Snow Avalanche Susceptibility Assessment Based On Ensemble Machine Learning Model In The Central Shaluli Mountain

The central part of the Shaluli Mountains is located in Ganzi,Sichuan,bordered by the Jinsha River and adjacent to Tibet.The natural environment is complex,the mountains are steep,and the avalanche develops,which poses a serious threat to human activities and engineering construction.Therefore,the evaluation of avalanche susceptibility in this area can not only help define the spatial pattern of avalanches on the Qinghai-Tibet Plateau,but also provide references for the prevention and early warning and emergency management of regional avalanche disasters.In this study,among 14 avalanche evaluation factors were selected through remote sensing interpretation supplemented by refined field surveys,GIS spatial analysis,and data mining.The two statistical models of Evidence Confidence Function(EBF)and Certainty Coefficient(CF)were used to combine the two methods.A machine learning model—logistic regression(LR)and multi-layer perceptron(MLP)to establish four integrated models of EBF-LR,CF-LR,EBF-MLP and CF-MLP,as well as the traditional frequency ratio model(FR)for avalanche easy developmental evaluation research.The main research conclusions are as follows.(1)Remote sensing interpretation characteristics and development characteristics of avalanches:Based on medium and high resolution multi-source remote sensing images,through visual interpretation combined with field survey verification methods,avalanche remote sensing interpretation signs in the study area were established,and a total of 536avalanches were obtained.The total area of avalanches is 114.79km~2,and the development density is about 0.075 per square kilometer.The relative height difference of avalanches is concentrated in the range of 300?600m;the dominant slope for avalanche development was20?35°;the slope direction of the study area was N-NE Avalanches are relatively developed,and avalanches with a slope of SE-S are relatively less developed;the snow area in the avalanche formation area was concentrated in the interval of 1×10~4m~2?5×10~4m~2,and the avalanche path length was the most widely distributed in the interval of 500?1000m.The development of avalanches in the study area had obvious spatial differentiation characteristics,showed a certain trend and unevenness.(2)Analysis and data extraction of the influencing factors:Based on GIS software,vectorized avalanche influencing factors and established a factor space layer,extracted a total of 18 influencing factors such as terrain factors,climatic factors,snow feature factors and other influencing factors,Using Person correlation analysis and variance expansion factor for multicollinearity analysis,and finally screened out altitude,slope,aspect,ground curvature,surface roughness,surface cutting,TWI,NDVI,water system,faults,roads,annual snowfall A total of 14 evaluation factors including the amount of snow,the maximum annual snow depth and the average temperature in January were finally supported by the basic environmental data of the avalanche susceptibility evaluation in the study area.(3)Evaluation of avalanche susceptibility:A total of five evaluation models including FR,EBF-LR,CF-LR,EBF-MLP and CF-MLP were constructed to evaluate the avalanche susceptibility of the central part of the Shaluli Mountains,and the results were studied.Evaluation index value range of avalanche susceptibility in the area:FR model[1.877,51.681],EBF-LR model[0.001,0.999],CF-LR model[0,0.999],EBF-MLP model[0.067,0.945],CF-MLP model[0,0.988].The accuracy of the results is tested by Kappa coefficient and ROC curve,and the results showed that these five models had appraisal value.Among them,the CF-MLP(Kappa=0.608,AUC=0.910)model is the best avalanche susceptibility evaluation model in this study.The accuracy of the FR(Kappa=0.584,AUC=0.894)model is second,and the CF algorithm and machine learning model combined performance of EBF was better than EBF.The most important influencing factors were altitude,aspect,topographic humidity index,and average temperature in January.In the process of susceptibility evaluation involving the three key factors of terrain,climate and snow cover,the sensitivity of the five models to terrain factors was highly consistent.The zoning map of avalanche susceptibility based on the CF-MLP model was obtained through the natural breakpoint method.The areas with very high susceptibility and high susceptibility accounted for about 10.01%and 15.33%of the total area respectively,mainly concentrated in Lingchang,Litang County.Around the mountain peaks at an altitude of about 5000m,such as the Gnei Mountains to the south of the village,and the alpine mountains on both sides of the line from Chaluo Township to Gaiyu Town in Batang County.The avalanche has little impact on the existing G318 and G215 national highways.The railway under construction mainly crosses the study area from east to northwest in the form of tunnels.The elevation is slightly lower than the current G318 national highway,and the impact of avalanches is smaller than that of the G318 national highway.However,the Yueling road section from Bomi Township to Batang County in the south of the study area,and Gaiyu to Shanyan Township in the north is mainly located in avalanche-prone areas.The land planning and construction of these areas and the life safety of residents and tourists need special attention.