment in the hills and mountains of southeast of Fujian and the complicated mechanism of typhoon rainfall landslides,the prediction and early warning of typhoon rainfall type landslides still need to be improved continuously.Therefore,it is of great theoretical and engineering significance to carry out statistics and analysis based on the big data of landslide disasters and establish the probability prediction model of typhoon rainfall landslides.Taking typhoon rainfall landslides in Anxi County as the research object,we analyze and summarize the distribution characteristics,development rules and genetic mechanism of this kind of landslide in the study area,and carry out landslide susceptibility evaluation and rainfall-induced landslide probability calculation.On this basis,the incremental Bayesian probability prediction mathematical model of the landslide induced by typhoons and rainstorms is derived by using Bayesian theory and incremental learning theory.The main results of this dissertation are as follows:(1)The landslides in the study area mainly affected by typhoons rainstorms.Restricted by rainfall and geological environment conditions,the distribution of landslides in the area is obviously affected by time and space.In terms of time,they mainly occur in the year and the month when typhoons and rainstorms frequently occur,and most of them occur from May to September;in terms of space,there are more geological disasters in the north-central part than the southern part due to the erosion and denudation of the hilly landscape and structural belt.Most of this type of landslides have a dual structure of slopes,with well-developed pipe network seepage systems,and most of them are induced by heavy rainfall.(2)The concept of spatial multi-scale is applied to the evaluation of landslide susceptibility,and the study area isdivided into three scales: Anxi County,Xiao lanxi watershed and individual landslide,and the evaluations are conducted respectively.Logical regression model and information content model are selected for large-scale Anxi County.Information content model is selected for Xiao lanxi watershed of slope type;The evaluation model of individual landslide is established based on the principle of information entropy.The results show that:(1)Both logical regression model and information content model are suitable for large-scale landslide susceptibility evaluation,but the former has higher accuracy and slightly better prediction ability than the latter.(2)The susceptibility zoning results of the information content model at local scale are better than those at regional scale.(3)Entropy weight method is suitable for individual scale landslide susceptibility evaluation,and the evaluation results are consistent with the actual landslide investigation.(3)The relationship between typhoon rainstorm type landslides and rainfall in the study area is analyzed,and three rainfall factors,namely,the rainfall on the day of the landslide,the daily rainfall of the 10 days before the landslide,and the cumulative effective rainfall in the early stage,are selected.Also,the time probability prediction model of the landslide induced by typhoon rainfall is established by using logical regression model.The accuracy of the model prediction is verified,and the results show that the model prediction effect is good and the accuracy is high,which can better predict the occurrence time of the landslide.(4)Based on the probability value of landslide susceptibility evaluation and rainfall-induced probability value,the Gaussian Bayesian network structure is constructed.According to Bayesian theory and incremental learning theory,the continuous variable incremental Bayesian classification prediction probability model of typhoon rainstorm type landslides is deduced,and the probability prediction system of typhoon rainstorm type landslides is built.Select some typhoon rainstorm type landslides in the study area as the samples to be predicted to test the learning performance and prediction accuracy of the model.The results show that the overall prediction accuracy of the model is high,and the combined prediction of disaster time and space can be achieved well,which can provide a basis for the subsequent prevention and treatment of this type of landslide.The results show that the overall prediction accuracy of the model is relatively high,and it can better realize the prediction of disaster spatio-temporal joint forecast which can provide a basis for the follow-up prevention and treatment of this type of landslides.