Skip to content
Cyberttak.
· 2 min read

Ai2 and Google partnered to build a more accurate climate prediction model

The new collaboration aims to create a long-term climate model using Google’s NeuralGCM architecture and Ai2’s expertise.

By CyberttakPublished · 23:39

In brief

Ai2 and Google build faster climate models

The partners are developing NeuralGCM+, an AI-driven tool for long-term predictions. It combines physical science with AI to help communities adapt to a changing climate.

A scientist adjusts a spinning globe in a climate research laboratory, with ocean current diagrams and atmospheric layer models nearby and a wall of weather cha
A scientist adjusts a spinning globe in a climate research laboratory, with ocean current diagrams and atmospheric layer models nearby and a wall of weather chaCredit: AI-generated · GPT Image 2

The Allen Institute for AI (Ai2) has entered a three-year partnership with Google.org and Google Research to advance Earth system modeling. This collaboration focuses on expanding the capabilities of NeuralGCM, an artificial intelligence model that blends physical equations with AI for climate simulation.

Traditional climate models are often limited by high computational costs. This makes running many different future scenarios difficult for researchers. NeuralGCM aims to change this by being faster and more computationally efficient than older systems.

Much of the progress in Earth system AI modeling has focused on weather forecasting, in the range of days to months.

While Google has previously used NeuralGCM for short-term weather forecasting, the new project targets climate trends spanning years to decades. The team intends to simulate how oceans and the atmosphere interact, which is a major factor in global heat redistribution.

Ai2 will also build upon its existing work with the Ai2 Climate Emulator (ACE), a series of models designed to speed up climate experiments. Integrating these methods into the new project is expected to make long-term projections more reliable.

The research agenda includes three primary goals:

  • Connect atmosphere and ocean models for stable, multi-decade simulations.
  • Improve the model's response to warming trends beyond current training data.
  • Make the tools open-source and easy for other scientists to use on common hardware.

Developing a model that can accurately project climates we have not yet experienced is a significant challenge. The project aims to bridge this gap by combining specific local atmospheric process insights with broader physical simulations.

Have you ever wondered if better data could change how our cities plan for water resources and climate changes?

This technology is still in development. The team has not yet proven that these AI models can reliably simulate future conditions that deviate significantly from historical observations.

Source: Ai2 Blog