Evaluation of a global ocean reanalysis generated by a global ocean data assimilation system based on a four-dimensional variational (4DVAR) method

Abstract

Japan Meteorological Agency (JMA) started to use a new global ocean data assimilation system for the operational seasonal predictions in February 2022. The system is composed of two subsystems with non-eddy-permitting (lower) and eddy-permitting (higher) resolutions. The lower-resolution subsystem adopts a four-dimensional variational (4DVAR) method to optimize the temperature and salinity fields, and the data-assimilated fields are downscaled into the higher-resolution subsystem using incremental analysis updates. The impact of introducing the 4DVAR method in the new ocean data assimilation system is investigated through the comparison of a regular reanalysis run of the system using the 4DVAR method with another run using a three-dimensional variational (3DVAR) method. A comparison of the temperature fields before the downscaling between the two reanalysis runs indicates that the 4DVAR method can more effectively reduce the misfits between the model field and assimilated observation data. However, the increase of the temperature root mean square difference (RMSD) relative to independent Argo float data, along with the larger variance, for the run with the 4DVAR method reveals that the 4DVAR method adjusts the temperature field more significantly but the adjustments are inconsistent with the independent data due to insufficient model physics and resolution. The increase of the RMSD is mitigated after the assimilated fields are downscaled into the higher-resolution subsystem. The 4DVAR method reduces the bias and RMSD of temperature relative to the independent data along the thermocline, as well as near the surface, in the equatorial vertical section, which is expected to affect the prediction of El Niño-Southern Oscillation (ENSO).