On this page we provide an overview of sub-seasonal forecast model data downloaded from the S2S Database. The data is downloaded at a 1.5 degree resolution and is freely-available from the ECMWF-hosted sub-seasonal forecast data portal. Throughout the AI Weather Quest, this data will be used to benchmark AI-based sub-seasonal forecasting systems. The following page describes full details regarding the latest model versions including frequency of forecasts/reforecasts and ensemble sizes. 

Dynamical forecast/reforecast configuration for AI Weather Quest competition-phase forecast visualisation and evaluation [IN DEVELOPMENT!]:

WMO Lead Centre Name

ECMWF

Montreal

Washington

Seoul

Tokyo

Beijing

Moscow

Teamname: Modelname on forecast portal

Dynamical_S2SDatabase: ECMWF

Dynamical_S2SDatabase: ECCCDynamical_S2SDatabase: NOAADynamical_S2SDatabase: KMADynamical_S2SDatabase: JMADynamical_S2SDatabase: CMADynamical_S2SDatabase: HMCR
Producing/Contributing CentreECMWFEnvironment and Climate Change CanadaNational Oceanic and Atmospheric Administration (NOAA)

Korea Meteorological Administration (KMA)

Japan Meteorological Agency (JMA)

China Meteorological Administration (CMA)

Hydrometeorological Centre of Russia
MARS catalogue nameecmfcwaokwbc

rksl

rjtd

babj

rums
Webpage describing model descriptionECMWF model descriptionECCC model descriptionNCEP model description

KMA model description

JMA model description

CMA model description

HMCR model description
Forecast selection
Forecast initialisation date selectionOnly forecasts from the initialisation date.Only forecasts from the initialisation date.Forecasts from current day and -1 day (i.e. Thursday and Wednesday start date).Forecasts from current day, -1 day and -2 days (i.e. Thursday, Wednesday and Tuesday start dates).Forecasts from current day, -1 day & -2 day (i.e. Thursday, Wednesday and Tuesday start dates).Only forecasts from the initialisation date.Only forecasts from the initialisation date.
Number of ensemble members (one control plus perturbed).1012116 per initialisation date.8 per initialisation date.5 per initialisation date.441
Reforecast selection
Hindcast date selection< 20241112: select initialisations -3 and 0 days [pre Monday and Thursday]. >= 20241112: if odd day use -2 and 0 days, if an even day use -1 and 1 days [as reforecasts are performed every odd day of month except Feb 29th].-3 and 0 days from initialisation date (i.e. previous Monday and the current Thursday)-4, -2, 0, 2 and 4 days from initialisation date.Select nearest hindcast date before and after initialisation date (i.e. two dates chosen).Select nearest hindcast date before and after initialisation date (i.e. two dates chosen).-3 and 0 days from initialisation date (i.e. previous Monday and the current Thursday)Single initialisation date
Hindcast years used.Last 20 years2001 - 2020 [20 years]1999 - 2010 [12 years]1993 - 2016 [24 years]1991 - 2020 [30 years]Last 15 years1991 - 2020 [30 years]
Number of hindcast members per initialisation (full number used to compute climatology).11 (x 20 years x 2 initialisations = 440)4 (x 20 years x 2 initialisations = 160)4 (x 12 years x 5 initialisations = 240)7 (x 24 years x 2 initialisations = 336)5 (x 30 years x 2 initialisations = 300)4 (x 15 years x 2 initialisations = 120)11 (x 30 years = 330)

Dynamical forecast/reforecast configuration for science blog analysis:

WMO Lead Centre Name

ECMWF

Montreal

Moscow

Tokyo

Washington

Seoul

Producing/Contributing CentreECMWFEnvironment and Climate Change CanadaHydrometeorological Centre of Russia

Japan Meteorological Agency

National Oceanic and Atmospheric Administration (NOAA)

Korea Meteorological Administration

MARS catalogue nameecmfcwaorums

rjtd

kwbc

rksl

Webpage describing model descriptionECMWF model descriptionECCC model descriptionHMCR model description

JMA model description

NCEP model description

KMA model description

Legend label for ECMWF science blogECMWF IFS CY49R1/CY48R1ECCC GEPS8HMCR EK40/EB40JMA CPS3NCEP CFSv2KMA GloSea6-GC3.2
Details regarding AI Weather Quest-led analysis
Forecast initialisation date selectionOnly forecasts from the initialisation date.Only forecasts from the initialisation date.Only forecasts from the initialisation date.Forecasts from current day, -1 day & -2 day (i.e. Thursday, Wednesday and Tuesday start dates).Only forecasts from the initialisation date.Forecasts from current day + -1 day (i.e. Thursday and Wednesday start dates).
Number of ensemble members (one control plus perturbed).10121415 (x3 = 15)167 (x2 = 14)
Hindcast date selection< 20241112: select initialisations -3 and 0 days [pre Monday and Thursday]. >= 20241112: if odd day use -2 and 0 days, if even use -1 and 1 days [as reforecasts are run every two days].-3, 0 and 4 days from initialisation date (Monday, Thursday, Monday)Single initialisation date (day 0)Select nearest hindcast date before and after initialisation date (i.e. two dates chosen).-4, -2, 0, 2 and 4 days from initialisation date.Select nearest hindcast date before and after initialisation date (i.e. two dates chosen).
Hindcast years used.2006-2016 [11 years]2006-2016 [11 years]

2006-2016 [11 years]

except < 16/10/2024 where 10 years 2006-2015


2006-2016 [11 years]2006-2010 [5 years]  2006-2016 [11 years]
Number of hindcast members per initialisation (full number used to compute climatology).11 (x11x2=242)4 (x11x3=132)

11 (x11=121)

110 members < 16/10/2024

5 (x2x11=110)4 (x5x5=100)7 (x11x2=154)


The AI Weather Quest would like to thank all those responsible for maintaining the S2S database. It's no easy task!

  • No labels