This suggests that at least for the variables most important for ocean carbon exchange, i.e., wind speeds, SST, and ice, the reanalysis products are either in general agreement, or that the differences among them are relatively unimportant

at the largest spatial scales. This finding is emphatically not true for regional analyses, where large differences in FCO2 are observed depending upon the reanalysis product used for forcing. pCO2 distributions are considerably less sensitive to the choice of reanalysis product. These findings have important implications for ocean modelers in choosing reanalysis products: namely that for global models it does not matter much, but for regional and local GDC-0199 ic50 model the selection can have important influences on carbon cycling and exchange estimates. see more The finding that different estimates of air–sea fluxes are produced by different reanalyses at regional scales reinforces the work by Otero et al. (2013), who used different reanalysis sources in the Bay of Biscay. Several other ocean carbon modeling efforts have utilized versions of NCEP forcing

data (e.g., Le Quéré et al., 2010, Doney et al., 2009 and McKinley et al., 2004). This effort provides a milepost for evaluating the use of different reanalysis forcing products for ocean carbon models, at least in a general sense. The overarching conclusion, i.e., that global estimate of carbon fluxes and pCO2 are insensitive to the choice of forcing is likely robust. Similarly the other conclusions that regionally and sub-regionally the choice of reanalysis has

successively more influence, is also likely to apply to other models as well. However the nature of the differences and sensitivities is likely to be different. The difference will be dependent upon Meloxicam the nature of the model formulation, but we hope the results provided here will be of help in the selection and use of reanalysis products for global and regional ocean carbon models. We thank the NASA/MERRA Project, the NOAA/NCEP Project and the ECMWF Project for the data sets and public availability. We also thank the Lamont-Doherty Earth Observatory for in situ pCO2 data and flux estimates. We thank three anonymous reviewers for insights. This work was supported by NASA Modeling and Analysis Program (MAP) and Carbon Monitoring System (CMS) Programs. “
“The bias of an estimator is formally defined as the difference between its expected value and the true value it is trying to estimate (e.g., Priestley, 1981). In the context of environmental modeling, biases are often approximated by the mean difference between simulated and observed quantities after averaging over certain temporal or spatial scales (e.g., WMO, 2008). Biases are a common problem in many environmental models (e.g., Randall et al.