Hydrological trends captured by assimilating GRACE total water storage data into the CLM5-BGC model

Terrestrial water storage (TWS) encompasses various components such as snow water, surface water, canopy water, soil water, and groundwater. Therefore, TWS is strongly affected by several factors including climate change, floods, drought, and anthropogenic activities. Predicting and quantifying TWS is critical in understanding terrestrial water cycles, as well as for managing water resources at the regional and global scale. TWS can be monitored with the Gravity Recovery and Climate Experiment (GRACE) satellites and estimated using land surface models (LSMs) on various spatial scales. The present study demonstrated how large-scale TWS data assimilation affects model performance by assessing the variations in TWS and related hydrological variables. Moreover, unlike traditional physical-based models, the assimilation of TWS data into an LSM enabled the assessment of anthropogenic impacts. TWS data from GRACE was assimilated into the Community Land Model version 5 (CLM5) with the CLM5-BGC biogeochemistry module covering the East Asia region. The model was forced using 40-ensemble meteorological forcing data from 2003 to 2010 and GRACE TWS data were assimilated using the ensemble adjustment Kalman filter (EAKF). Data assimilation markedly improved the performance of the model, as demonstrated by a decrease in the root mean square error from 0.71 mm/month in the free run (FR) without data assimilation to 0.48 mm/month in the data assimilation (DA) run, in addition to an increase in the correlation coefficient from 0.55 in FR to 0.69 in DA. Additionally, unlike the FR model, the DA model effectively captured the observed negative trends of the TWS, soil moisture, and evapotranspiration in Northern India and the Northern China Plain. Finally, our findings also demonstrated that the anthropogenic contribution to the negative trend of the TWS over Northern India and Northern China Plain was marginally increased in the DA model compared to the FR model. Collectively, our findings demonstrated that TWS data assimilation enables a more realistic evaluation of the long-term and large-scale changes in hydrologic variables.