TOC > MSC data > CanSIPS > CanSIPS data in GRIB2 on MSC Datamart
Canadian Seasonal to Inter-annual Prediction System (CanSIPS) Data in GRIB2 Format
The Canadian Seasonal to Inter-annual Prediction System (CanSIPS) is a long-term prediction system whose objective is to forecast the evolution of global climate conditions. CanSIPS is a multi-model ensemble (MME) system using two atmosphere-ocean-land coupled models (CanESM5 and GEM5.2-NEMO) developed by the Canadian Centre for Climate Modelling and Analysis (CCCma), Meteorological Research Division (MRD) and the Canadian Centre for Meteorological and Environmental Prediction (CCMEP). It is a fully coupled atmosphere-ocean-ice-land prediction system relying on the operational data assimilation infrastructure for the initial state of the atmosphere, sea surface temperature and sea ice. For further technical information about CanSIPS please refer to the technical note.
Principal components of CanSIPS
-
Initialization procedure:
- CanESM5: Initial atmospheric and oceanic conditions are obtained by running a coupled atmosphere-ocean-sec ice assimilation process constrained by CMC (GDPS) atmospheric analysis and GIOPS oceanic analysis through the same nudging procedure. Modelled sea ice thickness in the Northern Hemisphere is constrained near values from the SMv3 statistical model. Modelled sea ice concentration and sea surface temperature are constrained near values from GDPS G6. Land variables are initialized through the response of CLASS/CTEM to the assimilating atmosphere component.
- GEM5.2-NEMO: The atmospheric initial conditions of GEM5.2-NEMO come from those of the Global Ensemble Prediction System (GEPS).The ocean and sea ice initial conditions in the forecast come from the CMC GIOPS analysis whereas land surface initial conditions is from CMC (GDPS) analysis after adjusting it to the hindcast climatology.
-
Forecast mode: CanSIPS official forecasts are based on a 20-member ensemble of forecasts produced with each of the two models for a total ensemble size of 40. The 20 members, 12 month forecasts from each model are produced by combining the forecasts produced on the last day of each month (ensemble members 1-10) with forecasts issued four-days prior to that (ensemble members 11-20).
-
Hindcast mode: CanSIPS climatology is based on a series of retrograde forecasts (e. g. historical forecasts) covering the period 1990 to 2020. This climatology is very useful for interpreting realistic forecasts because real-time forecast anomalies are generated instead of raw forecasts.
How is the CanSIPS forecast configured ?
In CanSIPSv3, the ensemble size of the forecast has been increased from 20 members to 40 members (20 GEM-NEMO members + 20 CanESM5 members). This is achieved by combining the forecasts initialized on the last day of the month with the forecasts initialized four days prior to that. For each model, a 12-month forecast is produced with the first 10 members of each model obtained from the last day of the month and the second set of 10 members from the forecasts initialized four days prior to that. When the ensemble forecasts are completed, we construct seasonal mean anomalies with respect to the 30-year hindcasts for each ensemble member. Subsequently we implement deterministic (ensemble mean) and probabilistic (different categories with respect to the ensemble size) approaches to forecast the upcoming seasons.
Data location
MSC Datamart data can be automatically retrieved with the Advanced Message Queuing Protocol (AMQP) as soon as they become available. An overview and examples to access and use the Meteorological Service of Canada's open data is also available.
The data is available using the HTTPS protocol and resides in a directory that is plainly accessible to a web browser. Visiting that directory with an interactive browser will yield a raw listing of links, each link being a downloadable GRIB2 file.
The data can be accessed at the following URLs :
- https://dd.weather.gc.ca/ensemble/cansips/grib2/forecast/raw/{YYYY}/{MM}/ (forecast members and products at 2.5 degrees)
- https://dd.weather.gc.ca/model_cansips/100km/forecast/{YYYY}/{MM}/ (forecast members and products at 1 degree)
- https://dd.weather.gc.ca/model_cansips/100km/hindcast/{YYYY}/{MM}/ (hindcast)
where :
- forecast : Constant string indicating that the file contains the data from the forecast part of CanSIPS, in opposition to the hindcast part.
- hindcast : Constant string indicating that the file contains the data from the hindcast part of CanSIPS, in opposition to the forecast part.
- 100km : Indicate a 1 degree resolution (around 100km)
- MM : Month of the forecast start [01, 02, 03, ..., 12]
- YYYY : Year of the forecast start [Ex: 2012, 2013, ...]
Technical specification of the grid
Tables list the values of various parameters of the CanSIPS lat-lon grid, according to the resolution.
