The combined cost function for n buoys is: equation(8) cost=12∑i=1n(Ri-ri)2σri2+(Li-li)2σli2,where equation(9) σri2=ηi2+ϕi2, equation(10) σli2=ηli2+ϕi2+ϕL2,and each i is a buoy and the denominators are the summed uncertainties. The model output in the default configuration has zonally oriented bands in correlation and lead time, especially in the Central and Eastern Pacific. The model correlation, R, is enhanced along the equator and flanked by wider bands of very low R from about 1.5°N to 5°N and 2°S to 5°S ( Fig. 4). Model lead time, L, has a similar structure, with longer lead times

along Protein Tyrosine Kinase inhibitor the equator, flanked to the north and south by broad bands of lower lead times ( Fig. 8). While the network of buoys has a much lower spatial resolution, the same structure of enhanced r and reduced l is evident along the equator. Zonal bands of diminished r and enhanced l are evident along 2°N and 5°N and S, but are

difficult to resolve. Further from the equator, along 8°N and S, model correlation and lead time show little similarity to data. In all experiments, the model overestimates the magnitude of the average τ-SST correlation, ranging from 5.8% to 25.6%, and by 24.4% in the default configuration. All but two experiments reduce this bias relative to the default winds and parameters, yet none eliminate the bias ( Fig. 6). The correlation is highly sensitive to wind forcing product ( Figs. 6 and 7): the NOAA wind product (Exp. 2) reduced the correlation relative to the default experiment by 14.7%, while the greatest sensitivity to any parameter (the critical gradient Richardson number Rio) was a reduction in Selleck Ribociclib correlation by just 6.4% (Exp. 6). This is especially true in the Central and Eastern Pacific, as alternative wind products tend to

reduce the correlation relative to the default, bringing it close to observations ( Fig. 7). At 47 out of 65 buoys the model correlation with default KPP Sitaxentan parameters is greater than the observational correlation ( Fig. 4). A zonal pattern in misfit is also evident, as the overestimation is generally more significant for buoys farther from the equator ( Fig. 4). The overestimation is exaggerated from 180°W westward, due to a modeled increase in the magnitude of the correlation relative to the Eastern Pacific that is not as distinct in the observations ( Fig. 7). This may be related to the separation between the deep thermocline and the shallow mixed layer in the Western Pacific, which may act as a barrier to the entrainment of cooler water from the thermocline to the surface during wind events ( Lukas and Lindstrom, 1991). The lead time to maximum correlation has a meridional spatial pattern, increasing in the Eastern Pacific in both the model (L) and in observations (l) ( Figs. 8 and 9). The model also shows a slight decrease in lead time from the Western Pacific eastward toward the Central Pacific, but this is less evident in the observations ( Fig. 9).