Recently, the European Center for Medium-Range Weather Forecasts (ECMWF), together with the meteorological and hydrological departments of several European countries, jointly launched a collaborative plan called "Anemoi" to create a machine learning weather forecast system, aiming to provide the most advanced data-driven models. A key component that helps meteorological and hydrological departments in European countries use their own data to train models and run them in operations.
Anemoi builds on the experimental Artificial Intelligence Forecasting System (AIFS) developed by ECMWF and further extends the AIFS code base to meet the needs of a wider range of users. Its code is freely available on GitHub, and anyone can use it or contribute to its development with a license.
Anemoi contains multiple software packages written in Python language, covering different aspects of the artificial intelligence weather forecasting process - the Anemoi data set component can generate machine learning optimized data sets from meteorological data and observation data from different sources and different formats to ensure Providing high-quality, consistent and optimized data for model training can greatly simplify the data preparation process; the Anemoi training component provides a high degree of flexibility, and users can modify most parts of the training process through configuration files without modification. Modify the underlying code to ensure that meteorologists without deep programming expertise can experiment with data-driven weather forecast models; the Anemoi model component provides model code with the goal of efficiency and minimal dependencies to ensure a smooth transition from development to deployment of the model. ; The Anemoi reasoning component is based on ECMWF's experience in artificial intelligence models and can quickly deploy trained models in business; the Anemoi drawing component supports custom graph generation, and researchers can easily visualize charts.
Currently, Anemoi attracts representatives from the Spanish National Meteorological Institute, the Danish Meteorological Institute, the German Meteorological Office, the Finnish Meteorological Institute, the Italian Air Force Meteorological Service, the Royal Netherlands Meteorological Institute, the Norwegian Meteorological Service, Météo-France, the Swiss Meteorological Service and the Royal Meteorological Service of Belgium. Participation of the Institute. Some countries have made progress in creating machine learning models based on Anemoi. For example, the Norwegian Meteorological Office has created a regional model for Scandinavia, and the German Meteorological Office is using its global numerical forecast model (ICON) data to develop a model called AICON. Data-driven weather forecast model.