摘要
地质灾害预测预警在自然灾害防治中的地位日益凸显,而地质灾害易发性评估是地质灾害预测预警的基础。文章以北京西南地区的公路崩塌灾害为研究对象,利用GIS技术从多来源、多时相的数据源中提取了包括高程、地层岩性和降水在内的多种崩塌致灾因素;尝试将新型的XGBoost机器学习模型应用于地质灾害易发性研究,将452个公路崩塌灾害样本分为70%的训练集与30%的验证集进行训练,利用召回率、ROC曲线以及模型的不确定性等指标对模型结果进行评价,并与常用的SVM模型进行对比。结果表明,XGBoost模型的训练集和验证集的召回率分别为1.00和0.87,ROC曲线下面积分别为1.00和0.91,均优于SVM模型,并且其预测价值大的样本(崩塌发生概率为0.75~1和0~0.25)占比达到93%,远高于SVM模型。将训练得到的评估模型应用于北京西南地区,叠加区域路网数据,绘制北京西南地区的公路崩塌易发性图。用"8·11"房山崩塌灾害样本验证了易发性图的可靠性,并通过易发性图识别出需要重点防范的崩塌高易发性公路路段。
Geological hazard susceptibility assessment serves as the foundation of geological hazard prediction and early warning which plays an increasingly important role in disaster prevention and reduction.This study takes avalanche hazard along highway in the southwest Beijing as a case study.Various geological hazard inducers containing elevation,lithology,precipitation and so on are prepared in GIS from multi-source and-temporal data.The XGBoost model is first applied in geological hazard susceptibility assessment.The training of model is based on a total of 452 avalanche samples randomly divided into two subsets:a training subset comprising 70%and a validation one 30%.The model performance is evaluated using the recall,the area under the receiver operation characteristic curve(AUC)and the model estimation uncertainty in comparison with the popular SVM model.The results demonstrate a higher accuracy for the XGBoost than the SVM,with the recall and the AUC of the training and validation subsets for the XGBoost being 1.00 and 0.87,1.00 and 0.91,respectively.In addition,the XGBoost outperforms the SVM in the evaluation of the model estimation uncertainty,since the samples with susceptibility of 0.75-1 and 0-0.25 which are more valuable account for 93%of the total using the XGBoost.The trained assessment model is then applied to produce the highway avalanche susceptibility map of the southwest Beijing.The reliability of our model is verified by the‘8·11’avalanche hazard in Fangshan which is predicted as an extremely high susceptibility level.According to the susceptibility map,highway sections with high avalanche susceptibility in the southwest Beijing are recognized,which should be of great concern.
作者
林报嘉
刘晓东
杨川
尹航
Lin Bao-jia;Liu Xiao-dong;Yang Chuan;Yin Hang(China Highway Engineering Consultants Corporation Data Co.Ltd.,Beijing,100089,China;China University of Geosciences,Wuhan,430074)
出处
《公路》
北大核心
2020年第7期20-26,共7页
Highway
