基于信息量模型与随机森林模型的崩岗风险对比评估

Comparisons Between Benggang Risk Assessments Based on Information Model and Random Forest Model

崩岗是发生在我国南部地区一种典型的水土流失现象,对其进行风险评估有利于崩岗防控。建立崩岗风险评估体系和崩岗评估模型是崩岗风险评估的基础,而不同的评估模型对崩岗风险的评估结果存在差异。探究不同模型在崩岗风险评估中的应用差异,提高预测精度是目前亟需解决的问题。本研究以福建省安溪县龙门镇小流域为例,根据风险因子与崩岗发生之间的关系选择出主要因子,分别运用信息量模型与随机森林模型对崩岗进行风险评估,对比两种方法的优缺点,进一步探究较优的崩岗风险评估方法。结果表明:(1)信息量模型与随机森林模型均适用于研究区的崩岗发生风险评估,基于随机森林模型对崩岗发生风险的预测精度高于信息量模型(AUC值分别为0.89和0.81),混淆矩阵准确率达到84.09%,且泛化能力较好;(2)高程、地形起伏度、坡度和河网缓冲距离因子是崩岗发生的重要风险因子;(3)2个模型预测的崩岗风险空间分布大致相同,约83%的风险区等级面积趋于一致,且以中、高风险为主。研究结果表明随机森林模型的预测性能总体较信息量模型优,可用于未来崩岗发生的风险评估,为崩岗的防治工作提供参考。

Benggang represents a typical soil erosion phenomenon in the granite region of Southern China,characterized by considerable erosion and extreme landform changes.Different assessment models for Benggang risk exhibited different results,which requires a thorough comparison on their accuracy for the purpose of Benggang control.The establishment of a risk assessment system with a proper model is a prerequisite for Benggang risk assessment.And the differences in the applications of these models should be further determined for promotion of prediction accuracy.In this study,a small watershed in Longmen Town,Anxi county,Fujian province was targeted as case study for Benggang risk assessment.Main factors were selected in the case based on the relationship between risk factors and Benggang events.It compared the advantages and disadvantages between an Information model and a Random Forest model in the process of Bengang risk evaluation,to find out which model had a better performance.The results show that:(1)Both the Information model and the Random Forest model were applicable to the risk assessment of Benggang in the study area;however,the prediction accuracy of the Random Forest model for the risk of Benggang was higher than that of the Information model(AUC values were 0.89 and 0.81,respectively).The accuracy of the confusion matrix for the Random Forest model reached 84.09%with a better generalization ability.(2)Elevation,topographic relief,slope and river buffer distance mattered for the occurrence of Benggang.(3)The distribution of risk space of Benggang predicted by the two models was equivalent;approximately 83%of the risk areas were of the same level,dominated by medium-risk and high-risk.These study found that the prediction performance of the Random Forest model was generally better than that of the Information model.Random forest model is recommended to be used for Benggang assessment and control in the future.