Ai2 and Google fund a faster, open climate model for the coming decades

The Allen Institute for AI, known as Ai2, is building an open-source climate model called NeuralGCM+. It has multimillion-dollar, three-year funding from Google.org and technical support from Google Research. The goal is faster and more reliable projections of how climate could change over years and decades, including rain and water supplies.

What is being built

NeuralGCM+ builds on a Google Research model called NeuralGCM. It mixes two approaches. Physical equations handle large-scale movement of the atmosphere. AI handles smaller-scale events such as clouds and rainfall. Ai2 will extend the model to cover the atmosphere and the ocean together, so it can capture how oceans absorb and move heat. That heat has a major influence on long-term climate.

Why speed matters

Conventional climate models have shaped scientists' understanding of global warming for decades. But they need a lot of computing power, which limits how many experiments researchers can run. Faster, cheaper simulations could let them test more possible futures. As an example, an atmosphere-only version of NeuralGCM simulated a year of global atmospheric conditions in eight minutes on a single TPU, a computer chip, in a public benchmark.

Ai2 has also made progress with its own tool, the Ai2 Climate Emulator, or ACE. In one study, ACE2 ran experiments on how ocean-temperature patterns affect climate about 100 times faster than the conventional model it learned from.

The hard part

Speed is only part of the challenge. AI models trained on past data can match history well. But they struggle to project climates that have not happened yet. NeuralGCM's design may help, because its AI focuses on local processes inside slices of the atmosphere. That is a simpler job and may hold up better as the world warms. Ai2 says proving reliable future projections is still a central research task.

Who it could help

Google Research is working on NeuralGCM for nearer-term forecasts, such as how warm or wet an upcoming summer might be. Ai2 will focus on longer trends. One example is how rainfall and water availability could change through mid-century in regions such as the southwestern United States. Ai2 says it wants to put these tools within reach of more scientists and help communities make better decisions on water planning and adapting to climate change.

What is not known yet

• No release date for NeuralGCM+ has been given.
• The exact funding amount has not been shared.
• The model has not yet shown it can make reliable long-range projections.

Ai2 says its long-term aim is a shared platform that climate modeling labs can use and improve together.