滑坡灾害是一种常见且严重的地质灾害，在世界范围内广泛分布，降雨诱发的土质滑坡是滑坡中最常见、分布最为广泛的一类。土质滑坡的发生是地形、降雨、土体性质等多种因素综合作用的结果，土体性质决定了斜坡土体的初始强度和渗透能力，以及在外界触发因素作用时斜坡状态的变化方向，在降雨触发土质滑坡的过程中起到桥梁纽带的作用。在土体的不同粒径组分当中，粘粒组因具有较大的比表面积、较强的可塑性，对斜坡土体的强度和渗透性等性质具有相对粉砂以及更大颗粒而言更复杂的控制规律。且不同的粘粒矿物类型具有不同的晶格形态，晶面间距和晶面联结形式的差异导致矿物的颗粒大小以及吸水膨胀后结合水膜的厚度呈现较大差异，进一步复杂了粘粒组分对土体性质乃至土质滑坡的发生过程的影响。
滑坡的敏感性是对滑坡事件在特定地点发生可能性的表达，是滑坡预测预报和防灾减灾相关决策的重要参考。滑坡敏感性评价方法经历了从定性到定量，从单体到区域的过程，区域定量评价是趋势。土体性质在土质滑坡敏感性评价中的重要性已经在基于物理过程的单体斜坡稳定性模型和小流域数值模型中得到广泛证明，然而其在基于统计方法的区域滑坡敏感性评价中的却并未得到广泛应用，所使用的土体性质因子一般也以与斜坡失稳过程关系并不直接的土壤类型、质地分类、深度、容重等为主。粘粒含量及矿物组成影响的土体强度和渗透性在土质滑坡的区域敏感性评价中的作用，以及与当前得到应用的这些土体性质因子相比是否更优还不清楚，这可能导致局部地区的敏感性评价结果存在较大误差。
本文通过分析我国季风区土体粘粒含量和主要粘土矿物空间变化规律，人工配置了不同矿物组成和多种粘粒含量的样本进行了抗剪强度测定和渗透性参数计算，进而对粘粒矿物组成及其含量变化对斜坡土体抗剪强度和渗透性的影响规律进行了分析。在该规律的基础上，选择两个土质滑坡易发且在粘粒含量和矿物组成方面具有代表性的省份（陕西省和云南省）作为研究区，使用临界相对饱和深度（Mc）和饱和渗透系数（K）作为粘粒组影响的抗剪强度和渗透性的区域表征参数，结合已有的基础土体性质空间分布数据，获取了Mc和K的空间分布。随后在此空间分布基础上，结合研究区历史滑坡数据、基础地理环境数据和基础土体性质数据，利用广义加性模型（GAM）和随机森林模型（RF）在两地分别构建了区域滑坡敏感性评价模型，使用定量评估手段对粘粒含量因子以及粘粒组成含量影响的土体强度和渗透性表征因子（Mc和K）在区域滑坡敏感性评价模型中的表现以及对区域滑坡敏感性图的空间优化进行了比较和分析。主要研究结论如下：
（1）土体的抗剪强度和抗液化性随粘粒含量增加呈现幂数增长趋势。伊利石矿物的薄片状晶格形态使得伊利石试样拥有更低的有效内摩擦角，因而在相同粘粒含量条件下，以伊利石为主要粘粒矿物的土体具有更弱的抗剪强度和抗液化能力。在试验结果的基础上分别构建了以高岭石和伊利石为主要矿物的土体有效内摩擦角与粘粒含量的经验关系。
（2）土体的渗透性随粘粒的变化在粘粒含量20% - 25%附近出现拐点，拐点之后渗透性随粘粒矿物的增加而增加。粘粒含量增加导致的土颗粒比表面积的增加对渗透性的负效应与同时发生的比重降低、最小孔隙比增加对渗透性正效应的耦合作用是渗透性拐点出现的原因。伊利石矿物吸水膨胀降低了土体的孔隙连通率，使得以伊利石为主要矿物的土体渗透性比以高岭石为主要矿物的土体低约37.49% - 74.49%，且二者的渗透性差距在粘粒含量低于20%时随粘粒含量的增加而扩大。
（3）陕西和云南土体中的粘粒含量随着水热条件从北到南、从高到低呈现增加趋势。粘粒中的主要粘土矿物由北向南呈现伊利石>蛭石>高岭石的不稳定到稳定的空间变化特征。伊利石和高岭石分别为陕西和云南粘粒中的主要粘土矿物。Mc空间分布显示陕西和云南处于临界失稳状态的区域面积分别占全区面积的21%和23%，且主要分布在坡度大于40°的区域；处于稳定状态的区域分别占31%和38%，主要分布在坡度低于15°的区域。渗透性方面云南境内土体渗透性水平分布相对比较平均，渗透性最低的区域位于滇东喀斯特高原东南部。陕西境内的低渗透区域主要分布在关中平原和陕南的汉中盆地和巴山山区，陕北黄土区的土体渗透性相对较好。
（4）粘粒含量因子以及粘粒矿物组成及含量影响的土体强度和渗透性因子（Mc和K）参与构建的区域滑坡敏感性模型总体预测性能表现优秀。粘粒含量的参与使两地基于不同方法的模型总体预测性能提升了1.34% - 10.13%，与以往区域滑坡敏感性评价中通常使用的土体性质因子（深度和容重）相比也毫不逊色。Mc和K参与的模型则在单纯粘粒含量参与的基础上将模型性能进一步提升了0.34% - 13.91%。Mc和K参与生成的滑坡敏感性空间分布图具有更高的准确性和更低的过估比例，且低坡度区和城市集中用地区的评估结果拥有更好的空间合理性。而仅考虑粘粒含量的模型虽然整体预测性能有所上升，但对区域滑坡敏感性评估结果的空间优化则相对有限。
（5）粘粒含量是对滑坡敏感性模型准确性贡献度最高的基础土体性质因子，且在陕西GAM模型的所有13个预测因子中排在前半。而在Mc和K参与构建的模型中，Mc对模型预测准确性的影响在所有13个因子中居于首位，K因子也排在前半。Mc和K中额外的粘粒和矿物组成信息进一步拉开了其与模型中次一级因子的重要性差距，对模型预测的准确性产生了更多贡献。
（6）粘粒含量、容重、砂粒含量与滑坡敏感性的响应关系呈现先增后减的非单调特征，Mc、K以及沉积物厚度与滑坡敏感性呈现单调的负相关特征。各土体性质因子对滑坡的易感区间在陕西和云南存在比较显著的区域差异，云南滑坡敏感区域相比陕西具有相对更松、更粘、更粗糙的土体特性，这与两地本身土体性质的区域差异相一致。
