Soil moisture is a fundamental state variable in the terrestrial hydrological cycle and a key regulator of land–atmosphere interactions, particularly in alpine headwater regions. In land surface models, soil moisture simulations are highly sensitive to physical parameterization schemes and parameter values, leading to substantial uncertainties over high-elevation areas. In this study, an AMSO-based parameter optimization framework is developed to improve soil moisture simulation in the Noah-MP land surface model over the Sanjiangyuan Region of China, constrained by in-situ station observations. A set of sensitivity experiments with different physical parameterization schemes was first conducted to quantify their impacts on simulated soil moisture and to identify an optimal scheme at the regional scale. Subsequently, key soil-related physical parameters associated with soil hydraulic and thermal processes were optimized using the Adaptive Multi-Strategy Optimization (AMSO) algorithm, with station-based soil moisture observations serving as the optimization target. The optimized parameter set leads to a significant improvement in soil moisture simulation performance relative to the default model configuration, as evidenced by increased correlation coefficients and reduced error metrics at observation sites. Moreover, the optimized model exhibits enhanced robustness across different land surface conditions. The results demonstrate that coupling physical parameterization screening with observation-constrained intelligent optimization is an effective strategy for reducing parameter uncertainty and improving soil moisture simulations in alpine regions, providing a reliable parameter basis for hydrological and land–atmosphere coupling studies.