Researchers in Finland have deployed machine learning in meteorology to predict the potential levels of damage a storm can cause.
The study, a collaborative effort between Aalto University and the Finnish Meteorological Institute, found promising early results in using machine learning – where a computer identifies patterns in existing data in order to formulate predictions for future behaviour – to determine the severity of a storm, with particular reference to the potential of weather events to cause electricity blackouts or damage energy infrastructure. In order to implement machine learning in meteorology studies, the researchers collated data provided by three Finnish energy companies – Imatra Seudun Sähkönsiirto, Järvi-Suomen Energia and Loiste Sähkoverkko – which have all experienced weather-related power outages across energy grids in central Finland.
Researchers categorised storms in four classes, ranging from Class 0 – a storm which did not affect any energy infrastructure – to Class 3 storms, which cut power to more than 50% of the transformers observed in the study. Once the computer had received all the data on previous storms, its results in predicting the severity of future storms were described as ‘promising’, with a high level of accuracy. As more data is added to the program, researchers anticipate the accuracy of its predictions will improve further.
Finnish Meteorological Institute software architect and Alto University PhD researcher Roope Tervo said: “We used a new object-based approach to preparing the data, which [is what] makes this work exciting. Storms are made up of many elements that can indicate how damaging they can be: surface area, wind speed, temperature and pressure, to name a few. By grouping 16 different features of each storm, we were able to train the computer to recognize when storms will be damaging. Our next step is to try and refine the model so it works for more weather than just summer storms: as we all know, there can be big storms in winter in Finland, but they work differently to summer storms so we need different methods to predict their potential damage.”