基于InSAR负样本集成策略与BO-XGBoost的库区滑坡易发性评价OA
Landslide susceptibility assessment of a reservoir area based on an InSAR-derived negative sample integration strategy and BO-XGBoost
准确的滑坡易发性评价是大型水电工程区域地质灾害风险管控的关键.针对传统滑坡易发性评价中负样本选取存在可信度低、易引入预测偏差的问题,本文以某大型电站库区为例,提出了一种融合 InSAR形变信息与集成学习优化的评价新方法.研究构建了涵盖地形地貌、水文、地质环境条件及 InSAR形变速率的多源特征指标体系;设计并对比了五种负样本采样策略,重点构建了"基于 InSAR采样(InSAR-based sampling,IBS)、联合训练(joint training,JT)与权重叠加(overlay weighting,OW)"的集成策略(IBS-JT-OW).在此基础上,利用 GeoDetector剔除冗余特征,构建基于贝叶斯优化(Bayesian optimization,BO)的极限梯度提升(extreme gradient boosting,XGBoost)滑坡易发性评价模型.结果表明:IBS-JT-OW集成策略模型精度(AUC=0.969)显著优于传统采样方法及随机森林(random forest,RF)模型,该方法能更准确识别库区潜在风险,为区域滑坡风险评估提供了新的技术途径.
Accurate landslide susceptibility assessment is crucial for geological disaster risk management in large hydropower project areas.To address the low reliability and potential prediction bias associated with negative sample selection in conventional assessment methods,this study proposes a novel evaluation method that integrates interferometric synthetic aperture radar deformation information with ensemble learning optimization.A large hydropower reservoir area was selected as the case study.A multi-source feature index system was constructed,incorporating topography,hydrology,geological environmental conditions,and InSAR-derived deformation rates.Five negative sample selection strategies were designed and compared,with a focus on an integrated strategy based on"InSAR-based sampling,joint training,and overlay weighting(IBS-JT-OW)".On this basis,redundant features were then eliminated using GeoDetector,and a landslide susceptibility assessment model using Bayesian optimization and extreme gradient boosting was developed.Results indicate that the model based on the IBS-JT-OW integration strategy(AUC=0.969)significantly outperforms conventional sampling methods and the random forest model.This method more accurately identifies potential landslide-prone areas within the reservoir area and offers a new technical approach for regional landslide risk assessment.
范进勇;马彦飞;杨府学;周军;姜博;常鸣
中国长江电力股份有限公司,湖北 武汉 430014成都理工大学地质灾害防治与地质环境保护全国重点实验室,四川 成都 610059中国长江电力股份有限公司,湖北 武汉 430014中国长江电力股份有限公司,湖北 武汉 430014中国长江电力股份有限公司,湖北 武汉 430014成都理工大学地质灾害防治与地质环境保护全国重点实验室,四川 成都 610059
天文与地球科学
滑坡易发性采样策略InSAR技术极限梯度提升贝叶斯优化
landslide susceptibilitysampling strategyInSARextreme gradient boostingBayesian optimization
《沉积与特提斯地质》 2026 (3)
567-580,14
长江电力股份有限公司科研项目(合同编号:Z532502004)
评论