Analysis And Assessment Of Seismic Landslide Suspectibility Based On Newmark And DNN Combined Model

In recent years,seismic landslide risk assessment has attracted the attention of scholars.The exploration of reasonable and effective evaluation methods has important guiding significance for disaster prevention.This article takes 4834 landslides caused by the M_s7.0earthquake in Jiu Zhaigou as an example.First,the Newmark method and DNN model are used to evaluate the landslide hazards in this area and compared.Second,The limitation of traditional sampling method has been study and N-m sampling based on the Newmark model was proposed.Third,the method be used selecting the influence factors was proposed in combination with the frequency ratio model and analyze the factors extracted by the Newmark model.Finally,we present the N?DNN model combined by DNN and Newmark model and choose SVM model to compare,the results as following:(1)In the Newmark model,we use 1:20w geological maps to divide lithology and assign parameters to rock and soil.Then,the module of seismic landslide hazard assessment is produced based GIS.In the DNN model,we select seven factors to evaluation seismic landslide susceptibility,include slope angle,elevation,slope aspect,river,faults,and PGA.Finally,the landslide susceptibility index map and landslide susceptibility rank map was done.(2)This paper analyzes the limitations of the traditional sampling method and proposes N-m sampling based on the Newmark model.then,the two method has be compared.The research shows that the N-m sampling method is better than the traditional sampling method.(3)By studying the relationship between the influence factor dispersion and model performance,a new method combined the frequency ratio model was proposed to select influence factors.The results show that the method can effectively reflect the contribution of influence factors to model performance.Then,we using this method to analyze the impact factors extracted from the Newmark model,the results show that Dn has the highest contribution to the model performance.(4)After sampling and factor optimization,we combined the Newmark model and the DNN model to propose the N?DNN model,and selected the SVM model for comparison.the AUC value of the N?DNN model was 0.98,the accuracy rate was 0.93,and the recall rate was 0.91,all higher than the SVM,which shows that the N?DNN model is better than the SVM model.The research in this paper provides a new idea for the risk assessment of earthquake landslides.