Vertical displacements (dz) in permanent Global Positioning System (GPS) station positions enable estimation of water storage changes (ΔS), which historically have been impossible to measure directly. We use dz from 924 GPS stations in the western United States to estimate daily ΔS in California's Sierra Nevada and compare it to seasonal snow accumulation and melt over water years 2008–2017. Seasonal variations in GPS‐based ΔS are ~1,000 mm. Typically, only ~30% of ΔS is attributable to snow water equivalent (SWE). ΔS lags the snow cycle, peaking after maximum SWE and remaining positive when all snow has melted (SWE = 0). We conclude that seasonal ΔS fluctuations are not primarily driven by SWE but by rainfall and snowmelt stored in the shallow subsurface (as soil moisture and/or groundwater) and released predominantly through evapotranspiration. Seasonal peak GPS ΔS is larger than accumulated precipitation from the Parameter‐elevation Relationships on Independent Slopes Model and North American Land Data Assimilation System, indicating that these standard inputs underestimate mountain precipitation.

GPS monitoring of solid Earth deformation due to surface loading is an independent approach for estimating seasonal changes in terrestrial water storage (TWS). In western United States (WUSA) mountain ranges, snow water equivalent (SWE) is the dominant component of TWS and an essential water resource. While several studies have estimated SWE from GPS‐measured vertical displacements, the error associated with this method remains poorly constrained. We examine the accuracy of SWE estimated from synthetic displacements at 1,395 continuous GPS station locations in the WUSA. Displacement at each station is calculated from the predicted elastic response to variations in SWE from SNODAS and soil moisture from the NLDAS‐2 Noah model. We invert synthetic displacements for TWS, showing that both seasonal accumulation and melt as well as year‐to‐year fluctuations in peak SWE can be estimated from data recorded by the existing GPS network. Because we impose a smoothness constraint in the inversion, recovered TWS exhibits mass leakage from mountain ranges to surrounding areas. This leakage bias is removed via linear rescaling in which the magnitude of the gain factor depends on station distribution and TWS anomaly patterns. The synthetic GPS‐derived estimates reproduce approximately half of the spatial variability (unbiased root mean square error ∼50%) of TWS loading within mountain ranges, a considerable improvement over GRACE. The inclusion of additional simulated GPS stations improves representation of spatial variations. GPS data can be used to estimate mountain‐range‐scale SWE, but effects of soil moisture and other TWS components must first be subtracted from the GPS‐derived load estimates.