Norway and the rest of the world are experiencing more frequent floods and extreme weather due to climate change. Water damage caused by weather and natural events has cost 30 billion kroner over the past ten years, according to the Norwegian Natural Disaster Fund.
No local warnings
In Norway, we are fortunate that the Norwegian Water Resources and Energy Directorate (NVE) provides warnings about upcoming floods, but this mainly applies to larger rivers and waterways. The water level in smaller rivers can rise quickly, and these often run through densely populated areas.
Global challenge, local challenge
Flooding and preparedness are municipal responsibilities and are very demanding. It is especially challenging for smaller municipalities with limited resources to handle flood situations and ensure the safety of their residents. Warnings must be precise when there is a risk of flooding, so that we do not receive numerous warnings that turn out to be false alarms. It is crucial that the solution is easy to use, and that the warnings are delivered in a way that operational personnel can understand, allowing authorities to implement the correct measures and communicate effectively with residents.
Could the two summer students, Nils and Thanh, have solved the problem?
This summer, InfoTiles had the pleasure of working with business student Thanh Nguyen from the University of Stavanger, and Nils Thomas Doherty Midtbø, who studies machine learning and data analysis of watercourse data at the University of Tromsø. They have tackled practical challenges at InfoTiles’ customers and have developed and tested machine learning to predict river levels up to 12 hours in advance.
Several different machine learning algorithms were set up and tested to predict river height. By agreement with one of our development customers (a Norwegian municipality), the students were able to use data collected by InfoTiles from a specific point in a river. This was combined with upstream river data from the Norwegian Water Resources and Energy Directorate (NVE). The results were good, with an average absolute error of 5 cm, but they will be improved.