Articles | Volume 12, issue 5
https://doi.org/10.5194/bg-12-1483-2015
https://doi.org/10.5194/bg-12-1483-2015
Research article
 | 
09 Mar 2015
Research article |  | 09 Mar 2015

Comparison of ten packages that compute ocean carbonate chemistry

J. C. Orr, J.-M. Epitalon, and J.-P. Gattuso

Abstract. Marine scientists often use two measured or modeled carbonate system variables to compute others. These carbonate chemistry calculations, based on well-known thermodynamic equilibria, are now available in a dozen public packages. Ten of those were compared using common input data and the set of equilibrium constants recommended for best practices. Current versions of all 10 packages agree within 0.2 μatm for pCO2, 0.0002 units for pH, and 0.1 μmol kg−1 for CO32− in terms of surface zonal-mean values. That represents more than a 10-fold improvement relative to outdated versions of the same packages. Differences between packages grow with depth for some computed variables but remain small. Discrepancies derive largely from differences in equilibrium constants. Analysis of the sensitivity of each computed variable to changes in each constant reveals the general dominance of K1 and K2 but also the comparable sensitivity to KB for the ATCT input pair. Best-practice formulations for K1 and K2 are implemented consistently among packages. Yet with more recent formulations designed to cover a wider range of salinity, packages disagree by up to 8 μatm in pCO2, 0.006 units in pH, and 1 μmol kg−1 in CO32− under typical surface conditions. They use different proposed sets of coefficients for these formulations, all of which are inconsistent. Users would do well to use up-to-date versions of packages and the constants recommended for best practices.

Short summary
Basic marine carbonate system variables such as pH are often computed from others. Such calculations are made with many public software packages, but their results have never been compared. A new study compares 10 of these packages, quantifying differences, isolating causes, and making recommendations to reduce future discrepancies. This comparison effort has led to more than a 10-fold reduction in differences between packages for some computed variables.
Altmetrics
Final-revised paper
Preprint