Monitoring Scotland’s fresh waters from space: re-engineering the state-of-art using artificial intelligence - PhD

University of Stirling

Stirling, UK 🇬🇧

Scotland’s freshwater ecosystems are under significant pressure from a range of stressors including inputs of diffuse pollution from agriculture, wastewater spillages from combined sewer outflows, and the increasing occurrence of hydroclimatic extremes (Marsden & Mackay, 2001; Krokowski, 2022; May et al., 2022). Identifying and understanding the effects of these pressures on water quality is pivotal to inform management and restoration strategies, to apprise future regulatory and policy frameworks, and ultimately to protect biodiversity and human health. However, organisations across the water sector, from government regulators through to water utilities, are finding it increasingly challenging using traditional methods to collect the data they need to evidence decision-making. What is more, the move towards net zero operations is forcing the industry to re-think its use of resources and to seek ‘smarter’ ways of collecting and using data.

Monitoring water quality at scale has always been a challenge. Conventional sampling programmes are resource intensive and, while networks of high frequency autonomous sensors hold much potential, high capital and maintenance costs remain a barrier to deployments at scale. Conversely, satellite remote sensing provides a relatively cost-effective means of collecting data on surface water quality over large geographical scales – while also providing observations at a frequency that enables both short-term events (e.g., storm-induced sediment resuspension, algal blooms) and longer-term changes (e.g., those driven by catchment land-use and hydroclimate change) to be resolved. However, the accurate estimation of water quality parameters (e.g., turbidity, chlorophyll-a) remains a challenge for applications over inland waters – particularly for smaller waterbodies only observable from platforms such as the Copernicus Sentinel-2 constellation (Palmer et al., 2015).

Much effort has been invested over the last decade in the development, validation, and fine-tuning of algorithms for the retrieval of water quality parameters from satellite data. Various solutions have been proposed from relatively simplistic empirical models to more complex physics-based solutions (Neil et al., 2019). Recently, there has been renewed interest in the application of artificial intelligence and machine learning (AI/ML) approaches to overcome some of the common limitations of more conventional solutions to the inversion of satellite data. Numerous AI/ML methods have been used to assist water quality mapping from space (Pahlevan et al., 2021; Werther et al., 2022; Yang et al, 2022), some with notable success but, in spite the ever-increasing research base, there remains little consensus on what AL/ML methods perform best and how to apply them.

The studentship will be broadly structured into three inter-linked phases: (1) global model development and testing; (2) model validation for UK waters; (3) model fine-tuning and operationalisation – but with flexibility for the student to pursue other complementary research objectives as the project develops.
The studentship will capitalise on the wealth of existing data held by the Earth observation research group at the University of Stirling as well as community-owned databases (e.g., Limnades) and published datasets (e.g., Gloria). This will be augmented by in-situ monitoring data from SEPA and Scottish Water.

The ambition of this studentship is to fundamentally re-think how we use AI/ML methods for water quality mapping from space across the entire data processing pipeline, from pre-processing (e.g., atmospheric correction) through to the quality control post-processing, and possibly even for automating data interpretation. The student will work with experts in Earth observation and data science to re-engineer current approaches to the remote sensing of water quality with the prospect of the outputs of this research being embedded in future versions of the UK Lakes Observatory – our current operational water quality service. With support from key industrial partners, including the Scottish Environment Protection Agency (SEPA) and Scottish Water, this studentship has the potential to transform how we monitor Scotland’s rivers and lochs in the future.

Methodology

Phase 1: Global model development & testing
The student will explore a range of AI/ML approaches including convolutional neural networks, deep learning algorithms, recurrent neural networks and hybrid models for the retrieval of water quality parameters (i.e., turbidity, chlorophyll and, possibly, coloured dissolved organic matter) from Sentinel-2 satellite observations using existing global datasets (i.e., Limnades, Gloria). This will include a comparison of retrieval accuracies from top and bottom of atmosphere reflectances. Models will be benchmarked against the current state of the art with the best performing models advancing for further regional and case-study based validation in Phase 2.

Phase 2: Model validation for UK waters

The best performing models from Phase 1 will be evaluated against high-quality in-situ data from SEPA and Scottish Water – and other organisations monitoring water bodies in the UK. This validation will be undertaken at regional (Scotland/UK) scale and at a case-study scale for priority waterbodies including sites equipped with sensors for high-frequency data acquisition (e.g., Loch Leven, Carron Valley Reservoir). The main objective of this work phase will be to identify which models are suitable for water quality mapping in the UK. The student will also develop a UK-specific methodology for the dynamic selection of models (e.g., based on existing global optical-water-type concepts) to optimise overall performance if no single approach works adequately for all waterbody types.

