AI-powered weather and climate models are poised to change the future of forecasting, researchers say

A new weather forecasting and climate prediction system uses artificial intelligence (AI) to achieve results comparable to the best existing models while using much less computing power, its creators say.

In an article Published in NatureA team of researchers from Google, MIT, Harvard and the European Centre for Medium-Range Weather Forecasts say their model offers huge “computational savings” and can “improve large-scale physics simulations that are essential for understanding and predicting the Earth system.”

The NeuralGCM model is the latest in a continuing series of research models that use advances in machine learning to make weather and climate forecasting faster and cheaper.

What is NeuralGCM?

The NeuralGCM model aims to combine the best features of traditional models with a machine learning approach.

At its core, NeuralGCM is what is known as a “general circulation model.” It contains a mathematical description of the physical state of the Earth’s atmosphere and solves complex equations to predict what will happen in the future.

However, NeuralGCM also uses machine learning (a process of finding patterns and regularities in large volumes of data) for some less well-understood physical processes, such as cloud formation. The hybrid approach ensures that the results of the machine learning modules will be consistent with the laws of physics.

The resulting model can then be used to make weather forecasts several days and weeks in advance, as well as to predict the climate several months and years ahead.

The researchers compared NeuralGCM to other models using a standardized set of prediction tests called WeatherBench 2For three- and five-day forecasts, NeuralGCM performed about as well as other machine learning weather models such as Pangu And GraphCastFor longer-term forecasts, over ten and fifteen days, NeuralGCM was about as accurate as the best existing traditional models.

NeuralGCM has also had some success in predicting less common weather phenomena, such as tropical cyclones and atmospheric rivers.

Why machine learning?

Machine learning models rely on algorithms that learn patterns in the data they are given, and then use that learning to make predictions. Because climate and weather systems are extremely complex, machine learning models require large amounts of historical observations and satellite data for training.

The training process is very expensive and requires a lot of computing power. However, once the model is trained, using it to make forecasts is fast and inexpensive. This is one of the main reasons why they are attractive for weather forecasting.

The high training cost and low cost of use are similar to other types of machine learning models. GPT-4, for example, apparently It took months of training at a cost of more than $100 million, but it can respond to a query in moments.

One of the weaknesses of machine learning models is that they often struggle to perform in unusual situations, such as extreme or unprecedented weather conditions. To do this, a model must be able to generalize, or extrapolate beyond the data it was trained on.

NeuralGCM appears to outperform other machine learning models in this area because its physics-based core provides some grounding in reality. As the Earth’s climate changes, unprecedented weather patterns will become more common, and it’s unclear how well machine learning models will be able to keep up.

No one is using machine learning-based weather models for daily forecasts yet. However, this is a very active area of ​​research and, one way or another, we can be sure that future forecasts will rely on machine learning.

This article is republished from The conversation under Creative Commons license. Read more original article.