基于RFE与地理探测器因子优选的泸定县滑坡易发性评价OA
Landslide susceptibility assessment in Luding County based on RFE and GeoDetector factor selection
科学合理的滑坡因子筛选对于提升预测模型的准确性至关重要.本研究以泸定县为研究区,选取13个潜在诱发因子,分别采用预测驱动的递归特征消除(recursive feature elimination,RFE)和机制驱动的地理探测器(GeoDetector)进行因子优化,然后将筛选后的因子集输入极限梯度提升(XGBoost)、随机森林(RF)和逻辑回归(LR)三种机器学习模型进行训练和评价,最后通过受试者工作特征(ROC)曲线下面积(AUC)、准确率(accuracy)和 F1值等指标评估模型性能.结果显示,RFE迭代优化确定了最优的 10个因子,而 GeoDetector选取 11个因子集,旨在最大限度地降低冗余和噪声;模型性能对比中,基于 RFE优选因子的模型组在各项精度指标上均全面超越 GeoDetector组,其中 RFE-XGBoost模型表现出压倒性优势,其 AUC值高达 0.968,准确率(accuracy)为 0.842,F1值为 0.868,有力证明了 RFE策略在追求高精度预测方面的有效性;GeoDetector则在机制解释方面具有优势,揭示了地震与降雨的协同增强效应是该区域滑坡风险的主导驱动结构.在空间分布上,高易发性等级区域主要沿断层带分布,反映出研究区滑坡发育受特提斯构造演化影响.本研究证明了因子筛选对模型性能提升的重要作用,可为防灾规划提供科学依据.
Scientific and rational selection of landslide conditioning factors is essential for improving the accuracy of susceptibility prediction models.In this study,Luding County was selected as the study area,and 13 potential conditioning factors were initially considered.Two factor-selection methods were applied:Recursive Feature Elimination(RFE),which is driven by predictive performance,and GeoDetector,which is based on mechanism-based analysis.The selected factor subsets were then input into three machine learning models—XGBoost,Random Forest(RF),and Logistic Regression(LR)—for training and evaluation.Model performance was assessed using metrics such as the area under the receiver operating characteristic(ROC)curve(AUC),accuracy,and F1 score.The results show that RFE,through iterative optimization,identified an optimal subset of 10 factors,while GeoDetector selected 11 factors,aiming to minimize redundancy and noise.A comparison of model performance indicates that the models based on RFE-selected factors consistently outperformed those based on GeoDetector-selected factors across all accuracy metrics.Among all models,the RFE-XGBoost model demonstrated a dominant advantage,with an AUC of 0.968,an accuracy of 0.842,and an F1 score of 0.868,strongly validating the effectiveness of RFE for high-precision landslide susceptibility prediction.In contrast,GeoDetector excels in mechanism explanation,revealing that the synergistic effect of seismic intensity and rainfall represents the dominant driving structure controlling landslide susceptibility in the region.Spatially,areas with high landslide susceptibility are mainly distributed along fault zones,reflecting the influence of Tethys tectonic evolution on landslide development in the study area.This study highlights the important role of factor selection in improving model performance and provides a scientific basis for landslide disaster prevention and mitigation planning.
邓辉;胡仕荣;郑予;赵宇;普布扎西
成都理工大学地理与规划学院,四川 成都 610059||成都理工大学 地质灾害防治与地质环境保护国家重点实验室,四川 成都 610059成都理工大学地理与规划学院,四川 成都 610059成都理工大学地理与规划学院,四川 成都 610059||四川聚汇兴源工程设计有限公司,四川 成都 610081西藏地质三队地矿有限公司,西藏 拉萨 851400西藏地质一队地矿有限公司,西藏 拉萨 850007
天文与地球科学
滑坡易发性递归特征消除地理探测器机器学习泸定县
landslide susceptibilityrecursive feature eliminationGeoDetectormachine learningLuding County
《沉积与特提斯地质》 2026 (3)
581-594,14
西藏自治区科技计划项目"喜马拉雅山地区地质灾害隐患早期识别与风险动态评估关键技术研究"(XZ202501YD0004)四川省自然资源厅科研项目"突发性泥石流灾害应急技术支撑工作机制研究"(KJ-2024-011)
评论