Determination of Monthly Mixed Layer Depth Fields
In order to define monthly mixed layer depth (MLD), a weighted average based on two sources of MLD information was created, one source based on observations and the other based on a numerical ocean model. The first was the MLD product offered by the National Ocean Data Center (NODC). Specifically, the MLD fields computed via potential density at 1°×1° from gridded temperature/salinity (T/S) (Levitus and Boyer 1994a; Levitus et al. 1994) were used. This product is available at here. The second source was Fleet Numerical Meteorology and Oceanography Center (FNMOC) model mixed layer output at a resolution of 2.5°×2.5° (Clancy and Sadler 1992). Using daily FNMOC fields from March through December 1995 and January and February, 1996, monthly means were computed and then gridded to the same resolution as the NODC fields.
The T/S observations required for the NODC MLD product are highly non-uniformly distributed over the globe, and much of the ocean is completely unsampled (see Levitus and Boyer 1994a for methodology of filling the global 1°×1° grid). As a result, the MLD fields contain unrealistic spatial distributions, horizontal gradients, and magnitudes. This problem with definition of MLD from gridded T/S is known, and a developing approach is to define MLD from individual hydrographic profiles and to grid resultant MLD estimates only where observations exist (Monterey, G., Pacific Fisheries Environmental Laboratory, Pacific Grove, Calif., personal communication.). However, such MLD fields are not currently available. Therefore, a weighting function for the NODC MLD fields was defined based on observation density. Specifically, we used the monthly average number of salinity observations at NODC levels within the upper 50 m. Based on mapped observation density, a cutoff of 75 was chosen to define where salinity was well sampled and thus where the NODC MLD fields had a sufficient observational base. Above this cutoff, the weighting for NODC MLD was 1 (~7% of the grid points). Below the cutoff, the weighting for NODC MLD was the average number of observations divided by 75. Lastly, because some NODC MLD values are extremely and unrealistically deep where few observations exist, zero weighting was assigned where NODC MLD was > 400 m. This weighting procedure retained NODC MLD estimates in relatively well-observed regions and relied on the model (FNMOC) MLD estimates for poorly observed regions (in proportion to the paucity of observations).
Following this definition of the weighted average MLD product, there still remained grid points where neither input data set provided information. Missing grid points within the latitude range 65° N to 65° S were filled with a combination of spatial and temporal averaging (±2 months and 5° of latitude/longitude). Any points not filled by this procedure were filled with the mean of all valid monthly MLD values for that grid point. Finally, a 5°×5° median filter was applied to the monthly MLD fields to smooth the boundaries where missing data were filled in the last step.