基于加权信息量模型的滑坡灾害易发性评价--以灞桥区为例

Landslide Susceptibility Assessment Based on Weighted Information Values Model:Take Baqiao District as an Example

以灞桥区作为研究区,在查阅大量相关文献以及对区域地质环境条件、滑坡特征分析的基础上,选取坡度、坡向、高程、曲率、地形起伏度、地貌类型、地层岩性、断层距离、水系距离、降雨量、归一化植被指数(normalized difference vegetation index, NDVI)、道路距离12类影响因素作为评价因子。基于熵指数模型(index of entropy, IOE)与信息量模型(IV),提出一种新的加权信息量模型(WIV),分别将三种模型用于研究区滑坡灾害易发性评价中。借助ArcGIS软件计算得到整个研究区滑坡易发性指数,根据自然间断点法划分为5个易发等级区,分别为极低、低、中、高、极高易发区,生成最终的滑坡易发性区划图。最后采用ROC曲线对三种评价模型进行检验与比较,结果表明:IOE、IV、WIV三种模型的成功率曲线的训练精度分别为76%、84%、88%;预测率曲线的预测能力分别为72%、86%、90%,由此得出WIV模型较IOE、IV模型预测准确率更高。同时,对三种模型之间的独立性进行了检验,结果表明IOE模型与IV、WIV模型呈低相关度,IV模型与WIV模型呈显著相关。研究结果可为评价模型间的选取以及当地有关部门进行防灾减灾、工程部署以及土地利用规划提供参考。

Taking Baqiao district of Xi’an, Shaanxi, China as the study area, 12 landslide evaluation factors were selected based on literature review and geological analyses. The factors included slope, aspect, altitude, curvature, relief, landform, lithology, distance to faults, distance to rivers, rainfall, NDVI, and distance to roads. Based on index of entropy(IOE) and information values(Ⅳ), a new model of weighted information values(WIV) was proposed and used to evaluate landslide susceptibility in the study area. The landslide susceptibility indices of the whole study area were calculated by using ArcGIS software. According to the natural break method, five grades of landslide susceptibility were recognized, i.e., very low, low, moderate, high, and very high, respectively, based on which a map of landslide susceptibility was generated. Finally, the ROC(receiver operating characteristics) curve was used to validate and compare the three evaluation models. Result shows that the training accuracy of the success rate reached 76%, 84%, and 88% for IOE, Ⅳ, WIV models;and the prediction rate with the prediction curve reached 72%, 86%, and 90%, respectively. Thus, the prediction accuracy of the WIV model was higher than that of the IOE and Ⅳ model. At the same time, the independence of the three models was tested, and the results show that the IOE model had a weak correlation with the Ⅳ model and the WIV model, and the Ⅳ model had a strong correlation with the WIV model. This study can provide a reference for the selection of evaluation models, the disaster prevention and reduction, engineering deployment, and land use planning of local governments.