Researchers at the Massachusetts Institute of Technology (MIT) and their international colleagues have developed an artificial intelligence (AI) system that improves the accuracy of weather forecasts for periods of up to six weeks. The findings were published in the journal Nature. The development took first place in an international competition for meteorological models.
The algorithm learns from errors made by existing meteorological systems and then automatically adjusts their calculations. Testing demonstrated that forecasting accuracy for periods ranging from two to six weeks has been doubled—a result of particular significance for long-range forecasting.
The system also improves the prediction of extreme weather events, such as severe cold spells, floods, and droughts. Previously, forecasts looking several weeks ahead were considered highly unreliable due to the complexity and constant variability of atmospheric processes.
The developers believe this technology will enable energy companies, agricultural producers, and emergency services to prepare for weather-related threats in advance. It can be integrated with existing forecasting systems, and the algorithm does not require powerful supercomputers to operate.
Another advantage of the new system is its open-source code, which allows it to be incorporated into existing forecasting infrastructure. The study’s results are detailed in the journal Nature.
About the Author
