AUTHOR=Doumbia Cheick , Castellazzi Pascal , Rousseau Alain N. , Amaya Macarena TITLE=High Resolution Mapping of Ice Mass Loss in the Gulf of Alaska From Constrained Forward Modeling of GRACE Data JOURNAL=Frontiers in Earth Science VOLUME=7 YEAR=2020 URL=https://www.frontiersin.org/journals/earth-science/articles/10.3389/feart.2019.00360 DOI=10.3389/feart.2019.00360 ISSN=2296-6463 ABSTRACT=

The resolution of Gravity Recovery And Climate Experiment (GRACE) Terrestrial Water Storage (TWS) change data is too low to discriminate mass variations at the scale of glaciers, small ensemble of glaciers, or icefields. In this paper, we apply an iterative constraint modeling strategy over the Gulf Of Alaska (GOA) to improve the resolution of ice loss estimates derived from GRACE. We assess the effect of the most influential parameters such as the type of GRACE solution and the degree of heterogeneity of the distribution map over which the GRACE data is focused. Three GRACE solutions from the most common processing strategies and three ice distribution maps of resolutions ranging from 55,000 to 20,000 km2 are used. First, we present results from a series of simulations with synthetic data and a mix of synthetic/modeled data to validate the focusing strategy and we point out how inaccuracies arise while increasing the spatial resolution of GRACE data. Second, we present the recovery of the total GRACE-derived mass change anomaly at the scale of the GOA. At this scale, all solutions and distribution maps agree, showing ∼40 Gt/year of mean ice mass loss over the period 2002–2017. This result is similar to studies using GRACE solutions from the latest releases and time-series of more than 8 years. The first studies using GRACE data published during the 2005–2008 era generally overestimated the long-term ice mass loss. Third, we show results of the three resolutions tested to focus the mass anomaly. Using focusing units (mascon) of ∼30,000 km2 or larger, the focusing procedure provides reliable results with errors below 15%. Below this threshold, errors of up to 56% are observed.