Grasslands, integral to the geosphere-biosphere interface, regulate critical Earth system processes such as carbon sequestration, groundwater recharge, and soil stability. This study investigates how anthropogenic grazing strategies alter multi-layered soil properties (0-200 cm depth) in the Xilin Gol Grassland, Inner Mongolia, China, focusing on soil moisture, organic carbon (SOC), inorganic carbon (SIC), and total nitrogen (TN). Through mechanism-driven mathematical modeling, we quantify interactions between grazing practices and subsurface soil dynamics across lithospheric and hydrogeological layers. Our model simulates vegetation biomass responses to grazing intensity (e.g., continuous vs. rotational grazing) and predicts soil moisture variations under 11 strategy combinations. Field-validated results (R-2 = 0.82) reveal that heavy grazing disrupts soil hydrogeological balance, elevating salinity (15-20%) and depleting SOC (12-15%), whereas moderate grazing enhances SOC (6-8%) while stabilizing moisture levels critical for groundwater recharge. Seasonal moisture predictions (2022-2023) highlight risks of topsoil desiccation and deep-layer salinization under unsustainable practices. This work advances a geoscience-informed framework to optimize grazing policies, mitigate desertification, and sustain soil-water resources in semi-arid regions, aligning with global goals for geosphere resilience and resource sustainability.