摘要
以清江流域长阳西段滑坡为研究对象,采用SBAS-InSAR技术与随机森林算法相结合的评价方法,以斜坡单元为评价单元,对研究区滑坡危险性进行科学评价。选取地形类、地质类、水文类、人类工程活动等四大类共计12个评价因子指标作为研究区滑坡危险性评价指标,采用随机森林模型评价滑坡危险性,考虑评价模型中滑坡数据时效性差、不准确等特点,利用最新Sentinel-1A雷达数据,采用SBAS-InSAR方法获取最新的地面变形数据替代传统评价模型中的滑坡数据,结果表明,基于InSAR技术与随机森林算法的滑坡危险性评估AUC值为0.90,精确度较高。该方法有效地提高了地质灾害危险性评估的准确度,可以为政府部门的防灾减灾工作提供更加高效的决策支撑。
The Qingjiang River Basin,a typical karst mid-mountain geomorphic region where carbonate rocks constitute 72%of the lithology,has been extensively influenced by long-term geological processes.Quaternary loose deposit layers,mixed with soil and rock,are extensively distributed along both banks.These layers exhibit poor stability and are prone to frequent landslide disasters,such as the Pianshan landslide and Maoping landslide.Traditional landslide susceptibility assessments typically rely on static historical data and linear models,such as the information content method and the Analytic Hierarchy Process(AHP).However,these approaches are limited in their ability to capture the nonlinear characteristics of landslide evolution.To improve the timeliness and accuracy of the assessment,this study integrates Small Baseline Subset-InSAR(SBAS-InSAR)surface deformation monitoring technology with the random forest machine learning algorithm to conduct dynamic landslide susceptibility assessments 2 in Ziqiu town and Yuxiakou town,covering a total area of 251.89 km^(2) in Changyang county,Hubei Province.Based on the landslide development patterns in the study area,this study selects 12 evaluation indicators from four categories-topography,geology,hydrology,and human engineering activities-as the landslide susceptibility indicators for the region.It employs the Random Forest model for comprehensive susceptibility assessment.Due to the poor timeliness and inaccuracy of landslide data in traditional evaluation models,the study utilizes the latest Sentinel-1A radar data and applies the SBAS-InSAR method to obtain up-to-date surface deformation data to replace the conventional landslide data.The interpretation results show that the central and western parts of the study area are predominantly characterized by subsidence points,with a small number of uplift points,while the eastern region exhibits a greater distribution of uplift points.The density of surface deformation points decreases from the southern to the northern part of the study area.Along the Qingjiang River,surface deformation points are more densely distributed.From the western part of the study area to Ziqiu Town,subsidence points dominate,while from Ziqiu town to the eastern part of the study area,uplift points are more prevalent.Most of the surface deformation points are located around towns and villages.In Yuxiakou town,both subsidence and uplift points are present,while Ziqiu town mainly features uplift points.The surface deformation points are mainly located in hard carbonate rock formations,with densely distributed deformation points typically found on both sides of fault zones.The rock mass on either side of the fault is subjected to intense compression,leading to fracturing,which forms a fractured zone.The uneven distribution of stress on both sides can easily trigger landslides.Additionally,the comparison of surface deformation points with other evaluation factors reveals that these points are predominantly distributed in areas with slopes ranging from 8°to 25°,elevation differences between 10 m and 30 m,and proximity to construction land such as houses and roads,where human engineering activities significantly influence surface deformation.The 12 evaluation indicators and SBAS-InSAR interpretation results were used as training datasets,and the random forest method was employed to assess landslide susceptibility.The importance of the evaluation factors,as determined by the random forest classification prediction model,indicated that Quaternary thickness and engineering geological rock groups were significantly more influential than the other factors.This suggests that when the surface deformation rate is used as an indicator to evaluate landslide susceptibility levels,geological factors are predominant.The hardness and stability of the rocks,as well as the thickness of the Quaternary deposit layers,determine the scale and severity of landslides induced by surface deformation.Factors such as the influence range of river systems,elevation difference,and slope have relatively high importance values.Within a certain influence range of river systems,water level fluctuations may significantly affect landslide stability.Areas with greater elevation differences and steeper slopes exhibit higher surface deformation rates,consequently increasing the probability of landslides.The remaining factors have little impact on the prediction results;except for the distance to the fault,all are related to human engineering activities.This suggests that when the surface deformation rate is used as an indicator for landslide susceptibility assessment,human engineering activities may alter certain original landforms and potentially trigger surface deformation and landslides,although their influence is relatively limited.The calculation results indicate that the areas classified as extremely high-risk and high-risk for landslide hazards in the western section of Changyang,within the Qingjiang River Basin,account for a substantial proportion,reaching 32.22%.These areas are mainly concentrated near Ziqiu town on the eastern side of the study area and along the banks of the Qingjiang River,which aligns well with the spatial distribution of historical landslides.The Quaternary deposit thickness,engineering geological rock groups,and river systems are identified as the dominant controlling factors for landslide susceptibility.Areas characterized by thicker Quaternary deposits,interbedded with soft and hard rock layers,and the distance proximity to rivers(within 200 meters)demonstrate significantly higher probability of landslides.The ROC curve analysis of the hazard assessment model shows that the random forest model incorporating InSAR technology can effectively capture landslide susceptibility,achieving a high AUC value of 0.90.This model exhibits strong predictive performance and reliability,providing a novel approach to landslide susceptibility assessment and valuable decision-making support for governmental disaster prevention and mitigation efforts.
作者
孟小军
邢昭
MENG Xiaojun;XING Zhao(School of Environmental Studies,China University of Geosciences,Wuhan,Hubei 430074,China;College of Resources and Environment,Yangtze University,Wuhan,Hubei 430100,China)
出处
《中国岩溶》
北大核心
2025年第3期609-620,共12页
Carsologica Sinica
基金
鄂西山区典型顺层岩质边坡开挖条件下稳定性分析及影响范围预测研究(DQKJ2023-4)。
关键词
滑坡
SBAS-InSAR
随机森林算法
危险性评估
清江
landslide
SBAS-InSAR
random forest algorithm method
disaster assessment
the Qingjiang River
