Deriving Algorithms for the Remote Sensing of Carbon Dioxide Fugacity at the Ocean Surface

As concentrations of carbon dioxide in the atmosphere continue to rise, the capacity of the ocean to act as a carbon dioxide sink is of critical importance as it is the major sink of anthropogenic carbon dioxide. Uncertainties in our ability to quantify the role of the oceans in the carbon cycle, especially in computing the gas fluxes between atmosphere and ocean on global scales, leads directly to uncertainty in predicting the response of the of the climate system to increasing levels of carbon dioxide in the atmosphere. Here we report on a study to improve the accuracy of the retrievals of surface fugacity from earth observation satellites. A large data set of in situ measurements from equipment on the Royal Caribbean Cruise Lines ship Explorer of the Seas in the Caribbean Sea and western tropical Atlantic Ocean the relationship between the carbon dioxide concentration and variables measurable from space is explored using advanced computational techniques to improve on prior results derived by linear regression. Using natural selection as a conceptual model, the Genetic Algorithm approach maintains a population of “tentative” solutions that are subjected to “survival of the fittest” tests and to operators that implement mutation and recombination (mutual exchange of the “genetic information”). In our implementation, each specimen in the population represents one formula, expressed by a tree-like data structure. The fitness function that quantifies the individual's survival chances is defined as the mean square error scored by the given formula on the training data. We demonstrate in this case study that not only can the accuracy of satellite retrievals of surface fugacity of carbon dioxide be improved by using algorithms based on the information content of the data sets, but also the regions in which individual algorithms are applicable can also be determined. These regions align with the underlying dynamical oceanographic features. This approach can be applied to measurements taken elsewhere in the oceans, and of variables other than carbon dioxide.