随着经济的高速发展，人类活动愈加强烈、密集，地质灾害的发生频次亦是大大增加，给人们带来极大的人员和财产损失。开展地质灾害空间敏感性预测研究，对区域防灾减灾工作、地质灾害日常管理和后续国土空间规划具有相当重要的指导价值。同时，空间数据不确定性的存在直接或间接影响着地质灾害敏感性评价结果的数据可靠性。凤县地貌类型复杂，地质灾害频发，地质灾害类型齐全，本文以其作为研究区，进行面向地质灾害敏感性评价的空间数据不确定性研究，有着相当重要的科研价值与现实意义。本次研究取得了如下成果：

（1）基于前人研究成果及研究区地质环境特征，本文选取了坡度、坡向、高程、河流缓冲区、道路缓冲区、断层缓冲区、平面曲率、剖面曲率、归一化植被指数（NDVI）、降雨量、岩性和土地利用共计12个影响因子，利用频率比法（FR）分析了研究区地质灾害与影响因子各分级之间的空间关系。并经过多重共线性诊断表明各因子相互独立，皆可用于本次地质灾害敏感性建模。

（2）采用朴素贝叶斯树（Naive Bayes Tree ，NBTree）、J48决策树（J48）及二者与旋转森林（Rotation forest，RoF）构成的集成模型（RoF-NBTree、RoF-J48）等四种模型，分别进行4种栅格分辨率（12.5m、25m、50m、100m）和5种正负样本取样比例（1:1、1:2、1:3、1:4、1:5）下的地质灾害敏感性建模。模型精度比对分析表明：集成建模方法可以提高模型的性能；25m和50m分辨率的空间数据在研究区地质灾害空间预测中可以得到更好的结果；正负样本取样比例为1:1可以有效提高研究区敏感性建模的分类精度，正负样本取样比例为1:3可以有效提高研究区敏感性建模的预测精度；25m分辨率和正负样本比例为1:1所构建的数据集下的RoF-NBTree模型为本次凤县地质灾害敏感性建模的最优模型。

（3）研究区敏感性分区结果表明：敏感性分区面积与敏感性等级呈反比，极高敏感性面积占比最小、极低敏感性面积占比最大；高敏感性分区中地质灾害点密度较大而低敏感性区中的较小，且各模型的分区结果具有一致性；结合凤县河流、路网以及人员建筑的空间分布特征，发现预测结果的差异性分布规律与其高度一致，进一步验证了本次模型预测结果的可靠性。

With the rapid development of economy, human activities are more intense and intensive, and the frequency of geological disasters is also greatly increased, which brings great losses to people and property. The research on spatial sensitivity prediction of geological disasters is of great guiding value for regional disaster prevention and reduction, daily management of geological disasters and subsequent territorial spatial planning. At the same time, the uncertainty of spatial data directly or indirectly affects the data reliability of geological hazard sensitivity evaluation results. Fengxian County has complex geomorphic types, frequent geological disasters and complete geological disaster types. This paper takes Fengxian County as a research area to study the uncertainty of spatial data for geological disaster sensitivity evaluation, which has very important scientific research value and practical significance. The results of this study are as follows:

(1) Based on previous research results and geological environment characteristics of the study area, this paper selected 12 influencing factors, including slope, slope direction, elevation, river buffer, road buffer, fault buffer, plane curvature, profile curvature, normalized vegetation index (NDVI), rainfall, lithology and land use. The frequency ratio method (FR) was used to analyze the spatial relationship between geological hazards and each grade of the impact factors in the study area. The multicollinearity diagnosis shows that all factors are independent of each other and can be used in the geological hazard sensitivity modeling.

(2) Four models, namely Naive Bayes Tree (NBTree), J48 decision tree (J48) and the integrated model consisting of the two and Rotation forest (RoF) (RoOF-NBtree, RoOF-J48), are adopted. The geological hazard sensitivity modeling was carried out under four grid resolutions (12.5m, 25m, 50m, 100m) and five positive and negative sample ratios (1:1, 1:2, 1:3, 1:4, 1:5) respectively. The precision comparison analysis shows that the integrated modeling method can improve the performance of the model. The spatial data with resolution of 25m and 50m can get better results in the spatial prediction of geological hazards in the study area. The sampling ratio of positive and negative samples 1:1 can effectively improve the classification accuracy of sensitivity modeling in the study area, and the sampling ratio of positive and negative samples 1:3 can effectively improve the prediction accuracy of sensitivity modeling in the study area. The RoF-NBTree model with 25m resolution and 1:1 ratio of positive and negative samples is the optimal model for geological hazard sensitivity modeling in Fengxian County.

(3) The results of sensitivity zoning in the study area showed that the area of sensitivity zoning was inversely proportional to the sensitivity grade, and the area of extremely high sensitivity was the smallest and the area of extremely low sensitivity was the largest; The density of geological disaster points in the high-sensitivity zone is higher than that in the low-sensitivity zone, and the results of each model are consistent. Combined with the spatial distribution characteristics of rivers, road networks and human buildings in Fengxian county, it is found that the difference distribution law of the predicted results is highly consistent with that of the predicted results, which further verifies the reliability of the predicted results of this model.