基于频率比联接法和支持向量机的滑坡易发性预测

Landslide Susceptibility Prediction Based on Frequency Ratio Analysis and Support Vector Machine

区域滑坡易发性制图对滑坡灾害的防治非常有意义。以江西省上犹县滑坡为例,首先基于遥感(remote sensing,RS)和地理信息系统(geographic information system,GIS)平台获取11个滑坡评价因子;进一步利用频率比(frequency ratio,FR)联接方法和支持向量机(support vector machine,SVM)模型耦合出FR-SVM模型进行滑坡易发性预测,并对结果进行易发性分级;同时建立以原始评价因子作为模型输入变量的单独SVM模型,再次对上犹县进行滑坡易发性预测制图;最后通过受试者特征工作曲线下的面积(area under receiver operating characteristic curve,AUC)曲线开展FR-SVM和单独SVM建模工况下的精度验证分析。结果表明:FR-SVM模型对于区域滑坡易发性制图具有比单独SVM模型更好的预测性能。FR-SVM和单独SVM模型的AUC值分别为0.893和0.798,进一步表明FR-SVM模型在描述滑坡易发性指数分布及评价因子对滑坡发育影响特征方面要优于单独SVM模型。

Landslide susceptibility prediction is a meaningful method for landslides spatial prediction.The Shangyou County,Jiangxi Province of China was selected as a case study.Firstly,11 landslide evaluation factors based on remote sensing(RS)and geographic information system(GIS)platform were obtained.Furthermore,the frequency ratio method and support vector machine(SVM)model were coupled as FR-SVM model to predict the landslide susceptibility.Meanwhile,the single SVM model with original evaluation factors used for inputs was used to evaluate the landslide susceptibility again.Finally,through receiver operation characteristics(ROC)curve,the accuracy of the evaluation results of the FR-SVM and single SVM models were tested and compared.The results show that the FR-SVM model is more specific to the landslide susceptibility in Shangyou County.The area under receiver operating characteristic curve(AUC)values of the FR-SVM and single SVM models are 0.893 and 0.798,respectively,which indicates that the FR-SVM model is better than that of the single SVM in describing the distribution of landslide susceptibility indexes and the evolution rules of landslide disasters in Shangyou area.