Focus on Redouane Lguensat and Machine Learning for Oceanography

on the October 15, 2018

Redouane Lguensat works at the Institute of Environmental Geosciences (IGE) in the MEOM team (UMR of Grenoble Alpes University, IRD, Grenoble INP and CNRS). He is a Postdoc funded by the National Center for Space Studies (CNES) to work on the SWOT mission.
The Surface Water Ocean Topography (SWOT) mission aims to send a new satellite in space that will observe the Earth’s oceans and terrestrial waters. It is jointly developed by CNES and NASA with contributions from the Canadian Space Agency (CSA) and United Kingdom Space Agency. The satellite is expected to be launched in 2021 and is expected to help oceanographers and hydrologists improve their understanding of Earth’s surface water. For instance, it will provide measurements of Sea Surface Height (SSH) with a higher resolution and a larger field of vision than with current satellites.

Like any satellite, SWOT cannot perfectly reflect the reality. Satellite images are usually contaminated with noise and missing data. Redouane Lguensat’s first mission is to remove noises and fill missing data by using a popular machine learning technique, namely, deep convolutional neural networks. To train his algorithm he uses a simulator of what SWOT will yield (since it is not launched yet) to obtain images by couples: clean image and noisy image. With these couples, he will teach the machine how to go from a noisy image to the clean one.

The postdoc’s second mission is to predict the dynamics of the Sea Surface Height (SSH). Predicting SSH is of high interest for oceanographers and also for meteorologist (weather prediction) and marine scientists (fisheries, navigation, etc.)… Usually, SSH is predicted using physical-based numerical equations. Redouane Lguensat investigates whether machine learning methods can replace totally or partly the use of numerical equations. He uses data from the past to predict the future dynamics. He believes that the advancements made in computing power will make machine learning based methods appealing and thinks that combining both worlds (machine learning and numerical equations) can be an interesting research direction.

Finally, the young researcher works on eddy detection and classification. Using sea surface height data, ocean eddies are usually detected by geometrical or physical methods. By building a database of sea surface high images and eddy classification results (cyclonic, anticyclonic or no eddy), he trains a machine learning algorithm to detect and classify eddies automatically.


Redouane Lguensat works mainly with physical oceanographers and statisticians, particularly with Julien Le Sommer and Emmanuel Cosme, in the Institute of Environmental Geosciences (IGE). Redouane is very interested by the cross-fertilization between geoscientists and data scientists and believes that it will open many new research avenues. For him, machine learning methods are very powerful tools that should be carefully applied when dealing with physical sciences. Collaboration between multidisciplinary researchers is the key.

Published on October 15, 2018