× Close
In the future, summer houses and agricultural land in Jamarbugat Municipality will be flooded after heavy and continuous rains. Credit: Jamer Bugat Municipality
Suzanne Nielsen realizes it’s probably only a matter of time before her parents’ summer house in Slatstrand in northern Jutland is affected by the flood. Because below the house, which is just 400 meters from the bay of Jämerbügüt in the municipality of the same name, the groundwater level is now often so high that there is a risk that large amounts of rain cannot run off. , but rather enter the house.
“It’s a concern if we get a lot of rain,” she admits.
To give residents and decision makers the best chance to protect themselves from flooding in the area, DTU researchers have helped Jammerbugt Municipality develop an early warning tool. It can provide 48-hour notice of localized flooding along rivers, streams and coastal areas in the municipality. It is the first of its kind to provide localized flood warnings.
“It will give us time to react if needed, so it will be a big help,” says Suzanne Nielsen from her home in Aalborg, about 40 kilometers from the summer house where she lives. Cares for parents, who live in Norway.
Complex nature, complex calculations
The tool — a so-called “weight index” — is based on artificial intelligence trained on freely available data on dynamics affecting flood risk. The data comes from satellite images and weather forecasts, as well as land and sea water levels and landscape topography.
However, calculating water movement and accumulation in open landscapes is difficult because many parameters affect the way water moves and accumulates. To handle this complexity, artificial intelligence was used in the development of the model behind the wet index.
According to Roland Löwe, by using specific design principles in building the model and feeding it with carefully selected data, researchers can improve understanding of water movement, distribution, and interaction with the surrounding environment. What is added? He is one of the developers of the wetness index and an associate professor at DTU specializing in water behaviour.
Both fluctuations
The Jammerbugt municipality tested the device in 2023. The results show better-than-expected predictions for the wet spring months. However, during the summer, when Denmark was almost in a drought, the tool incorrectly predicted flooding in the same areas that flooded during the rainy spring.
The incorrect predictions were due to the tool being trained with little data from the summer months. This is because satellites cannot register water under vegetation and given that fields are covered with vegetation during the summer, the data at this time of year is sparse.
“Early warnings need to be relatively accurate for citizens to trust the system. That’s why we chose to do a test run, where only selected citizens checked it regularly—and where as a municipality But we had drones in the air to validate the predictions, explains Heidi Egberg Johansson, project manager from the Jämerbügüt municipality.
However, she emphasizes that the overall experience is that the project partners have created a tool with great potential. Therefore, the municipality is looking for funding to retrain and possibly adjust the model, which will remain offline until the work is done, says Heidi Egberg Johansen.
Faster calculations and decisions
Accurate calculations are important—not only when citizens and emergency services need to prepare water tubes and sandbags, but also when, for example, municipalities need to decide what the future holds. How to improve your drainage system to handle the wet climate of Traditional simulations can easily produce rock-solid calculations of a system’s ability to divert water under various scenarios—but they take forever to complete, according to Roland Löwe.
“In practice, this means that every time planners need to analyze something, they have to hire consultants who disappear into a box for two months before That they can come back with results. And it’s very painful,” he explains.
To reduce computation time while maintaining physical accuracy, researchers rely on scientific machine learning, a branch of artificial intelligence that combines two different approaches.