Improving the simulation of terrestrial water storage anomalies over China using a Bayesian model averaging ensemble approach

• Main Authors: Jian-Guo LIU, Bing-Hao JIA, Zheng-Hui XIE, Chun-Xiang SHI
• Format: Article
• Language: English
• Published: KeAi Communications Co., Ltd. 2018-07-01
• Series: Atmospheric and Oceanic Science Letters
• Subjects: Terrestrial water storage anomalies; multi-forcing and multi-model ensemble simulation; Bayesian model averaging; spatiotemporal variation; uncertainty
• Online Access: http://dx.doi.org/10.1080/16742834.2018.1484656

The ability to estimate terrestrial water storage (TWS) is essential for monitoring hydrological extremes (e.g., droughts and floods) and predicting future changes in the hydrological cycle. However, inadequacies in model physics and parameters, as well as uncertainties in meteorological forcing data, commonly limit the ability of land surface models (LSMs) to accurately simulate TWS. In this study, the authors show how simulations of TWS anomalies (TWSAs) from multiple meteorological forcings and multiple LSMs can be combined in a Bayesian model averaging (BMA) ensemble approach to improve monitoring and predictions. Simulations using three forcing datasets and two LSMs were conducted over mainland China for the period 1979–2008. All the simulations showed good temporal correlations with satellite observations from the Gravity Recovery and Climate Experiment during 2004–08. The correlation coefficient ranged between 0.5 and 0.8 in the humid regions (e.g., the Yangtze river basin, Huaihe basin, and Zhujiang basin), but was much lower in the arid regions (e.g., the Heihe basin and Tarim river basin). The BMA ensemble approach performed better than all individual member simulations. It captured the spatial distribution and temporal variations of TWSAs over mainland China and the eight major river basins very well; plus, it showed the highest R value (> 0.5) over most basins and the lowest root-mean-square error value (< 40 mm) in all basins of China. The good performance of the BMA ensemble approach shows that it is a promising way to reproduce long-term, high-resolution spatial and temporal TWSA data.