Orr¹, James C., Olivier Aumont², Kenneth G. Caldeira³, Karl E. Taylor⁴ and OCMIP Members

¹IPSL-LSCE, CEA Saclay, Bat. 709, L’Orme, F-91191 Gif-sur-Yvette, France, Tel: (33) 1 69 08 77 23, Fax: (33) 1 69 08 77 16, E-mail: orr@cea.fr, ²IPSL-LODyC-UPMC, Paris, France, ³LLNL/CCCM, Livermore, CA, USA, ⁴LLNL/PCMDI, Livermore, CA, USA, and OCMIP Members:

R. Schlitzer and M.-F. Weirig (AWI, Germany),

R. Matear (CSIRO, Australia)

Y. Yamanaka and A. Ishida (IGCR, Japan),

M. Wickett (LLNL, USA),

M. Follows (MIT, USA),

E. Maier-Reimer (MPIM, Germany),

S. Doney and K. Lindsay (NCAR, USA),

H. Drange and Y. Gao (NERSC, Norway),

F. Joos, K. Plattner (PIUB, Switzerland),

J. Sarmiento, R. Slater, R. Key, and A. Gnanadesikan (Princeton, USA),

I. Totterdell and A. Yool (SOC, UK),

A. Mouchet, E. Deleersnijder, and J.-M. Campin (UL, Belgium),

C. Sabine (PMEL, USA),

R. Najjar, F. Louanchi (PSU, USA),

N. Gruber (UCLA, USA)

Data-model comparison of modern air-sea CO₂ fluxes

We have compared simulated results from 13 OCMIP-2 models to the data-based seasonal climatology of modern air-sea CO₂ fluxes. Data-based fluxes were obtained by multiplying the climatological ΔpCO₂ fields times the OCMIP-2 fields of gas exchange. To assess differences, we used a “Taylor” diagram, which simultaneously plots the correlation coefficient R, the simulated and observed variances, and the centered pattern RMS difference. All OCMIP models used the same simple biogeochemistry; we also analyzed results from another ocean model with more sophisticated biogeochemistry (PISCES). All models were skillful in simulating the annual zonal mean distribution ($R > 0.9$). Yet, models showed little skill in simulating the seasonal cycle ($0.3 < R < 0.7$ for the monthly zonal mean comparison). Even the best model (PISCES) explained only 40\% of the overall spatiotemporal variability. The seasonal correlation coefficient R approaches 1 in the subtropics, but it reaches nearly -1 in the high latitudes and tropics. This model-data anticorrelation is most disconcerting in the high northern latitudes, where data are abundant. Elsewhere the strong anticorrelations are less useful: in the Southern Ocean pCO₂ data are sparse, and there is little seasonal variability in the tropics. The cause for the northern anticorrelation is not caused by the biogeochemical component, based on our sensitivity test with PISCES. Nor is it caused by the circulation component, given the model diversity and this consistent problem. Instead, the northern anticorrelation appears due to the annual mean boundary condition for atmospheric pCO₂, a simplification which has been used by all ocean carbon models. Therefore, for studies of seasonal and interannual variability, ocean carbon models will need to account for seasonal and latitudinal variability of atmospheric pCO₂ (e.g., the seasonal amplitude of surface atmospheric pCO₂ at 60°N is about 15 ppm, and its phase, in terms of the air-sea flux, opposes that from the current ocean carbon model simulations).