Data-driven weather models: A new era in meteorology
Year: 2025
Type: conference-abstract
Abstract: Over the last few years, machine learning (ML) has transformed the field of weather forecasting, with state-of-the-art data-driven models offering enhanced accuracy and efficiency, compared to traditional methods. The rapid advancement of ML in weather forecasting is often described as revolutionary: since around 2018, progress has accelerated rapidly and models can now achieve a level of skill that is comparable and, for a wide range of metrics, better than traditional numerical weather prediction models. Moreover data-driven models can generate forecasts in minutes, while consuming up to 1000 times less energy. ECMWF is embracing this revolution and has developed the first operational data-driven weather forecasting system, AIFS (Artificial Intelligence Forecasting System). Together with collaborators across Europe, ECMWF is also working to democratise access to machine learning methods through the development of Anemoi, an open-source framework to integrate ML with meteorological forecasting. In this talk, we will provide an overview of the performance of AIFS and other data-driven weather models, not only in terms of their overall forecast accuracy, but also their ability to predict extreme events. We will also discuss developments in data-driven ensemble systems and using ML models to directly learn and forecast from observations. For both applications, ML opens up huge opportunities, due not only to the speed at which data-driven weather forecasting models can make forecasts, but also their ability to learn from novel observation datasets, facilitating the exploitation of Earth System data. Finally, this talk will conclude with a forecast of the future — a glimpse at what’s on the horizon for this fast-moving field.
DOI: 10.5194/ems2025-519
Authors: Mariana Clare, ECMWF Colleagues
Language: English
Subfield: Atmospheric Science
Field: Earth and Planetary Sciences
Domain: Physical Sciences
SDG: Climate action
Open Access status: gold
OpenAlex ID: W4411805347
