摘要
滑坡灾害危险性区划在滑坡灾害管理工作中具有重要意义。该文以深圳市为研究区,将随机森林方法应用于滑坡灾害危险性区划和因子评价工作。研究中采用随机森林方法进行建模,结合ROC曲线和AUC值对模型性能进行评价;通过预报效率曲线对全区的滑坡危险性区划总体效果进行评价;根据错误率变化和节点不纯度变化,对不同因子的重要性进行分析,结合作用曲线,探讨不同因子对滑坡的影响规律。实例研究说明随机森林方法适用于滑坡危险性区划工作,具有良好的建模效果。
Landslide susceptibility mapping plays an important role in landslide risk management.This paper applies random forest to landslide susceptibility mapping and factor importance evaluation,with a case study carried out in Shenzhen.By calculating the area under the ROC curve(AUC)as well as the forecast efficiency curve,the performance of the model is evaluated.The importances of the factors are discussed based on the error rate and tree node impurity output by random forest,while the partial effect curves demonstrate the detailed influences of factors on landslide susceptibility.It can be concluded from the case study that random forest,stable and accurate,is suitable for landslide susceptibility mapping.The paper may provide direct reference for future studies on landslide susceptibility mapping.
出处
《地理与地理信息科学》
CSCD
北大核心
2014年第6期25-30,F0002,共7页
Geography and Geo-Information Science
关键词
随机森林方法
滑坡灾害危险性区划
集成学习方法
模型评价
因子评价
random forest
landslide susceptibility mapping
ensemble learning
model assessment
factor importance evaluation