Using Machine Learning Techniques for Polar Vortex Predictions
Source: Journal of Geophysical Research: Machine Learning and Computation
The polar vortex is a large swirl of cold air that circles the North Pole. Sometimes, it is disturbed by sudden stratospheric warming (SSW) events. These events can send waves of very cold air farther south than usual. When this happens, temperatures across North America and Europe can drop very quickly. But these events are still difficult to predict, so it is hard to warn people before an SSW event is about to happen.
Boscu et al. used a machine learning (ML) method to learn how SSW events behave. They trained it using a computer simulation based on the laws of physics that captures important movements in the atmosphere. This allowed them to run polar vortex simulations faster. They also examined the ML model itself to understand how it represents different features of the polar vortex.
The authors trained their model using the Holton-Mass model, which simulates changes in the stratosphere, a layer of Earth's atmosphere. This model produces two main states of the polar vortex: one weak and one strong. When the vortex changes from the strong state to the weak state, an SSW event occurs. First, the authors checked that their model, called a conditional variational autoencoder, could reproduce both the strong and weak states. They then used principal component analysis to better understand how the model represents these states. They found four groups of patterns that scientists could explain in terms of real atmospheric behavior. These groups represented the strong and weak states and the paths between them as the vortex changed from one state to the other.
The authors say these groups contain a clear signal that can predict whether the polar vortex is about to be disrupted. This suggests that scientists might be able to adapt the ML model to better predict when future SSW events will happen in the real world. They also say their work shows that artificial intelligence can help scientists model and understand complex features of the atmosphere. (Journal of Geophysical Research: Machine Learning and Computation, https://doi.org/10.1029/2026JH001387, 2026)
—Nathaniel Scharping (@nathanielscharp), Science Writer