ECMWF is pleased to announce Discover Anemoi - a six-part series of training webinars covering the use of the Anemoi framework.
Anemoi is a collaborative and open-source framework for developing machine learning weather forecasting models. Named after the Greek gods of the winds, the goal of Anemoi is to provide the key building blocks to train state‑of-the‑art data-driven weather forecasting models and run them in an operational context. As a framework it seeks to handle many of the complexities that meteorological organisations will share, allowing them to easily train models from existing recipes but with their own data.
Find out more about Anemoi in our recent news article.
Discover Anemoi will feature six webinars, demonstrating the use of key components of the Anemoi ecosystem. The webinars will cover:
- An introduction to Anemoi
- Creating datasets
- Building graphs
- Training models
- Inference
- Contributing as a developer
The lectures will build upon each other, following the full life-cycle of a model from conception to inference, but can also be followed independently if you are already familiar with some parts of the Anemoi framework.
Please note that the webinars assume some familiarity with machine learning and weather forecasting.
The webinars are open to all, however some products and services that will be showcased may only accessible to ECMWF's Member and Co-operating States.
All the webinars will be recorded and recordings will be made publicly available after the event.
To attend the webinars, you need to register separately for each one - see the links below. The registration for each webinar closes two days before the event, so please register in advance.
Webinars
Webinar 1 - Introduction to Anemoi
Monday 13 January 14:00 UTC - 15:00 UTC
Get started with Anemoi, ECMWF's open-source framework for machine learning weather forecasting. In this introductory session, you'll learn about the framework's core philosophy, key components, and how it fits into the operational weather forecasting landscape. We'll walk through practical examples demonstrating how Anemoi streamlines the development of data-driven weather models using graph neural networks while incorporating meteorological expertise. Perfect for both newcomers and experienced practitioners interested in modern data-driven weather forecasting approaches.
Speaker: Jesper Dramsch (ECMWF)
Webinar 2 - Anemoi Datasets
Wednesday 15 January 14:00 UTC - 15:00 UTC
This webinar will introduce anemoi-datasets, which manages datasets for training data-driven weather models. We will explore how to create, format and transform data sets from various sources. The webinar will also showcase how datasets can be combined and subsetted, allowing users to select exactly the data they need to train their models.
Speaker: Baudouin Raoult (ECMWF)
Webinar 3 - Anemoi Graphs
Tuesday 21 January 14:00 UTC - 15:00 UTC
In this session, we introduce anemoi-graphs, the tool within the Anemoi environment for creating graphs, which serve as the data structure for anemoi-models. This session is designed for users interested in working not only with global models but also for researchers focused on limited area models or stretched grids. It includes a presentation on how to build customised graphs with best practices and shows the inspection tools available.
Speaker: Mario Santa Cruz (ECMWF)
Webinar 4 - Anemoi Training
Thursday 23 January 14:00 UTC - 15:00 UTC
Dive deep into the model training capabilities of Anemoi. This session covers the training of weather forecasting models, from configuration and hyperparameter tuning to monitoring and validation. You'll learn how to leverage Anemoi's training infrastructure to develop robust models. We'll explore practical examples of training workflows and discuss strategies for achieving optimal performance, including distributed training for high-resolution models.
Speaker: Jesper Dramsch (ECMWF)
Webinar 5 - Anemoi Inference
Tuesday 28 January 14:00 UTC - 15:00 UTC
After successfully training a model and saving it to disk, we can run the model in inference mode and obtain predictions from it. While training typically takes place on a historic dataset, inference often takes place on recent data in order to produce useful forecasts. This session will teach you the basics of inference, how to integrate it into your workflows and some pitfalls to look out for when deploying models operationally.
Speaker: Gert Mertes (ECMWF)
Webinar 6 - Anemoi for Developers
Thursday 30 January 14:00 UTC - 15:00 UTC
Join us for a deep dive into the Anemoi ecosystem: Anemoi equips developers with modular tools to build, train, and deploy ML-powered weather models using their own data. This seminar is geared towards developers who would like to contribute and collaborate within the Anemoi-ecosystems. We will look at design choices and how we can leverage the existing software architecture to incorporate new components such as new model architectures and graph structures.
Speakers: Helen Theissen and Jesper Dramsch (ECMWF)