An update to ECMWF's machine-learned weather forecast model AIFS

2025/09/23 by Moldovan, Gabriel, Pinnington, Ewan, Nemesio, Ana Prieto +18 · 1 citation
#Atmospheric and Oceanic Physics (physics.ao-ph) #FOS: Physical sciences

paper · doi:10.48550/arxiv.2509.18994

Abstract

We present an update to ECMWF's machine-learned weather forecasting model AIFS Single with several key improvements. The model now incorporates physical consistency constraints through bounding layers, an updated training schedule, and an expanded set of variables. The physical constraints substantially improve precipitation forecasts and the new variables show a high level of skill. Upper-air headline scores also show improvement over the previous AIFS version. The AIFS has been fully operational at ECMWF since the 25th of February 2025.

Cited by

• AIFL: A global daily streamflow forecasting model using deterministic a LSTM pre-trained on ERA5-land and fine-tuned on IFS
• FlowCast-ODE: Continuous Hourly Weather Forecasting with Dynamic Flow Matching and ODE Solver

Related

• AIFS -- ECMWF's data-driven forecasting system
• WP-MIP: An Artificial Intelligence, Hybrid, and Physically Based Model Intercomparison Project for Weather Prediction
• Statistical post-processing yields accurate probabilistic forecasts from Artificial Intelligence weather models
• OMG-HD: A High-Resolution AI Weather Model for End-to-End Forecasts from Observations
• Improving AI weather prediction models using global mass and energy conservation schemes