Using ECMWF’s Forecasts (UEF2024)

Large Language Models (LLMs) have revolutionized the way we interact with data and information. We will examine the potential of LLMs in enhancing the interpretation of complex weather and climate datasets, as well as generating new insights through data processing and analysis. We will begin with an overview of LLMs, outlining fundamental approaches to developing systems that utilize them,...

Machine Learning (ML) is playing an increasingly significant role across ECMWF, both through hybrid approaches (helping to improve existing forecasting systems) and data-driven approaches (resulting in new models such as our data-driven forecasting system, AIFS). Within data-driven approaches, we are training from reanalysis/analysis datasets, whilst also exploring how to train models directly...

ECMWF has developed a data-driven forecast model, the Artificial Intelligence/Integrated Forecasting System (AIFS), which is now run in experimental mode alongside our NWP model IFS. AIFS’s forecasts are highly skilful and now available to the public. The talk will describe the current state of AIFS, its architecture and framework, and discuss some of the design decisions. Furthermore, we will...

In this presentation we will give an overview of the evaluation work for AIFS, with a focus on different kind of extreme events. As for other machine-learning based weather forecasts, we see very good scores both in terms of surface and upper-air variables. We will explore if there is any pattern in the performance relative to IFS (seasonal, regional, etc.). Regarding extremes we will look at...