Chen Li: Assessment of Global Data-Driven Models for Weather and Sub‑Seasonal Prediction over Australia
Recent advances in machine‑learning (ML)‑based forecasting systems are opening new pathways for numerical weather prediction (NWP) and sub-seasonal forecasting. ECMWF’s Artificial Intelligence Forecasting System (AIFS) represents a significant step towards operationally relevant, purely data-driven prediction for both deterministic and probabilistic NWP, with ongoing developments extending its applicability to longer lead times including sub-seasonal time scales.
For Australia, evaluating these emerging systems is critical for understanding their potential value for regional prediction. In this contribution, we present evaluation results of various AIFS model for weather and sub-seasonal predictions over Australia. We examine both deterministic and ensemble configurations, including AIFS‑Single and AIFS‑ENS, and explore early results from extensions of AIFS that incorporate ocean information, such as AIFS‑marine. Where available, we also discuss progress towards an Australian adaptation of the regionally enhanced machine learning weather model Bris, and its value for forecasting applications.
To place these results in a broader context, we additionally consider select models from the Weather Prediction Model Intercomparison Project (WP‑MIP), providing insights into the relative strengths and limitations of a variety of ML‑based forecasting systems, and highlighting both their potential and the current challenges for operational use.