Potential future changes of terrestrial water storage based on climate projections by ensemble model simulations

•Historical and future TWS changes were constructed using the BMA method based on multi-model ensemble simulations and GRACE data.•BMA-based TWSA estimates agreed better with GRACE observations than single simulations and simple arithmetical averaging.•Combined effects of global warming and human activities would induce significantly decreasing trends in future TWS over most river basins in China. An accurate estimation of terrestrial water storage (TWS) is crucial for water resource management and drought monitoring. However, the uncertainties in model physics, surface parameters and meteorological data often limit the accuracy of land surface hydrological models in estimating TWS. In this study, a multi-model-based framework was developed to predict TWS in China by 2050 using a Bayesian model averaging (BMA) method and GRACE satellite observations. Compared to GRACE observations, our BMA-based TWS anomaly (TWSA) estimations reduce root mean square errors by 10–16% and increase correlation coefficients by 26–46% over semi-humid and semi-arid basins than simple arithmetical averaging for the validation period (2008–2016). At the same time, BMA shows decreasing root mean square differences (10–12%) over humid basins. The calibrated BMA weights were then applied to future projections of TWSA under two Representative Concentration Pathways (RCP): RCP 2.6 and RCP 6.0. The overall rate of TWSA for the future period (2021–2050) was detected with the same direction as that from past decades (2003–2016), but with larger decreasing values. Especially for the Haihe basin in North China, BMA-based TWSA would decrease faster by about 19% for RCP 2.6 and 26% for RCP 6.0. These results suggest a decreasing trend in future TWS over most of the basins in China due to combined effects of global warming and human activities, which suggests likely aggravated risk of water shortage and a growing need for adaptive water resources management.