Soil moisture constants like saturated water content and field water capacity are crucial in soil science, but can be hard to measure directly. This study in Qinghai Province used different methods to predict these constants from soil properties like organic matter and texture. The results showed that the BP neural network method was the most accurate for estimating saturated water content and capillary water-holding capacity, while the regression method was best for field water capacity. The Rosetta model didn't perform as well. This research helps understand soil water in the Three-River Headwaters Region and can be useful for managing water resources in the area.