Exploring different iron regimes in the global ocean by an unsupervised clustering technique

Abstract:

Collaborative efforts over the past decade between observationalists and modelers have rapidly improved our understanding of the ocean micronutrient iron (Fe) cycling, which limits biological productivity in about half of the world ocean. However, process studies that measure the magnitude of different Fe sources and sinks are still lacking. As a consequence, ocean biogeochemistry models are free to manipulate their representation of Fe mechanisms to match the available observations. These deficits lead to a significant uncertainty on the relative roles of various mechanisms regulating the ocean Fe distribution and limit our confidence in model projections of ocean biogeochemical changes. In this study, we applied an unsupervised machine learning technique (K-means clustering) to analyze the global ocean Fe budget simulated by a series of model simulations, which have different levels of complexity in their Fe representation. Without using any latitude or longitude information, K-means reveals various Fe regimes in the global ocean, reflecting dominant mechanisms control the ocean Fe pattern at different ocean basins. The results show that while external sources set the overall magnitude for the mean ocean dissolved Fe (dFe) concentration, the upper ocean dFe pattern is regulated by internal cycling processes, where the delicate balance between scavenging and dissolution of Fe is modulated by the availability of ligands. We suggest that unsupervised machine learning techniques can serve as a new tool to diagnose the outputs from different ocean models and interpret observations.