（7）本研究从物理机制和区域统计两个角度的研究结果都印证了土体中粘粒组分在土质滑坡敏感性分析中的重要意义。以土体抗剪强度和渗透性为切入点，分析了土体中粘粒含量及矿物组成变化对土质斜坡稳定性的影响规律，进一步加深了对降雨引发的土质滑坡的过程和机制的理论认识。鉴于降水导致的土质滑坡的普遍性和易发性，建议在以后的区域滑坡敏感性评价实际工作中尽量纳入粘粒组分相关因子，以提升评价结果的可靠性和物理可解释性。与斜坡失稳过程关系更直接更密切的因子（比如粘粒组成和含量影响的土体强度和渗透性因子）可以更多地提升评价结果的准确性和空间合理性。此外，各土体性质因子的复杂响应关系反映了在使用土体性质因子进行区域滑坡敏感性评价时，使用可同时反映线性和非线性关系的模型的必要性。

Landslides are common and serious geological disasters and are widely spread around the world. Rainfall-induced soil landslides are the most common and widely distributed type of landslides. The occurrence of soil landslides is the result of the combined effect of various factors such as topography, rainfall, and soil properties. The soil properties play a role of bridge in the triggered process of rainfall-induced soil landslides by determining the initial strength and permeability of the soil, as well as the change direction under the action of external triggering factors. Among the different particle size components of soil, clay particles have more complex control laws on the strength and permeability of soil than silt, sand, and larger particles due to its large specific surface area and strong plasticity. Moreover, different basal spacings and interplanar connection forms lead to significant differences in the particle size of clay minerals and the thickness of the bound water film after water swelling, which further complicates the influence of clay on soil properties and the occurrence process of soil landslides.
Landslide susceptibility refers to the possibility of landslide occurrence in a specific environment, which is an important reference for decision-making of landslide prediction, prevention, and mitigation. The evaluation methods of landslide susceptibility have experienced a process from qualitative to quantitative, from single slope evaluation to regional assessment, and the regional quantitative evaluation is the dominant trend. The importance of soil properties in the susceptibility assessment of soil landslides has been widely proved by physical single slope stability models and small watershed numerical models. However, its application in the statistical regional landslide susceptibility assessment is still insufficient. The soil properties factors that have been considered are usually not directly related to the slope instability process, such as soil types, texture classification, soil depth, and bulk density. The role of soil strength and permeability controlled by clay content and mineral composition in regional soil landslide susceptibility assessment is still unclear, which may lead to large errors in some local susceptibility results.
