Ensemble generation of regional ocean physics and biogeochemical model uncertainties, empirical consistency and suitability for probabilistic forecasting

The study builds upon two Copernicus marine projects, SCRUM and SCRUM2, focusing on ensemble forecasting operational capabilities to better serve coastal downscaling. Both projects provided coupled physics-biogeochemistry ensemble generation approaches, tools to strengthen CMEMS in the areas of ocean uncertainty modelling, empirical ensemble consistency and data assimilation, including methods to assess the suitability of ensembles for probabilistic forecasting. The study is conducted by performing short- to medium-range ensembles in the Bay of Biscay, a subdomain of the IBI-MFC. Ensembles were generated using ocean stochastic modelling and incorporating an atmospheric ensemble. Sentinel 3A data from CMEMS TACs and arrays were considered for empirical consistency, using innovation statistics and approaches taking into account correlated observations. Finally, several properties of ensembles were estimated as components of known probabilistic skill scores: the Brier score (BS), and the CRPS. This was done for pseudo-observations (Quasi-Reliable test-bed) and for real verifying observations in a coastal upwelling test case.