9). Lead time proves largely insensitive to changes in the KPP parameters, but it responds very strongly to changes in wind product, which tend to increase lead time basin-wide. The NOAA wind product especially causes increased lead times ( Fig. 9). Implicit in the assumption that the differences between wind products represent uncertainty in wind forcing is that each of those products is equally valid. However, the wind products are unequal in their impact on model lead time. The NOAA wind experiment tremendously increases the estimate

of the uncertainty in wind forcing because it is so different from the other three products. In reality, no wind product is entirely independent buy Osimertinib from another, and they may not be equally valid estimates of the wind forcing. All the reanalysis products are based on the same atmospheric data sets (the NASA Raf tumor wind includes additional QuickSCAT scatterometer data), but differ in data assimilation method and in the model used in their generation. However, because of concerns over the integrity of the NOAA wind, it was not included in the mixing model to create the 20 blended wind products. The two components of the cost function (Eq. (8)) – maximum lead correlation and lead time to maximum correlation – show

different degrees of sensitivity to changes in wind forcing and KPP parameters. The correlation-based cost term [cost(R, r)] shows comparable sensitivity to some KPP parameters relative to the sensitivity to wind. The largest changes in cost(R, r) from the default for a single 6-phosphogluconolactonase experiment belong to Exps. 5, 1, and 7, corresponding to perturbations to the critical bulk Richardson # (Rib), wind product (ECMWF), and critical gradient Richardson # (Ri0) ( Fig. 10b). The sensitivity to Ri0 (Exps. 7, 8) is larger than the spread in cost(R, r) between any of the wind products. The lead time-based

cost [cost(L, l)] appears far more sensitive to wind forcing than changes to the KPP parameters ( Fig. 10d). Notably, the NOAA winds (Exp. 2) cause a 252% increase in cost(L, l) from the default experiment. In order to emphasize the sensitivity in lead time L to the NOAA wind product, it is represented by the unfilled diamonds in Fig. 9. The overwhelming sensitivity in cost(L, l) to the NOAA winds even dominates the combined cost [cost(R, r, L, l)] ( Fig. 10e). Therefore, lead time appears to worsen, rather than improve, the signal to noise ratio. Because of the known bias between the model correlation R   and the observed correlation r  , a second cost function is calculated in which each experiment is compared to the model mean, R¯, instead of observations, r  : equation(11) costR¯=12∑i=1n(Ri-R¯i)2σri2,where R¯i is the mean model correlation of the 19 KPP experiments (Exps.