Weather forecasts are often unreliable. However, to reduce the impact of extreme weather, such as drought or heat, it is important to know in advance when extreme weather is approaching. Van Straaten uses a combination of physical and statistical weather models for this purpose. He demonstrates that the distribution of cold and warm air can be better predicted when looking at longer periods and larger areas.
Using machine learning techniques, Van Straaten and his fellow researchers discovered that processes in the Pacific Ocean are essential in influencing the distribution between cold and warm air. This link in the climate system was found to be missing in the model used by many weather apps for predictions. Together with the Royal Netherlands Meteorological Institute (KNMI) they developed a method that statistically corrects that model and makes it more useful.
Improved with AI
This method also provides more insights into the errors in weather predictions, including by comparing forecasts with actual measured weather. According to Van Straaten, this method is a clear way in which classical, physics-based weather models can be improved with AI. "All users of weather forecasts, from water boards to farmers to ordinary citizens, can benefit from it. They will have access to reliable information earlier."