ARIES project 2027

Integrating machine learning and remote sensing to track pressures on migratory fish in UK rivers and estuaries    

Many migratory fish in the UK that move between marine and freshwater systems are in serious decline, with some species at risk of extinction. Water deficits, urbanisation, agricultural pressures, and climate change are altering river habitats and water quality and quantity, with consequences for ecologically, commercially, and culturally important species including Atlantic salmon, brown trout, and European eels. As commercially important species and indicators of river and estuarine ecosystem health, they provide valuable economic and ecological insights into environmental change. However, conventional monitoring is often spatially and temporally limited, hindering understanding of how environmental pressures affect ecosystems and fish populations over time.

This project will develop a novel, high-resolution approach combining remote sensing and machine learning to investigate environmental change in UK rivers and estuaries. The River Stour, Kent, and River Frome, Dorset, will provide initial case studies before the methodology is scaled to rivers across the UK that support important migratory fish populations.

Methodology: 

Approximately ten years of monthly, high-resolution satellite imagery will be combined with LiDAR data. Using Python and GPU-accelerated computing, deep-learning approaches, including U-Net convolutional neural networks and YOLO, will extract, segment, detect and classify environmental features.

Anthropogenic pressures will be tracked through changes in river course, intra- and inter-annual variation in water colour, urbanisation, tree cover, and crop colouration. Remotely sensed indicators will be integrated with official migratory fish and water-quality datasets to investigate relationships between environmental change and fish populations, while accounting for imperfect detection. Fieldwork alongside the Environment Agency and GWCT will provide additional fish and water-quality data from the Frome and Stour. The validated methodology will then be applied to other UK rivers.

Training:

The PGR will receive interdisciplinary training in remote sensing, machine learning, spatial analysis, environmental modelling, and quantitative ecology. They will develop expertise in programming, image analysis, large environmental datasets and statistical modelling.

Person Specification:

We seek an enthusiastic, motivated individual interested in environmental science, freshwater or marine ecology, remote sensing, data science, or conservation. Programming experience is required, with further training provided. Strong analytical and problem-solving skills and enthusiasm for interdisciplinary research are essential. Acceptable first degree subjects: BSc in computer or data science, ecology, biodiversity or related subjects.

Supervisory Team: 

• Dr Jake Bicknell, University of Kent 
• Dr Sophie Elliott, University of Kent  
• Dr Marek Grzes, University of Kent