IAP-24-123

Automated Continuous In-situ Dissolved Metal Monitoring & Modelling (ACID M3)

In the UK, historic metal mining has led to the contamination of 9% of the rivers in England and Wales (Environment agency, 2008). Monitoring is carried out by the environment agency – individual water samples are taken and analysed at intervals throughout the year. However, individual data points at a limited number of locations do not give a full picture of the sources and contributions to metal pollution in UK rivers. As part of ongoing monitoring and remediation efforts, an extensive water sampling programme is run by UK government departments (namely the Environment Agency and the Department for Environment Food & Rural Affairs), charities and trusts (such as The Rivers Trust), and private entities who may be required to monitor their pollution. Water bodies across the country are sampled, and these are sent to laboratories for testing (Environment Agency, 2021). A wide range of tests are carried out, including for metal concentrations and various physicochemical properties.

Despite a breadth of historical research and intervention efforts, water quality persists as a major barrier to the environmental health of the UK – with only 16% of rivers classed as having good ecological health. Monitoring this water is the first crucial stage in combating this issue, acting as the foundation of all analysis and research. However, high costs and long turn-around times of traditional water quality testing means that data is often sporadic and inconsistent, severely limiting the quality of analysis. Furthermore, the use of laboratory testing means that data is reactive, and pollution events may only be identified sometime after they have occurred. In essence, the expense of sampling is high and sampling frequency may be low and sporadic and not capture high flow (i.e. extreme weather events, which may become more frequent with global climate change).

Much of the data thus far obtained is publicly available on the ‘Water quality data archive’ – although it is noted that not all the data from private entities and third parties are included (Department for Environment Food & Rural Affairs, 2024). Data is available from January 1st 2000 until the present day and includes the result of fifty-eight million measurements from four million samples, taken from fifty-eight thousand sampling points. Concentrating on the Northumbria River Basin (an area of previous study, currently unpublished), the frequency of sampling can be plotted (Figure 1). Figure 1 highlights the variable nature of data sampling, which when assessing pollution sources, especially those which may be affected by weather (such as discharge from metal mines).

The need for more frequent sampling has come to the forefront owing to recently changed legislation mandating the clean-up of 50% of the metal mine impacted waters in England by 2038. Schematic Figure 2 highlights the sources of such pollution for an abandoned metal mine site. There is therefore a present and urgent need to continuously monitor sites in and around rivers which are affected by metal mine pollution so that appropriate remediation efforts can be made, and action can be quickly taken in the event of a change in conditions. For reliable continuous monitoring, especially when metal mine treatment schemes are installed in the future, there are no presently available methods other than collection of water samples and shipping to laboratories for sample preparation and subsequent analysis.

While a number of probes are currently used in water quality measurement (pH, conductivity and total dissolved solids are particularly readily available and frequently used), for individual metal contaminants, however, owing to a limited number of metals that can be analysed by the use of ion selective electrodes (ISEs), especially for Zn, in situ continuous monitoring, though highly desirable, is not possible. For other elements, such as Ba, Cd, and in some cases Pb, the concentration of these elements often falls below the minimum detection of currently available ion selective electrodes (see Table 1 and Table 2).

In recent work on developing treatment technologies for dissolved metals in legacy metal mine pollution, we have noted that (i) conductivity and pH give insight into dissolved metal content (and associated anions), especially when taken either side of a treatment technology; (ii) metals often co-vary as groupings; (iii) the timespan on which metal content varies is often linked to weather events and not rapid. This allows us to propose a way to monitor metals in real-time, continuously and with greatly reduced need for manual sampling.

Underpinning our approach is the use of machine learning, an artificial intelligence-based method, which has started show promising results for water quality prediction. For example, Desbureaux et al (2022) and Xu et al (2022) both explored the use of algorithms such as neural networks and randomForests to model water quality from a reduced number of easier-to-monitor water properties (such as pH and conductivity).

In a recent project run by the Greenwell Group machine learning was utilised to predict metal pollution in UK rivers from pH, water temperature, conductivity and weather data, with an accuracy of 88% (Trowers, 2024). However, the limited and sporadic water monitoring data serves as a major limitation of these studies. Through using continuous monitoring via electrodes, data can instead be collected consistently in real-time, and streamed directly into such algorithms for live processing and analysis. Further research into the viability of detection methods, analytical and machine learning techniques, and the importance of alternative parameters is therefore a crucial next step in addressing monitoring metal mine water discharge, and the health of our environment.

