The complete weather forecast on a PC: AI to democratize forecasts
When it comes to weather forecasts, AI technology can do more and more work. A system has now been presented that can produce competitive forecasts on a PC.
(Image: eskystudio / Shutterstock.com)
A research group from the UK claims to have developed an AI technology that can take over the entire weather forecasting process and not just individual parts. Whereas not so long ago several models were needed for such forecasts, all of which required a supercomputer, the system called "Aardvark Weather" now runs entirely on a desktop computer. It only takes minutes to calculate, the team writes. With 10 percent of the measurement data of conventional systems, the new technology already produces better results than the GFS weather forecasting model of the US weather service. It should soon be of particular help in the Global South.
Competitive forecasts with a fraction of the effort
AI-based weather forecasts are not new; Google DeepMind, among others, is conducting research into this, and just a few days ago the European Centre for Medium-Range Weather Forecasts (ECMWF) began making the forecasts of an AI-based weather model available. However, as the research group led by Richard Turner from the University of Cambridge now explains, so far only parts of the entire process have been outsourced to an AI. With Aardvark, however, the entire pipeline has now been replaced by machine learning. The model is fed with data from satellites, weather stations and other measuring devices. It generates global and local forecasts.
Videos by heise
By using AI technology throughout the process, competitive forecasts can already be produced and this is just the beginning. The approach could also be easily applied to other forecasts, such as hurricanes, forest fires or tornadoes. Until now, it has taken years to develop special systems for this. At the same time, the technology works with a fraction of the resources and is therefore many times cheaper. It could therefore democratize weather forecasts and make them available in developing countries and regions without many measuring points. This is exactly what those responsible now want to work on. They present their work in the scientific journal Nature.
(mho)
