摘要
Abstract
Accurate prediction of landslides is a critical aspect of geological hazard prevention and mitigation.In this study,a landslide susceptibility assessment approach was developed through the quantification of evaluation factors and the application of a coupled modeling framework.The results show that the accuracy of landslide susceptibility evaluation of the three models(random forest model(RF),convolutional neural network model(CNN),and coupled model(RF-CNN))is 78.4%,83.6%,and 86.4%,respectively.The study demonstrates the following key findings:1)The use of a hybrid modeling approach significantly improved the accuracy of landslide susceptibility assessment.Specifically,the RF-CNN model outperformed the standalone RF and CNN models,with accuracy improvements of 8.0%and 2.8%,respectively.2)Continuous evaluation factors were categorized using the frequency ratio model.Subsequently,variable selection was performed based on the Spearman rank correlation coefficient,tolerance and variance inflation factor(VIF),and contribution rate of each factor.This process led to the elimination of less influential variables and resulted in a 4.3%increase in model accuracy.3)Areas classified as very high and high susceptibility zones were primarily located in the northwestern and southern parts of the study area.These results are positively correlated with key contributing factors such as surface deformation rate,topographic relief,and slope gradient.关键词
滑坡灾害/易发性评价/频率比模型/耦合模型Key words
landslide hazard/susceptibility assessment/frequency ratio model/coupled model分类
天文与地球科学
