旋转森林模型在滑坡易发性评价中的应用研究

以三峡库区万州段为研究区，从多源空间数据中提取29个致灾因子作为区域滑坡易发性分析的评价指标，在数字高程模型基础上采用集水区重叠法划分斜坡单元，构建旋转森林集成学习模型，定量预测滑坡空间易发性，并生成滑坡易发性分区图。在易发性分区图中，高易发区占11.6%，主要分布在万州主城区和长江及支流两岸；不易发区占45.6%，主要分布在人类工程活动低、植被覆盖度高的区域。采用受访者工作特征曲线和曲线下面积对旋转森林模型的滑坡易发性进行评价，结果显示该模型的预测精度为90.7%，其预测能力优于C4.5决策树。研究表明，应用旋转森林进行滑坡易发性评价具有预测能力强、精度高等优点。

Application of the Rotation Forest Model in Landslide Susceptibility Assessment

Focusing on Wanzhou region of the Three Gorges Reservoir, 29 hazard factors were extracted from the multi-source spatial data used in evaluation factors in landslide susceptibility analysis. The study area was partitioned into slope units from digital elevation model to resample the conditio-ning factors. A rotation forest model was trained and used to map landslide susceptibility with the best accuracy being 90.7%, according to the receiver operator characteristic (ROC) curve and area under the curve (AUC). The higher susceptibility zones were about 11.6% of the total area, and primarily distributed in the main Wanzhou city zone, and along both sides of the Yangtze River and its tributaries. The stability zones are accounted for about 45.6%, mainly distributed in the areas of low human engineering activities and high surface cover degree. The results show that application of rotation forest in the landslide susceptibility assessment exhibits both excellent prediction ability and high accuracy.