A Deep Learn Emulator for Ocean Biogeochemical Modelling

The ocean's biogeochemistry is crucial for understanding the global ocean carbon cycle. Within the climate ocean model Nemo, the PISCES module (Pelagic Interactions Scheme for Carbon and Ecosystem Studies), is based on the numerical calculation of 24 different biological, physical and chemical variables which contribute to a complex bio-geo-chemical relationship to be able to estimate the net source and sinks of primary carbon production. In this work, we want to present the first steps toward using the Deep Neural Networks as a multi-variate problem trained on the model output to predict the next sequences and replace the module with an emulator solely based on machine learning (ML).