Using simulations to plan the drinking water supply of the future

For cities and municipalities, securing the water supply means not only responding to periods of drought but also looking decades ahead to ensure that pipeline networks, storage facilities, wells, and treatment plants operate reliably even during shortages. “When planning new facilities, we must estimate how much drinking water a municipality will need in 2040, 2050, or beyond. We combine data with mathematical models, simulations, and machine learning methods in order to provide a basis for robust decision-making—even under uncertain future conditions,” says Prof. Wolfgang Nowak of the Institute for Modelling Hydraulic and Environmental Systems (IWS) and the Stuttgart Center for Simulation Science (SC SimTech).

Water suppliers have traditionally based their planning on historical consumption data. However, given the realities of climate change and ongoing changes in population, settlement patterns, industrial use, and consumer behavior, past trends cannot simply be extrapolated into the future. The way these factors interact also varies from municipality to municipality.

Another challenge is the amount and quality of data available. Small and medium-sized water suppliers in particular often have access to only limited and incomplete data, whereas AI methods require large amounts of training data. The researchers are therefore investigating how to establish a reliable basis for decision-making even when only limited data is available.

The researchers combine information from different municipalities, time periods, and climate regions in order to account for both short-term weather fluctuations and long-term climate trends. The models are designed to identify broader patterns while incorporating the local conditions of each service area. This allows information to be transferred more effectively between comparable supply systems without overlooking local characteristics. A municipality with limited data of its own can thus benefit from insights gained from comparable regions.

“With the right methods and a systematic approach to integrating data, we can develop reliable models even from an incomplete dataset,” says Nowak. The goal is not to provide an exact figure for future water demand. “We want to show which developments are plausible, how much uncertainty they involve, and which decisions remain robust under changing conditions,” he continues.

These methods are not developed exclusively on a computer. Together with RBS wave, EnBW, and municipal water suppliers in Baden-Württemberg, the researchers work with actual consumption and operational data. The researchers draw on the industry partners’ knowledge of regional supply systems, existing infrastructure, and specific planning processes to develop models and forecasting methods that reflect the actual requirements of water supply.

These tools could show under which climate, population, and consumption scenarios existing capacities would reach their limits and help water suppliers determine sooner when to consider additional storage facilities, new wells, connecting pipelines, or other adjustments. “Simulation-based methods cannot eliminate these uncertainties. However, they can be used to quantify them, make different future scenarios comparable, and provide a better basis for long-term decisions,” says Nowak.

The research is being conducted through two interconnected projects. In the project “Machine learning for planning water supply infrastructure in the face of climate change”, which is funded by EnBW from 2024 to mid-2027, researchers are developing automated forecasting and early-warning methods for water suppliers with limited data.

The project “Data-driven, long-term forecasts of water demand under the influence of climate change”, which is funded by the German Research Foundation from 2025 through the end of 2027, focuses on systematically incorporating changes in climate, variations in relationships over time, and uncertainties in forecasts. Both projects are being carried out in collaboration with RBS wave.