As an important function for determining both the hydraulic and mechanical properties of soils, the soil–water characteristic curve (SWCC) based on the particle size distribution underestimates the water content at high suction head. In this study, the quantitative relationship between the clay content and residual pore fraction is established based on soil gradation parameters (uniformity coefficient $C_u$ and curvature coefficient $C_c$); the van Genuchten equation is employed to describe the soil–water characteristic curve; and then a semi-physical and semi-statistical method based on clay content is proposed to predict water-holding of unsaturated soils, especially for the soil with higher clay contents or low water contents (high matric suctions). Eighty-seven soil samples containing nine texture classes from the unsaturated soil hydraulic database (UNSODA) are used to verify the proposed approach. Two traditional models proposed by Hwang et al. (2011) and Meskini-Vishkaee et al. (2014) are introduced to compare the predictive effect with the improved method. Although three methods all show good performance, the improved prediction method based on uniformity coefficient ($C_u=d_{60}/d_{10}$) has a better estimation accuracy than the two others according to statistical parameters, including root-mean-square error (RMSE), sum of square errors (SSE), and determination coefficient (R²). The ranking order of average uniformity coefficients in the three soil classes (sandy, silty, and clayey) agrees well with the ranking order of clay contents. Meanwhile, various $d_y/d_x$ are used to characterize soil gradation and have similar goodness of fitting to residual pore fraction. In summary, the improved prediction model can effectively predict soil–water characteristic curves, especially for soil with low water contents (high matric suctions) or high clay contents. The model benefits the predictions of mechanical properties, permeability coefficient, shear strength and slope stability of unsaturated soil at low water contents.

作为确定土壤水力和力学性质的重要函数，基于粒度分布的土水特征曲线（SWCC）低估了高吸水头下的含水量。本研究基于土壤级配参数（均匀系数），建立了粘土含量与残余孔隙率之间的定量关系。$C_{你}$ 和曲率系数 $C_C$); 用van Genuchten方程描述土壤-水特征曲线；然后提出了一种基于粘土含量的半物理和半统计方法来预测非饱和土的持水量，特别是对于粘土含量较高或含水量较低（高基质吸力）的土壤。使用来自非饱和土壤水力数据库 (UNSODA) 的 87 个包含九个纹理类别的土壤样本来验证所提出的方法。Hwang 等人提出的两种传统模型。(2011) 和 Meskini-Vishkaee 等人。(2014) 被引入比较预测效果与改进方法。虽然三种方法都表现出良好的性能，但基于均匀系数的改进预测方法（$C_{你}=d_{60}/d_{10}$) 根据统计参数，包括均方根误差 (RMSE)、误差平方和 (SSE) 和决定系数 (R ² ) ，具有比其他两个更好的估计精度。三个土壤类别（砂质、粉质和粘土）的平均均匀度系数排列顺序与粘土含量排列顺序非常吻合。同时，各种$d_{是}/d_X$用于表征土壤级配，并具有与残余孔隙分数相似的拟合优度。综上所述，改进后的预测模型可以有效地预测水土特征曲线，特别是对于低含水量（高基质吸力）或高粘土含量的土壤。该模型有利于预测低含水量非饱和土的力学性质、渗透系数、抗剪强度和边坡稳定性。