Northumbrian Water’s project partners are: ADAS, Anglian Water, Cognizant, Northern Ireland Water, South West Water, Stream, The Rivers Trust, Google, WRc, Wessex Water and Xylem.

Funded by Ofwat’s Water Breakthrough Challenge and led by Northumbrian Water, with Spring Innovation as the knowledge-sharing partner, RDMAI is a cross-sector initiative building open-source, scalable AI tools to tackle waterbody pollution and improve river health.

Data from a range of sources, including citizen science and satellites, has been used to build the models.

The release of AI/ML and remote-sensing models on the open-source platform GitHub is the project’s first major milestone, following completion of the development and initial testing phases. Throughout this period, the project team collated datasets from within and outside the sector, ran experiments with AI/ML models and held co-creation sessions with partners and stakeholders.

The resulting models and datasets aim to support:
• River flow predictions
• Pollution source tracking
• Pollution hotspot mapping (Feedback is invited at this stage to help refine and enhance the models as the project progresses).

The UK’s water environment is under pressure from population growth, climate change, pollution from multiple sources and nutrient overload. Just 14% of English rivers are meeting Water Framework Directive standards for good ecological status.

River Deep Mountain AI aims to address this challenge by developing open-source, scalable AI/ML models to uncover pollution patterns and unlock actionable insights for protecting waterbodies.

George Gerring, project lead, Northumbrian Water, said, “We have built a set of capabilities that use artificial intelligence, machine learning, generative AI and remote sensing to understand and predict different variables impacting waterbodies health.

“The open-source release of these models on GitHub means they are available for citizens, researchers, water organisations and NGOs to use. Any feedback on the early releases will help us refine and build on what we’ve achieved so far.”

Angela MacOscar, head of innovation, Northumbrian Water, said: “Useable data on waterbody health is disparate and hard to access, which is why the RDMAI team is working to squeeze as much actionable information out of existing data as possible.

“By integrating data from various sources, including environmental sensors, satellite imagery and citizen science, the project is bridging the data gaps in waterbody health and empowering better, faster and more effective interventions. Open-sourcing these models marks a major shift in how we collaborate to tackle environmental challenges.”

Stig Martin, global head of ocean, Cognizant, said: “This project is a testament to the power of research and development and daring to use technology to solve complex, large-scale environmental problems.

“We believe in transparency and are proud that this project is open-source, allowing everyone to see how the system is built and how it generates its insights. It has been incredibly rewarding to be part of a collaboration that is not just talking about change but is actively building the tools to make it happen.”

Cognizant RDMAI write: The scope of RDMAI model development has been selected based on environmental impact, stakeholder needs, scalability and the potential of AI to provide benefits over current approaches. As part of an early release, we have open sourced the first batch of AI/ML and remote sensing models.

This includes:
• Open Flow Model (v2): Using AI/ML to estimate daily mean river flow in ungauged rivers.
• Open E. coli Model (v2): Using AI/ML to predict E. coli concentrations and risks in bathing waters.
• Open Slurry Tank Detection (v2): Using remote sensing and computer vision to identify and map circular slurry tanks across river catchments.
• Open Bare Cropland Detection Model (v2): Using remote sensing to identify bare cropland.
• Open Risk Map (v2): Collating catchment data to map pollution hotspots and risk areas.
• Open Orthophosphate Model (v2): Using AI/ML to predict Orthophosphate concentrations in river catchments.
• Open Poached Land Detection: Leveraging machine learning and remote sensing to detect soil degradation.

In River Deep Mountain AI we believe that solutions to big challenges come through collaboration, transparency and a systemic approach. These principles underpin the project throughout and we have implemented several ways of working that help to support this. These include; cross-sector partnerships, co-creation with stakeholders, building for scale, and most importantly knowledge-sharing.

For more information about River Deep Mountain AI, visit: https://github.com/Cognizant-RDMAI For more information about Xylem, visit: www.xylem.com