An operational Eddy resolving 1/16° global ocean nowcast/forecast system

The first real-time eddy resolving nearly global ocean nowcast/forecast system has been running daily at the Naval Oceanographic Office (NAVOCEANO) since 18 October 2000 and it became an operational system on 27 September 2001. Thirty-day forecasts are made once a week. The system, which was developed at the Naval Research Laboratory (NRL), uses the NRL Layered Ocean Model (NLOM) with 1/16° resolution and seven layers in the vertical, including a Kraus–Turner type bulk mixed layer. Sea surface temperature (SST) from satellite IR and satellite altimeter sea surface height (SSH) data from TOPEX/POSEIDON (T/P), ERS-2 and Geosat-Follow-On (GFO), provided via NAVOCEANOʹs Altimeter Data Fusion Center (ADFC), are assimilated into the model. The large size of the model grid (4096×2304×7) and operational requirements make it necessary to use a computationally efficient ocean model and data assimilation scheme. The assimilation consists of an optimum interpolation (OI) based scheme that uses an OI deviation analysis with the model as a first guess, a statistical inference technique for vertical mass field updates, geostrophic balance for the velocity updates outside of the equatorial region and incremental updating of the model fields to further reduce inertia–gravity wave generation. A spatially varying mesoscale covariance function determined from T/P and ERS-2 data is used in the OI analysis. The SST assimilation consists of relaxing the NLOM SST to the Modular Ocean Data Assimilation System (MODAS) SST analysis, which is performed daily at NAVOCEANO. Real-time and archived results from the model can be viewed at the NRL web site http://www.ocean.nrlssc.navy.mil/global_nlom. This includes many zoom regions, nowcasts and forecasts of SSH, upper ocean currents and SST, forecast verification statistics, subsurface temperature cross-sections, the amount of altimeter data used for each nowcast from each satellite and nowcast comparisons with unassimilated data. The results show that the model has predictive skill for mesoscale and other types of variability lasting at least 1 month in most regions and when calculated globally.