New Insights on AI in Water: Highlights from the Hydroinformatics Knowledge Exchange

The presented applications covered different parts of the water cycle and made use of various AI techniques. While many of the presented applications were still in an early phase, the discussions highlighted both the opportunities and the challenges of translating AI research into operational use. The participants also discussed the need for further collaboration with academic and knowledge institutions, the development of clear AI policies, and that AI applications need to further mature, before they can be widely adopted in the sector.

Using AI to model water consumption

At the start of the meeting, Pieter Jan Haest (De Watergroep) and Pieter Robberechts (KU Leuven) presented two ongoing projects of De Watergroep regarding the use of AI for the understanding and forecasting water consumption patterns. aims to explore the application of AI techniques to establish general water consumption profiles of customers. This work is based on previous research applied in the electricity sector. The translation of the research in the context of the water sector poses challenges, due to the differences of the two sectors (e.g. water consumption is particularly event-driven and individual). The second project (i.e. Waterdata) concerns the development of a water consumption forecast model for Flanders by sharing data from digital water meters. This project is carried out under the umbrella of AquaFlanders and is supported by the Flemish government and Digitaal Vlaanderen. The project highlights the importance of interoperability (i.e. the ability of different systems, devices, or organisations to exchange, interpret, and use data seamlessly), by aiming to make data from digital water meters widely available for research purposes.

Towards AI driven water quality assessment

Frederic Béen presented some of the joint work that KWR and the Vrije Universiteit Amsterdam are carrying out in the field of advanced chemical screening and data science. More specifically, the team is developing methods to detect various contaminants, including unknown or transformed substances, in water samples. The analysis of water samples usually involves high-resolution mass spectrometry. This technique produces thousands of signals per sample, and thus extensive data processing is required, as well as prioritisation of signals belonging to known toxic compounds. In this context, the application of machine learning techniques can facilitate this data analysis, by analysing the spectra and identifying and classifying the detected chemical compounds.

A very big challenge are unknown substances, where for example only a small fraction of negative effects and impacts on biological life are explained by known substances. Accordingly, the use of AI to predict the presence of structural alerts may aid in the understanding of previously undetected contaminants and their potential toxic effects, helping to reveal patterns and risks that conventional analysis would miss.

Waternet collaborations with academic institutions in the field of AI

The meeting closed with Jim Odenhoven (AGV/Waternet) presenting collaborative work with Dr. Riccardo Taormina, where the experiences of Waternet regarding practical AI applications developed in collaboration with universities were shared. One project investigated whether large language models (LLMs) could be used to query complex asset documentation and operational data. By combining metadata from installations with generative AI, researchers tested whether AI could answer questions about plant performance or support reporting tasks. While initial results showed that AI could reproduce calculations and generate outputs similar to official reports, providing the right context and dealing with imperfect sensor data remained challenging.

Additionally, student projects explored predictive models for pipe failures, to help utilities prioritise replacement strategies more effectively, as well as a hackathon that produced creative tools such as dashboards and automated data collection. Overall, the collaboration highlighted both the potential of AI and the value of working with students and researchers on real-world water management challenges, even if meaningful results require time and iteration.

Looking forward

During the closing discussion participants emphasised the value of continued collaboration with academic and research institutions to translate state-of-the-art research in practical tools for water utilities. Moreover, they underlined that the exchange of experiences and early results (e.g. in the context of the Hydroinformatics Knowledge Exchange meetings, which are organized on the basis of the water utilities’ joint research programme (Waterwijs) helps utilities better understand the opportunities and limitations of emerging AI applications. However, due to the novelty and rather low level of maturity of the developed applications, further research and pilot projects are required, before a wider operational adoption of such applications. Finally, the need for clear AI policies and governance structures was discussed, to ensure responsible, transparent, and reliable use of AI within the water sector.