Informal Seminar: Insights from the AI Weather Quest: An international machine-learning competition for sub-seasonal prediction

In light of rapid advances in using machine learning (ML) for weather prediction, we launched the ECMWF AI Weather Quest, an international competition designed to benchmark ML-based sub-seasonal forecasts within a realistic operational framework. The competition challenges participants to forecast global probabilisti quintile forecasts of near-surface temperature, mean sea level pressure, and precipitation at three- and four-week lead times, a particularly challenging timescale where both dynamical and statistical approaches show limited skill. By requiring real-time forecasts, it has promoted the development of robust, deployable systems capable of delivering actionable probabilistic information.
Six months into the Quest, 40 teams have already participated, spanning academia, public institutions, and industry. ML-based post-processing of ECMWF IFS forecasts has achieved the highest skill, underscoring the continued value of dynamical models. In this seminar, we will provide an overview of the AI Weather Quest design and present an initial assessment of sub-seasonal forecast skill across ML-based and dynamical systems. We will also discuss emerging fully data-driven systems, lessons learned so far, and opportunities for wider engagement as the competition progresses.