An ensemble solution for the Earth's time-varying gravitational field from the NASA/DLR GRACE mission

Several groups produce estimates of the Earth's time-varying gravitational field with data provided by the NASA/DLR Gravity Recovery and Climate Experiment (GRACE) mission. These unprecedented highly accurate global data sets track the time-variable transport of mass across and underneath the surface of the Earth and give insight into secular, seasonal, and sub seasonal variations in the global water supply. Knowledge gained from these products can inform and be incorporated into ocean and hydrological models and advise environmental policy planning. Therefore, a complete understanding of the accuracy and variations between these different fields is necessary, and the most accurate possible solutions desired. While the various gravity fields are similar, differences in processing strategies and tuning parameters result in solutions with regionally specific variations and error patterns.

This study analyzed the spatial, temporal, and spectral variations between four different gravity field products. The knowledge gained in this analysis was used to develop an ensemble solution that harnesses the best characteristics of each individual field to create an optimal model. Multiple methods were used to combine and analyze the individual and ensemble solutions. First a simple mean model was created; then the different solutions were weighted based on the formal error estimates as well as the monthly deviation from the arithmetic mean ensemble. These ensemble models as well as the four individual data center solutions were analyzed for bias, long term trend, and regional variations between the solutions, evaluated statistically to assess the noise and scatter within the solutions, and compared to independent hydrological models. Therefore, the form and cause of the deviations between the models, as well as the impact of these variations, is characterized. The three ensemble solutions constructed in this analysis were all effective at reducing noise in the models and better correlate to hydrological processes than any individual solution. However, the scale of these improvements is constrained by the relative variation between the individual solutions as the deviation of these individual data products from the hydrological model output is much larger than the variations between the individual and ensemble solutions.