Shallow landslide susceptibility assessment in Jinsha River Basin based on machine learning models
[Objective]The hazard-pregnant environment in Jinsha River Basin is complicated,and shallow landslides occur fre-quently,which seriously threatens the safety and property of local residents and the operation and maintenance of infrastructure.It is urgent to construct a suitable and accurate regional shallow landslide susceptibility map to guide the layout of disaster preven-tion measures and the planning of infrastructure construction.[Methods]Compared with the logistic regression model,two ma-chine learning models(i.e.,the gradient boosting decision tree model and random forest model)were selected to evaluate the susceptibility of shallow landslides in the Zhaotong City of Jinsha River basin.Based on 2 369 historical landslides,14 landslide-related factors such as slope,slope direction,geomorphic type,soil type,distance to rivers,distance to roads,NDVI,seismic intensity and annual rainfall were selected to construct three shallow landslide susceptibility evaluation models of the Zhaotong City.[Results]The result showed that(1)the AUC values of the three models were all larger than 0.800,and the performance of the two machine learning models were better than that of the logistic regression model;(2)the accuracy of the random forest model was the highest,and its AUC value and Kappa coefficient were 0.910 and 0.907,respectively.The high-prone and ex-tremely high-prone areas identified in various regions were highly consistent with the actual landslide distribution,and the over-fitting phenomenon was weak;(3)the relative importance of each evaluation index in the random forest model was well reflected in such areas with complicated and diverse disaster-pregnant mechanisms,and all kinds of disaster-causing environments were more comprehensively considered in the random forest model.[Conclusion]The result indicated that the machine learning mod-els can better evaluate the susceptibility of shallow landslides in the Jinsha River Basin under complex disaster-pregnant environ-ments,which is helpful to guide the disaster prevention and mitigation practice in this area.
万方数据
维普
