【作者】 肖桐；

【导师】 马金辉；

【作者基本信息】 兰州大学 ， 地图学与地理信息系统， 2007， 硕士

【摘要】 滑坡是一种严重的自然灾害，其发生的时间、地点、规模和方式具有很大的不确定性。这就要求我们在对地质灾害因果分析的基础上，寻求对区域滑坡地质灾害进行危险性评价，从而对其进行更为有效的控制和管理。本文从影响滑坡发生的内因和外因两方面出发，深入讨论了影响区域滑坡发生与发展的各个因素，提取了坡度等15个定量、定性因子，基于似然比原则，计算了滑坡在每个因子中的似然比，探讨了滑坡灾害与各个因子之间的关联关系，并对各个因子进行了分类。在此基础之上，采用CF确定性系数，计算得到了每个因子层中各个分类的CF值，评价了不同因子类别对滑坡的影响。按照CF值的特点，对研究区的危险程度进行了初步划分，得到了不同因子类别应属于的危险区域。之后，对各个因子类别进行了CF值的合并，评价出了对滑坡具有正反馈的因子为坡度、水平曲率、剖面曲率和坡长坡度因子；另一方面，从栅格数据的角度出发，利用二元逻辑回归模型，随机选择了区域内一定数量的样本，得到了区域内滑坡发生概率的二元逻辑回归模型，并计算了标准化回归系数(β~*)，按照其排序结果对影响因子进行了提取，发现降雨量、高程和最大60分钟雨强是滑坡的重要影响因素。本文最终对滑坡的空间预测是基于均一单元的划分和基于栅格两种方法。以矢量数据为基础提取了研究区的滑坡因素均一单元，利用栅格数据，建立了在20m空间分辨率下的滑坡因子空间，在似然比分类的基础上，分别利用多元线性回归模型和二元逻辑回归模型对区域的滑坡分布进行了模拟预测研究。最后对两种方法的精度进行了评估，发现基于栅格因子空间的划分方法比基于均一因子属性单元的划分方法更有利于模拟、预测滑坡的空间分布。更多还原

【Abstract】 Landslide is one of typical environment geological hazards, whose harms become more and more serious. There are plenty of uncertainty of the occurrence, location and type of the landslide. Based on the analysis for cause and effect of the geologic factors, we must predict the landslide and grade it so that we can control and manage the slope more efficiently.In this paper, we discuss the two facet of the landslide factor, the first is inner factor and the other is outer factor. Based on the GIS, we extract 15 factors of landslide: lithology, stratum, altitude, slope, aspect, plan curvature, profile curvature, slope length factor, slope location, average rainfall, highest rain intensity in 10 minutes, highest rain intensity in 60 minutes, NDVI, river cost distance and landuse. We divide these factors into two types: qualified factors and normal factors. For qualified factors, we calculate the likelihood ratio between landslide and them, and discuss the relationship between landslide and each factor. Afterward, we classified each factor based on the likelihood ratio.After classified the qualified factors, we calculate the certainty, factor (CF) of each class of each factor, and evaluate the most important factor class. Based on CF, we divide the study area into several regions with different hazard degree, and sort the factor into the different regions. After this, we integrate the different class of one factor into one value according to some rules. So, we get four factors, which have positive response for landslide. In the end, we use multi-variable regression model to build an equation to describe the spatial frequency of the occurrence of landslide.On the other hand, we start with grid data. We use logistic model (before we calculate the model, we select part of the data randomly). According to the result, we get the coefficient, namedβ*. Based on this index, we order the coefficient and sort the different class. Based on the rule of likelihood ratio, we get the likelihood ratio between landslide probability and landslide. And then, we classified the landslide probability and get the landslide hazard distribution map. Finally, we compare the predict precision of the two methods. We find that logistic model is more accurate than multi-variable regression model.Throughout the study, we predict the landslide according to homogeneous unit based on vector data model and grid unit based on grid data model. Based on likelihood ratio classified factor, we use multi-variable regression model and logistic regression model to simulate the hazards of landslide. But the regression model based on homogeneous unit is not as accurate as we expect. The non-linearity characteristic is the most important reason. Secondly, logistic regression model can predict the landslide probability effectively, but it is not as effective as CF in evaluating the relationship between landslide and each factor.更多还原