基于信息量-随机森林模型的地震带地质灾害易发性评价:以松潘-较场地震带为例

Evaluation of Geological Hazard Susceptibility in Seismic Zone Based on Information Data-RF Model:A Case Study of Songpan-Jiaochang Seismic Zone

地震会引起地表振动及破坏,同时加大滑坡、崩塌、泥石流等次生灾害的发生概率,对位于地震带区域城市进行地质灾害易发性预测,是地质灾害防治的有效措施。为了探究地震带区域地质发育程度对地质灾害的影响,以松潘-较场典型地震带的平武县为例,从地形地貌特征、地层地质条件、气象水文、地震带发育特征、土壤植被、人类工程活动影响6个方面选取地质灾害的诱发因子,采用信息量模型、信息量-层次分析法(analytic hierarchy process,AHP)和信息量-随机森林(random forest,RF)3种评价模型对平武县地质灾害进行易发性评价,结果表明信息量-RF模型的对比分析结果优于其他两种模型,接受者操作特征(receiver operating characteristics,ROC)曲线精度评估信息量-RF模型的曲线下的面积(area under curve,AUC)(0.991)高于信息量-AHP模型(0.931)和信息量模型(0.920),说明基于信息量耦合随机森林的综合易发性评价模型更适用于地震带地区的地质灾害易发性评价,具有良好的预测精度。

Earthquake will cause surface vibration and damage,and increase the probability of landslide,collapse,debris flow and other secondary disasters.It is an effective measure to predict the vulnerability of geological disasters in cities located in seismic zones.In order to explore the influence of regional geological development degree on geological disasters,Pingwu County in the Songpan-Bianchang typical seismic belt was taken as an example,and the inducement factors of geological disasters were selected from six aspects:topographic and geomorphic features,stratigraphic geological conditions,meteorology and hydrology,development characteristics of seismic belt,soil and vegetation,and the influence of human engineering activities.Three evaluation models,information-information model,information-analytic hierarchy process(AHP)and information-random forest(RF),were used to evaluate the vulnerability of geological disasters in Pingwu County.The results show that the comparative analysis results of information-RF model are better than the other two models.The area under curve(AUC)value(0.991)of the information-RF model evaluated by receiver operating characteristics(ROC)curve accuracy is higher than that of the information-AHP model(0.931)and the information-AHP model(0.920),indicating that the comprehensive vulnerability assessment model based on the information-coupled random forest is more suitable for the vulnerability assessment of geological disasters in seismic zones and has good prediction accuracy.