Study On Landslide Risk Warning Modeling Based On Sensitivity And Critical Rainfall Threshold

Landslide is defined as the rock mass or soil mass on the slope,which is affected by groundwater activity,rainwater immersion,river erosion,earthquake or engineering cutting slope,etc.,and slides down the slope as a whole or separately along a certain weak surface or weak zone under the action of the slope body's own gravity.Once the landslide disaster occurs,it will not only directly threaten the safety of people's life and property,but also indirectly lead to land planning,production and life,and the use of environmental resources and other far-reaching issues.In recent years,the machine learning model of landslide vulnerability evaluation has made remarkable progress,but it is still necessary to adopt more advanced machine learning model to improve the accuracy of landslide vulnerability.At the same time,the existing critical rainfall threshold method for predicting the time probability of rainfall-type landslide is mainly based on cumulative rainfall-duration or rainfall-intension-duration,and it often fails to consider the effect of the early effective rainfall on the slope stability,so the calculation accuracy of critical rainfall threshold needs to be improved.In addition,the time probability of landslide obtained by different critical rainfall threshold methods is different,which brings great uncertainty to the calculation of critical rainfall threshold.In view of the problems existing in the above risk warning,this paper takes the typical mountainous and hilly area in China--Xunwu county,Jiangxi province as the research area to carry out the landslide risk warning based on the vulnerability and critical rainfall threshold.The specific research contents and research results of this paper are as follows:(1)the data of 325 rain-type landslides in the study area in the past 40 years were obtained,and the distribution of topography,geomorphology,geological conditions,hydrological environment,soil vegetation and human engineering activities in the study area were investigated and analyzed.The environmental influencing factors of landslide development in the study area were analyzed by combining with the historical landslide catalogue information of the study area.(2)based on the environmental impact factors of the landslide in the study area,selection of topography factor,geological factor,hydrological environment factor,surface coverage factor of these four kinds of basic environmental factor 10 landslide prone evaluation factors,including altitude,slope,slope to curvature,plane,section curvature,relief,distance drainage,formation lithology,NDVI(normalized difference vegetation index),NDBI(normalized building index),through the study of the classification of various environmental factors or classification,statistical classification or frequency ratio to analyze the landslide of the relationship between the development and evaluation of the selected factors.(3)the data in the research area were normalized,and the training test data were divided in a ratio of 7:3.Three supervised machine learning models,namely support vector machine,multi-layer perceptron and random forest,were selected to evaluate the landslide vulnerability in the research area,and the grading diagram of landslide vulnerability evaluation of the corresponding model was obtained.ROC curve,statistical accuracy and comparison of classification results were used to compare the modeling effects of the three models,and the results showed that the random forest model could more accurately evaluate the landslide susceptibility in the study area.(4)according to the daily rainfall data and the record date of the landslide recordedby the meteorological bureau of Xunwu county over the years,the rainfall data of the day and 15 days before the landslide were obtained statistically;By analyzing the correlation between rainfall,rainfall duration and landslide events,the effective rainfall coefficient in the early period,which is used to calculate the effective rainfall intensity in the early period,and the effective rainfall intensity in the early period,which is in line with the actual situation of the study area,is obtained.On the basis of the effective rainfall intensity in the early stage,the models of cumulative effective rainfall-duration,effective rainfall intensity-duration and cumulative effective rainfall-intensity were calculated based on the rainfall duration of the landslide event,and the modeling results of the three threshold models were compared and analyzed.The results show that the effective rainfall intensity-diachronic model can better reflect the time probability of the occurrence of rainfall landslide in the study area.(5)the landslide susceptibility(reflecting the spatial probability of landslide occurrence)and the time probability of rainfall-induced landslide in Xunwu county were coupled to realize the spatial-temporal joint warning of regional landslides.In this paper,10 typical landslide cases in the study area are analyzed,and the warning classification chart and corresponding warning measures are obtained.The results show that the warning classification results are consistent with the actual situation.