基于数据驱动的降雨型浅层滑坡易发性时空建模方法  

Spatiotemporal modeling of rainfall-induced shallow landslidesusceptibility based on data-driven methods

作　　者：王毅[1] 陈曦 方志策 杜宝裕 Wang Yi;Chen Xi;Fang Zhice;Du Baoyu(School of Geophysics and Geomatics,China University of Geosciences,Wuhan,Hubei 430074;Hubei Geological Survey,Wuhan,Hubei 430034;Remote Sensing Application Technology Center of Hubei Geological Bureau,Wuhan,Hubei 430034)

机构地区：[1]中国地质大学(武汉)地球物理与空间信息学院,湖北武汉430074 [2]湖北省地质调查院,湖北武汉430034 [3]湖北省地质局遥感应用技术中心,湖北武汉430034

出　　处：《资源环境与工程》2024年第5期612-619,共8页Resources Environment & Engineering

基　　金：国家自然科学基金联合基金项目“地质环境遥感大数据智能解译”(U21A2013);湖北省地质局科技项目“基于多源数据的自然资源遥感监测体系研究”(KJ2022-7)、“多灾种应急动态风险区划展示平台建设”(KJ2022-58)。

摘　　要：传统滑坡易发性方法在计算空间概率时,通常忽视了影响因子的动态变化特征及地形单元之间的空间依赖关系。为了解决该难题,提出一种联合时空关系的广义加性模型,在江西省南部区域开展降雨型浅层滑坡易发性时空建模,预测给定时间范围内特定地形单元发生滑坡灾害的概率。首先,利用皮尔逊相关系数和基于赤池信息准则的序列前向特征选择方法对滑坡易发性影响因子进行评价与筛选;随后,联合时空关系构建伯努利广义加性时空模型开展易发性动态预测;最后,通过时空交叉验证方法评估模型的时空预测性能。结果表明,该模型具备优异的拟合性能和预测能力,在不同训练数据百分比下的预测性能非常稳定,其平均预测精度能够达到0.881。Traditional landslide susceptibility methods often ignore the dynamic characteristics of influencing factors and the spatial dependencies between terrain units.To address this issue,a generalized additive model combining spatiotemporal relationships is proposed for spatiotemporal modeling of rainfall-induced shallow landslide susceptibility in southern Jiangxi Province.This method predicts the probability of landslide occurrence within specific terrain units over a given time frame.Initially,Pearson correlation coefficients and a sequential forward feature selection method based on the Akaike Information Criterion(AIC)are employed to evaluate and select impact factors of landslide susceptibility.Subsequently,a Bernoulli generalized additive spatiotemporal model is constructed by integrating spatiotemporal relationships to conduct dynamic susceptibility predictions.Finally,the spatiotemporal cross-validation method is used to assess the model s spatiotemporal predictive performance.The results indicate that the model demonstrates excellent fitting and predictive capabilities.The spatiotemporal random cross-validation results show that the model s predictive performance remains stable across different percentages of training data,with an average prediction accuracy of 0.881.

关 键 词：降雨型浅层滑坡 滑坡易发性 时空建模 广义加性模型 时空随机交叉验证 

分 类 号：P642.22[天文地球—工程地质学]