A brand new system for forecasting weather and predicting future climate uses artificial intelligence to realize results comparable with the perfect existing models while using much less computer power, in response to its creators.
In a paper published in Nature yesterday, a team of researchers from Google, MIT, Harvard, and the European Center for Medium-Range Weather Forecasts say their model offers enormous “computational savings” and might “enhance the large-scale physical simulations which are essential for understanding and predicting the Earth system.”
The NeuralGCM model is the most recent in a gentle stream of research models that use advances in machine learning to make weather and climate predictions faster and cheaper.
What Is NeuralGCM?
The NeuralGCM model goals to mix the perfect features of traditional models with a machine-learning approach.
At its core, NeuralGCM is what’s called a “general circulation model.” It incorporates a mathematical description of the physical state of Earth’s atmosphere and solves complicated equations to predict what’s going to occur in the longer term.
Nonetheless, NeuralGCM also uses machine learning—a strategy of looking for patterns and regularities in vast troves of information—for some less well-understood physical processes, resembling cloud formation. The hybrid approach makes sure the output of the machine learning modules can be consistent with the laws of physics.
The resulting model can then be used for making forecasts of weather days and weeks prematurely, in addition to looking months and years ahead for climate predictions.
The researchers compared NeuralGCM against other models using a standardized set of forecasting tests called WeatherBench 2. For 3- and five-day forecasts, NeuralGCM did about in addition to other machine-learning weather models resembling Pangu and GraphCast. For longer-range forecasts, over 10 and 15 days, NeuralGCM was about as accurate as the perfect existing traditional models.
NeuralGCM was also quite successful in forecasting less-common weather phenomena, resembling tropical cyclones and atmospheric rivers.
Why Machine Learning?
Machine learning models are based on algorithms that learn patterns in the information fed to them after which use this learning to make predictions. Because climate and weather systems are highly complex, machine learning models require vast amounts of historical observations and satellite data for training.
The training process could be very expensive and requires lots of computer power. Nonetheless, after a model is trained, using it to make predictions is fast and low cost. It is a large a part of their appeal for weather forecasting.
The high cost of coaching and low price of use is comparable to other forms of machine learning models. GPT-4, for instance, reportedly took several months to coach at a price of greater than $100 million, but can reply to a question in moments.
A weakness of machine learning models is that they often struggle in unfamiliar situations—or on this case, extreme or unprecedented weather conditions. To enhance at this, a model must generalize, or extrapolate beyond the information it was trained on.
NeuralGCM appears to be higher at this than other machine learning models because its physics-based core provides some grounding in point of fact. As Earth’s climate changes, unprecedented weather conditions will develop into more common, and we don’t understand how well machine learning models will sustain.
No one is definitely using machine learning-based weather models for day-to-day forecasting yet. Nonetheless, it’s a really lively area of research—and a technique or one other, we might be confident that the forecasts of the longer term will involve machine learning.
