降雨诱发的滑坡灾害是全球最常见的灾害链类型之一，在全球变暖和极端气候事件 增加、人口快速增长与经济高速发展的背景下，降雨—滑坡灾害链对社会经济带来的威 胁也进一步加剧。充分认识降雨—滑坡灾害链的危害并准确评估其可能的社会经济影 响，是防灾减灾工作的重要部分之一，建立准确的降雨—滑坡灾害链风险定量评估模型 极为必要。然而受制于降雨—滑坡灾害链本身的特性以及滑坡数据可用性不足，现有的 对于降雨—滑坡灾害链的研究多数仍然聚焦于致灾因子和孕灾环境本身，没有形成统一 的风险评估逻辑体系。因此，综合考虑影响区域灾害风险的多元影响因素，并建立起准 确可靠的风险评估体系，仍然是降雨—滑坡灾害链研究中亟需解决的问题。本文以贵州 省毕节和六盘水两市的降雨诱发滑坡事件为例，以灾害系统功能体系为逻辑框架和理论 依据，开展的工作与主要结论如下： 

    （1）基于研究区 103 个降雨滑坡点位数据，选取 4 个类别共 10 个滑坡敏感性评价 指标，采用信息量、频率比、频率比—随机森林耦合方法进行敏感性评估，并采用 ROC 曲线（AUC 值）进行评估效果验证。结果表明三种方法 AUC 值均大于 0.5，评估结果具 有可信度，其中信息量方法的评估效果最好（AUC=0.737）。评估结果显示，研究区降雨 滑坡敏感性分区与水系和道路的分布密切相关。根据评估结果的对比，选取基于信息量 方法的敏感性评估结果作为后续研究中脆弱性部分环境因子的表征变量。

    （2）在 103 个滑坡事件中选取有明确发生时间的降雨滑坡事件 66 起，根据 MSWEP 降水数据获取滑坡点的降雨序列，采用 I-D 和 E-D 阈值方法建立研究区降雨滑坡阈值曲 线（R2 分别为 0.2 和 0.85）。之后采用核密度估计进行事件连续概率的计算，获得连续概 率密度分布函数，采用 Logistic 模型进行曲面拟合，得到 I-D 和 E-D 阈值方法的降雨—滑 坡连续概率阈值曲面，结果均具有较高的信度（R2 分别为 0.944 和 0.923；RMSE 分别为 0.065 和 0.078）。 

    （3）选取已有滑坡清单中 50 起具有直接经济损失记录的滑坡事件，提取灾害点的 GDP 和人口密度值，建立脆弱性评估滑坡清单。利用线性回归方法（广义线性回归）和 非线性回归方法（随机森林，XGBoost，GBDT）选取不同的致灾因子变量建立区域直接 经济损失脆弱性评估模型。结果显示不同模型对于不同区间的损失数据评估存在效果上 的差异，但总体评估效果上基于随机森林模型的脆弱性评估模型较为优良（R2 =0.523， RMSE=6.45）。基于评估结果，选取随机森林模型作为风险评估的损失预测模块方法。 II 

    （4）基于对致灾因子危险性，孕灾环境敏感性以及承灾体脆弱性的分析，构建了针 对贵州省毕节、六盘水两市的降雨—滑坡灾害链直接经济损失定量风险评估模型。结果 表明，研究区内直接经济损失风险的空间分布格局呈条带状和点状相结合的分布特征， 与道路、河流以及山区地区人口相对稀疏的分布格局一致。直接经济损失量在单位网格 （1km2 ）的最大值为 34.769 万元。不同县（市、区）之间也存在显著的差异，其中网格 最大直接经济损失风险出现在六盘水钟山区与毕节市七星关区，20 年重现期县域内网格 平均直接经济损失风险分别为 19.96 万元/km2和 18.66 万元/km2 。不同县（市、区）的直 接经济损失风险变化趋势也存在差异。 

      本文基于灾害系统的理论框架，融合降雨滑坡阈值、敏感性和脆弱性评估，提出了 区域降雨—滑坡直接经济损失风险评估模型，该模拟既能实现降雨—滑坡灾害链直接经 济损失风险的定量评估，同时考虑的因素比传统的滑坡风险评估因素更为全面，完善了 对滑坡风险评估影响因素的认识，具有一定的创新性。同时建立了以滑坡清单为核心的 风险评估体系，与灾害管理与风险管理部门的实际工作相适应，在滑坡清单的基础上建 立的风险评估方法，可为防灾减灾和灾害风险管理工作提供有价值的参考。

