确定性系数与随机森林模型在云南芒市滑坡易发性评价中的应用

编制科学的滑坡易发性分区图，可以有效降低灾害带来的损失。以云南省芒市为研究区，利用确定性系数模型（certainty factor，简称CF）方法计算各个因子的敏感值，作为随机森林（random forests，简称RF）的分类数据，选取合适的训练数据和最优化的模型参数进行模型预测，从而对研究区进行滑坡易发性评价分区。采用频率比方法将连续性因子离散化，从而通过确定性系数计算因子不同区间的滑坡易发性，同时利用CF先验模型，对研究区负样本进行选取。通过计算袋外误差得到最优化的RF参数，随后利用RF模型对研究区模型进行训练及预测。绘制ROC曲线和三维遥感影像对预测模型结果分别进行定量和定性评价，结果表明，所得到的模型精度为91%，优于随机抽样得到的结果。最后，采用平均基尼不纯度减少和平均准确度下降两种计算方法计算、评价了研究区各个因子的重要性。基于以上对研究区进行的滑坡易发性评价结果，可以为该区灾害风险评估和管理提供依据。 

Application of certainty factor and random forests model in landslide susceptibility evaluation in Mangshi City,Yunnan Province

Drawing up scientific zoning maps of landslide susceptibility can effectively reduce the loss caused by disasters.Taking Mangshi City, Yunnan Province as the research area, the researchers used certainty factor (CF) method to calculate the sensitive values of each factor, and used them as classified data of random forests (RF), selected appropriate training data and optimized model parameters, and finally established the prediction model of susceptibility in the research area.In this paper, the frequency ratio method is adopted to discretize the continuity factor, so as to calculate the landslide susceptibility of different sections of the factor through the deterministic coefficient.Meanwhile, CF prior model is used to select negative samples in the research area.The optimized RF parameters are obtained by calculating the out-of-pocket errors, and then the RF model is used to train and predict the research area model.ROC curve and 3D remote sensing image were drawn to evaluate the prediction model results quantitatively and qualitatively, and the results showed that the accuracy of the model was 91%, which was better than that of random sampling Finally, the importance of each factor in the study area was calculated and evaluated by using two calculation methods of average Gini impurity reduction and average accuracy reduction.Based on the above, the landslide vulnerability assessment is carried out in the study area to provide a basis for disaster risk assessment and management in this area.