【摘要】 通过地理信息系统(GIS)技术,结合神经网络(NN)和逻辑回归模型(LR)开展汶川县范围内地震诱发滑坡易感性评价,并对两种模型结果进行比较研究。基于2008年5.12 Ms8.0级地震,选取高程、坡度、坡位、坡向、岩性、微地貌、距断层距离、距水系距离、距道路距离、年平均降雨量、归一化植被指数、地震峰值加速度共12个因子作为地震滑坡影响因子,基于ARCGIS10.1平台将这些影响因子专题图层栅格化;采用提取的模型训练样本,由R软件对神经网络(NN)和逻辑回归模型(LR)进行训练;将训练好的模型对整个汶川县地震滑坡易感性进行仿真,并将仿真结果划分为五类滑坡敏感区域:极低,低,中,高和极高,分别得到LR与NN模型仿真的滑坡易发性分区图;根据汶川县实际地震滑坡分布图进行统计分析,以及采用ROC曲线对两种模型的仿真结果进行对比分析,神经网络(NN)和逻辑回归模型(LR)的AUC值分别为0.930和0.941。研究表明两种模型的滑坡易感性评价图与实际滑坡发育基本吻合,评价结果较好,且LR模型预测精度相对较高。更多还原

【Abstract】 Its purpose is to comparative study on the susceptibility mapping model of landslides at meizoseismal area 2008 Wenchuan earthquake by two statistics method: neural network and logistic regression. Based on the landslides from f ield investigation of in Wenchuan County, main influencing factors of the earthquake triggered landslides were studied, including 12 factors: peak ground acceleration(PGA), distance from faults, distance from drainage, distance from highway, slope gradient, average annual rainfall, normalized difference vegetation index(NDVI), lithology, slope position, elevation, slope aspect and micro-landform. Using ArcGIS 10.1 set up factors layers and corresponding raster data. NN model and LR model were achieved to training, testing and predicting by R software. The study area was reclassified into five categories of landslides susceptibility: very low, low, medium, high and very high. By statistics analysis, the landslide data from field investigation are good agreement with the evaluation results. The overall prediction capability are about 0.930 and 0.941 shown by AUC values of NN model and LR model, respectively. Both NN model and LR model are satisfactory for susceptibility mapping of landslide triggered by earthquake, but all results showed that the LR model seems to be slightly superior to the NN model.更多还原