Don't miss our weekly PhD newsletter | Sign up now Don't miss our weekly PhD newsletter | Sign up now

  Project S2121: A Machine Learning Accelerated Spectral Matching Algorithm to Identify Harmful Algal Blooms and Other Threats to Aquaculture Operations. (Dr David McKee, University of Strathclyde)


   Department of Physics

This project is no longer listed on FindAPhD.com and may not be available.

Click here to search FindAPhD.com for PhD studentship opportunities
  Dr David McKee, Prof Paulo Prodohl  No more applications being accepted  Funded PhD Project (Students Worldwide)

About the Project

The Food and Agriculture Organisation of the United Nations estimates that total aquaculture production in 2022 reached ~131 million tonnes, with a farm-gate value of USD $312.8 billion. In the same year, aquaculture production of animal species exceeded animal capture from the wild for the first time. Harmful algal blooms (HABs) present a severe threat to the aquaculture industry, with reports suggesting annual losses in regional sectors of amounts varying between tens and hundreds of millions of dollars. More recently, there has been a growing appreciation of the threat posed by swarms of microjellyfish, and there are situations where other factors, in particular the introduction of anoxic waters, can lead to fish kills.

To mitigate these threats, there have been several attempts to develop early warning schemes using ocean colour satellite data as a primary source of information, with varying degrees of success. Some limitations are unavoidable, including the absence of data under cloudy skies. However, a more tractable limitation has been the quality of ocean colour data collected for optically complex coastal and shelf seas. Two major problems arise. The first is poor quality atmospheric correction of reflectance spectra over turbid waters, where the quality of the optical signal is degraded by inability to properly remove the effect of atmospheric scattering. The second is poor performance of ocean colour algorithms in the presence of significant concentrations of sediments (MSS) or coloured dissolved organic material (CDOM), as is often the case in coastal waters. More recently, our group at UoS has shown that there is scope for other material classes, particularly zooplankton, to significantly contribute to ocean colour reflectance signals.

The UoS group has developed a methodological framework based on meticulous radiative transfer modelling of surface remote sensing reflectance signals that allows for the identification of anomalous reflectance signals that deviate from the standard bio-optical model consisting of phytoplankton (Chl), sediments and CDOM. Key features of this approach are that it allows selection of an appropriate atmospheric correction on a per pixel basis and it is easily ported to different satellite sensors. It has significant potential to address fundamental issues in retrieval of basic water quality parameters (Chl, MSS and CDOM) and to provide important new information on the presence and concentration of potentially toxic red algae blooms, cyanobacteria blooms and other anomalous events. There are a couple of limiting features of the approach, the most significant of which is the computational requirements. We propose to address this through the use of machine learning techniques to emulate the spectral matching results, providing an accelerated version of the algorithm for operational use.

While optical oceanography provides valuable insights into the presence and concentration of various algal species and other optically significant constituents in marine environments, integrating a complementary genomics component can significantly enhance the early warning capabilities for detecting harmful algal blooms (HABs). The QUB group works on genomics-based approaches, such as metabarcoding or metagenomics of environmental DNA (eDNA) derived samples, which have shown promise in providing species-specific information about the composition and dynamics of microbial communities, including harmful algae, in real time. By analysing genetic material directly from water samples, these methods can detect HAB species at low concentrations before they become visible to remote sensing tools, thus providing an earlier warning.

The genomics component of this project will involve the collection of water samples from key aquaculture sites affected by HABs and microjellyfish. Metabarcoding techniques will be used to amplify and sequence specific genetic markers (e.g., 18S rRNA, ITS regions) from these samples, allowing for the rapid identification and quantification of harmful algal species and other potentially toxic microorganisms. The results will be integrated with the optical data to validate remote sensing anomalies and potentially refine the spectral matching algorithms.

The PhD project will:

  • WP1. Extend the library of optically significant constituents focusing on known toxic algal species and targeting microjellyfish species known to impact on Scottish aquaculture operations.
  • WP2. Develop and apply an extended version of the current spectral matching algorithm incorporating results from WP1 and results from on-going work on coccolithophores and krill.
  • WP3. Develop and validate an accelerated version of the algorithm using machine learning-based emulation.
  • WP4. Identify case studies of known HAB events and assess performance of the spectral matching approach against standard algorithms and other published approaches.
  • WP5. Carry out comparative analyses of known HAB events using both genomics and optical approaches to assess the performance of the combined method against traditional methods. This aligns with WP4, where case studies are used to evaluate the spectral matching approach.

The project builds upon a growing body of evidence that the spectral matching approach is technically sound. We have a well-established network of partners that will facilitate access to samples e.g. microjellyfish and have good links with other experts in this domain. Satellite data is open access and we have the measurement and software tools needed to run this type of analysis. 

This project is offered as part of the UKRI AI Centre for Doctoral Training in Sustainable Understandable agri-food Systems Transformed by Artificial Intelligence (SUSTAIN). To learn more about SUSTAIN (including eligibility requirements and our fully-funded studentship package), and to apply, please visit https://www.sustain-cdt.ai/.


Biological Sciences (4) Computer Science (8) Engineering (12) Environmental Sciences (13) Geography (17) Physics (29)
Search Suggestions
Search suggestions

Based on your current searches we recommend the following search filters.

 About the Project