EGU2020-349
https://doi.org/10.5194/egusphere-egu2020-349
EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Revisiting the Identification of Wintertime Atmospheric Circulation Regimes in the Euro-Atlantic Sector

Swinda Falkena1, Jana de Wiljes2, Antje Weisheimer3, and Theodore G. Shepherd4
Swinda Falkena et al.
  • 1Department of Mathematics and Statistics, University of Reading, Reading, United Kingdom (s.k.j.falkena@pgr.reading.ac.uk)
  • 2Institute for Mathematics, University of Potsdam, Potsdam, Germany
  • 3European Centre for Medium-Range Weather Forecasts (ECMWF), Reading, United Kingdom
  • 4Department of Meteorology, University of Reading, Reading, United Kingdom

A number of methods exist for the identification of atmospheric circulation regimes. The most commonly-used method is k-means clustering. Often the clustering algorithm is applied to the first several principal components, instead of the full field data. In addition, many studies use a time-filter to get rid of high frequency oscillations before the clustering is executed. We discuss the consequences of these filtering techniques on the identified circulation regimes for the Euro-Atlantic sector in winter. Most studies identify four regimes: the Atlantic Ridge, the Scandinavian Blocking, and the two phases of the North Atlantic Oscillation. However, when k-means clustering is applied to the full field data of a reanalysis dataset, the optimal number of regimes is not found to be four, but six. This optimal number is based on the use of an information criterion, together with consistency arguments. The two additional regimes can be identified as the opposite phases of the Atlantic Ridge and Scandinavian Blocking, since they have a low-pressure area where the original regimes have a high-pressure area. Furthermore, the incorporation of a persistence constraint within the clustering algorithm is found to preserve the occurrence rates of the regimes, and thus maintains the consistency of the results. In contrast, applying a time-filter to enforce persistence of the regimes changes the occurrence rates. We conclude that care must be taken when filtering the data before the clustering algorithm is applied, since this can lead to biases in the identified circulation regimes and their occurrence rates.

How to cite: Falkena, S., de Wiljes, J., Weisheimer, A., and Shepherd, T. G.: Revisiting the Identification of Wintertime Atmospheric Circulation Regimes in the Euro-Atlantic Sector, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-349, https://doi.org/10.5194/egusphere-egu2020-349, 2019

Comments on the presentation

AC: Author Comment | CC: Community Comment | Report abuse

Presentation version 2 – uploaded on 07 May 2020 , no comments
Included email address.
Presentation version 1 – uploaded on 23 Apr 2020
  • CC1: Trends in regimes over 1979-2018 or in the future?, Florentin Breton, 05 May 2020

    Hi, thank you for presenting this interesting research. Do you think there could be trends in the occurrence, spatial pattern, or persistence of the regimes over 1979-2018 or in the future?

  • CC2: Comment on EGU2020-349, Nikolaos Mastrantonas, 05 May 2020

    Hello Swinda et al,

    Thank you for this nice work!

    Based on your findings it would be advisable to use the actual data without prior smoothing, and also calculate deviations from a fixed state, without incorporating seasonality information. Will the second part provide informative results, also when all daily data are used for identifying the weather regimes, instead of only the Dec-Mar ones? Or in such a case, it would be useful to have at least 2 background states (summer/winter)?

    Thanks,

    Nikos

     

    • AC1: Reply to CC2, Swinda Falkena, 06 May 2020

      Hi Nikos,

      Thank you for your interest in our work! In general I would say less smoothing before clustering is better, however we do know seasonality has an effect on the regimes. Therefore I believe that a fixed background state is suited when studying only one season (e.g. winter or summer, not sure about spring or fall), but when considering a full year I think the signal you will pick up when using a fixed background state is seasonality itself. In that case I do think subtracting the (smooth) seasonal cycle is the best way to go, as using only a discrete (e.g. 2) number of background state could introduce discontinuities in the result. I hope this answers your question.

      Thanks,

      Swinda