In arid environments, shallow hypersaline lakes are critical to regional ecological stability. However, accurately monitoring suspended particulate matter (SPM) in these waters remains challenging for conventional optical remote sensing. The primary obstacles include signal saturation during high-turbidity events, interference from bottom reflectance in shallow zones, and insufficient satellite revisit frequency. To address these limitations, we developed a physics-informed machine learning (PIML) framework to isolate the hydrodynamic drivers of SPM in Ebinur Lake. Unlike purely data-driven approaches, we constructed a feature space grounded in wave mechanics, incorporating variables such as bottom shear stress, effective fetch, and temporal memory into a Random Forest regressor. Crucially, we employed physically downscaled ERA5 instantaneous wind gusts to capture the nonlinear threshold behavior of sediment entrainment. The model demonstrated robust performance, achieving a five-fold cross-validated R-2 of 0.91 (RMSE = 81.14 mg/L; RRMSE = 32.3%) while overcoming optical saturation issues. Feature attribution analysis identified instantaneous wind gusts as the dominant factor (>90% importance), significantly outperforming mean wind speed. We further quantified a critical physical threshold of similar to 12.0 m/s, confirming that sediment resuspension is an energy-limited process triggered by extreme wind events. Additionally, the model functioned as a "virtual geostationary sensor," successfully reconstructing hourly SPM dynamics typically missed by polar-orbiting satellites. This study presents a transferable and physically interpretable paradigm for high-frequency water quality monitoring in data-scarce inland lakes.