基于物理过程不确定性的降雨诱发浅层滑坡易发性快速区划:GIS-FORM技术开发与应用

Fast zoning of rainfall-induced shallow landslide susceptibility based on physical process uncertainty:development and application of GIS-FORM

高效绘制降雨诱发山体滑坡易发性分布图是区域尺度滑坡预测和地质灾害早期预警的重要技术手段。考虑降雨导致地表径流情况,采用无限长斜坡模型耦合孔隙水压力(PWP)变化规律,提出一种能够模拟区域边坡湿润锋深度随时空变化的简化瞬时降雨入渗模型PRL-STIM(physically-based probabilistic modelling of rainfall landslide using simplified transient infiltration model)。该模型通过引入一阶可靠度方法(FORM)开展高效概率计算,模拟区域尺度内地质条件的内在不确定性,实现区域尺度滑坡失效概率空间分布快速评估。以地理信息系统(GIS)为平台,采用Python程序语言,开发一种能够自动执行区域滑坡易发性分析的PRL-STIM v1.0工具。以我国甘肃娘娘坝镇于2013年7月发生的降雨诱发区域浅层滑坡为例,进行所提模型在区域滑坡易发性评估中的工程应用。结果表明,考虑50%失效概率阈值可有效地表征区域滑坡灾害易发性分布;简化的瞬时降雨入渗模型能够有效识别高风险滑坡区域,设置20 m缓冲区后的确定性和概率预测精度分别达到79%和81%;此外,相比于TRIGRS模型72%的预测精度,PRL-STIM预测精度达到75%,且服从相关非正态分布的岩土参数对降雨滑坡预测与区域尺度滑坡易发性有显著影响。

Efficient mapping of rainfall-induced landslide susceptibility is crucial to the success of regional-scale landslide prediction and the early warning of geological hazards.In this paper,a physically-based probabilistic modelling tool,herein named the probabilistic rainfall-induced landslide using simplified transient infiltration model(PRL-STIM),was proposed to deal with the fast mapping of landslide susceptibility at regional scales.This modelling tool integrates the infinite slope model with considerations of rainfall-induced pore water pressure(PWP)and surface runoff.The first-order reliability method(FORM)for efficiently performing probabilistic computations is employed to simulate the geotechnical and geological uncertainties.The proposed PRL-STIM v1.0 tool is developed based on the Python programming language integrating with the Geographic Information System(GIS)framework.Validation of the proposed model is illustrated by an engineering case study of the rainfall-induced regional shallow landslides that occurred in July 2013 in Niangniangba,Gansu Province,China.The analysis results demonstrate that the adoption of the 50%failure probability threshold can effectively characterize the region’s landslide hazard susceptibility distribution.High-risk landslide areas can be well identified,with deterministic and probabilistic prediction accuracies reaching 79%and 81%,respectively,when a 20m buffer zone is used.Furthermore,it is shown that the probabilistic prediction accuracy of the rainfall-induced landslide susceptibility by PRL-STIM achieves 75%,surpassing the 72%prediction accuracy of the TRIGRS model based on the Richards equation,and it is worth noting that non-normal distributions of random geotechnical parameters may exert a significant influence on the predicted landslide susceptibility.