<p>Accurate soil moisture simulation is essential for understanding regional hydroclimate variability and improving climate predictions. This study evaluates the performance of FGOALS-g3 under the Atmospheric Model Intercomparison Project (AMIP) configuration in simulating surface soil moisture (SSM) over southeastern China during 1980–2014. Compared to the ERA5 and ESA-CCI reference datasets, the model exhibits a consistent dry bias across spatial distribution, seasonal cycle, and interannual variability. Replacing the model’s default soil texture with the Global Soil Dataset for Earth System Modeling (GSDE) significantly reduces this bias. The improvement stems from a shift toward finer soil texture, characterized by a 41.25% reduction in sand content, which enhance soil water retention through increasing soil water suction and decreasing hydraulic conductivity. However, the efficacy of this approach is region-specific and cannot be generalized to other areas such as Northeast China. Furthermore, despite SSM improvement, a dry bias remains. Our findings demonstrate that merely refining soil texture data is inadequate. This limitation arises primarily from a spurious strong land-atmosphere coupling regime, blocking the translation of improved land surface states into better precipitation simulation. Therefore, achieving realistic hydroclimate simulations requires prioritizing the correction of this coupling bias.</p>