Deriving the four winter weather regimes

Summary. The large-scale atmospheric circulation in the Euro-Atlantic region is comprised of repeating patterns known as weather regimes. This post explains how they were obtained from reanalysis data using a simple process: remove seasonal means, apply PCA to daily fields, then cluster with k-means into the four standard winter weather regimes. The result is a set of four weather regimes and the distance of every day to each of the regimes.

Why look at regimes?

Regimes are quasi-stationary, persistent large-scale patterns that have an effect on surface weather, which in turn is associated with the generation of renewable energy and the electricity demand (temperature, wind speed, presence or absence of clouds). They add structure at the sub-seasonal range and support consistent discussions of risk and planning. Their identification is sensitive to analysis choices, so methodological consistency is important. For more information on weather regimes, refer to the ECMWF Newsletter 165 [1].

Scope and data

• Dataset: ERA5 reanalysis
• Region: 90°W–30°E, 20–80°N (Euro-Atlantic region)
• Time period: Winters from 1979 to 2018
• Temporal and spatial resolution: Daily data at 1° resolution
• Field: 500 hPa geopotential height, which may need to be transformed from geopotential (the variable available in ERA5) by dividing it by the acceleration of gravity at surface (other studies also use 700 hPa or MSLP)

Method

1. Pre-processing: Compute daily anomalies (remove winter mean) and apply latitude-cosine weighting
2. PCA: Reduce dimensionality; keep 14 components to preserve close to 90% of variance and suppress small-scale noise
3. Clustering: k-means with k=4, initialized from the first four PCs assigns each day to its nearest cluster
4. Back-projection: Invert PCA and remove the latitude weighting to obtain regime centers in regular map space

 

 

Figure 1. Cluster-center 500 hPa geopotential height anomalies for the four winter weather regimes (1979–2018) based on the representation in Van der Wiel et al. (2019) [2].

Outputs

• Regime maps: Four cluster-center anomaly fields describing typical large-scale flow states
• Daily diagnostics: Distances from each day to each regime center; useful as a “strength of match” metric. Often, the only factor considered is which one is the closest cluster to the daily average for any given day

Some factors to consider

• Results depend on domain, season, variable choice, and number of regimes; these should be kept fixed if comparisons are being made
• PCA reduces computational cost and filters smaller, non-repeating patterns, but also smooths local features; PCA may not be suitable for cases where local features are of interest
• Clusters summarize variability, but no single day will look like the cluster mean

Minimal code skeleton (for reproducibility) 

import xarray as xr
import numpy as np
from sklearn.decomposition import PCA
from sklearn.cluster import KMeans

# Assume z500 is an xarray.DataArray of the 500hPa geopotential height with dimensions time, latitude, longitude
z500 = z500 * np.cos(np.radians(z500['latitude']))
z500 = z500 - z500.mean(dim='time')
z500_2d = z500.stack(space=('longitude','latitude'))

pca = PCA(n_components=14)
Z = pca.fit_transform(z500_2d)

kmeans = KMeans(n_clusters=4, init=[
  [1,0,0,0,0,0,0,0,0,0,0,0,0,0],
  [0,1,0,0,0,0,0,0,0,0,0,0,0,0],
  [0,0,1,0,0,0,0,0,0,0,0,0,0,0],
  [0,0,0,1,0,0,0,0,0,0,0,0,0,0]], verbose=1)

dist = kmeans.fit_transform(Z)
centers_geo = pca.inverse_transform(kmeans.cluster_centers_)

da_distance_to_centers = xr.DataArray(dist,
coords={'time':z500_2d['time'], 'cluster':np.arange(4)})
da_cluster_centers = xr.DataArray(centers_geo,
coords={'cluster':np.arange(4), 'space':z500_2d['space']}).unstack()

da_cluster_centers =
da_cluster_centers/np.cos(np.radians(da_cluster_centers['latitude']))

# The output da_distance_to_centers is an xarray.DataArray that contains the distance from the z500 of each days to each of the cluster centers
# The output da_cluster_centers is an xarray.DataArray that contains the cluster centers for each of the four clusters 

Sample view of the internal python workflow

References

1. ECMWF Newsletter 165 (Autumn 2020): How to make use of weather regimes for extended-range predictions in Europe. https://www.ecmwf.int/en/newsletter/165/meteorology/how-make-use-weather-regimes-extended-range-predictions-europe ↩︎
2. van der Wiel, K., Bloomfield, H. C., Lee, R. W., Stoop, L. P., Blackport, R., Screen, J. A., & Selten, F. M. (2019). The influence of weather regimes on European renewable energy production and demand. Environmental Research Letters, 14(9), 094010. DOI: https://doi.org/10.1088/1748-9326/ab38d3 ↩︎

 

NexSys Away Day in Wicklow

On May 27th, NexSys held an away day at the beautiful Glenview Hotel, near the Glen of the Downs in Wicklow. The day began with an opening presentation from NexSys director, Terrence O'Donnell, who gave us an update on the group's progress and outlined the schedule for the day. The event was divided into two parts: a morning session focused on data and model integration, which included group discussions led by James O'Donnell, followed by small group talks to share insights with the wider team.

 

Following an enjoyable lunch at the hotel’s restaurant, where I had the opportunity to engage in discussions with senior researchers from NexSys. The afternoon session was dedicated to developing greater research collaboration. Led by Michelle Carey from the School of Mathematics and Statistics at UCD, the session allowed us to learn about each other’s work, share challenges, and identify ways to support one another’s research. For instance, during my discussion, I learned that a fellow PhD student was developing a power system model that incorporated electricity prices under different scenarios. He mentioned his need for data on solar and wind generation over extended periods, something that I could provide based on my own work.

 

To close out the day, we embarked on a guided hike through the stunning Glen of the Downs, offering a perfect opportunity to unwind and connect with colleagues. The day concluded with a lovely BBQ, leaving us all with new insights and strengthened connections among the researchers in the NexSys group. 

PhD Masterclass by Dr. John Bell

 On Monday, November 4th, I had the opportunity to attend a masterclass on the intersection of research and policy, led by Dr. John Bell, the European Commission's Director for a Healthy Planet. This event, organized by the UCD Earth Institute, brought together PhD students interested in climate, environmental, and sustainability research.

The session began with a brief introduction by Imelda Maher, the event's organizer, who is the Sutherland Full Professor of European Law and Director of the UCD Dublin European Institute. Following this, Dr. Bell shared his reasons for joining us, explaining his interest in discussing the role of research in policy-making with us, as early-career researchers. He highlighted how our work, as emerging scientists, can have an impact on policies enacted by the European Union and other governmental bodies.

After the introductions, each of us had the chance to briefly present our research topics and pose a question to Dr. Bell. The research areas represented were impressively diverse, ranging from studies on biological systems and social sciences to energy systems, which is my area of focus. I also found it inspiring to see participants of different ages, including some older PhD students, something I don’t often encounter.

My question for Dr. Bell focused on the perception of scientific research by the public. I pointed out that while science aims to deepen our understanding of the world, research often involves hypotheses that may lead to inconclusive or evolving results. This nuance is not always well understood outside academic circles, which can lead to mistrust. I asked him how we might help the public understand that scientific findings are not absolute truths but are instead part of a broader journey of discovery.

Dr. Bell responded by emphasizing the need to keep "universities as lighthouses" that guide the future of policy-making. He stressed that for research to truly influence policy, we as researchers need to become better communicators, making our findings more accessible and comprehensible to a wider audience.