基于信息量和逻辑回归耦合模型的滑坡易发性评价

目前,滑坡易发性评价大多只采用单一模型进行研究,而单一模型存在缺陷,如只采用信息量模型则不能反映各因子对滑坡发生的权重。通过将两个模型进行耦合分析可以很好地发挥各模型的优点和弥补各模型的不足,从而达到模型优化的目的。针对滑坡易发性常用的信息量模型和逻辑回归模型,提出信息量-逻辑回归耦合模型。以江西省宁都地区为例,获取研究区共297个滑坡,提取高程、坡向、坡度、平面曲率、剖面曲率、地形起伏度、距水系距离、岩性、植被覆盖率、地表建筑物指数共10个因子建立评价指标体系,再分别采用上述3个模型开展易发性评价,最后采用预测率曲线(the prediction rate curve, ROC)评价各模型精度。结果表明:信息量模型、逻辑回归模型和信息量-逻辑回归耦合模型预测率曲线与坐标轴围成的面积(area under ROC, AUC)值分别为0.838、0.864和0.876,可见信息量-逻辑回归耦合模型的评价精度更高,建模更为合理。研究区内滑坡主要沿水系两侧分布,高程和岩性对滑坡的发生起主要作用。

Landslide susceptibility assessment based on the coupling model of information value and Logistic regression

Nowadays, the landslide susceptibility prediction is usually conducted by using a single model, however, the single prediction model often has some disadvantages. For example, only adopting the information value model cannot reflect the weight of each factor on the landslide. The advantage of each model can be fully achieved and the shortcomings of each model can be avoided, through the coupling analysis of two single model, so as to achieve the purpose of model optimization. Hence, for the information value model and logistic regression model which are commonly used for landslide susceptibility prediction, information-logistic regression coupled model is proposed in this study. Then taking Ningdu district of Jiangxi province as an example, this study obtains a total of 297 landslides in the study area and selects ten conditioning factors (the elevation, slope aspect, slope, profile curvature, plane curvature, relief, distance from drainage, lithology, Normalized Difference Vegetable Index (NDVI), Normalized Difference Building Index (NDBI)). Then the above three models are used to predict the landslide susceptibility in Ningdu district. The prediction rate curve is adopted to assess the prediction accuracy of each landslide susceptibility model. The results show that the AUC values of prediction rate curves of information model, logistic regression model and information-logistic regression coupled model are 0.838, 0.864 and 0.876, respectively. It can be seen that the information-logistic regression coupled model has the greatest prediction performance. In addition, landslides in Ningdu district are mainly distributed along both sides of river system, and elevation and lithology play major roles in the occurrence of landslides.