Note: These regimes are defined for December-March, based on wintertime modes of variability, but I will leave them running for the intervening warm season since I’m curious to see how they evolve and what features of the flow are represented.
The 500 hPa geopotential height anomalies are transformed into 12-dimensional principal component space (based on the leading 12 principal components during December-March in ERA5 reanalysis) and averaged over each 7-day period. Each ensemble members is then assigned to one of four regimes (Arctic High, Arctic Low, Alaskan Ridge, Pacific Trough; for more information see here) by the minimum Euclidean distance to the cluster centroid in principal component space. The maps below show the average of all ensemble members assigned to each regime and the number of members in each regime. The bold font and coloured axes indicate the regime(s) with the largest ensemble support.
Please note that currently the 500 hPa anomalies are computed with respect to ERA5 climatology. Biases in the GEFS Z500 field may cause incorrect regime attribution, particularly when biases are co-located with the cluster centroids and exceed the magnitude of the true flow anomaly.