how a.i. can help us predict the weather
Traditional weather forecasting is slow and expensive. AI can offer a more efficient and accurate alternative.
Weather and climate are extremely difficult to model. There are two reasons why. The first is that they’re complex, interconnected systems where a tiny change at the right time and in the right place can start a massive cascade that changes everything. The second is math. Navier–Stokes equations, which are used in describe the motions of water and air, are extremely difficult to solve at scale without using massive compute power and resources. And even then, there’s a degree of uncertainty.
Enter an AI model called Aurora. It was trained on 1.3 billion variables around pollution, temperatures, historical record of storms and anomalies, pressures, and even climate simulations and records, then let loose to create forecasts on its own.
The results were impressive to put it mildly. It outperformed existing models for both five and ten day forecasts and its predictions for storms and hurricanes were not just faster, but more accurate. Computing the likelihood and path of these storms from a set of records treated as if they were new information, Aurora was better at plotting likely courses, and calculating wind speeds, pressures, and air quality.
Okay, but how exactly can it do that? The most likely answer is that we still have a lot to learn about applying existing mathematical formulas to weather and a massive AI model provides convenient shortcuts, focusing squarely on the physics embedded in its architecture and historical records showing what happens when the inputs align a certain way. In other words, our existing models may be getting sidetracked and lack some context that Aurora can apply.
Next for the model is continuing to run it and generate forecasts to make sure it stays that accurate on completely new raw data, then connecting it directly to satellites and weather stations to generate predictions in virtually real time at incredible resolutions, down to square meters compared to its current limit of 10 square kilometers.
If it works, it could revolutionize weather forecasting, storm tracking, early warnings, and climate modeling by contextualizing and processing more data than ever before, saving lives, optimizing agriculture, and guiding infrastructure projects.
Aurora is a perfect example of how we can use AI not just to generate clicks and save money on automation of entry level corporate jobs, but to actually improve our lives, just like its counterparts currently diagnosing cancer, discovering new antibiotics, and solving universal mysteries. And we will definitely want to keep an eye on how well it’ll continue to work in the near future.
See: Bodnar, C., Bruinsma, W.P. et al. A foundation model for the Earth system. Nature (2025), DOI: 10.1038/s41586-025-09005-y