Convergence of Daily GRACE Solutions and Models of Submonthly Ocean Bottom Pressure Variability

Knowledge of submonthly variability in ocean bottom pressure (p(b)) is an essential element in space-geodetic analyses and global gravity field research. Estimates of these mass changes are typically drawn from numerical ocean models and, more recently, GRACE (Gravity Recovery and Climate Experiment) series at daily sampling. However, the quality of p(b) fields from either source has been difficult to assess and reservations persist as to the dependence of regularized GRACE solutions on their oceanographic priors. Here, we make headway on the subject by comparing two daily satellite gravimetry products (years 2007-2009) both with each other and with p(b) output from a diverse mix of ocean models, complemented by insights from bottom pressure gauges. Emphasis is given to large spatial scales and periods p(b) signals (similar to 2 cm root-mean-square error) over Southern Ocean abyssal plains in the present GRACE de-aliasing model. These and other imperfections in baroclinic models are especially apparent at periods <10 days, although none of the GRACE series presents a realistic ground truth on time scales of a few days. A barotropic model simulation with parameterized topographic wave drag is most commensurate with the GRACE fields over the entire submonthly band, allowing for first-order inferences about error and noise in the gravimetric mass changes. Estimated p(b) errors vary with signal magnitude and location but are generally low enough (0.5-1.5 cm) to judge model skill in dynamically active regions.

Automated summary

Knowledge of submonthly variability in ocean bottom pressure is crucial for space-geodetic analyses and global gravity field research. Comparisons between various satellite gravimetry products and ocean models showed imperfections in short-term variations, with a parameterized barotropic model simulation proving to be more consistent with GRACE fields. Estimated errors in bottom pressure varied but were generally low enough to assess model skill in dynamically active regions.