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.
k-means clustering with k=4 is performed on the first 12 principal components of daily 500 hPa geopotential height anomalies in December-March months during 1979-2018 in the region 180-30°W 20-80°N, following the method of Lee et al. 2019 (GRL). This yields 4 weather regimes: the Arctic High (ArH), Arctic Low (ArL), Alaskan Ridge (AkR) and Pacific Trough (PT). Click here to see a spatial map of each regime.
Each ensemble member on each day is assigned to a regime based on its minimum Euclidean distance to the centroid of the cluster in principal component space. Please note that at current there is no bias correction on the Z500 anomaly field, so inherent biases in GEFS may be evident with lead-time, and thus this product should be considered experimental.
The regime attribution of the ensemble-mean, in principal component space, is indicated by the square at the top of each bar. Similar results are typically obtained using the modal regime.
Individual ensemble members
Regime of weekly-mean field
Forecast evolution of ensemble-mean regime