Rainfall-induced landslides are one of the most common geological hazards worldwide, with a wide range of socio-economic impacts. In the context of global warming and increasing extreme weather events, with the concentration of population and development of economics, rainfall-induced landslide disaster-chains will pose a greater threat to socio-economic security. Adequate understanding of the hazards of rainfall-landslide and accurate assessment of their possible socio-economic impacts is one of the important parts of disaster prevention and mitigation, so an accurate quantitative risk assessment model of rainfall-landslide needs to be established. However, due to the characteristics of the rainfall-landslide and the lack of landslide data, most of the existing studies on rainfall-landslides still focus on the causative factors and the disaster-inducing environment, and no unified risk assessment approaches has been formed. Therefore, it is still an urgent problem in the study of rainfall-landslide to consider multiple influencing factors affecting regional disaster risk and to establish an accurate and reliable risk assessment system. Taking the rainfall landslide event in Bijie and Liupanshui, Guizhou Province as an example, the work carried out and the main conclusions of this paper, using the functional system of the disaster system as the logical framework and theoretical basis, are as follows: (1) Based on the data of 103 rainfall landslide pixels in the study area, a total of 10 landslide susceptibility assessment factors in 4 categories were selected, and the information value, frequency ratio, and frequency ratio-random forest methods were used for susceptitbility assessment. ROC curve (AUC value) was then used to verify the assessment effect. The results show that all three methods have AUC values greater than 0.5, indicating that the assessment results have credibility, among which the information value method has the best assessment effect (AUC=0.737). The assessment results show that the rainfall landslide susceptibility zoning in the study area is closely related to the distribution of water systems and roads. Based on the comparison of the assessment results, the susceptibility assessment results of the information value method were selected as the characterization variables for the environmental factors of the vulnerability assessment part in the subsequent study.

(2)Among 103 landslide events, 66 rainfall landslide events with clear occurrence time were selected, with the use of MSWEP precipitation data to obtain rainfall sequences at landslide pixels. I-D and E-D threshold methods were used to establish rainfall landslide threshold curves (R2 =0.2&0.85) in the study area. Then the kernel density estimation was used for the calculation of the continuous probability of events to obtain the continuous probability density distribution function, and the logistic model was used for surface fitting to obtain the continuous probability threshold surfaces of rainfall landslides for the I-D and E-D threshold methods, and the results both had high confidence (R2 =0.944&0.923; RMSE=0.065&0.078). (3) Fifty landslides pixels with direct economic losses recorded in the existing landslide inventory were selected, and the GDP and population density values of the disaster pixels were extracted to establish a vulnerability assessment landslide inventory. Linear regression (generalized linear regression) and nonlinear regression (random forest, XGBoost, GBDT) methods were used to establish regional direct economic loss vulnerability assessment models using selected assessment factors. The results show that there are differences in the effectiveness of different models for the assessment of loss data in different intervals, but the overall assessment effect of the random forest model is better (R2 =0.523, RMSE=6.45). Based on the assessment results, the random forest model was selected as the loss prediction method for risk assessment module in latter research. (4) A quantitative risk assessment model of direct economic loss risk in the rainfall-induced landslides was constructed for the cities of Bijie and Liupanshui in Guizhou Province based on the analysis of the factors of hazardouness, susceptibility and vulnerability. The results show that the spatial distribution pattern of direct economic loss risk in the study area is characterized by a combination of strip and point distribution, which is consistent with the relatively sparse distribution pattern of roads, rivers and population in mountainous areas. The maximum value of direct economic loss in the unit grid (1km2 ) is 347.69 million yuan. There are also significant differences among different counties (cities and districts), where the maximum direct economic loss risk of the grid occurs in Zhongshan District, Liupanshui, and Qixingguan District, Bijie, with the average direct economic loss risk of the grid in the county of 199.6 and 186.6 thousand yuan/km2 , respectively, during the 20-year recurrence period. the change trend of direct economic loss risk also varies among different counties (cities and districts). The main innovation of this study is to integrate the original assessment modules of rainfalllandslide threshold study, susceptibility assessment, and vulnerability assessment into the theoretical framework of disaster functional system, and propose a complete rainfall-induced landslide direct economic loss risk assessment model considering various factors, which improves the quantification of rainfall-induced landslide risk assessment research and improves the understanding of the influencing factors of landslide risk. At the same time, an inventorybased risk assessment system is established, which is compatible with the practical work of disaster management and risk management departments. The inventory-based risk assessment is expected to provide valuable reference for practical disaster prevention and mitigation and disaster risk management work.