基于逻辑回归模型的汶川地震灾区崩塌滑坡易发性评估

Research on the risk of collapse and landslide in Wenchuan earthquake disaster area based on logistic regression model

选取了汶川地震灾区的22个县作为研究区域,基于6007条崩滑数据和多源信息,利用GIS技术手段,选取高程、坡度、坡向、断层、岩性、降雨量和土地利用7个因子作为崩滑易发性评价的影响因子,基于逻辑回归模型进行崩滑易发性评估并生成易发性分区图。结果表明,历史崩滑点分布与崩滑易发性分区图拟合程度较好,该地区的汶川县、茂县、北川县、绵竹市、青川县、平武县为崩滑易发性较高的区域。采用ROC曲线对模型成功率和预测率进行评价,求得训练样本集的AUC(ROC曲线下与坐标轴围成的面积)为0.84,验证样本集的AUC为0.79,表明采用的逻辑回归模型预测准确率较高。

22 counties in the Wenchuan earthquake disaster area were selected as the study area.Based on 6007 collapse and land⁃slides data and multi-source information,using GIS technology,seven factors including elevation,slope,aspect,fault,lithology,rainfall,and land use were selected as the influencing factors for the evaluation of collapse and landslide susceptibility.Then based on the logistic regression model,the susceptibility to collapse was evaluated,and a susceptibility zone map generated.The results showed that the distribution of historical collapse and landslide points fitted well with the zoning map of collapse and landslide susceptibility.Wenchuan county,Mao county,Beichuan county,Mianzhu city,Qingchuan county,and Pingwu county in this area were areas with high susceptibility to collapse and landslides.The ROC curve was used to evaluate the success rate and prediction rate of the model,and the AUC(Area under curve)of the training sample set was 0.84,and the AUC of the verification sample set was 0.79,indicating that the logistic regression model used in this paper had a high prediction accuracy.