New machine learning methods set to improve forecasts

A University of Reading mathematician and meteorologist is to develop new tools to improve weather and climate forecasting. 

Dr Eviatar Bach, who holds a joint appointment across the Department of Mathematics and Statistics and the Department of Meteorology, has been awarded £1.25 million by UK Research and Innovation to become a Future Leaders Fellow. UKRI's Future Leaders Fellowships allow universities and businesses to develop their most talented early career researchers and innovators. 

Weather forecasters build their predictions by combining real-world observations, such as satellite and weather station data, with computer models of the atmosphere. This process, called data assimilation, has limits. It can struggle to capture the full range of possible outcomes, and can lose accuracy during severe weather. 

Dr Bach's fellowship will use machine learning to make data assimilation more accurate and better able to capture that uncertainty. Her work will also use efficient machine learning weather models to generate large numbers of forecasts at once, and combine machine learning with traditional forecasting to improve accuracy from two weeks to several months ahead.  

Dr Bach said: "A better forecast can mean the difference between a good harvest and a failed one. For a farmer deciding when to plant, or a family bracing for a storm, knowing what the weather holds can change everything. 

"This project is about using new tools to get more out of the huge amounts of weather data we already collect, so that people making those decisions, whether they're planning a season's crops or preparing for extreme weather, can trust the forecast a little more."  

Partners on the project include researchers at the European Centre for Medium-Range Weather Forecasts, New York University, the Alan Turing Institute, and TomorrowNow, an agricultural decision-support platform used by farmers across Africa.