Regional ocean model uncertainties using stochastic parameterizations and a global atmospheric ensemble

Abstract

A Bay of Biscay model configuration is used as a test case to assess the data-based consistency of ensemble-based ocean model uncertainties of several types: [A] built-in stochastic parameterizations at regional ocean scales, [B] ocean model response to a global atmospheric model ensemble and [C] both A and B simultaneously. Ensembles of varying length were generated. In addition to a seasonal-range ensemble, three medium-range ensembles were carried out over successive overlapping segments permitting to compare consistency metrics for different lead times. The largest spread was obtained for the C case, although most of the model uncertainties were attributable to the stochastic ocean parameterizations in A. We addressed the question of which ensemble type and lead time was able to provide the most realistic model uncertainties given observations of SST, sea level, and Chlorophyll a, using a theoretical and diagnostic consistency analysis framework expanded from Vervatis et al. (2021a). In our results, consistency was satisfactory for the stochastic ensembles of types A and C, for the “aged” error cases (but only marginally with respect to the “young” error cases), and whenever physical and biogeochemical uncertainty processes were active in the region and could be detected by the observational networks, such as the onset of the spring shoaling of the thermocline and the phytoplankton abundance primary bloom. Sea level empirical consistency was improved when a wide range of low- to high-frequency errors were included in the signal of dynamic atmospheric process in the data and in the model inverse barometer. These findings provide additional insight that can help configure ensemble-based methods in academic studies and in operational ocean forecasting systems.