Hello, I am reaching out in order to get some more information/guidance on the process of clustering in the ECMWF medium-range product. As someone can read in this documentation (https://www.ecmwf.int/en/forecasts/documentation-and-support/ens-cluster-products), the clustering is performed over the following time windows: 3-4 days, 5-7 days, 8-10 days, 11-15 days. Although it is described that the members remain in the same cluster within one time window, I am wondering how the synoptic consistence can be guaranteed when going from one time window to the other. And why is there not a continuation of the period, i.e. days 3-4, 4-6, 6-8 etc.

Thank you in advance for your time!

4 Comments

  1. Rebecca Emerton

    Hi Vasilis Pappas - thanks for your question! There is some further information regarding the methods for the medium-range clustering in our forecast user guide at this link: Clustering - ENS Medium-Range. I hope this will provide more insight into your question regarding the synoptic consistency, which is discussed in more detail in the forecast user guide than on the website. If this doesn't answer your question please let us know here, and in the meantime I'll see if I can pass on your question to someone who works more closely with these products and might be able to answer regarding the time window continuation. Thanks, Rebecca. 

  2. Laura Ferranti

    Hi Vasilis - thanks for your question. As you mentioned, the medium range clustering products describes the forecast ensemble distribution for 4 sections of the forecast trajectories. The calculation for each trajectory section (or time window) is carried out independently and therefore there is no constraint to garantee synoptic consistency among clusters belonging to different trajectory sections. That is why we use 4 well separated sections. The differences in spread between the early and late part of the forecast is too large to allow the clustering computation to be effective when using the whole forecast trajectory. I hope this answer your question, best regards, Laura.

  3. Vasilis Pappas

    Thank you to both Rebecca Emerton and Laura Ferranti . Two questions please: 1) is there a similar link/methodology description on the EC46 clustering?, 2) i agree that the trajectory sections (time windows) are well separated with each other. How should someone then use the in-between periods if they wanted to develop a continuous time-series based on the ensemble forecast, i.e. would it be considered an error to consider time windows such as 1-4, 4-7, 7-10, 10-15? Thank you, Vasilis.

  4. Laura Ferranti

    Hi Vasilis - yes you can find the descriptions of the WR probabilities fro the extended range in the user guide. Please note that for this products we use predefined climatological regimes and we attribute the daily values of each enesembe member to one of climatological regime. The approach is quite different to the one used for the medium range product.  There is nothing wrong in computing the clustering for subsequent time windows (1-4, 4-7). However you will still have no constraint for synoptic consistency between 1-4 and 4-7.  I think that it will be very difficult to create a continuos time-serie for the whole trajectory with this approach. Please note that the synoptic consistency is garanteed only within the limit of the time window. 

    I hope this answer your questions

    best regards

    Laura