In this study, samples with different mineral compositions and various clay contents were artificially configured to simulate the spatial variation of clay content and main clay minerals in Chinese monsoon region. The shear strength and the permeability parameters were then tested and calculated. The changing in soil shear strength and permeability with clay mineral composition and content were summarized. Based on the experimental results, two representative landslide-prone provinces (Shaanxi and Yunnan) in terms of clay content and mineral composition were selected as the study areas, and the spatial distribution of the critical relative saturation depth, Mc, and the soil saturation permeability coefficient, K, were obtained combined with the existing basic soil spatial data. The regional landslide susceptibility assessment models were finally constructed in the two study areas based on the generalized additive model (GAM) and random forest model (RF). Through a series of quantitative evaluation methods, the performance of clay content, Mc, and K in the model and their spatial optimization in landslide susceptibility maps were compared and analysed based on a series of quantitative evaluation methods. The following main conclusions could be drawn:
(1) The shear strength and liquefaction resistance of samples in this study showed a power increase trend with the increase of the clay content. The lamellar lattice morphology of illite minerals makes the illite samples have a lower effective internal friction angle, thus the illite samples performed relatively weaker shear strength and liquefaction resistance. Based on the test results, the empirical relationship between the effective internal friction angle and the clay content of soils with kaolinite and illite as the main minerals was constructed respectively.
(2) The change of soil permeability with the change of clay showed an inflection point near the clay content of 20% - 25%, and the permeability increased with the increase of clay after the inflection point. The coupling effect of the negative effect on permeability brought by the increase of the specific surface area and the positive effect brought by the decrease of the specific gravity and the increase of the minimum void ratio could be one of the reasons for the appearance of the permeability inflection point. The water-swellable illite mineral reduces the pore connectivity of the sample, so that the permeability of the illite samples were about 37.49% - 74.49% lower than that of the kaolinite samples under the same mineral addition ratio. Furthermore, the permeability gap between samples with different main mineral types widened with the increase of clay content when the clay content is lower than 20%.
(3) The clay content in the soils of the SX and YN showed an increasing trend from north to south and from high to low with the change of hydrothermal conditions. The main clay minerals in the clay show the unstable to stable spatial variation of illite > vermiculite > kaolinite from north to south. Illite and kaolinite are the main clay minerals in SX and YN, respectively. The spatial distribution of Mc showed that the critical instability areas in SX and YN account for 21% and 23%, respectively, and were mainly distributed in the region with a slope greater than 40°. The areas in the stable state accounted for 31% and 38%, respectively, and were mainly distributed in areas with slopes less than 15°. The distribution of permeability in Yunnan is relatively evenly, and the region with the lowest permeability is located in south-eastern YN. The low permeability areas in Shaanxi province are mainly distributed in Guanzhong Plain, Hanzhong Basin, and Bashan mountain, while the permeability of soil in northern SX is relatively higher.
(4) The overall prediction performance of the regional landslide susceptibility models constructed with the participation of clay content, Mc and K showed excellent. The participation of clay content improved the overall prediction performance of the models based on different methods in the two places, ranging from 1.34% to 10.13%, which is not inferior to the soil property factor commonly used in the previous regional landslide susceptibility evaluation (soil depth and bulk density). Mc and K further improve the model performance by 0.34% - 13.91% on the basis of clay content. The landslide susceptibility map generated with Mc and K performed higher accuracy and lower overestimation ratio. The assessment results of low-slope areas and urban intensive use areas performed better spatial rationality. However, the spatial optimization of landslide susceptibility map that considering the pure clay content was limited.
(5) The clay content was showed the soil factor with the highest contribution in the model accuracy, and even ranked in the first half of all 13 predictors in the SX-GAM models. For models generated with Mc and K, the contribution of Mc on the model predictive accuracy came on the top, and K ranked the first half in the SX-GAM models and the YN-RF models. The additional mineral information in Mc and K further widened the contribution gap between them and the second-order factors in the models.
(6) The response relationship between clay content, bulk density and sand content and landslide susceptibility presented non-monotonic characteristics of first increase and then decrease, and Mc and K, as well as the soil depth showed monotonic negative characteristics. There are significant regional differences between SX and YN in the susceptibility interval of soil properties to landslides. Compared with SX, the landslide-sensitive area in YN has relatively looser, stickier and rougher soil properties, which is consistent with the regional differences in soil properties between the two regions.
(7) The results in this study from physical experiments and regional statistics both confirmed the importance of soil properties, especially the clay component in the susceptibility assessments of soil landslides. This study analysed the influence of clay content and mineral composition on the stability of soil slope based on the shear strength and permeability of soil, and further deepened the theoretical understanding of the process and mechanism of soil landslides induced by rainfall. In view of the ubiquity and susceptibility of rainfall-induced soil landslides, it is suggested that clay related factors should be included in the actual work of regional landslide susceptibility evaluation in order to improve the reliability and physical interpretability of the evaluation results. Factors that are more directly and closely related to the slope instability process (such as soil strength and permeability factors affected by clay composition and content) can further improve the accuracy and spatial rationality of the evaluation results. Moreover, the complex response relationships of individual soil properties highlight the importance of using models that reflect both linear and nonlinear relationships when using soil factors for regional landslide susceptibility assessments.