Analysis Of Causes And Critical Threshold Of Landslide Hazards In Sichuan Province Based On SPARK

Landslides are the most common geological disasters in Sichuan Province.Every year,landslides cause a lot of casualties and property damage.In order to effectively prevent the landslide,it is necessary to analyze the cause of the landslide,predict the possibility of the landslide based on its characteristics,and study the critical rainfall that triggers landslides.The causes of landslides are very complicated.They are caused by internal factors such as topography and geology,as well as external factors such as rainfall and earthquakes.Current research on critical rainfall that triggers landslides is not good enough,most of them focus on specific areas,the research results are not practical.Therefore,this thesis analyzes the causes of landslides based on Sichuan landslide disaster information over the years,builds a landslide probability prediction model,and improves the traditional rainfall threshold research method based on recommendation system.The main work of this thesis is as follows:(1)Based on the landslide disaster information from 2009 to 2017,various factors leading to the occurrence of landslides were extracted.The factors extracted in this thesis include a total of 13 factors including elevation,slope,slope position,terrain fluctuations,lithology,seismic strength,land coverage,rainfall,etc.,which comprehensively characterize the characteristics of the landslide.(2)The distribution of various factors that lead to the formation of landslides is analyzed,and the importance of factors is analyzed using F-test,chi-square test,and mutual information,and the less important impact factors are eliminated according to the analysis results.(3)On the SPARK platform,the landslide probability prediction model was implemented based on the random forest and neural network.,and fused the model using Blending fusion algorithm.The performance of the model was evaluated using multiple indicators such as TPR,FPR,ROC curve,and AUC.(4)The rainfall threshold for landslide initiation in Wangcang County,Linshui County and Huidong County,which are distributed in different areas of Sichuan,were fitted using traditional critical threshold research methods.Based on the recommendation system,the traditional research methods are improved,and the problems of inaccurate threshold estimation and poor scalability are solved.In order tomake the recommendation system more practical,its architecture is implemented in parallel based on the SPARK platform.The conclusions of this thesis are as follows:(1)The leading factors of landslides in Sichuan Province are rainfall and human activities.Among them,rainfall is the most important influencing factor,and landslides in Sichuan are caused by the combined effects of previous rainfall and critical rainfall.(2)The fusion model has better performance than the primary model,and it is more effective in predicting the probability of landslide occurrence.When the judgment threshold is set to 0.8,the model's TPR reaches 86%,and the FPR is less than 2%.(3)The performance of the landslide rainfall threshold research method based on the recommendation system is superior to traditional methods.The method based on the recommendation system can better estimate the critical threshold for landslides with different regions,and the prediction error is within 10 mm.