Faster computational models, smarter decisions
In a world where climate change, water safety, and sustainable infrastructure are becoming increasingly urgent, swift action is essential. But how can we make rapid decisions when reliant on complex computational models that may take hours, days, or even weeks to process? Deltares is investing in model reduction – an innovative technique that enables us to tackle societal challenges both faster and more intelligently.
The challenge: gaining faster insights into complex systems
In disciplines such as fluid dynamics and sediment transport, computational models are often so detailed that even supercomputers can take weeks to simulate a single scenario – such as an extreme storm or a dike breach. This hampers swift decision-making, particularly during severe weather events. The question “can’t it be faster?” is more relevant than ever, especially as we move towards real-time decision-making.
As part of Deltares’ Enabling Technology programme, Kelbij Star and Joost Kranenborg are exploring how model reduction can help shift the paradigm.
What is model reduction?
Model reduction, also known as Reduced-Order Modelling (ROM), simplifies complex computational models without losing the essence of the simulated system – whether it’s a river, coastal zone or subsurface environment. By focusing on dominant patterns rather than every detail, models can be run hundreds of times faster. “We reduce millions of equations to just a few dozen, while still retaining 99.99% of the system’s behaviour – such as flow patterns or changes in water levels,” explains Kelbij, applied physicist at Deltares.
Smart integration of data, physics and experience
At Deltares, we combine simulations, laboratory data and field measurements to train our models. “We often start with simulations, which we then validate using lab data from controlled environments. After that, we confirm their applicability with field measurements,” explains Kelbij. “This makes our models not only fast, but also physically grounded and practically applicable.”
Peter Vermeulen, senior advisor and groundwater modeller in the groundwater and water security team at Deltares, devoted his PhD research nearly 25 years ago to model reduction in groundwater modelling. His work focused, among other things, on using ROMs for model optimisation problems, where parameters are fine-tuned to achieve the best possible match with measurements. “After my PhD in 2006, there was little enthusiasm for ROMs, as models were still relatively small and computational power was increasing rapidly. But I now see growing momentum, as current modelling approaches often struggle to meet the demands of real-time modelling,” says Peter.
Practical applications
We apply ROM techniques to flow and groundwater models, including their use in digital twins. These digital replicas require fast-running models to respond instantly to changing conditions in the physical environment. Think of applications in water management, climate adaptation and subsurface governance.
Imagine a severe storm sweeping across the Netherlands, with water levels rising rapidly. Decision-makers must quickly determine where temporary water storage is needed, or which dikes require reinforcement. Traditional models take too long to simulate all possible scenarios. Thanks to ROMs, real-time simulations become feasible, providing immediate insight into the most effective measures. This not only speeds up decision-making – it can save lives.
Joost, expert in detailed morpho dynamic modelling at Deltares, adds: “By combining physics and data, we gain a better understanding of the discrepancies between lab and field measurements. That’s essential for robust decision-making.”
Deltares forms a unique bridge between academic research and societal application. It is through collaboration that we can harness Reduced Order Models to calculate complex water and soil challenges more quickly and sustainably.
Dr ir Shobhit Jain, researcher at TU Delft
Faster response to flooding
“A great example of ultra-fast flood simulation is the SFINCS-model. While it’s not a classic Reduced Order Model, its reduced-complexity nature provides a solid foundation for further acceleration using ROM techniques or AI surrogates,” says Kelbij. “This model, developed at Deltares, can efficiently simulate storm surges combined with river discharge. What I find particularly powerful is its balance between computational speed and accuracy. That makes it ideal for rapid scans, early warnings and testing adaptation measures. SFINCS has already been applied globally – from Bangladesh to the United States and across various small island nations.”
Erosion around offshore wind turbine foundations
Another example is erosion around the foundations of offshore wind turbines. Traditional models are so computationally intensive that only one scenario can be simulated per week. Joost Kranenborg explains: “With ROMs, we can gain much faster insights into how different scenarios affect erosion. This helps us design more robust structures and communicate more effectively with clients. It also allows us to test design variants and analyse uncertainties more quickly – which is crucial for sustainable offshore energy projects.”
More examples
ROMs also prove their value in a wide range of applications beyond flood modelling.
- For instance, they’re used for (near)real-time leak detection in urban pipeline systems. “Traditional models often take too long to locate a leak,” says Kelbij. “ROMs enable faster and more accurate pinpointing the location and spread of a leak. That helps minimise damage to infrastructure and the environment – and makes pipeline systems more resilient in crisis situations.”
- In maritime incidents such as oil spills, every minute counts. “With ROMs, we could rapidly simulate how oil spreads under the influence of currents, wind and tides,” Kelbij adds. “This enables emergency services to take immediate action – from deploying clean-up resources to temporarily closing ports or protecting vulnerable natural areas.”
- We also aim to use ROMs as digital twins of our experimental facilities. “We create virtual replicas of test setups that simulate flow or soil processes,” Joost explains. “In such a digital twin, we could link a fast-running version of the computational model – the ROM – to live measurement data. That would allow us to run scenarios in real time and respond instantly to changing conditions.”
Smart use of data
Beyond computation time, data volume is a hidden bottleneck in modelling complex systems. Terabytes of field measurements, satellite imagery and simulations can strain both processing power and storage capacity. ROMs offer a solution not only by simplifying the model itself, but also through data efficiency – focusing on the most relevant patterns. While training a ROM still requires substantial data, it’s often a one-time investment. Once trained, the reduced model can be deployed rapidly and repeatedly to simulate a wide range of scenarios.
This efficiency demands collaboration between domain experts, data scientists and software developers. By combining subject-matter expertise, intelligent algorithms and robust software, we create models that are both fast and reliable – essential for decision-making under pressure.
Deltares works closely with universities and partners to achieve this. “Our ultimate goal is to make real-time models available for climate adaptation, sustainable energy and water safety,” says Kelbij. Innovation doesn’t happen in isolation – it thrives through collaboration across disciplines.
Machine learning and ROM give us the opportunity to make simulations not only faster and more energy-efficient, but also smarter – by combining data and physics. At Deltares, these technologies are applied with a critical eye: not as a replacement for physical models, but as a powerful complement.
Dr Alexander Heinlein, onderzoeker TU Delft
Collaborating on ROMs
In early September, Deltares hosted a session on ROMs, bringing together experts from various fields to explore the potential of Reduced Order Modelling (ROM) for complex water and soil challenges. TU Delft researchers Dr Shobhit Jain and Dr Alexander Heinlein were invited to present their work on model reduction, following their earlier organisation of the 4TU.AMI SRI Workshop on Model Reduction under the SRI theme 'SRI ‘Model Reduction for Industrial Applications’.
The session yielded valuable insights into where ROMs can make a difference, what knowledge gaps still exist, and how we can accelerate the transition to practical applications.
At Deltares, we’re not only committed to further developing ROMs – we’re focused on applying them. That’s why we welcome input from our partners: where do computation time or data volume pose a challenge? Where can faster models truly make an impact? We’re keen to explore the possibilities together.
