Smarter rainfall forecasting with artificial intelligence
Expert(s) |
Ruben Imhoff
Extreme rainfall and urban flooding are becoming increasingly common — not only in the Netherlands, but across the globe. Climate change is intensifying downpours, accelerating runoff, and heightening flood risks in cities and vulnerable regions. So how can we better prepare? Ruben Imhoff, researcher at Deltares, is exploring innovative ways to improve rainfall forecasting using artificial intelligence (AI). His work sits at the intersection of meteorology, hydrology, and data science.
From passion to practice
Ruben’s fascination with rainfall forecasting began during his studies. Towards the end of his degree, he was offered the opportunity to pursue a PhD in rainfall nowcasting — which is forecasting rainfall for the coming hours ahead. “Rainfall is often the decisive factor in flood forecasting and hydrology,” he explains. “Bringing together meteorology and hydrology is incredibly compelling, and I didn’t want to miss that chance.”
Societal relevance
Improved rainfall forecasts can make a real difference in water management — from optimising pump operations to adjusting weir levels. But the benefits extend far beyond water systems. Accurate forecasts can help prevent damage by temporarily shutting down wind turbines, adjusting motorway speed limits, or informing outdoor events. “It’s all about being able to respond in time to what’s coming,” says Ruben. Timely evacuations can help avert disasters. Besides, aviation also benefits — enabling adjustments to landing schedules, rerouting flights, or implementing safety protocols during heavy rainfall.
AI as a gamechanger
Rainfall is notoriously difficult to predict due to its variability in time and space. Traditional weather models perform well over longer timescales (days ahead) but are less effective for the next few hours. “Nowcasting uses recent observations, such as weather radar imagery, and extrapolates them intelligently into the near future,” Ruben explains. “Because we’re already working with video-like data, it’s a natural fit for AI. AI can detect patterns that conventional methods often miss.”
Reliability in extreme conditions
AI could be a breakthrough in forecasting extreme rainfall — but it’s not straightforward. An AI model can only simulate what it has seen before. That’s why Ruben is developing models that incorporate physical laws and learn relationships, such as how humidity influences thunderstorms. Deltares is also investing in hybrid models, where AI complements existing physics-based models. “This way, we retain insight into the underlying processes and the proven physical approach, while using AI to enhance the aspects our models currently struggle to simulate,” Ruben explains.
Collaboration is key
“Rainfall is where meteorology and hydrology converge,” says Ruben. “Collaboration with data scientists is essential to apply AI effectively. But dialogue with end users is just as important — a forecast only has value if it’s usable in practice.”
Integrating AI into software
At Deltares, AI techniques are increasingly embedded into existing software tools to enhance forecast quality. One example is short-term rainfall prediction, where recent radar data is combined with numerical weather models such as Harmonie Cy43, or models from the European Centre for Medium-Range Weather Forecasts (ECMWF).
This approach enables faster and more accurate responses to extreme rainfall — both in urban settings and in international projects (such as in Ghana and Ethiopia), where heavy rainfall can have severe consequences for vulnerable communities.
Deltares develops and refines these methods within an open-source environment, adding functionality to existing tools. A key example is the use of the open-source Python library PySTEPS
for probabilistic rainfall nowcasting. Thanks to its modular design, researchers and practitioners can develop and integrate new methods into operational workflows. This aligns with the Deltares philosophy: working in open knowledge communities and applying technology in socially relevant contexts.
Although Deltares is not a meteorological institute, we play a crucial bridging role between weather services, water authorities, and hydrological experts. That position allows us to accelerate innovation.
Ruben Imhoff, Deltares
Advice for governments
Ruben advocates for a clear distinction: what can be prevented, and when is timely warning the best option? “Some extremes simply can’t be avoided. In those cases, you need to be able to issue warnings and activate response protocols.” He stresses that governments should invest not only in infrastructure, but also in intelligent forecasting systems and rapid decision-making.
Learning from past events is crucial: how did we respond, what worked well, and where were the bottlenecks? Translating these lessons into concrete improvements in modelling and response protocols can strengthen the resilience of cities and regions. Ruben emphasises that forecasts only become valuable when they lead to action — and that requires clear communication, strong inter-agency collaboration, and trust in technology.
He also calls for the involvement of end users — such as water authorities, safety regions, and municipalities — in the development of AI-supported systems. “They know what’s needed at the local level. By combining their expertise with technological innovation, we can truly make a difference.” It’s also the only way to build shared trust in AI solutions.
Looking ahead
AI in rainfall forecasting is still in its early stages, but Ruben sees great potential. “Especially for localised, intense showers, there are real opportunities. I hope AI becomes a reliable tool in our forecasting toolkit.” He highlights the importance of explainable AI — systems that allow us to understand why a model makes a certain prediction. “AI is a means, not an end goal. Domain expertise and mission remain the guiding principles.”
Delft-FEWS User Days
Would you like to learn more about combining AI for flood forecasting? Then visit the International Delft-FEWS User Days on 5 and 6 November 2025, where this topic is also on the agenda.
