The Impact of Atmospheric Modeling Errors on GRACE Estimates of Mass Loss in Greenland and Antarctica

Abstract

AbstractSystematic errors in Gravity Recovery and Climate Experiment (GRACE) monthly mass estimates over the Greenland and Antarctic ice sheets can originate from low‐frequency biases in the European Centre for Medium‐Range Weather Forecasts (ECMWF) Operational Analysis model, the atmospheric component of the Atmospheric and Ocean Dealising Level‐1B (AOD1B) product used to forward model atmospheric and ocean gravity signals in GRACE processing. These biases are revealed in differences in surface pressure between the ECMWF Operational Analysis model, state‐of‐the‐art reanalyses, and in situ surface pressure measurements. While some of these errors are attributable to well‐understood discrete model changes and have published corrections, we examine errors these corrections do not address. We compare multiple models and in situ data in Antarctica and Greenland to determine which models have the most skill relative to monthly averages of the dealiasing model. We also evaluate linear combinations of these models and synthetic pressure fields generated from direct interpolation of pressure observations. These models consistently reveal drifts in the dealiasing model that cause the acceleration of Antarctica's mass loss between April 2002 and August 2016 to be underestimated by approximately 4 Gt yr−2. We find similar results after attempting to solve the inverse problem, recovering pressure biases directly from the GRACE Jet Propulsion Laboratory RL05.1 M mascon solutions. Over Greenland, we find a 2 Gt yr−1 bias in mass trend. While our analysis focuses on errors in Release 05 of AOD1B, we also evaluate the new AOD1B RL06 product. We find that this new product mitigates some of the aforementioned biases.