SODA3	A reanalysis of ocean climate

Background The goal of SODA is to reconstruct the historical physical (and eventually biogeochemical) history of the ocean. As its name implies, the Simple Ocean Data Assimilation ocean/sea ice reanalysis (SODA) uses a simple architecture based on community standard codes with resolution chosen to match available data and the scales of motion that are resolvable. Agreement with direct measurements (to within observational error estimates) as well as unbiased statistics are expected. While SODA remains a university-based research project, we want to be helpful to potential users by providing a reliable, well-documented, source of seasonal climate time-scale ocean reanalysis to complement the atmospheric reanalyses available elsewhere (NOAA/EMC, NASA/GMAO, and ECMWF, for example).

SODA3 (SODA Version 3) represents a major upgrade of SODA. The model has been switched to GFDL MOM5/SIS numerics with eddy permitting 1/4°x1/4°x50lev resolution (28km at the Equator down to <10km at polar latitudes) . This model is similar to the ocean component of the GFDL CM2.5 coupled climate model, and includes the same SIS1 active sea ice model. A number of improvements have been included in the sequential DA filter, but SODA3 still has a pre-specified flow-dependent error covariance. We now offer an ensemble set of ocean reanalyses, the spread of which provides information about uncertainty. A description of SODA3 and its comparison to SODA2 is provided in a manuscript: Carton, Chepurin, and Chen (2018) which is currently in review.

For most SODA3 reanalysis ensemble members the surface fluxes are calculated from specified hourly-to-daily downwelling solar and longwave radiation, precipitation, 2m air temperature and specific humidity, 10m winds, along with SLP, SST, and ocean currents into a set of bulk formulas to compute surface heat, freshwater, and momentum fluxes. In a few cases we use the surface fluxes as provided by the atmospheric reanalysis centers. Because of the sensitivity of the ocean state to the choice of bulk formulas we conduct experiments with two different bulk formulas: Large and Yeager and COARE4. The up to ±0.05 N/m² systematic differences in stress, and ±10 W/m² differences in latent, and sensible heat fluxes due to the choice of bulk formula is illustrated in this picture (produced by Senya Grodsky). To quantify the uncertainty in the reanalysis due to uncertainty in surface forcing we are carrying out an ensemble of ocean reanalyses, each using the same reanalysis system, data sets, and initial conditions, but with forcing provided by different atmospheric reanalyses. (see Table below. Experiments with names in grey have not been released yet. The reanalysis fields for these experiments are available through the pulldown tabs above or through the live links in the Table). We suggest that you choose the ensemble member associated with your choice of atmospheric reanalysis. When we have completed the ensemble we will construct an 'ensemble-average' ocean reanalysis.

Flux Bias Correction An emphasis for SODA3 has been to reduce the systematic component of error (known as bias). Bias enters ocean reanalysis in four ways: 1) through initial conditions, 2) through surface forcing, 3) through model biases (including deficient resolution and missing physics), and 4) through measurement and sampling bias. We think the worst of these is currently the bias introduced through surface forcing, while the second worst is the measurement/sampling bias in the years prior to the deployment of Argo. Here we address the worst source of bias. Uncorrected, the seasonal fluxes (as produced by the bulk formulas) leaving the atmosphere in an atmospheric reanalysis are not consistent with the fluxes ocean dynamics and observations expect. If this systematic imbalance is ignored then the ocean reanalysis is degraded and much of the effort put into data assimilation is simply used to correct for the error introduced by flux biases. A better approach is to attempt to debias the fluxes. For the sequential data assimilation approach used here we can do this by conducting a preliminary ocean reanalysis, collecting the statistics of the observation - model misfits (the analysis increments), and then repeating the reanalysis using the error statistics to bias-correct the atmospheric fluxes (basically this is an iterative approach). If you are curious to know how we implement the procedure and what impact it has we refer you to a new paper (Carton, Chepurin, Chen, and Grodsky, 2018). The experiments marked with an (*) have had this seasonal bias correction procedure applied to the fluxes. Two atmospheric reanalysis surface meteorology data sets (CORE2 and DFS5.2) have had their own bias adjustement applied so we have not attempted to apply our own de-biasing procedure.