Soil is an indispensable component of ecosystems, serving as the foundation for plant growth and development. Soil water content directly affects the water and nutrient cycles within the soil-vegetation ecosystem, thereby determining the growth status of plants, which leads to is a strong correlation between vegetation growth and soil water content. To further investigate this relationship, this study selected ground hyperspectral data of three high-quality forage grass species (Artemisia frigida willd, Goosegrass and Tall fescue) in Zhaosu County as research samples. Fourteen types of spectral transformation methods, including inverse, derivative, and logarithmic, were applied to identify the characteristic spectral bands that are highly correlated with the soil water content by correlation analysis. Subsequently, a predictive model for soil water content was constructed based on vegetation hyperspectral data, with the aim of screening sensitive wavelengths and identifying optimal inversion models. Additionally, the CASA model was employed to calculate the actual net primary productivity (ANPP) of grassland, aiming to analyze the degree of correlation between soil water content and vegetation productivity. The results showed that: (1) The spectral reflectance characteristics of the three high-quality forage grass species in Zhaosu County were similar in the visible light bands but differed significantly in the near-infrared bands. Moreover, the vegetation spectral reflectance showed a clear decreasing trend with the increase of soil water, except for the 25%-30% soil water content range. (2) After undergoing 14 types of spectral transformations, the correlation between spectral reflectance of the three high-quality forage grass species in Zhaosu County and soil water content has been identified. Particularly, under the spectral transformation forms of R″, (1/R)', (log R)', (log 1/R)', (R^1/2)″, the correlation with soil water content was relatively high, with a large number of characteristic bands. (3) Among the eight spectral transformation forms such as (1/R)', (1/R)″, (log 1/R)″, and (R^1/2)″, the models for estimating soil water content showed a good accuracy under the transformations of (R^1/2)″ and (log 1/R)″. (4) Over the past 20 years, the ANPP of grassland in Zhaosu County exhibited a significant decreasing trend, with the degraded areas of grassland concentrated in the plains. However, the decrease in vegetation productivity was a result of multi-factorial influences, with soil water content being only a weak influencing factor. In conclusion, this study provides a new approach for soil water content inversion and its relationship with vegetation productivity. These findings offer valuable guidance for the designation of strategies for sustainable land use and management.