Welcome to the AI Weather Quest confluence hub.

The AI Weather Quest (AI WQ), organised by the European Centre for Medium-Range Weather Forecasts (ECMWF), is an ambitious international competition designed to harness artificial intelligence (AI) and machine learning (ML) in advancing weather forecasting. It challenges participants to produce and submit sub-seasonal weather forecasts – covering the critical weeks between medium-range and seasonal predictions – using AI/ML models.

This Confluence hub offers access to the latest dynamical forecast models evaluation and preliminary analysis on AI-based sub-seasonal forecasts, detailed how-to guides for using the forecast portal and navigating the leaderboards, as well as updates to the Python package. It also lists the collection of online AI Weather Quest outputs.

Important links:

To view all AI Weather Quest outputs and links, access this page.

Competition overview:

Teams are challenged to submit weekly, real-time sub-seasonal forecasts of at least one of the following variables:

  • Near-surface (2m) temperature (tas)
  • Mean sea level pressure (mslp)
  • Precipitation (pr)
  • Madden–Julian Oscillation (MJO) - introduced for testing during the SON 2026 bridge period, public evaluation of forecasts begins with the DJF 2026 period
  • Tropical storm days (TS) - introduced for testing during the SON 2026 bridge period, public evaluation of forecasts begins with the DJF 2026 period

For temperature and pressure forecasts, teams are required to produce weekly averages, which will be evaluated against weekly means calculated using six-hourly data (00, 06, 12 and 18 UTC).

For precipitation forecasts, the focus is on weekly accumulations, which will be compared against corresponding reanalysis totals.

For the Madden–Julian Oscillation, teams are required to submit probabilities for each of the eight active MJO phases and for an inactive state, for the Thursday within each forecasted week.

For tropical storms, teams submit basin-based tercile probabilities of tropical storm activity. Activity is defined by the total number of tropical storms present on each day, aggregated over the forecasted week.

Forecasts for tas, mslp and pr must provide global quintile probabilities at a 1.5-degree latitude–longitude resolution. Tropical storm forecasts must provide basin-based tercile probabilities. Forecasts for these variables are submitted for one or both of the following inclusive lead-time windows:

  • Days 19–25 (week 3)
  • Days 26–32 (week 4)

MJO forecasts are submitted at lead times of:

  • 22 days
  • 29 days

To ensure flexibility for AI/ML innovation, participants can:

  • Submit up to ten forecasted variables per AI model each week, corresponding to five variables across two lead times.
  • Use up to three different AI/ML models, allowing a maximum of 30 submissions per team each week.
  • Develop AI/ML models using any observational or forecast datasets (which may include ECMWF-supported datasets).
  • Develop AI/ML models using any programming language.

Submissions are welcome from various types of ML/AI models, including (but not limited to):

  • Models that post-process numerical weather prediction data.
  • Machine-learning based models specifically designed for weather prediction.
  • Statistical models that focus primarily on generating quintile probabilities.
  • Hybrid models that combine physical simulations with machine-learning techniques.

Recent space activity

Show More

Space contributors

{"mode":"list","scope":"descendants","limit":"5","showLastTime":"true","order":"update","contextEntityId":540944258}


  • No labels