Application Of Logical Regression Model In Assessment Of Landslide Hazard Susceptibility

Landslide geo-hazards occur frequently in the south of China,which have the characteristics of short duration and large damage degree,and seriously threaten the safety of human life and property.How to evaluate the susceptibility of such geo-hazard and take prevention measure to reduce the losses has become an important topic in the disaster reduction and prevention research.This paper,taking Ganzhou including Nankang and Zhanggong Districts in southwest Jiangxi as an example,based on data from the field survey historical landslide development types and their distribution characteristics were investigated and a number of landslide hazard-causative geo-environmental factors were selected to compose the hazard assessment indicator system.The relationship between the impact factors and historical landslides was analyzed by data-driven statistical method,and the multiple logistic regression(MLR)analysis was employed to model the landslide hazard susceptibility.The outcomes of this research are shown as follows:(1)Landslide is mainly affected by both human activity and other induced factors.Most of the landslides in the region are distributed along with engineering activities,and the landslide bodies occur in the strongly weathered crust and Quaternary slope deposits,and heavy and continuous rainfall is the triggering factor of landslide hazards.(2)In terms of the occurrence of landslides in the study area and the conclusions of other authors,13 geo-environmental factors,including elevation,slope,aspect,distance from roads,distance from faults,lithology,NDVI,land use type,average annual rainfall,etc.,were employed as landslide-causative factors to compose the hazard assessment indicator system.(3)Two logistic regression models(MLR-1 and MLR-2)were built based on two data-driven statistical analysis methods,namely random sampling(RP)and certainty factor(CF),to assess the susceptibility of landslide hazards.According to the model verification results,the accuracy of the MLR-1 model is higher than that of the MLR-2 model.The proportion of the historical landslides in the areas with high and extremely high susceptibility accounts for 88.71%.19 newly occurred landslide hazards after 2017 were also used for case verification and the proportion of landslides in the areas with high and extremely high susceptibility takes up 84.21%.This means that the prediction by the MLR-1 model is reliable.The areas with high susceptibility of landslide hazard are mainly distributed along the roads,indicating that these landslides are greatly affected by human activities.This is consistent with the field survey.(4)Regarding sampling or processing approaches,different sampling will lead to different results even though the same modeling algorithm is used.An appropriate disaster susceptibility evaluation model is the premise of realizing reasonable disaster zoning and evaluation.The results show that the MLR-1 model looks more suitable than MLR-2 for zonation of landslide hazard susceptibility in Ganzhou area,and this method can be extended to other areas with similar geological and environmental conditions.