The Danish Centre for AI Innovation A/S (DCAI) has announced a multi-year agreement with the Danish Meteorological Institute (DMI) to develop the next generation of weather models on Gefion. The agreement follows breakthrough results in a recent pilot project, which demonstrated that new AI-powered models trained on Denmark’s AI supercomputer Gefion approach the accuracy of DMI’s long-standing model in predicting temperature and wind while significantly improving efficiency — running in minutes rather than hours.
A specialized team of researchers at the Danish Meteorological Institute (DMI) has, within a few months, developed a regional AI weather model using the Gefion supercomputer. This new model is competitive with an internationally developed weather model in key parameters such as temperature and wind predictions.
The existing weather model is based on physical equations. Using an initial set of observations, it calculates how the weather will progress. This weather model has been under international development for over 20 years, whereas the new AI model has been trained on a large dataset for just a few months.
The AI model requires only a fraction of the computing power and takes minutes to run, compared to the traditional model, which takes several hours to compute. Preliminary results show that in addition to being more efficient, the new AI weather model approaches the accuracy of the current model in predicting temperature and wind conditions.
This promising outcome has paved the way for a three-year agreement between the Danish Centre for AI Innovation (DCAI), the company operating the AI supercomputer Gefion, and DMI, with a goal of advancing next-generation AI weather models.
“The pilot project has shown that DMI can leverage AI very quickly and efficiently. We have several AI models in our pipeline for predicting precipitation and cloud cover, which will directly benefit from the experience we have gained in just a few months,” says DMI meteorologist Kasper Hintz.
“The collaboration between DCAI and DMI marks a significant step forward in the application of AI to the very important field of meteorology,” says Nadia Carlsten, CEO of the Danish Centre for AI Innovation.
“With AI supercomputing, we are unlocking new opportunities for weather forecasting, and DMI’s groundbreaking work is already demonstrating how impactful this technology can be, including for sustainability,” she adds.
Problem solving
Gefion is ranked 21st on the prestigious Top 500 list of the world's most powerful supercomputers. Its powerful advanced computing infrastructure can be applied to solve problems in various sectors, from pharmaceutical and biotech applications to AI-driven innovation.
In combination with traditional weather models, AI technology like Gefion’s is expected to deliver significant improvements to daily weather forecasts and warnings, which are crucial for national preparedness against extreme weather. Faster models that can be run more frequently could enable emergency services and other authorities to respond more quickly, potentially saving critical infrastructure and, ultimately, lives during extreme weather events.
Better and faster forecasts, such as cloud cover predictions, will also be directly beneficial to the green transition. Solar energy production for example is directly tied to the weather, and ultra-short-term predictions of sunlight are an important factor for electricity delivery to the consumers.
Three-year collaborative agreement
While DMI’s new AI model has already shown significant promise, researchers emphasize that the journey is far from over. DCAI and DMI have signed a three-year agreement that allows DMI researchers to continue developing the next generation of weather models on Gefion. The extended collaboration is expected to provide the necessary momentum to refine and expand AI-driven weather models, integrating additional variables like precipitation and uncertainty estimation.
“DMI’s mission is to protect the Kingdom of Denmark from hazardous weather and provide the best forecasts. That’s why we explore new technologies like artificial intelligence, and our pilot project on Gefion has delivered significant results. Our partnership with DCAI now enables more DMI projects to train on Gefion. At the same time, we are strengthening a shared ambition with DCAI to establish a Danish AI community that can develop sought-after AI skills and build national expertise in the field,” says Thomas Kjellberg, Deputy Director at DMI.
DMI is engaged in various research projects that incorporate AI and machine learning in different ways. Several of these projects require large-scale GPU training and the ability to train on multiple GPU nodes — a dedicated computing unit within the supercomputer with memory and storage — enabling fast data exchange between nodes.
“The DMI pilot project has shown that Gefion can deliver fast and tangible results for those at the forefront of innovation. We are delighted that DMI has decided to enter into a long-term collaboration with us following the project’s outstanding results. And we look forward to supporting their continued advancements in AI-driven meteorology,” says Nadia Carlsten, CEO of the Danish Centre for AI Innovation.
