基于综合智能分析的广东省崩塌易发性评价

Evaluation of collapse susceptibility in Guangdong Province based on comprehensive intelligent analysis

本文旨在使用智能分析模型提高广东省崩塌易发性评价精度。为此,收集该省7171个崩塌数据以及13个基础环境因子,并用分批分块的方式解决了海量数据处理的问题。利用随机森林模型(RF)和耦合频率比-随机森林模型(FR-RF)开展全省1∶5万精度的崩塌易发性评价,采用受试者工作曲线和易发性分布特征来评估模型性能,并与传统评价方法(层次分析法(AHP)、频率比法(FR)和耦合频率比-层次分析法(FR-AHP))进行比较。结果表明:(1)13个基础环境因子无显著相关且都具有较高贡献度,地形起伏、高程、多年平均降雨量是影响广东省崩塌发生的最主要环境因子;(2)广东省崩塌极高、高易发区主要分布在东北部、中部和西南部地区;(3)5种模型中RF的预测精度最高,AUC值为0.884,其次为FR-RF(0.860),FR-AHP(0.797),FR(0.794)和AHP(0.601),且RF和FR-RF的易发性指数不确定性较低。除AHP外,4种模型都能得到合理的崩塌易发性评价结果,而RF是最适宜评价广东省崩塌易发性的方法。总体而言,智能分析方法性能优于传统评价方法。本文成果可为大尺度低海拔区域崩塌易发性评价提供参考。

This study aims to improve the evaluation accuracy of collapse susceptibility in Guangdong Province using an intelligent analysis model.To this end,7171 collapse data and 13 basic environmental factors in the province are collected,and the problem of massive data processing is solved by batch and block methods.Random Forest model(RF)and the coupled Frequency Ratio-Random Forest model(FR-RF)are used to carry out 1∶50000 accuracy evaluation of collapse susceptibility.The performance of the model is evaluated by the receiver operating curve and the susceptibility distribution characteristics.The evaluation results were compared with those of traditional evaluation methods(Analytic Hierarchy Process(AHP),Frequency Ratio(FR)and Coupled Frequency Ratio-Analytic Hierarchy Process(FR-AHP)).The results show that:(1)The 13 basic environmental factors have no significant correlation,and they all have a high degree of contribution.Topographic relief,elevation and annual average rainfall are the most important environmental factors affecting the occurrence of collapses in Guangdong Province;(2)The very high and high susceptibility areas are mainly distributed in the northeast,central and southwestern regions;(3)Among the five models,the prediction accuracy of RF is the highest,with an AUC value of 0.884,followed by FR-RF(0.860),FR-AHP(0.797),FR(0.794)and AHP(0.601).The uncertainty of the susceptibility index of RF and FR-RF is low.Except for AHP,all four models can be used to obtain reasonable evaluation results of collapse susceptibility,and RF has been found to be the most suitable method to evaluate the collapse susceptibility of Guangdong Province.Overall,the performance of the intelligent analysis method is better than that of the traditional evaluation method.This study can be considered as a reference for the evaluation of collapse susceptibility in large-scale low-altitude areas.