Projected Changes in Northern Hemisphere Weather Regimes Using a Deep Learning–Based Classification Approach

The accelerated warming of the Arctic relative to the rest of the globe has sparked ongoing debate about its influence on Northern Hemisphere atmospheric circulation. Many studies suggest that this warming may alter large-scale circulation through changes in temperature gradients, storm tracks, and planetary wave dynamics. From a weather regime perspective, which describes preferred and recurrent large-scale circulation patterns, this study investigates the projected changes in Northern Hemisphere atmospheric circulation across different seasons. First, the ability of CMIP6 models to reproduce observed circulation regimes is evaluated against ERA5 reanalysis. We then assess the projected response of these regimes under climate change scenarios in terms of their frequency of occurrence and persistence. The analysis focuses on mean sea level pressure and applies a physically informed convolutional autoencoder combined with k-means clustering. This data-driven climate classification workflow uses unsupervised deep learning to reduce the dimensionality of spatiotemporal climate simulation data into compact representations.

Results show that CMIP6 models generally reproduce the main Northern Hemisphere circulation patterns and their seasonal behavior, particularly in winter and spring, although performance varies among models. The ensemble mean slightly underestimates the amplitude of mean sea level pressure anomalies in all seasons, most notably in summer. Despite this bias, the main circulation patterns and their seasonal characteristics are reasonably well reproduced. Based on this present-day evaluation, projections toward the end of the twenty-first century indicate that changes in regime frequency are stronger and more robust under SSP5-8.5. Zonal regimes, such as the NAO+ pattern, as well as regimes associated with negative pressure anomalies over the Arctic, tend to become more frequent, in agreement with previous studies, while blocking regimes exhibit a systematic decline under warming. Finally, the weather regime framework provides the basis for an ongoing investigation of the associated impacts of projected circulation shifts on the regional climate system.