Picking Our Battles: Early Insights from the Battle of the Water-Futures

In June 1985, at the “Computers in Water Resources” conference in Buffalo, New York, the organisers proposed a friendly competition to solve a simplified drinking water system sizing problem. Just like that, the “Anytown” benchmark was born. Fast forward forty years, and countless PhD students and researchers are still developing new methodologies based on that very same problem.

Over the years, these competitions have become a tradition in the drinking water research community. More recently, whenever the WDSA (Water Distribution Systems Analysis) and CCWI (Computing and Control for the Water Industry) communities join forces, a new “Battle” special session takes place. For example, you can read about what we learned from our participation in the Battle of the Water Demand Forecasting in Ferrara two years ago.

This year, it was a bit closer to home. The 4th WDSA/CCWI conference was organised by our Water-Futures project partners at KIOS (University of Cyprus), with Prof. Marios Polycarpou sharing the conference chair duties with our own Prof. Dragan Savić. Naturally, to justify our presence at the wonderful venue in Paphos, the entire Water-Futures project team had to step up. While KIOS carried the heaviest load of the conference organisation (amazing, by the way!), we jumped in as reviewers, chaired sessions (a first for many of us!), and even hosted a short course on modern water systems optimisation (you can find our material here.

But without a doubt, our biggest effort went into curating this year’s competition: The Battle of the Water-Futures.

If you were to read the (admittedly long and complicated) problem statement, you’d see that the name was more than a reference to our ERC-funded Synergy Project Water-Futures (WF); it captured the nature of the challenge itself. This time, the participants’ goal was to devise a master plan for a national water grid and ensure its continuous operation until the end of the century. This sits right at the heart of the core theme of both the conference and the WF project: Designing the next generation of urban water systems.

Covering such a long horizon on a high-level system naturally pushes you into uncharted territory, or what we researchers like to call “deep uncertainties”. This describes a situation so volatile that we not only disagree on what the future will look like, but we can’t even agree on the likelihood of different scenarios. Will the population boom or bust? How will weather patterns shift? Will new technologies curb per-capita water demand or accidentally increase it? Looking back, could anyone 100 years ago have truly predicted where we are now?

To make the challenge tangible, we designed a completely new case study featuring fictitious regional water utilities interacting within a national context. We heavily based the entire model on the Netherlands – because how could it be any other way, right? – but used only synthetic and open-source data. Fortunately, the Netherlands is incredibly rich in open data, and we owe a massive thank you to institutions like the CBS for population data, KNMI for weather projections, and Vewin for realistic water demand information.

Participants had to make decisions about major infrastructure interventions (such as opening or closing water sources and building massive transport pipelines) alongside high-level policies (such as setting water prices or managing debt levels).

The competition unfolded over several weeks across three distinct stages, each representing a 25-year window. In each stage, participants sent us their master plan for the future. We then simulated those plans against a specific, secret future that we had cooked up behind closed doors. Once each 25-year block “passed,” we sent the teams a data update and a narrative explaining what had just happened to their world. With these new observations at their disposal, they had to adapt, learn from experience, and steer the course of their strategy for the next quarter-century.

Uncertainties were everywhere: climate impacts, economic shifts, fluctuating electricity prices, and even random construction delays for new treatment plants. To make things beautifully chaotic, public expectations were also a moving target. Teams had to constantly balance and prioritise conflicting goals: keeping financial debt low, ensuring system reliability, maintaining water affordability, and slashing greenhouse gas (GHG) emissions.

Our First Takeaways

What we saw on the final day of the conference, when the participants presented their approaches, was truly impressive. Twelve teams showcased their methodologies, which ranged from classic, rule-based engineering heuristics to ultra-modern, fully autonomous AI agents (spoiler alert: an autonomous AI agent actually took second place, though it did require a fair bit of ad-hoc engineering baked into it to handle the chaos!).

While we are saving the final scores and official rankings for an upcoming peer-reviewed paper, a few major high-level lessons already stand out:

– Flexibility and adaptivity are everything: Some participants openly lamented that they “couldn’t go back” once they had implemented certain interventions. Imagine wasting millions of euros of taxpayer money on a massive infrastructure project, only to discover a decade later that it was way oversized or not even remotely enough. Ouch!

– Predicting the future is a losing game: Our data showed a stark example of this: 72% of the time, participants who tried to predict inflation underestimated it. That means they were off-track almost three out of every four years.

– The political traps are real: Balancing the “public perspective” proved incredibly difficult. For instance, participants who successfully kept their debt levels low often had to compromise on water affordability for consumers. Since no one knew what the public’s actual preference was ahead of time, we saw quite a diversity in what different teams chose to prioritise.

– Sometimes, enough is enough: More infrastructure isn’t always better. Not every intervention brought an improvement, and we purposely designed the problem this way. One participant masterfully presented this concept.

– Pick your battles (and embrace the unknown): We cannot perfectly model the world. We intentionally over-designed and detailed the problem, making it so massive and complex that it was impossible to “do it all”. We assumed compromising on the level of detail would be a necessity for the teams. To our surprise, however, everyone tried incredibly hard anyway – and paradoxically, some teams actually wanted more!

It was incredibly satisfying to see so many people treating what was supposed to be a fun hypothetical project with such immense effort. In the end, it was all driven by a pure, scientific desire to just do a bit better.

Ultimately, our hope is that those who had the opportunity to see the presentations live—and those who will work on this challenging problem in the future—will grasp the sheer size of this extremely complicated challenge. We are barely scratching the surface.

But if this battle showed us anything, it’s that the future of water management isn’t about building the perfect static defence; it’s about building adaptability and flexibility into our systems and our minds.

A massive thank you to the conference organisers and the Water-Futures team who made this possible, and to the participants whose dedication made this whole thing so meaningful.