When i download data from the monthly seasonal forecast dataset at:
https://cds.climate.copernicus.eu/cdsapp#!/dataset/seasonal-monthly-single-levels?tab=form
e.g.
Then for each lead time month I have 28 members. I believe that these data are derived from daily seasonal forecast data, but for the Met Office, there are only data for 4 dates each month:
How are these daily data used to form the monthly data? and how is the ensemble numbering used?
Thanks,
Kevin
For Met Office hindcast, monthly statistics or anomalies of month M include individual members numbered from 0 to 27 (so 28 in total). These members are actually initialized on the 1st of the month M and 25th, 17th, 9th of previous month M-1 (with 7 realizations for each of these start dates).
The corresponding initialization dates for the individual members of monthly products are as follows:
member 0: 1st of month M realization 1
member 1: 1st of month M realization 2
............
member 6: 1st of month M realization 7
member 7: 25th of month M-1 realization 1
member 8: 25th of month M-1 realization 2
............
member 13: 25th of month M-1 realization 7
member 14: 17th of month M-1 realization 1
member 15: 17th of month M-1 realization 2
............
member 20: 17th of month M-1 realization 7
member 21: 9th of month M-1 realization 1
member 22: 9th of month M-1 realization 2
............
member 27: 9th of month M-1 realization 7
Usually, the Met Office forecast include monthly products with individual members from 0 to maximum 61 (so 62 in total). This depends on the number of days in the previous month: for example products for August will have 62 members in total, while products for March will have only 56 members in total (non-leap years). There are 2 realizations per day and the start dates are 1st of the current month M and all days of previous month M-1 except the 1st:
member 0: 1st of month M realization 1
member 1: 1st of month M realization 2
member 2: last day of month M-1 realization 1
member 3: last day of month M-1 realization 2
............
last member: 2nd of month M-1 realization 2
So the numbering of monthly products members starts with the most recent initialization date and goes back in time.
However, there are (rare) exceptions when for Met Office forecast we don't have the "usual" 2 realizations per day. This is explained in footnote (c) on page https://confluence.ecmwf.int/display/COPSRV/Seasonal+forecasts+and+the+Copernicus+Climate+Change+Service#SeasonalforecastsandtheCopernicusClimateChangeService-production_schedules
(c) Due to the flexibility of the Met Office forecasting system, forecast failures on a given date are not usually recovered by re-running the missed forecasts at a later date, but by running more members with initial conditions of the day of recovery.
Example: An incident affected the 22 August 2017 forecast so no members are available for that date. Instead, 4 members were initialised on 23 August 2017.
Also, please note that only the first most recent 50 members of Met Office forecasts are used to generate the graphical products for Met Office or C3S multi-model combination displayed on C3S web, as explained in footnote (b) on the same page:
(b) The production schedule of the Met Office forecasting system doesn't prescribe how to build an ensemble for a specific nominal start date. The following choices are currently in use for the C3S products based on GloSea5 data:
FORECASTS: the most recent 50 members starting on or before the 1st of the month. (NOTE: The data from all the daily members, and not just those 50 used to produce the C3S forecast products, is processed and made available through the CDS)
HINDCASTS: 28 members (7 starting on the 1st of the month and 7 on each of the 9th,17th and 25th of the previous month)
In the future, we will include new metadata in the monthly products that will provide details on the initialization of a specific ensemble member.
Thanks Simona, that is very helpful!
Kevin
Dear Simona,
I am going to download data of total precipitation from: "Seasonal forecast monthly statistics on single levels from 2017 to present"
I don't know exactly what are the differences between "Ensemble" and "Individual members" under the "Product type" menu.
I am going to download hindcast of each of six models for period 1993-2016. All leads are needed. Also the members are not important for me so I like to average over the members.
When I select "Ensemble mean", selecting the hindcast years disables. Please help me.
Dear Morteza,
The "Individual members" include statistics computed using data only from a particular ensemble member (we can have monthly mean, monthly minimum, monthly maximum or monthly standard deviation). This individual member statistics is computed for both hindcast and forecast start dates. For example, extracting ECMWF system 5 monthly mean will result in 51 values, one for each individual member.
The "Ensemble mean" is the average over the monthly means from all individual ensemble members. This is computed only for the forecast start dates.
The "Hindcast climate mean" is the average over the monthly means from all individual ensemble members (so it is an ensemble mean), but averaged also over all years in the hindcast (climate) period. That's why this is not available for each hindcast start date in part. To retrieve it, you have to associate it with a particular forecast start date. For example, retrieving ECMWF system 5 January 2020 start date will give a Hindcast climate mean which is an average over the monthly means from all individual ensemble members and over all January start dates in the hindcast period 1993-2016.
For most recent systems of the models, hindcast period is 1993-2016, with the only exception of Met Office January start date which has 1994-2016. But some of the past systems had hindcast period ending in 2014 or 2015. For more info you can see the page Summary of available data.
If Hindcast climate mean is not useful for you, and you want each start date individually, then I suggest you extract the Individual members monthly means and after you compute the ensemble mean for each of the six models.
I hope this helps,
Simona
