Project Rationale
The Mozambique channel and Madagascan waters are a region of the southwestern Indian Ocean with complex circulation patterns that play a key role in the distribution of marine species and regulation of local ecosystems [1]. Unsupervised machine learning techniques can help classify the nonlinear interactions of these circulation patterns and determine those impacting biological productivity [2], where local economies are dependent on marine living resources. Although the flows around Madagascar are known to affect the southern Mozambique channel, they may also influence dynamics farther north because of high eddy activity within the channel. The anticyclonic and cyclonic eddies result in warmer and cooler conditions [3], affecting nutrient distributions. This then impacts phytoplankton, zooplankton, fish and higher trophic levels, and the northern part of the channel that is on the annual migration routes of the Indian Ocean tuna stocks [1]. The biological effects of the eddies aren’t fully understood but, in a changing climate, could affect the fisheries on which local populations depend both for food and livelihoods. This project will apply machine learning to long-term satellite and in situ observations to unravel the effects of the mesoscale dynamics around Madagascar and in the Mozambique Channel on biological productivity there.
Methodology
This project will combine the latest satellite ocean datasets (Chlorophyll-a [Chl-a], Sea Surface Temperature [SST], Sea Surface Salinity [SSS], Sea Surface Height [SSH], Currents, Winds) in synergy with historical in-situ observations and will apply unsupervised machine learning techniques to the observations.
High-resolution satellite Chl-a (for 1997-2019) and SST (for 1981-2018) datasets from the Climate-Change Initiative (CCI) project (http://www.esaoceancolour-cci.org/ and http://www.esa-sst-cci.org/) will be used to investigate the variability of phytoplankton biomass and cold/warm regional conditions. 10-years of continuous SSS from space (http://cci.esa.int/salinity) will be exploited to help understand changes in circulation features. Satellite winds and altimetry derived SSH and currents (from the 1992 to present) will also be exploited to inform on physical controls.
The connectivity between the Mozambique Channel and northern Madagascar and regions farther north, will be explored using in-situ drifter data. The in-situ drifter data from both the Global Drifter (https://www.aoml.noaa.gov/phod/gdp/index.php) and Argo (https://argo.ucsd.edu/data/) programmes will be used in this study.
The main methods of unsupervised learning to be applied are primarily deterministic approaches, the K-means clustering and Self Organizing Maps which will be used for feature detection and investigating relationships between biological and physical parameters. A stochastic deep learning model, the Boltzmann Machines, will be explored.
Training
The INSPIRE DTP programme provides comprehensive personal and professional development training alongside extensive opportunities for students to expand their multi-disciplinary outlook through interactions with a wide network of academic, research and industrial/policy partners. The student will be registered at the University of Southampton and hosted at NOC, Southampton. Specific training will include:
• computational, statistical and machine learning methods required for the analysis of the remotely sensed and in situ data
• how to access, analyse and interpret satellite observations of the ocean
• there may be an opportunity to take part in a research cruise
Continue with Facebook
