AI General Circulation Model

Machine learning as a tool to innovate weather forecasting systems
Thematic area: Earth/Climate, Projects
Financing: IFAB call for projects
Enabling Technology: Image recognition, Machine Learning, Super-resolution

A new model for the forecasting of meteorological phenomena based on modern machine learning techniques, to greatly enhance forecasting systems, by considerably reducing calculation times and energy expenditure.


  • Creation of a new approach to the Climate General Circulation Model (GCM) by exploiting recent developments in the field of Machine Learning (ML) and intra-seasonal atmospheric signals.
  • Significant reduction in the time required to obtain short-term forecasts and, if possible, increase in medium-term forecasting capabilities.

The initial challenge

Today’s weather forecasts are based on GCM – Global Circulation Model – computer systems, which are designed to reproduce the behaviour of the global meteorological and climatic system by applying physical and mathematical models. These models, which allow scientists to better understand and consequently predict the mechanisms that govern the atmosphere and oceans, have significant costs in terms of calculation resources (large infrastructures), energy/manpower and time: a ten-day forecast can require many hours of calculation and use hundreds of supercomputer nodes.

The “AI General Circulation Model” project (AIGCM) aims to develop a Proof-of-Concept (POC) of a new meteorological model based on machine learning that is potentially competitive with current models. The new “General Circulation Model” would overcome some of the limits of today’s forecasting systems, as it would significantly reduce both the infrastructure costs and the time required for forecasting, and could, for example, provide data that help to predict energy demands and/or production well in advance.

The solution

This innovative approach is now possible thanks to the recent developments in machine learning (ML) and deep learning based on neural networks (NN) and thanks to the availability of a huge amount of data on atmospheric behaviour, data that have been reconstructed by means of reanalysis products (using classic GCM) and have a temporal depth of up to the last 70 years.

Neural networks are automated learning models that are part of the broader set of machine learning algorithms and are based on the assumption that it is possible to simulate the complexity of the problem thanks to the abundant and vast amount of data. The neural network is thus “trained” and develops a system that makes it possible to make forecasts for the future based on the acquired behaviour of the object of interest (in this case the fluid air/water). Neural networked-based methods are very fast in their estimations and constitute an ideal balance between model complexity, forecast resolution and estimation accuracy. Daily forecasts using these techniques can take between a few tenths of a second and a few minutes, making them several orders of magnitude faster than conventional techniques!

The “AIGCM” project model will have the following characteristics:

  • Calculation speed: minutes and not hours;
  • Forecast scope: up to 5 days;
  • Geographical precision of forecasts: 2.2 km grid
  • Geographical area: national (Italy)

The Project consists of three phases:

    • Phase 1 – Setting of the infrastructure environment and choice of the machine learning technique (type of neural network)
    • Phase 2 – Neural network training using historical data (satellite images and measurements). This will be the most time-consuming and computationally-intense phase.
    • Phase 3 – Refinement of the model in order to make it specialised in the prediction of certain data, e.g.: specific temperature forecasts, 48-hour forecasts, etc.


The resulting model prototype will serve as a starting point for future advances in the construction of a fully-operational AI-GCM that can potentially be integrated with conventional GCM models.

The benefits will be felt particularly in those decision-making processes that are greatly influenced by the availability of fast and accurate meteorological information: energy companies, for example, or even energy communities, will be able to make decisions regarding the production and storage of energy well in advance; again as an example, farmers will be able to implement resilience actions against extreme meteorological phenomena, public protection services will be able to raise the alert level in the presence of certain peaks, insurance companies will be able to customise their services based on the geographical area and seasons.


For further information, please contact:

Sustainable Development Goals


Share on:

You may also be interested in:

Stay updated on the latest IFAB events and projects.Subscribe to our monthly newsletter