Researchers from the University of Cambridge, the Alan Turing Institute, Microsoft Research, and the European Centre for Medium-Range Weather Forecasts (ECMWF) have presented a new AI-based weather forecasting system - Aardvark Weather. The know-how allows forecasts to be formed much faster and with fewer resources than traditional weather systems. This is reported by UNN with reference to the publication The Guardian.
Details
It is noted that thanks to the development, a specialist with a desktop computer will be able to independently provide accurate weather forecasts. The new system, which is based on artificial intelligence, analyzes and collects data dozens of times faster and uses thousands of times less computing power than traditional systems.
According to the publication, Aardvark uses raw weather data from ground stations, satellites, aircraft and ships to create forecasts without traditional multi-stage processing. The system has already demonstrated superiority over the American GFS model and competes with official forecasts from the US National Weather Service.
Professor of Machine Learning Richard Turner (University of Cambridge) noted that the system allows creating personalized forecasts, for example, for agriculture in Africa or wind energy in Europe. It is also expected that Aardvark will be able to form forecasts for 8 days ahead instead of the usual 5.
This is a radically new approach to forecasting, which is likely to become the new standard in this field
In turn, Dr. Scott Hosking (Turing Institute) noted that the system can democratize forecasting - make powerful tools available to developing countries. It also has practical implications for civil defense services, government agencies and businesses.
Co-author of the study, Dr. Anna Allen, emphasized the potential of the model in predicting natural disasters - hurricanes, tornadoes, forest fires - as well as phenomena such as air quality or changes in sea ice.
The development is based on previous research by Huawei, Google and Microsoft, which have already demonstrated the possibility of replacing traditional numerical models with AI solutions.
The European Centre for Medium-Range Weather Forecasts (ECMWF) has already begun implementing this approach in its operations.