Using Machine Learning to improve the representation of Phytoplankton dynamics in Earth System Models

The ocean carbon cycle and global climate are intricately connected as organic matter sinking into the deep ocean (the biological carbon pump) stores carbon in the deep ocean. Without this storage, atmospheric carbon dioxide would be 20-30\% higher than it is today. As the biological pump is affected by marine plankton abundance, it is vital to understand what controls plankton abundance. Plankton are grouped into size classes (PSCs) which impact photosynthetic efficiency, sinking rate, and marine food chain. Therefore, discerning the causes of spatio-temporal variability of PSCs is a scientific priority for understanding the ocean’s role in and response to climate change. Earth System Models (ESMs) are used to predict PSCs from environmental drivers by modelling biogeochemical and physical processes. ESMs’ representations of processes are limited by simplifying assumptions and exhibit significant biases. It is difficult to know if the relationships established by the ESMs are representative of the natural world. This study intends to decipher the relationships between the abundance of PSCs and environmental predictors using machine learning (ML), interpretable ML (XAI) and satellite products. The aim is to determine how the relationships between environmental drivers and PSCs found in nature differ from those used by ESMs for their predictions. Subsequently, we aim to use scientific machine learning to alter the underlying equations used by ESMs for predictions so that they obey the relationships found in nature. This will help improve predictions of PSCs by ESMs and increase our understanding of the marine carbon cycle’s response to climate change.