Better forecasts with AI for water management

Looking ahead on the Rhine with LobithNN

The Netherlands lives with water. Every day, decisions are made about water distribution, weirs, locks, navigation, and freshwater availability. To make these decisions effectively, water managers need to know what is likely to happen tomorrow and the day after.

Deltares is increasingly exploring and applying AI as part of applied scientific research. Not as a replacement for existing knowledge and models, but as a valuable addition. A good example is LobithNN, an AI model that predicts the water entering the Netherlands at Lobith over the next 48 hours.

These forecasts are now used operationally by Rijkswaterstaat and support experts in the daily management of the Dutch water system.

For hydrologist and model developer Jurian Beunk, the project shows how new technologies can create tangible value. “AI allows us to improve discharge forecasts at Lobith, enabling better decision-making: better forecasts, better decisions.”

Why Lobith matters for water management

Lobith, located in the province of Gelderland, is where the Rhine enters the Netherlands. What flows into the country at this point influences a large part of the Dutch water system.

“Many water managers in the Netherlands look to Lobith to understand what is coming their way,” says Jurian. “The Rhine is a large and relatively slow-moving river system. That means measurements taken upstream can already provide indications of what will happen in the Netherlands hours or even days later.”

LobithNN uses observations from multiple monitoring stations along the Rhine in Germany as well as observations at Lobith from the past days. Based on this information, the AI model calculates how much water will flow into the Netherlands over the next two days. These insights help water managers look ahead and take timely action.

Supporting day-to-day water management

When people think of water models, they often think of floods or droughts. In reality, a large part of Dutch water management revolves around everyday operational decisions. Water is continuously allocated, water levels are monitored, and infrastructure such as weirs and locks is adjusted based on expected conditions. Reliable forecasts are essential for all these decisions.

“Many decisions are made based on information about what will happen tomorrow,” says Jurian. “The better you can anticipate the future, the better you can act today.” This is exactly where LobithNN adds value. The model supports experts who prepare daily forecasts and helps them build a more complete picture of the situation more quickly.

What makes AI different?

LobithNN is based on deep learning, a technology that enables models to learn complex relationships from large volumes of historical data. The model is trained on past water-level observations at stations upstream in Germany together with discharge observations at Lobith, and learns how these observations relate to future discharge at Lobith.

Importantly, the model does not need to be given an explicit description of the underlying hydrological, hydraulic, and hydrodynamic processes. From the patterns in the observations, it can learn representations of the evolving state of the river system. For example, how changes in water levels propagate downstream and how the response at Lobith develops over time.

This is an important difference from the traditional hydrological, hydraulic, and hydrodynamic models that Deltares has been developing for decades. Traditional models explicitly represent the physical processes that govern the water system, whereas AI models learn relationships and representations directly from observations. As Jurian puts it: “An AI model learns relationships from data. A hydrologist understands why those relationships exist. That combination is what makes it powerful.” For Jurian, the future is therefore not about choosing one approach over the other, but about combining both.

Data can teach us a great deal. Physical knowledge helps us understand processes. Together, they provide a strong foundation for the next generation of models.

Jurian Beunk, Hydrologist and Model Developer at Deltares

From research to operational practice

One of LobithNN’s key achievements is that it has moved beyond research and into operational use within Rijkswaterstaat’s forecasting systems (RWsOS). Its forecasts are now part of the organisation’s operational workflows and support the daily activities of Dutch water managers.

For Jurian, perhaps the most rewarding outcome is seeing the model used in practice. “What I am most proud of is that this work did not end up sitting on a shelf. Together, we developed something that is now genuinely used in the day-to-day management of water in the Netherlands. It shows that research does not stop at publishing a report; it can make a real difference in practice.” This transition from research to application is closely aligned with Deltares' mission: developing and applying knowledge to address societal challenges.

Innovation through collaboration

LobithNN was developed through close collaboration between researchers, hydrologists, AI specialists at Deltares and Rijkswaterstaat. Bringing together different areas of expertise resulted in a solution that is both scientifically robust and practically applicable.

According to Jurian, this collaboration is essential for successful innovation. “Good AI does not emerge from technology alone. You also need knowledge of the system, an understanding of operational practice and people who know which decisions ultimately need to be made.”

“The AI model learns complex relationships from large volumes of monitoring data across the Rhine basin, says Jing Deng, AI-specialist at Deltares. "The challenge is to turn what it learns into forecasts that are not only accurate, but also robust under different hydrological conditions. That is where collaboration with hydrologists and domain experts is essential. They bring knowledge of the river system and help us understand whether the model’s behavior is consistent with what we know about the underlying processes. At the same time, AI model can reveal relationships in the data that may be difficult to capture explicitly in a traditional model. It is this combination of machine learning, hydrological knowledge and operational expertise that makes LobithNN so powerful."

It is this combination of machine learning, hydrological knowledge and operational expertise that makes LobithNN so powerful.

Jing Deng, AI-specialist at Deltares

AI as a tool

The future remains uncertain, even with AI. Although AI can support people in decision-making, responsibility for decisions remains with individuals and organisations. A computer cannot be held accountable.”

Reflecting on the importance of expertise, Jurian adds: “Hydrological processes and water systems are highly complex. Deep knowledge and expertise are essential for making the right decisions.”

Within water management, people remain responsible for interpretation, judgement, and decision-making. AI supports that process by delivering insights more quickly and making information more accessible. Especially in applications that directly affect society, this requires a careful and responsible approach.

Looking to the future

LobithNN illustrates how artificial intelligence can contribute to the societal challenges that Deltares seeks to address. From improved forecasting and faster analysis to new ways of extracting knowledge from data, AI is creating new opportunities to support researchers and professionals.

The greatest value, however, does not lie in technology itself. Real impact emerges when technology, scientific knowledge, and human expertise come together to enable better decisions. As Jurian concludes: “AI helps us look ahead. It is up to people to turn those insights into better decisions.”

Memo 'Redevelopment of LobithNN'

The technical background, model choices and results are described in the memo ‘Redevelopment of LobithNN’. In it, you can read about how the model was developed and tested, what improvements have been achieved, and what opportunities the researchers see for further development.