摘要
本文以湖北省远安县为研究区,利用采集的资料,提取出了与滑坡发生相关的8类指标因子:高程、坡度、坡向、地层岩性、斜坡结构、断层、水系、公路。针对连续型致灾因子,选取定性等间距划分和频率比法划分得到两类指标因子体系,分别带入人工神经网络模型和随机森林模型,绘制得研究区易发性评价区划图。最后,利用ROC曲线图对4个模型的精确性进行分析,得到ANN模型的成功率和预测率分别为0.899和0.901,FR-ANN模型的成功率和预测率0.934和0.935; RF模型的成功率和预测率分别为0.886和0.886,FR-RF模型的成功率和预测率分别为0.928和0.929。以上说明,无论对于人工神经网络还是随机森林模型,基于频率比法的因子分级均表现出了更高的精确性。
Taking the Yuan’an County of Hubei Province as the research area,we extract 8 types of index factors related to landslide occurrence.These factors are elevation,slope,aspect,stratum lithology,slope structure,faults,water systems,and highways.For the continuous hazard factors,two types of index factor systems were selected by the qualitative equal interval division and the frequency ratio method.The two systems were brought into the artificial neural network model and the random forest model,respectively,and the susceptibility evaluation maps in the study area were obtained.Finally,the accuracy of the four models was analyzed by using ROC graphs.The success rate and predictive rate of the ANN model were 0.899 and0.901,the FR-ANN model were 0.934 and 0.935,the RF model were 0.886 and 0.886,and the FR-RF model were 0.928 and 0.929,respectively.The result shows that both the artificial neural network and the random forest model exhibit higher accuracy based on factor grading of the frequency ratio method.
作者
闫举生
谭建民
YAN Jusheng;TAN Jianmin(Wuhan Center of China Geological Survey,Wuhan,Hubei 430205,China)
出处
《中国地质灾害与防治学报》
CSCD
2019年第1期52-60,共9页
The Chinese Journal of Geological Hazard and Control
基金
中国地质调查局项目(1212011014020)
关键词
滑坡易发性
频率比
人工神经网络
随机森林
landslide susceptibility
frequency ratio
artificial neural network
random forest