Phase 3: Model fine-tuning & operationalisation
The final phase of the project will focus on fine-tuning of the validated models from Phase 2 for operational readiness. This could involve exploration of solutions for tackling important but often overlooked components of the data processing chain including methodologies for spatio-temporal data aggregation and gap-filling as well as the automated identification of outliers and anomalies. Computational efficiency of algorithms will also be explored. The final component of the studentship will be to explore whether AL/ML can be used to assist the use of satellite-derived water quality information by exploiting the ‘big data’ potential of satellite datasets to infer and forecast system behaviour at timescales of interest to responsible authorities.

Project Timeline

Year 1

• Development of literature review and project proposal
• Basic familiarisation with in-situ datasets and methodologies
• In-depth training in AL/ML techniques and programming frameworks
• Collation and screening of satellite and in-situ datasets
• Evaluation of AI/ML approaches used for satellite water quality estimation and selection of subset to develop.

Year 2

• Review of progress in Year 1
• Development, validation and refinement of AI/ML models using UK datasets
• Development of AI/ML framework for dynamic model selection
• Draft of first paper from Year 1 results
• Short-term placement with SEPA
• Develop career development plan

Year 3

• Review of progress in Year 2
• In-depth case-study and validation of models
• Development of AL/ML approaches for data post-processing- anomaly detection and classification
• Draft of second paper from Year 2/3 results and thesis chapters

Year 3.5

• Draft of final paper from Year 3 results
• Draft additional thesis chapters
• Submit thesis for examination
• Undertake viva examination

Training & Skills

In addition to the training provided through the IAPETUS DTP, the student will receive in-house and external training in all key components of the project including:

• Satellite data processing and analysis
• AI/ML methodologies and applications
• Scientific programming in Python and/or R
• Field sampling and laboratory analysis of water samples (mainly for experience)

The student will present their findings annually within a postgraduate research symposium specific to the universities of Stirling and Glasgow and international conferences. The student’s progress will be subject to annual progress reviews. All research students are members of Stirling Graduate School and are encouraged to attend seminars (that are particularly relevant to them) in addition to the generic training skills provided by the IAPETUSII DTP. Students also take advantage of the opportunities for networking with external visitors and students from other academic areas to promote interdisciplinarity. The student will also spend some time on placement with SEPA.

References & further reading

Linda May, Philip Taylor, Iain D. M. Gunn, Stephen J. Thackeray, Laurence R. Carvalho, Peter Hunter, Mairéad Corr, Anne J. Dobel, Alanna Grant, Gemma Nash, Emma Robinson and Bryan M. Spears (2022). Assessing climate change impacts on the water quality of Scottish standing waters. CRW2020_01. Scotland’s Centre of Expertise for Waters (CREW).

Krokowski, J.T. (2022). Update on incidences of cyanobacteria (blue-green algae) in Scottish freshwaters. The Glasgow Naturalist 28, Part 1.

Marsden, M. W., & Mackay, D. W. (2001). Water quality in Scotland: the view of the regulator. Science of the Total Environment, 265(1-3), 369-386.

Neil, C., Spyrakos, E., Hunter, P.D. and Tyler, A.N., 2019. A global approach for chlorophyll-a retrieval across optically complex inland waters based on optical water types. Remote Sensing of Environment, 229, pp.159-178.

Pahlevan, N., Mangin, A., Balasubramanian, S.V., Smith, B., Alikas, K., Arai, K., Barbosa, C., Bélanger, S., Binding, C., Bresciani, M. and Giardino, C., 2021. ACIX-Aqua: A global assessment of atmospheric correction methods for Landsat-8 and Sentinel-2 over lakes, rivers, and coastal waters. Remote Sensing of Environment, 258, p.112366.

Palmer, S.C., Kutser, T. and Hunter, P.D., 2015. Remote sensing of inland waters: Challenges, progress and future directions. Remote sensing of Environment, 157, pp.1-8.
Werther, M., Odermatt, D., Simis, S.G., Gurlin, D., Lehmann, M.K., Kutser, T., Gupana, R., Varley, A., Hunter, P.D., Tyler, A.N. and Spyrakos, E., 2022. A Bayesian approach for remote sensing of chlorophyll-a and associated retrieval uncertainty in oligotrophic and mesotrophic lakes. Remote Sensing of Environment, 283, p.113295.

Yang, L.; Driscol, J.; Sarigai, S.; Wu, Q.; Lippitt, C.D.; Morgan, M. Towards Synoptic Water Monitoring Systems: A Review of AI Methods for Automating Water Body Detection and Water Quality Monitoring Using Remote Sensing. Sensors 2022, 22, 2416. https://doi.org/10.3390/ s22062416


POSITION TYPE

ORGANIZATION TYPE

EXPERIENCE-LEVEL

DEGREE REQUIRED

You ad could be here!