Abstract: The soil-water characteristic curve (SWCC) is an important characterization of the water-holding performance and water-vapor transport law of unsaturated soil. Due to its complicated testing process and many influencing factors, it is difficult to fully express it through a series of tests and mathematical models. The aim of this study is to explore the influences of soil type and physical state on SWCC. The abundant test data from worldwide research are collected as the database, and three characteristic values of SWCC (air-entry value, dehumidification rate and residual volumetric water content) are selected as the analytic target. Then, some data analyses are adopted to reveal the influence and sensitivity of different occurrence conditions on the characteristic values, and machine learning methods are employed to analyze the sensitivity. In the database，the material composition (particle gradation, particle size, plasticity index) and occurrence state (compactness, saturated water content, dry-wet cycle and ambient temperature) of the soil are common indicators that affect its water holding capacity. The influence mechanism of above factors presents some great differences but also obvious interrelation. The results of sensitivity analysis indicates that the plasticity index that reflects the clay content and the dry density that represents compactness) are two dominant factors influencing the water retention capacity of unsaturated soils. The distribution ranges of three characteristic values are also provided with considering the influences of two dominant factors, and can present most of test results and guide the engineering application.

Key words: unsaturated soils, soil-water characteristic curve, data mining, influence mechanism, sensitivity analysis