16.06.2025
We recently sat down with project coordinators Ron McTaggart-Cowan (ECCC), Linus Magnusson and Inna Polichtchouk (both ECMWF) to get their thoughts on the WP-MIP goals, scope, and how you can get involved.
I’ve heard that you’re running an AI intercomparison project. Is that true?
We’re planning a project that’s broader than just AI intercomparison. The Weather Prediction Model Intercomparison Project (WP-MIP) is designed to include traditional physical models, AI models, and all forms of hybrids between the two.
So is this WP-MIP a bilateral ECMWF-ECCC initiative?
Not at all! One of the essential elements of the project is that we’ve got broad engagement of the operational weather prediction community. We currently have 17 centres from 6 different continents that plan to contribute forecasts. This project came out of discussions between WGNE and WIPPS, then merged with a planned follow-up to the successful DIMOSIC project that Linus ran a few years ago. In this first phase we’re focusing on medium-range guidance from global deterministic prediction systems, but we’re hoping to expand that scope in the future to include ensembles, limited area models, and more.
With such broad engagement, what do you hope to achieve?
As a first step, we’re hoping to build a centralized archive for distributed diagnostics and development. The important second step focuses on model evaluation, pulling in the expertise of the JWGFVR and PDEF and providing the community. We think that we’ll be able to accelerate the development of new evaluation tools and techniques that effectively compare AI, hybrid and physical model predictions. An important practical outcome of WP-MIP will be guidelines for evaluation best-practices that allow developers and operational centres to fairly assess strengths and weaknesses across the different modelling paradigms.
Are these goals similar to other projects in this domain?
The active involvement of so many national meteorological services and WMO working groups distinguishes WP-MIP from other efforts like WeatherBench and projects within NOAA. Those projects have played the important role of introducing AI techniques to the weather prediction community; however, a broader base of engagement is needed going forward. We believe that the WMO is the right organization to lead such an effort given its global reach and the broad range of expertise contained in its working groups.
What stage is the project at right now, and how would someone join the group?
Our contributors are in the process of setting up their integrations and submitting prototype data to the MARS system at ECMWF where the database will be hosted. We hope that early datasets will be available by the end of the summer. That being said, we’ll be more than happy to welcome new groups to the project at any time. Anyone can access the project Whitepaper or get in touch with either of us via email. Remember that you don’t have to be running a model to have a place in WP-MIP: diagnostics and evaluation are central to the project. We send out regular status updates to those who have expressed interest in the project, so please let us know and we’ll gladly add anyone to our distribution list!
For more information and access, check out the WP-MIP page linked here below
