GIS-based Logistic Regression Nodel For Rainfall-induced Landslide Susceptibility Assessment

China is a country prone to landslide hazards.In recent years,frequent earthquakes,extreme rainfall conditions and nonstandard engineering activities have aggravated the occurrence of landslide hazards.Construction planning in mountainous areas,route selection and the increasing need of social safety have put forward new demands on disaster prevention while regional landslide assessment has become an important method for landslide prevention in the world.Yueqing is one of the most serious areas in China that suffered from rainfall-induced landslide hazards,and the landslides caused by rainfall account for the most of geological disasters in Yueqing every year,which has caused great loss of life and property to Yueqing,and also threaten the social stability of Yueqing.In this paper,the key theory and technology of landslide susceptibility assessment were discussed firstly,then the key problems in GIS-based regional landslide susceptibility assessment were summarized and Logistic regression model was introduced.Then statistical analysis was carried out to select the landslide controlling factors for Yueqing area.After that,the influence of cell size on landslide susceptibility assessment was investigated by comparing the assessment results of 5 m,10 m,15 m,20 m,25 m and 30 m cell size.Additionally,categorical and continuous variables were applied to assess landslide susceptibility respectively to explore the influence of variable type on landslide susceptibility assessment.Moreover,the importance of different controlling factors to landslide in Yueqing was analyzed based on single factor regression model.The main research contents and conclusions are summarized as follows:(1)Time and spatial distribution laws of landslide in study area were summarized.Based on the statistical analysis on environmental factors and triggering factors of study area,8 factors such as elevation,slope,aspect,lithology,rainfall,NDVI,distance to rivers and roads were selected as landslide controlling factors.(2)Using Logistic regression model,landslide susceptibility were assessed with categorical variables under different cell size such as 5 m,10 m,15 m,20 m,25 m and 30m to investigate the influence of cell size on landslide susceptibility assessment.The results were validated by ROC curve and cross-validation method respectively and validation results shown that cell size has an certain influence on landslide susceptibility assessment.With the increase of cell size,the accuracy of results increase firstly and then have a significant decrease.20 m cell size is the best cell size for this study and the order for assessment accuracy from high to low is 20 m>15 m and 25 m>5 m and 30 m>10 m.(3)Elevation,slope,rainfall and NDVI were set as categorical variables and continuous variables respectively to investigate the influence of variable type on landslide susceptibility assessment.The results show that variable type has an significant influence on landslide susceptibility assessment and Logistic regression model with categorical variables performed better than its with continuous variables.Also,variable type play more important role than cell size when the gaps between different cell sizes are insignificant.(4)10 landslide assessment samples were set for each regression situation and their results were compared to explore the influence of random sampling method on landslide susceptibility assessment.The results show that random sampling is a suitable landslide sampling method and repeated tests can decrease the random errors.(5)Single factor regression method was applied to explore the importance of each landslide controlling factor to landslide by comparing the AUC of each regression model.The results show that rainfall is the most important factor to landslide occurrence in Yueqing while distance to rivers and roads play unimportant role to it.Besides,it is an applicable way to replace cumulated precipitation by annual precipitation when assessing rainfall-induced landslide susceptibility if accurate rainfall data such as daily precipitation or hour precipitation are unavailable.