Click on an image to expand

Image Captions

Figure 1. Water quality data sampling frequency of occurrence in the Northumbria river basin. a) by day of the month, b) by day of the week, c) by month and d) by day of the month and month,Figure 2. Schematic diagram of routes of metal contamination into rivers from abandoned metal mines. From Gibson et al. (2022),Table 1: Commonly available cationic ion selective electrodes, their measurement conditions and indicative price. Highlighted are electrodes for the project. Typical cost for meter for ISE: £3000,Table 2: Typical metals of interest concentration in 2024 (2023 for Parys Mountain) in sites of interest. Highlighted values suggest ISEs may be applicable for measurement.

Methodology

In this project we propose to address the problem of lack of specific metal electrodes through data science approaches and to develop software tools coupled to in situ monitoring in real time to allow the first real-time analysis of dissolved metals in water at legacy metal mine sites. We will aim to secure IP for the technology and develop the tools via our CASE partner, X-Ray Mineral Services Ltd. The student will spend a 3-month placement working with X-Ray Mineral Services Ltd to develop a database of mine-water data using public information (e.g. Environment Agency and Natural Resources Wales, and highlight areas where treatment using methods developed by X-Ray Minerals Services Ltd may particularly be efficiently deployed.

Metals of particular concern in the UK rivers are lead, zinc, cadmium, nickel and copper. Reviewing the commonly available ISEs yields a number of potentially useful probes (Table 1). Commonly in the UK, metal mine drainage tends to be circum-neutral due to the presence of limestone formations in the vicinity of the mine workings, with a couple of notable exceptions where the pH may drop to 3. It would therefore seem to be appropriate to use the Ca, Pb, Cd and Cu ISEs for remote and continuous monitoring alongside total conductivity and pH.

Our approach will be to build upon a recently completed masters in data science project (Trowers, 2024) to:

i) Obtain data from metal mine site water chemistry from the Environment Agency, Natural Resources Wales, the Coal Authority and X-Ray Mineral Services Ltd.
ii) Clean, characterise and convert the data to SI units.
iii) Identify and quantify correlations between pH, Eh, dissolved oxygen, conductivity and reported levels of all metals at UK metal mine sites.
iv) Identify 2 relevant field sites in proximity to Durham with good data. One with a treatment scheme (e.g. Nent Haggs), and another (e.g. Killhope or Cambokeels mine) without a treatment scheme.
v) Use the correlations from the data analysis, data obtained from conductivity, pH, temperature, dissolved oxygen, flow and from ion selective electrodes that are within range (see Table 1 and Table 2) for Pb, Cu, Ca, and potentially Cd, as well as spot samples for ICP to train the machine learning algorithm.
vi) Run the machine learning algorithm for both field sites and for above and below the treatment system as treatment will result in a new water metal composition equilibrium.
vii) Analyse for ion selective electrode redundancy and refine model with fewest necessary electrodes.
viii) Test model on 3 further sides including with significantly different water chemistries: i) Cwmystwyth, Pugh’s Adit (circumneutral, high Zn and some Fe); ii) Woodend Mine, Threkeld (very acidic, extreme Zn levels), Parys Mountain, Dyffryn Adda discharge (very acidic, high Cu levels) (Table 2).

Project Timeline

Year 1

• Months 1-3: Literature review; order electrodes and equipment; training on mine site working (Nenthead Mines Trust); brief project partners and gain all permissions needed for field work. Complete training needs analysis and identify skills gaps.
• Months 4-6: Get feedback on literature review and refine.
• Months 7-9: Undertake calibration and testing of electrodes on single metal solutions vs conductivity. Assess effect of anions (Cl, NO3- and SO42-) on conductivity vs ISE. Develop calibration database. Gain good understanding of performance of ISE.
• Month 9: 1st Review with 2 internal assessors.
• Months 10-12: Publish extended critical literature review in Earth Sciences Reviews or similar.

