基于IV-RF模型地质灾害易发性及可解释性分析

Susceptibility and Interpretability Analysis of Geological Hazards Based on the IV-RF Model

摘要: 云南省地质构造复杂，地质灾害多发易发，开展地质灾害易发性预测对防灾减灾意义重大。本研究以镇雄县为研究区域，采用信息量模型(Information Value, IV)量化了各因子属性值对灾害的影响，并耦合随机森林模型(Random Forest, RF)建立评价模型同时，采用沙普利可加和解释方法(Shapley Additive Explanations, SHAP)对模型进行可解释性分析，以明确各因子对地质灾害的影响。研究收集了镇雄县地质灾害点及16个影响因子数据，经处理后用于模型构建与验证。结果表明，IV-RF复合模型ROC曲线下面积(AUC)达到0.93，具有较高的预测精度和可靠性，且地质灾害易发性呈现显著空间差异，极高和高易发区集中在县域中部、南部及东部部分区域。SHAP分析显示，高程(DEM)、距道路距离(Distance to Roads, DistRd)、归一化植被指数(NDVI)等因子对地质灾害影响显著，各因子信息量值增大通常与地质灾害易发性增加相关。本研究为区域地质灾害风险管理提供了科学有效的工具和重要参考。

Abstract: Yunnan Province is characterized by complex geological structures and a high frequency of geological hazards. Accurate prediction of geological hazard susceptibility is of great significance for disaster prevention and mitigation. Taking Zhenxiong County as the study area, this study uses the Information Value (IV) model to quantify the influence of attribute values of each factor on hazards, and couples it with the Random Forest (RF) model to construct an evaluation model. Meanwhile, the Shapley Additive Explanations (SHAP) method is employed to conduct interpretability analysis of the model, so as to clarify the impact of each factor on geological hazards. Data on geological hazard locations and 16 influencing factors in Zhenxiong County were collected, processed, and then used for model construction and validation. The results show that the area under the ROC curve (AUC) of the IV-RF hybrid model is 0.93, indicating high prediction accuracy and reliability. In addition, geological hazard susceptibility shows significant spatial heterogeneity, with very high and high susceptibility zones concentrated in the central, southern and parts of eastern regions of the county. SHAP analysis reveals that factors such as elevation (DEM), distance to roads (DistRd), and Normalized Difference Vegetation Index (NDVI) impose significant impacts on geological hazards, and the increase in information values of each factor is generally positively correlated with the increase in geological hazard susceptibility. This study provides a scientific and effective tool and an important reference for regional geological hazard risk management.