Reconstructing the carbon dioxide absorption patterns of world oceans using a feed-forward neural network: software implementation and employment techniques

Abstract

Oceans play a major role in the global carbon budget, absorbing approximately 27% of anthropogenic carbon dioxide (CO2). As the degree to which an ocean can serve as a carbon sink is determined by the partial pressure of CO2 in the surface water, it is critical to obtain an accurate estimate of the spatial distributions of CO2 and its temporal variation on a global scale. However, this is extremely challenging due to insufficient measurements, large seasonal variability, and short spatial de-correlation scales. This paper presents an open source software package that implements a feed-forward neural network and a back-propagation training algorithm to solve a problem with one output variable and a large number of training patterns. We discuss the employment of the neural network for global ocean CO2 mapping.