Focus on... Adekunle Ajayi and the variability of oceans

on the November 15, 2020

Adekunle Ajayi is a machine learning research engineer at the Grenoble Alpes Data Institute. His research project focuses on applying deep learning techniques to predict the atmospherically-forced component of the ocean. Adekunle has a PhD in Ocean Physics from Université Grenoble Alpes.
The ocean is a large body of water that covers approximately 70% of the planet Earth. Understanding the dynamic of oceans is of great interest to oceanographers. Societal benefits from studying in the oceans include but are not limited to (i) accurate weather forecasting, (ii) met-ocean forecasting for offshore enterprise (oil and gas, shipping and fishing companies, etc.), (iii) improving climate change projection.

Ocean dynamics are driven mainly by atmospheric winds (forced variability) and cross-scale turbulent interaction between oceanic flows (intrinsic variability). To better characterize the ocean's forced and intrinsic variability, numerical experiments have been designed to simulate the ocean's dynamics at different statistical states. One of such experiments is the OCCIPUT3 (Oceanic Chaos - ImPacts, strUcture, predicTability) project. OCCIPUT3 is a 50-member ensemble of 1/4° global ocean/sea-ice simulations (covering the period 1960-2015), starting from perturbed initial conditions then driven by the same atmospheric forcing for all the 50 members. The quick divergence and loss of correlation between all the members demonstrate the chaotic behavior of the ocean. This experiment allowed the atmospherically-forced component (variability of the ensemble mean), and the intrinsic component (the difference between any given member and the ensemble mean), characterized by the ensemble standard-deviation, to be estimated. The result of this numerical experiment led to the generation of a massive amount of datasets.

Adekunle’s project focuses on using artificial intelligence to estimate the ocean's atmospherically-forced component by using the simulated seas surface height (SSH) fields of the OCCIPUTS experiments. This project follows Mickael et al. 2019, where the authors estimated the ocean's forced variability by training neural networks with spatial snapshots of oceanic SSH fields. Adekunle's current project aims to improve the prediction of the foed variability by taking into account both the spatial and temporal evolution of the ocean (t,x,y). Upon completion, his project would provide a robust method (based on machine learning) for filtering intrinsic oceanic variability.?

Published on November 16, 2020