Ocean data assimilation guidance using uncertainty forecasts

This paper discusses preliminary tests on using predicted forecast errors to estimate the impact of observations in correcting the Naval Research Laboratory (NRL) tide resolving, high resolution regional version of the Navy Coastal Ocean Model (RNCOM) assimilating local observations processed through the NRL Coupled Ocean Data Assimilation (NCODA) system. Since there will always be a shortfall of data to constraint all sources of uncertainty there is an obvious advantage to optimally guide observations to reduce model errors that could be producing the most negative impacts. The importance of this topic has been further heightened in oceanic applications by the advent of Underwater Automated Vehicles (UAVs) that can bring persistent observations but need to be told where to go and when, following regular schedules. This works tests a technique named the Ensemble Transform Kalman Filter (ETKF) that can be used to automate such adaptive sampling guidance and has been successfully applied for atmospheric modeling optimization. The ETKF uses an ensemble of state-fields from a certain initialization time and rapid low rank solutions of the Kalman filter equations to estimate integrated predicted error reduction for selected target ensemble variables, or combinations of variables, over areas and forecast ranges of interest. The error estimates are produced through independent RNCOM runs using perturbed forcing and initial conditions constrained at each analysis time by new estimates of the analysis errors as provided by NCODA, using a technique named Ensemble Transform (ET). The skills of these systems in tracking the RNCOM forecast errors and predicting the reduction in forecast error from a set o possible observations were tested using local profile measurements off the East Philippines. Results show areas of larger uncertainty close to the major spatial gradients as one could anticipate and a good accuracy of error estimates with an high spread-skill (i.e. ensemble est- imates had the ability to correctly separate the small ensemble spread well correlated with the smaller observed errors from the larger ensemble spread well correlated with the larger observed errors). This consistency is a necessary condition to allow the ETKF to accurately predict the impact of the observations in reducing model errors. These ETKF skills were then tested by comparing the vertically averaged predicted temperature corrections based on the local measurements with the vertically averaged magnitude of the observed changes between two consecutive forecasts (before and after assimilating the data). Results showed the system had skills to accurately predict RNCOM errors and the impact of observation networks in reducing the error of model state-variables.