山区公路沿线地震滑坡易发性区划研究

基于地理信息系统（GIS）技术，建立逻辑回归评价模型，对汶川县和芦山县公路周边8 km范围的地震滑坡开展易发性区划研究。在研究过程中，首先，从地震诱发滑坡灾害的触发机理出发，参考现场工程地质勘察资料及现有的研究成果，选取包括高程、坡度、坡位、坡向、曲率、微地貌、岩性、距断层距离、距水系距离、归一化植被指数（NDVI）、地震峰值加速度（PGA）、距离道路距离、年平均降雨量在内的13个地震诱发滑坡的影响因子；然后，以汶川县地理空间数据库为基础，通过10组样本在R软件中的训练，建立逻辑回归评价模型；最后，应用上述建立的模型，在ARCGIS平台下将汶川县公路沿线地震滑坡的易发性划分为极低、低、中、高、极高5个级别。为了进一步检验所建立评价模型的稳定性，将上述模型仿真到芦山县公路沿线，形成芦山县公路沿线地震滑坡易发性区划图。通过2个研究区域的ROC曲线，可以得到AUC值分别为0.912和0.783。研究结果表明：预测结果与实际地震滑坡分布吻合度较高，建立的评价模型稳定性较好，可以推广用于地形、地质环境相似山区公路沿线地震滑坡易发性区划评价。

Research on Susceptibility Mapping of Earthquake-induced Landslides Along Highway in Mountainous Region

Earthquake-triggered landslides was a key factor influencing highway capacity in mountainous area. An accurate susceptibility mapping along highway was necessary for the adoption of proper preventive measures and normal operation of highway. In this paper, on the basis of geographic system (GIS) technology, logistic regression (LR) model was established to investigate susceptibility in an area of 8km around highway in Wenchuan Country and Lushan Country. In the process of research, firstly, based on triggering mechanism of earthquake-induced landslides, the research achievements and results of site investigation, thirteen main factors that influence earthquake-induced landslides were identified, including elevation, slope gradient, slope position, slope aspect, curvity, micro-landform, lithology, distance from streams, distance from faults, normalized differential vegetation index (NDVI), peak ground acceleration (PGA), distance from highways, and mean annual rainfall. Then, based on the geospatial database of Wenchuan Country, 10 sets of samples were trained by R software, and the LR model was developed. Finally, a map of the earthquake-induced landslides susceptibility classes for Wenchuan Country was obtained by dint of the ARCGIS platform. The susceptibility values of earthquake-induced landslides were divided into five categories, from very low to very high. The application in Lushan Country tested the validity and stability of the LR model, a map of the earthquake-induced landslides susceptibility classes for Lushan Country was obtained. The AUC values of the ROC curve for the model were 0.921 and 0.783 respectively, corresponding to Wenchuan Country and Lushan Country. The results show that the susceptibility map generated by the LR model is identical with the true distribution of earthquake-induced landslides, and the LR model is with high stability. Therefore, the LR model can be applied to predict the earthquake-induced landslides along highways in other mountainous regions with similar geological environment conditions.