In frozen soils, a portion of pore water remains unfrozen due to the effects of capillarity, adsorption, and possibly solute. The variation of the amount of unfrozen water and ice in a frozen soil, which is primarily influenced by subzero temperature, has great impacts on the physical and mechanical behavior of the soil and is critical for broad applications ranging from engineering to climate change. In the present study, the various methods that have been used for determining unfrozen water (and ice) content are comprehensively reviewed. Their principles, assumptions, advantages, and limitations are discussed. It is noted that there is yet no perfect way to accurately quantify unfrozen water content in frozen soils. In addition, the soil-freezing characteristic curve (SFCC) of a typical volcanic soil sampled in the Hokkaido prefecture of Japan is investigated. The unfrozen water content of the prepared soil specimens was measured using a cheap moisture sensor, which is based on the frequency domain reflectometry technique. The temperature of the specimens was determined by a rugged temperature sensor. Different numbers of freeze-thaw (F-T) cycles and different freezing/thawing methods (i.e., one- and three-dimensional) were considered, and their effects on the SFCC were investigated. The experimental results suggest that neither the F-T cycles nor the freezing/thawing methods had significant influence on the measured SFCC. The presented comprehensive review and experimental investigations are of importance for both the scientific and engineering communities.

An indirect method is nowadays considered as an efficient way to obtain soil‐water characteristic curve (SWCC) in engineering application. However, existing indirect models often oversimplify the soil pore and accumulation structure, which are not consistent with the natural soil. For this purpose, a novel granular packing state is obtained based on the relative compaction determined by porosity. A conceptual SWCC model (WANG24) is then established with particle size distribution (PSD) and the equivalent novel granular packing. 62 soils from 7 soil texture classes in the UNSODA database were used to validate WANG24. When comparing with the Mohammadi and Vanclooster (MV11), Arya and Heitman (AH15), and Arya and Paris (AP81) models, WANG24 shows its highest accuracy with lowest average root mean square error (RMSE) of 0.0243 (g·g⁻¹). The capillary and adsorption on SWCC are also analyzed. The absolute errors between the soil water content predicted by equivalent novel granular packing and measured data are smaller than those of other packing states, mostly in the range of 0–0.015 (g·g⁻¹). The soil packing states tend to be closer as the particle size decreases. In addition, the effect of initial void ratio to soil water content and matric head is explained. The model can reasonably describe the complexity of soil accumulation structure and improve prediction accuracy. It can provide a basis and reference for the subsequent hydraulic characterization of unsaturated soils.