Accurate and timely weather forecasting is a famously difficult task. However, recent advances in neural networks and large-scale compute has resulted in noteworthy improvements. A senior researcher at the University of Miami has recently conducted some preliminary analysis of the accuracy of numerous weather forecasting models through the 2025 US storm season, including the system operated by the US National Weather Service (the Global Forecast System (GFS) model), and the results are fascinating. This analysis was picked by an editor at Ars Technica who wrote an interesting article that grabbed my attention.
The difference in errors between the US GFS model and Google’s DeepMind is remarkable. At five days, the Google forecast had an error of 165 nautical miles compared to 360 nautical miles for the GFS model, more than twice as bad. This is the kind of error that causes forecasters to completely disregard one model in favor of another.
Not only did the DeepMind model outperform the GFS model...
But there’s more. Google’s model was so good that it regularly beat the official forecast from the National Hurricane Center (OFCL), which is produced by human experts looking at a broad array of model data.
The DeepMind model not only predicted hurricane tracks, but also hurricane intensity.
It’s worth noting that DeepMind also did exceptionally well at intensity forecasting, which is the fluctuations in the strength of a hurricane. So in its first season, it nailed both hurricane tracks and intensity.
A Google DeepMind blog expands on this:
In physics-based cyclone prediction, the approximations required to meet operational demands mean it’s difficult for a single model to excel at predicting both a cyclone’s track and its intensity. This is because a cyclone's track is governed by vast atmospheric steering currents, whereas a cyclone’s intensity depends on complex turbulent processes within and around its compact core.
The DeepMind blog is particularly interesting.
What is fascinating to me, is not so much weather prediction specifically, but more generally the potential that these sorts of neural network models may be able to improve conflict and geopolitical risk forecasting; which much like the weather is famously difficult to predict.
While these sorts of models are improving forecasting accuracy on challenging topics like the weather, predicting the next conflict before it happens, along with it's intensity, remains unsolved. However, study into this area remains ongoing. See here, here and here for a taste.