In

this paper we assume that the spread between the NCEP, ECMWF, NOAA/CIRES, and NASA wind products that are used to force the ocean surface represents the uncertainty in wind forcing. The differences between them are largely due to the methodology of constructing wind estimates from the same observational datasets. However, the NASA wind velocity also incorporates QuickSCAT scatterometer. KPP parameter perturbations are coarse, with adjustments of 50–100% in most parameters (Exp. 4–22 [Table 2]). In cases in which the parameter is actually a structure function, e.g. Exp. 9–14, adjustments to constants within those functions have an effect on the parameter of approximately 50%, although this is depth-dependent and perturbations approach zero as the Monin–Obukhov click here length approaches zero. The perturbations are not designed to test the viability of alternative KPP parameter values, but serve instead as sensitivity tests. The

perturbations are large because the intention is to discover whether there is any sensitivity in the model to that particular parameter. Wind speed and SST are observed at buoys in the TAO/TRITON Array. Observed Obeticholic Acid chemical structure wind speed is converted into neutral wind stress τ (N/m2) at the ocean surface by a drag coefficient calculated according to Large and Pond (1981). For inclusion in the model-data comparison, a buoy must have at least one continuous 30-day or longer period with no missing wind speed and SST data during the Nov. 1st, 2003–Oct. 13,

2007 modeling period. Only those continuous intervals are included in the study. Sixty-five buoys in the TAO/TRITON array satisfy that criterion in the domain spanning 8°S to 8°N and 180°W to 90°W. Of these, twelve have hourly data and 53 have ten-minute data. SST observations are matched for comparison to the output from the top ocean model layer (2.5 m beneath the Unoprostone sea surface) at the model grid point nearest each buoy, a maximum distance of 0.24° (about 26 km). The cost function takes the form of a likelihood function, which is a calculation of the probability of making an observation given a model. From this perspective, observations and model output are distributed with variances that are a function of their uncertainty. Model time series are complete for the entire 2004–2007 simulation period, but the first 1.5 yr are removed to allow for model spin up. Missing buoy data prevent the calculation of a single lagged correlation on an entire observational time series. Instead, separate lagged correlation calculations are conducted on each time series of continuous observations of 30 days or longer (separate time series distinguished by color in Figs. 2 and 3).