Year 2

• Months 13-15: Visit first 2 field sites and collect mine waters; initial testing of electrodes in laboratory (no flow) and subsequently in flow reactor to understand performance limitations.
• Months 16-18: First deliverable of algorithm developed on laboratory ICP, conductivity and ISE data and written report.
• Months 19-21: Three month position with X-Ray Mineral Services Ltd to develop database structure and populate with publicly available mine water data. Allow correlation with wider Hafren Group companies (X-Ray Mineral Services Ltd parent company) data sets on geology and soil chemistry. Collection of waters at the Welsh metal mines (Cwmystwyth and Parys Mountain).
• Months 22-24: Laboratory testing of Welsh metal waters for conductivity, pH, and ISE to populate database and refine the model used; tests in batch and under flow; use of dilution and spiking with known metals to further develop models and test predictive capacity of algorithm.
• CONFERENCE during Y2: Attend MineEX conference (Run in Wales to demonstrate advances in Metal Mine pollution understanding, measuring and treatment) and deliver an oral/poster presentation.

Year 3

• Months 25-27: Produce field portable ion selective electrode array, coupled to conductivity, pH, flow and dissolved O2 meter connected via Raspberry Pi (conductivity, pH and O2 meter already in Greenwell Group); testing of field array for robustness in laboratory with all 5 mine waters. Work on methods paper to underpin technology and allow impact case study development. Outreach/public awareness briefing at Nent Head visitor centre.
• Months 28-30: Deploy array in field for 1 month at Nent Haggs and then 1 month in field at Killhope/Cambokeels. Assess data, refine models and algorithm. Identify any weaknesses in field array design and refine. Collect samples for ICP analysis weekly for comparison with ISE/algorithm fitting.
• Months 31-33: Deploy modified array in field for 1 month each at Cwmystwyth, Parys Mountain and Woodend Mine. Collect samples weekly (or arrange with CASE sponsor to) for ICP analysis. Potential to spend further 1-2 months at CASE sponsor offices for Wales based work. Develop paper on the 5 different 1 month field trials to demonstrate continuous data and fit between correlated ISE data/algorithm and ICP results.
• Months 33-36: Complete any last adjustments to field array. Investigate IP options. Write up initial impact study and continue to draft field trial paper and thesis. Clean laboratory, dispose of samples safely.
• CONFERENCE during Y3: Attend International Mine Water Association conference and present an oral contribution.

Year 3.5

• Months 37-39: Completion of paper on application of algorithm to 5 mine water case-studies and minimisation of electrode usage. Handover/training of X-Ray Mineral Services Ltd (CASE sponsor) staff. Stakeholder briefing event (Held at Nenthead Mines Trust community room (Durham Uni partner for teaching metal mine pollution, with mine tours).
• Months 40-42: Completion and submission of thesis and final updates to database and algorithm with company report to X-Ray Minerals Services Ltd.

Training
& Skills

The student will receive training, as needed depending on background, on metal mine chemistry (through attending relevant module in Earth Sciences degree programme, including Level 2 Environmental Fieldwork; Environmental Geochemistry); statistical methods and machine learning (attending relevant modules in Mathematics/Computer Science) as well as specific training in use of electrodes. X-Ray Mineral Services will provide training on commercial applications of data and database methods. Training will be arranged with local providers (via Nenthead Mines Trust) on safe working on metal mine sites. Fieldwork training and first aid/safety will be provided via the Department of Earth Sciences. Newcastle University will provide further training in metal mine data and use of external (Environment Agency, British Geological Society) databases.

References & further reading

• Department for Environment Food & Rural Affairs. (2024). Open WIMS data. Retrieved from Water Quality Archive: https://environment.data.gov.uk/water-quality/view/landing
• Desbureaux, S., Mortier, F., Zaveri, E., Vliet, M. T., Russ, J., Rodella, A. S., & Damania, R. (2022). Mapping global hotspots and trends of water quality (1992–2010): a data driven approach. Environmental Research Letters.
• Environment Agency. (2008). Abandoned mines and the water environment, Science project SC030136-41
• Environment Agency. (2021). Water Quality Archive Documentation. Retrieved from Water Quality Archive: https://environment.data.gov.uk/water-quality/view/doc/reference
• Gibson, Gillian & Stewart, Alex. (2022). Unnatural Cycles: Anthropogenic Disruption to Health and Planetary Functions. Geosciences. 12. 137. 10.3390/geosciences12030137.
• Trowers, S. (2024), Heavy Metal Machine Learning: A data-led investigation of metal pollution in the Northumbria River Basin, Master’s Thesis, Durham University
• Xu, X., Lai, T., Jahan, S., Farid, F., & Bello, A. (2022). A Machine Learning Predictive Model to Detect Water Quality and Pollution. Future Internet.

Apply Now