A Week of Water Futures at KWR

The Water Futures project tackles this ambitious research problem that cannot be addressed by a single team alone. This project is enabled by the ERC Synergy Grant funding (2021-2027), which is one of the most prestigious funding opportunities within the EU Horizon Europe programme. In Water Futures, the problem is tackled by the combined expertise of four research groups from the Athens University of Economics and Business (AUEB, Greece), Bielefeld University (BU, Germany), KIOS Research Innovation Centre of Excellence (KIOS, Cyprus) and KWR Wate Research Institute (KWR, the Netherlands). They combine expertise from the environmental economics, machine learning, control engineering and water systems analysis fields to untangle the complex socio-technical challenge of planning flexibly for the future.

On the 17^th and 18^th of October, the Water Futures team was brought together for the seventh time – this time in Nieuwegein at the premises of KWR Water Research Institute. However, the meeting actually started on Monday of that week when a group of researchers from Bielefeld University and one from TU Delft came to KWR to discuss collaboration and joint opportunities with KWR’s five PhD students employed on this project.

The Pre-Meeting (14-16 October)

During those first three days of meetings, those early-career researchers discussed various Artificial Intelligence (AI) topics of interest, including the use of Graph Neural Networks (GNN) for developing model emulators (surrogate models) and metamodels, through generating ideas on the use of Reinforcement Learning (RL) and GNNs for modelling and optimisation, to the AI techniques for image analysis and optimisation of water systems. They also enjoyed sharing the experience of jointly participating in an international competition to predict short-term water demand at various scales. The team came second in the competition, but true to the project vision, our solution showed the least uncertainty in its results – proving that sometimes it’s not about winning, but about being the most reliable solution.

A great many ideas were generated during those three days of discussions, but more importantly, conversations were held in a relaxed and open atmosphere. The dynamic and open interactions over three days have undoubtedly energised our joint team and reinforced our commitment to achieving our shared goals.

Day One (17 October)

On the first day of the official meeting, the four Principal Investigators (PIs) – Professors Phoebe Koundouri (AUEB), Barbara Hammer (BU), Marios Polykarpou (KIOS) and Dragan Savić (KWR and Exeter University) – were joined by over 30 Water Futures researchers for an intense day of presentations and discussions. This time, the meeting focused more on the latter than on the former, proving that sometimes it’s the conversations that steal the show. The four teams have published a number of seminal peer-reviewed papers, including Prof Koundouri’s paper in Springer Nature, and delivered invited and keynote talks at various conferences. Notably, Prof Hammer has an important paper on explainable AI accepted for presentation at AAAI-25 in Philadelphia next year. The PIs have also received some of the most prestigious awards and recognition, such as the 2023 IEEE Frank Rosenblatt Technical Field Award to Prof. Polykarpou (with two previous winners of this award winning the 2024 Nobel Prize for this year) and the 2024 IWA Distinguished Fellowships to Prof. Savić. Clearly, our PIs are not just leading the field – they’re practically running a victory lap!

After the introductory overview of the project progress by each of the four PIs, a much longer discussion ensued about possible synergistic ideas in the project. For example, we discussed how can uncertain economic parameters and human decision-making attitudes be included in the long-term design and operation of water systems. Another interesting topic covered by the discussion was how can the notion of fairness be included in the decision-making about water systems, which draws upon legal, economic, technical and machine-learning perspectives.

In the afternoon, we invited two external expert speakers, Professors Riccardo Taormina (TU Delft) and John List (University of Chicago) to present their work on GNNs in water distribution systems and the economics of scaling. Why is this important to us? Because AI and GNNs have made a huge impact on water distribution system research, and also because scaling of innovation and research project outcomes is essential for projects such as ours, and for achieving wider SDG6 goals. It was incredibly exciting and inspiring to have two world-renowned experts present on their respective topics. Their insights not only broadened our understanding but also sparked a lively and engaging discussion among all participants. The exchange of ideas and perspectives that followed each presentation was evidence of the synergy and intellectual curiosity that drives our project forward.

Networking in a social setting is crucial for building strong personal, research and professional relationships and fostering collaboration. Our project dinner in the historic centre of Utrecht was a perfect example of this, as it provided an informal environment for team members to connect and share ideas. Despite a mix-up with the restaurant reservations (we are not perfect), the evening turned out to be a fantastic experience. The relaxed atmosphere allowed for meaningful conversations and strengthened our team bond, proving that sometimes the best moments come from unexpected situations.

Day Two (18 October)

During the second day (actually half a day) of the Water Futures meeting, the focus was on evaluation and validation of solutions and developing a system that would allow the sharing of data and a fair and effective comparison of methods and results. The discussion then involved benchmarks and the necessary data and information needed to ensure fair usage and comparison of algorithms and methodologies. The session wrapped up with an interactive discussion about the desired features of the system, use cases and expectations, involving all the participants.

Luca Hermes from the Bielefeld group then delivered the primer on physics-informed machine learning. He reminded us that we normally have a lot of knowledge about the processes we model, such as data, the laws of physics, governing equations, constraints, etc. Therefore, the best way to create credible models is to make them match the observation data and obey the physics. He concluded that this type of machine learning offers flexibility by incorporating both constraints and data and can be applied beyond physics problems (e.g., economics and social dynamics systems).

The meeting finished around lunchtime, as most of the delegates had to catch flights or trains to get back home. It was a race against time, but thankfully, no one had to resort to Marty McFly’s kind of teleportation into the future – yet!