Scientists from FIT are working on a project to forecast the weather for 24 hours in advance using artificial intelligence. Together with meteorologists from Meteopress, they are looking for ways to make machines learn to read weather maps and make sense of the vast amount of data from satellites and radars. Compared to meteorologists, AI has one huge advantage - it can evaluate radar data in a snap. The research is based on the final thesis of the student of FIT Bc. Matej Choma.
"We are working on a 24-hour forecast. At the moment we can predict the weather 15% better than the standard forecast," says Mgr. Petr Šimánek from FIT, one of the project's researchers, adding: "We use data from standard forecasts, from which the neural network learns how to make a more accurate weather forecast."
In the past, the faculty has already completed a project with Meteopress for short-term weather forecasting, i.e. for tens of minutes in advance. Together, they have developed an efficient system that can predict the next hour's precipitation with high accuracy based on current meteorological data. This is used to predict thunderstorms, torrential rainfall and other extreme weather events. This project is also largely based on the thesis work of Matej Choma.
More information about the project can be found at the following links:
Neural networks in weather radar nowcasting | by Matej Choma | researchsummer | Medium
Interpolation and Extrapolation of Subsequent Weather Radar Images (cvut.cz)