环境因子空间特征约束的区域滑坡敏感性模糊逻辑分析方法

Fuzzy Logic Approach for Regional Landslide Susceptibility Analysis Constrained by Spatial Characteristics of Environmental Factors

中国西部山区灾难性滑坡事件频繁发生,滑坡敏感性分析已成为灾前科学预警和主动防范的必要手段。传统滑坡敏感性分析方法中单一知识驱动模型对滑坡灾害环境因子定权主观性强,数据驱动模型过分依赖样本数据的质量及数量。针对上述问题,提出了一种环境因子空间关联特征与启发式模糊逻辑模型耦合的区域滑坡敏感性分析方法,通过灾害环境因子滑坡频率比与信息熵权等空间统计指标,显式描述滑坡灾害环境因子的贡献度与空间分布特征,以此约束多因子耦合的区域滑坡敏感性计算。选择中国重庆市奉节县内的灾害多发地带进行验证评估,实验结果表明,所提方法优于单一的信息量模型、信息量-逻辑回归模型方法。

Objectives:The complex geological conditions in the mountainous areas of western China,with strong internal and external dynamics effects,make catastrophic landslides frequent.The analysis of landslide susceptibility has become a necessary means for scientific early warning and active prevention before disasters.In the traditional landslide susceptibility analysis method,the general calculation accuracy of the single knowledge-driven model is limited,and the weight of the impact factor is highly subjective.The data-driven model also relies too much on the quality and quantity of sample data,and the heterogeneity of the landslide disaster environment is prominent.Methods:In order to overcome the problems of limited quantity and quality of sample data and large differences in landslide disaster environment,we propose a regional landslide susceptibility method that couples the contribution weight of landslide disaster environmental factors and heuristic knowledge fuzzy logic model.The proposed method uses spatial statistical indicators such as the historical landslide frequency ratio and the information entropy weight of the landslide disaster environmental factors to explicitly describe the contribution and spatial distribution characteristics of the landslide disaster environmental factors,which measures the constraint relationship and the mapping structure between multI-factors and landslides,and realizes multi-factor coupling regional landslide susceptibility.Results:The experiment selects the disaster-prone areas in Fengjie,Chongqing for verification and evaluation;The results show that the proposed method has a more uniform and reasonable partition area,with an area under curve(AUC) value of 0.854,and the best prediction accuracy,than single information value(IV)model and information value and logistic regression(IVLR) model,which ensures the reliability and accuracy of the method.Conclusions:The proposed method overcomes the strict requirements of landslide susceptibility analysis on the number of historical observation samples,improves the accuracy of landslide susceptibility analysis through a hierarchical stacking strategy,and provides reliable technical support for the susceptibility analysis of large-scale.