<p>GNSS recovery of terrestrial water storage changes using elastic mass loading theory has gained attention in hydrogeodesy. Traditional methods use equally-weighted Laplacian smoothing constraints to stabilize inversion results, ignoring the uneven distribution of GNSS stations. To improve the inversion accuracy, we propose a density-weighted strategy to realize non-uniform Laplacian smoothing constraints, applying weaker constraints in dense station areas and stronger ones in sparse regions. Using the Pacific Northwest as a study area, simulations and real-data inversions show that density-weighted inversion improves accuracy in dense GNSS regions while performing similarly to the traditional method in sparse areas. It more accurately captures sharp gradient changes in water loads compared to the equally-weighted scheme, which often underestimates strong seasonal hydrological signals in high-altitude mountainous regions and overestimates weaker signals in low-altitude areas. GNSS inversion results closely align with NLDAS data, detecting strong seasonal signals in major mountain ranges and demonstrating its capability to detect small-scale variations. Both significantly diverge from GRACE data, which, due to its coarse resolution, fails to reappear fine details. This study highlights the potential of GNSS-based density-weighted inversion to improve hydrological monitoring across diverse terrains, supporting climate adaptation and resource planning.</p>