IAP-24-038

Understanding the role of coastal habitats to fisheries using bioacoustics and machine learning

​​Food security is at risk with a rapidly growing human population. Therefore, there is a need for marine and coastal ecosystems that can maximise fisheries’ productivity1. However, global and local anthropogenic stressors, such as ocean warming, aquaculture, and pollution, have degraded coastal habitats and the fauna that reside in them1. Coastal habitats, such as seagrass beds, oyster beds and seaweed aggregations, provide invaluable ecosystem services, including food by acting as a feeding, shelter and nursery area for important commercial fish species such as pollock, plaice and herring1. However, some of these habitats, such as seagrass, have declined by 92% in the last 100 years in the UK. The effects of this degradation on the utilization by fish remain challenging to understand and are largely undocumented, as is the effectiveness of habitat restoration. The latter is important, as Blue Growth industries such as aquaculture and wind farms are rapidly expanding, which could lead to further degradation of coastal habitats and/or investments in restoration projects. Understanding fish use of coastal habitats across the UK and the economic value associated with this is important to ensure the sustainable development of our oceans.

​Passive Acoustic Monitoring (PAM) and soundscape metrics are new approaches which are being used to understand how habitat health and marine biodiversity changes in time and space2. Short-and long-term recordings have been used for rapid and cost-effective assessments of biodiversity, habitat health and restoration success2,3. However, so far, there is no research on the biogeography of coastal soundscapes in the UK and the extent to which commercially important fish utilize these habitats. Since analysing species-specific sounds from acoustic datasets can be time-consuming, there is a need to develop new machine-learning-based solutions to increase the effectiveness and efficiency of coastal habitat health monitoring4,5.

​Understanding habitat use is vital to maintaining healthy fish populations, which in turn supports food security and the livelihoods of millions of people who depend on fishing. It is critical that we detect early signs of changes in habitat use by fish as a consequence of habitat degradation or restoration, so appropriate actions can be undertaken to help or further improve the preservation of biodiversity, maintain ecosystem functions, and protect marine ecosystems and the human communities that rely on them.

​Aim: This interdisciplinary PhD project will contribute to filling in gaps in our understanding regarding coastal habitat use by fish, by answering the following research questions.

​What fish and fish sounds are associated with coastal habitats in Scotland?

​What environmental and anthropogenic factors drive spatiotemporal differences in the occurrence fish sounds?

​Which machine-learning based method is most effective at automatically identifying fish sounds.

​What is the economic value of coastal habitats in relation to fish use?

​​To address these questions the PhD project will focus on three objectives and their respective methodological approaches:

​Objective 1: Gain an understanding of the environmental and anthropogenic factors that may impact the presence of fish and their sound production by conducting a literature review, and analysis existing data.

​Approach: To get familiar with fish sounds associated with coastal habitats, first existing data collected as part of a project with Sail Britain, led by the primary supervisor Laurence De Clippele, will be analysed. Acoustic properties such as the frequency range, length and peak frequency will be extracted using Raven Pro and/or Python/R software and used to create a dichotomous key for sound type identification (months 1-3). Environmental, fishery and anthropogenic impact/activity GIS data layers will be extracted from open-source databases (e.g. EMODnet, gov.scot). These will, for example, include layers on the distribution of coastal habitats, essential fish habitats, terrain variables (e.g. depth, slope), oceanographic conditions (e.g. wave exposure) anthropogenic impacts/activities (e.g. aquaculture). Based on the literature review and datasets, a sampling plan for fieldwork will be designed (months 3-6). New PAM and underwater image data will be collected across a range of coastal habitats across the west coast of Scotland, and additional data on habitat complexity will be measured in terms of rugosity and cover of substrates and species such as kelp and seagrass. There will be an opportunity to join Sail Britain to collect data along the west coast, and to collaborate with SeaWilding. These newly collected datasets will manually be annotated for fish sounds and related to the datasets found as part of the literature review to unravel what impacts fish sound occurrences, and therefore functional habitat use, across the West coast of Scotland.

​Objective 2: Develop new machine learning-based monitoring solutions to be applied in monitoring fish presence in coastal habitats

​Approach: Signal processing and deep learning neural network approaches will be tested to develop models which can be used to automatically annotate species-specific fish sounds in acoustic recordings. The fish sounds, annotated as part of the first objective, will be used to train and test the models. If needed, additional data can be collected as part of a second field season.

​Objective 3: Quantify the value of coastal habitats for fisheries using the ecosystem goods and services framework

​Approach: Use quantitative methods to estimate and compare the direct economic impacts as a consequence of the revenue of fishing, based on the frequency of occurrence of fish in coastal habitats and the distribution of these habitats, along with the estimated indirect benefits of restoration and improved marine protection.

Year 1

​​Literature review on the status, distribution and fish use of UK’s coastal habitats; Annotate and characterize fish sounds using existing acoustic recordings; Plan and conduct fieldwork to collect acoustic and environmental data from a range of coastal habitats; Annotate newly collected datasets; Prepare/submit manuscripts for publication; attend conferences

Year 2

​​Spatial analysis, using GIS and statistical approaches to understand what drives differences in fish sound occurrences across different habitats; Develop machine learning models to automate fish identification from acoustic recordings; Conduct additional fieldwork if needed; Prepare/submit manuscripts for publication; attend conferences

Year 3

​​Use quantitative methods to estimate and compare the direct economic impacts of coastal habitat conservation and restoration on the revenue of fisheries; Prepare/submit manuscripts for publication; attend conferences

Year 3.5

​​Writing-up of PhD, attend conferences, submit-peer reviewed publications​

The student will receive interdisciplinary training in key skills relevant to the data collection and analysis across multiple disciplines, including (i) acoustic analysis (De Clippele/ Park/Mallik), (ii) ecological experimental design & snorkeling (De Clippele/Park/Bailey), (iii) machine-learning-based data modelling skills, including using signal processing and deep learning approaches (De Clippele/ Mallik), (iv) environmental economic valuation (De Clippele/Simpson/Bailey) and (v) combined ecological-economic analyses approaches (De Clippele/Simpson/Park/Bailey).