Development of a new 4DVAR Global Ocean Data Assimilation System (MOVE/MRI.COM-G4) and a new bias correction scheme in the Japan Meteorological Agency

The Japan Meteorological Agency (JMA) has been operating a Four-Dimensional Variational (4DVAR) global ocean data assimilation system, named MOVE/MRI.COM-G3 (G3) since February 2022. It is used for monitoring the El Niño Southern Oscillation and the Indian Ocean Dipole and providing initial conditions to the oceanic part of the JMA coupled prediction system, named JMA/MRI-CPS3 (CPS3). G3 is composed of the lower-resolution (0.5˚×1˚) model, G3A and the higher-resolution (0.25˚×0.25˚) model, G3F. Temperature, salinity, and sea surface height observations are assimilated into the lower-resolution model through a 4DVAR method in G3A. The temperature and salinity fields analyzed by G3A are dynamically downscaled into the higher-resolution model through Incremental Analysis Updates (IAU) in G3F. By combining G3A and G3F, a flow-dependent analysis through the 4DVAR method and the initialization of the higher-resolution model are enabled with limited computational resources.

We are currently developing a new ocean data assimilation system named MOVE/MRI.COM-G4, which runs 4DVAR directly on the 0.25˚ model. The increased analysis resolution for the 4DVAR is expected to improve the depiction of Kuroshio and mid-latitude eddies and reduce initial shocks as it matches the resolution of the oceanic part of CPS3. To balance the increased computational requirements of the higher resolution analysis, the 4DVAR analysis window is shortened from ten to two days and the number of 4DVAR iterations is reduced to one-third with a preliminary 3DVAR analysis providing the first guess of the 4DVAR analysis to ensure its quality. Preliminary experiments show improvements as expected in the mid-latitude, while there is still room for improvements in the analysis quality in the tropics and sea ice regions.

We also plan to modify the model-bias correction scheme to introduce seasonal variation of biases. The current scheme corrects first guesses by removing the model biases updated online using analysis increments smoothed with a 2-year time constant and represents inter-annual variation in model biases. Additionally, in the new scheme, the monthly offline term is estimated a priori using analysis increments in preliminary reanalysis. A low-resolution test experiment using the new scheme shows that temperature and salinity fields are improved globally with respect to the independent Argo data. We aim to apply the new scheme to G4.