, 2010). On the 40–160-h time scale the correlation Angiogenesis inhibitor relationship is cleaner for the model than observations, as model SST generally has less variability than observations ( Figs. 1b and 2). To examine the statistics

of that relationship, the lagged correlation is calculated for the filtered time series of both model and observations. Each value of the lagged correlation series is a calculation of correlation with the time series of SST and τ offset from one another by a different lead/lag time. We consider only the correlations in which the τ time series leads SST, because the ocean model is forced by a prescribed atmosphere that has no response to the ocean, rendering lag time meaningless. For comparison between model and observations, we select the largest magnitude correlation for any lead time less than 48 h. The lead time itself is examined separately. The KPP turbulent mixing scheme is implemented in a version of the Massachusetts Institute of Technology general circulation model (MITgcm) (Adcroft, 1995, Marshall et al., 1997a and Marshall et al., 1997b), in hydrostatic configuration

with a 1/3° resolution C-grid on a domain encompassing the Tropical Pacific, from 26°S to 30°N and 104°W to 290°W (Table 1). The model is run for approximately four years, from Nov 1st, 2003 to October 13th, 2007 with a 15-min timestep. The model configuration is based on Hoteit et al., 2008 and Hoteit et al., 2010. Initial and lateral boundary conditions for the ocean temperature, salinity, and velocity come from the OCean Comprehensible Atlas (OCCA) (Forget, 2010). Surface forcing selleck inhibitor for temperature, specific humidity, shortwave and longwave radiation, wind (unless otherwise noted), and precipitation are interpolated to the model grid size and time step from the NCEP/NCAR 1.8°, six-hourly Reanalysis (Kalnay et al., 1996) and prescribed at the ocean surface. The MITgcm calculates

heat fluxes between the ocean and atmosphere. The default experiment (Exp. 0 [Table 2]) uses the NCEP/NCAR forcing and default KPP parameter values. An ensemble of 42 additional experiments is conducted (Table 2). In the first three experiments the KPP parameters are held at their default values while wind forcing is replaced with alternatives: ECMWF (Gibson et al., 1997), NOAA/CIRES Twentieth out Century reanalysis (Compo et al., 2011), and NASA Cross-Calibrated Multi-Platform Ocean Surface Wind Velocity (Atlas et al., 1996) (Exp. 1–3 [Table 2]). In the next 19 experiments, KPP parameters are perturbed to artificially large and small values (Exp. 4–22 [Table 2]). An additional 20 experiments are conducted using wind forcing that is blended from the NCEP/NCAR, ECMWF, and NASA products (Exp. 23–42 [Table 2]). The blending is done using a mixture model to weight the contribution from each of the three wind products, resulting in a Dirichlet distribution of weighting with the highest probability being an equal weight for each product.