CR2025_44 Cod in crisis in the Celtic Sea: Predicting the Cod distribution using Satellite Ocean Colour and Machine Learning

Lead Supervisor: Shovonlal Roy, Department of Geography and Environmental Sciences, University of Reading

Email: shovonlal.roy@reading.ac.uk

Co-supervisors: Robert Thorpe, Centre of Environment Fisheries and Aquaculture Science (CEFAS); Paul Dolder, Centre of Environment Fisheries and Aquaculture Science (CEFAS); Hong Wei, Department of Computer Science, University of Reading

Ensuring a sustainable supply of food for the nation is a major priority for the UK Government. This includes access to sustainable marine-sourced food, particularly fisheries. The Celtic Sea is an important component of the UK shelf seas (Figure-1), and home to a complex mixed fishery where multiple species are caught together, with fisheries targeting one stock but catching others concurrently. The Celtic Sea has been subject to overfishing in the past, leaving cod in a parlous state. Managers hope to recover the situation by enforcing quotas, limiting catches of major commercial species to preserve them, with particularly low quotas for the highly impacted cod. To conserve cod, fishing on many other stocks is kept below sustainable levels since boats unintentionally catch cod and exhaust its quota whilst fishing for other stocks, limiting the fishery below its true potential.

With better cod forecasts in near-real-time (NRT), boats would be able to avoid unintentional catches of cod, and could increase their fishing on other stocks without increasing the risks to the cod population, resulting in a more productive and sustainably managed fishery. Given the high mobility of cod within a heterogeneous and data-poor environment, such a forecast has hitherto been impossible, but developments in satellite remote sensing (RS)-based monitoring and machine learning (ML) alongside increased computing power for big data are making short-term operational forecasts possible. Hence, this project will build a novel data-driven modelling framework that can provide NRT prediction of the distributions of the highly mobile Celtic Sea cod, which is sought after by fisheries managers and policy makers in the UK/EU, as well as by the International Council for Exploration of the Seas (ICES). 

Fishing vessels provide data on their operations in NRT, showing if quota is being used inefficiently with unfavourable catch ratios (e.g. too much cod relative to say haddock). Leveraging these information, we will develop a novel amalgamation of RS and ML techniques to generate a new product, a short-term forecast for cod in the Celtic Sea which can be used by the fishery to improve its operational efficiency. The project outcome can be used to better manage the fishing vessels e.g. areas to avoid, reducing the rate at which cod is caught relative to other stocks with more generous quotas, and improving the overall catch of the fishery.

The proposed NRT forecasts would require near-real-time information of the drivers that determine the dynamics and distribution of marine fish. RS of ocean can routinely provide the advanced information on marine ecosystems and local ecological and environmental drivers of fish distribution, which has previously been utilized by the supervisors in their joint projects (e.g. [2, 3]). Marine environmental drivers such as sea-surface temperature, incident light, ocean salinity, wind direction and climate indices are currently available as standard products of RS on NRT at a high spatial resolution. Additionally, recent advances in RS led by the supervisors can provide estimates of biomass, species composition and nutritional status of the microscopic autotrophic primary producers (e.g. [4-5]), which regulate the dynamics of marine fish through food-web connections. This project will utilize these large datasets on RS-derived products and cod distribution through advanced ML techniques (e.g. [6]) and develop a novel NRT prediction system focused on Celtic Sea cod.

The student will:

• Collate historical cod catch data in the Celtic Sea from various data archives, process high-resolution RS images from ESA/NASA archives, and generate advanced ocean-colour products as environmental drivers. (Year 1)
• Develop an ML-based model, e.g. based on a variant of the Deep Neural Network (DNN) incorporating multiple marine environmental drivers (e.g. plankton composition, light, temperature, salinity, climate indices), and train the model with historical cod data (Year 2).
• Apply the ML model to provide NRT forecasts of cod in the Celtic Sea, which will then be validated using forthcoming catch data. (Year 2-3)
• Apply the NRT forecasts in collaboration with fisheries managers and policy makers to improve the efficiency of cod fisheries management in the Celtic Sea. (Year 3)

This project includes CASE sponsorship from CEFAS, UK.

Training opportunities: 

The project is multidisciplinary, involving large fish-catch data, satellite remote sensing, ML, statistical methods and biological oceanography. The student will get an opportunity to visit the CASE partner DEFRA-CEFAS, and learn to access their data archives and in-house lab facilities and will learn about fisheries data collection and tagging. Training will be given on processing of satellite remote sensing and ecological modelling (in GES, UoR), building ML models from the AI/ML experts in Computer Science (UoR), model calibration and prediction (UoR, Cefas). The student will learn coding for data visualization and ML modelling using python, FORTRAN/ MATLAB/R. 

Student profile: 

This project would be suitable for students with a degree in Physics, Mathematics, Meteorology, Oceanography or a closely related environmental or physical science. Training/support will be available for motivated students.

Co-Sponsorship details:

This project will receive a CASE award from CEFAS. 

References:

1. Hernvann et al. (2020) The Celtic Sea Through Time and Space: Ecosystem Modeling to Unravel Fishing and Climate Change Impacts on Food-Web Structure and Dynamics. Front. Mar. Sci. 7:578717
2. Boyd, R., Roy, S., Sibly, R., Thorpe, R. , Hyder, K. (2018) A general approach to incorporating spatial and temporal variation in individual-based models of fish populations with application to Atlantic mackerel. Ecological Modelling , 382, 9-17, https://dx.doi.org/10.1016/j.ecolmodel.2018.04.015
3. Boyd, R., Sibly, R,, Hyder, K. , Walker, N., Thorpe, R. , Roy, S. (2020) Simulating the summer feeding distribution of Northeast Atlantic mackerel with a mechanistic individual-based model. Progress in Oceanography , 183, https://dx.doi.org/10.1016/j.pocean.2020.102299
4. Roy, S. (2018) Distributions of phytoplankton carbohydrate, protein and lipid in the world oceans from satellite ocean colour, The ISME journal, 12(6), 1457-1472.
5. Roy, S. et al.(2017) Size-partitioned phytoplankton carbon and carbon-to-chlorophyll ratio from ocean colour by an absorption-based bio-optical algorithm. Remote Sensing of Environment , 194, 177-189.
6. Yun Bai, et al. (2021), “Regression modelling for enterprise electricity consumption: a comparison of recurrent neural network and its variants”, International Journal of Electrical Power & Energy Systems, Part A, 126(3):106612