Data at 2.5x2.5 degrees resolution
| Parameter | Value |
|---|---|
| ni | 145 |
| nj | 73 |
| resolution | 2.5° |
| coordinates of the first grid point | 90° S 0° E |
Data at 1.0x1.0 degree resolution
| Parameter | Value |
|---|---|
| ni | 360 |
| nj | 180 |
| resolution | 1.0° |
| coordinates of the first grid point | 89.5° S 0.5° E |
File name nomenclature
The files have the following nomenclature:
-
Members and products at 2.5 degrees:
- Individual members:
cansips_forecast_raw_latlon2.5x2.5_{VAR}_{LVLTYPE}_{LVL}_{YYYY}-{MM}_allmembers.grib2 - Probability products:
cansips_forecast_prob-{StatProcess}_latlon2.5x2.5_{VAR}_{LVLTYPE}_{LVL}_{YYYY}-{MM}_{PPP}.grib2
- Individual members:
-
Members and products at 1 degree:
- Individual members:
{YYYYMM}_MSC_CanSIPS_{Var}_{Level}_LatLon1.0_P{Month}M.grib2 - Probability products:
{YYYYMM}_MSC_CanSIPS_{Var}-{StatProcess}_{Level}_LatLon1.0_{Month}.grib2
- Individual members:
-
Hindcast:
{YYYYMM}_MSC_CanSIPS-Hindcast_{Var}_{Level}_LatLon1.0_P{Month}M.grib2
where :
- cansips|CanSIPS : Constant string indicating that the data is from the CanSIPS system
- MSC : Constant string indicating that the data is from the Meteorologcal Service of Canada (MSC)
- forecast : Constant string indicating that the file contains the data from the forecast part of CanSIPS, in opposition to the hindcast part
- CanSIPS-Hindcast : Constant string indicating that the file contains the data from the hindcast part of CanSIPS, in opposition to the forecast part
- raw : Constant string indicating that the file contains raw data without bias correction
- VAR : Variables included in the 2 degrees files [TMP, HGT, PRATE, SSHG, PRMSL, UGRD, VGRD]
- Var : Variables included in the 1 degree files [AirTemp, GeopotentialHeight, PrecipRate, SeaSfcHeight-Geoid, Pressure, WindU, WindV]
- StatProcess : Statistic process [prob-near-normal, prob-below-normal, prob-above-normal, ProbNearNormal, ProbBelowNormal, ProbAboveNormal]
- LVLTYPE : Vertical level type [SFC for the surface, TGL for height above the ground, ISBL for pressure level, MSL for mean sea level]
- LVL : Vertical level value
- Level : Vertical level [Sfc for the surface, AGL-2m for 2m above ground level]
- MM : Month of the forecast start [01, 02, 03, ..., 12]
- YYYY : Year of the forecast start
- allmembers : Constant string indicating that all members [01, 02, 03, ..., 20] are grouped into the file
- PPP : Forecast product time length ex: P3M is for a product with forecast a period of 3 months
- Month : Months concerned from the beginning of the forecast, either 1 month [ex: P07M for December if the beginning of the forecast is May] or several months starting from the beginning of the forecast [ex: P01M-P03M for June to August if the beginning of the forecast is May]
- grib2 : Constant string indicating the GRIB2 format is used
Examples :
- cansips_forecast_raw_latlon2.5x2.5_HGT_ISBL_0500_2012-10_allmembers.grib2
- cansips_forecast_prob-below-normal_latlon2.5x2.5_TMP_TGL_2m_P3M_2018-12.grib2
- 202309_MSC_CanSIPS_AirTemp_AGL-2m_LatLon1.0_P00M.grib2
- 202305_MSC_CanSIPS_AirTemp-ProbBelowNormal_AGL-2m_LatLon1.0_P06M-P09M.grib2
- 202010_MSC_CanSIPS-Hindcast_WaterTemp_Sfc_LatLon1.0_P10M.grib2
Internal Structure of the Files
The internal structure of the forecast and hindcast files is the following :
Each file contains 40 records (corresponding to the 40 ensemble members) and the first 20 ensemble members belong to GEM5.2-NEMO whereas the last 20 ensemble members are from CanESM5. Please, find below the detailed classification of ensemble members with respect to model and start date.
- ensemble 1-10 : from GEM5.2-NEMO with forecasts initialized on the last day of the month
- ensemble 11-20: from GEM5.2-NEMO with forecasts initialized four days prior to the last day of the month
- ensemble 21-30: from CanESM5 with forecasts initialized on the last day of the month
- ensemble 31-40: from CanESM5 with forecasts initialized four days prior to the last day of the month
List of variables
Individual members:
Ensemble products:
The files contain probability products by member counting above or below different thresholds, percentiles, means, medians and standard deviations. Note that products and thresholds are defined in the file metadata. The following products are available for each of the variables listed below.
APCP
- Probability above 0 kg/(m2*s)
- Minimum (0 percentile), maximum (100th percentile), standard deviation and mean
TMP
- Probability above 0 K
- Minimum (0 percentile), maximum (100th percentile), standard deviation and mean
Tips for Computing Anomaly Forecasts
It is recommended to use anomaly forecasts instead of the raw forecasts. Anomaly forecasts can be obtained by subtracting from the forecast a climatology based on the hindcasts. The recipe is as follows for a given variable:
For a specific forecast month (e.g. 2016-02) an ensemble mean file needs to be created and we can call it ensm_for_02_2016 for the purpose of this example. This file now contains only 12 temporal records since the mean of 40 ensemble members is performed. The temporal record of this file starts in the month 02 of the year 2016 and stretches until the month 01 of the year 2017.
Subsequently, the same procedure is repeted for the hindcast files but separately for each of the hindcasts that start in 1981 and stretch until the year of 2010. Each year will have an ensemble mean file for the month 02, ensm_hin_02_YYYY, but for the particular hindcast year (YYYY stands for a particular hindcast year). By making the mean of all of the 30 ensm_hin_02_YYYY files, the climatology of the ensemble mean hindcast for the month of February is obtained, which can be called ensm_hinclim_02 in this example.
The subtraction of ensm_hinclim_02 and ensm_for_02_2016 allows for anomaly forecast production for the month 02 of the year 2016. Since this anomaly forecast now contains 12 temporal records, starting from the month of February 2016, we can say that the anomaly forecast for the month of February has zero lead time, for the month of March 2016 a one month lead time and finally for the month of January 2017 (last of the 12 record) an eleven month lead time.
Similar approach applies for the seasonal forecasts only with the seasonal means (e.g. DJF, JFM, FMA, or any other running season) being calculated before the anomaly forecast is computed.
Support
If you have any questions about this data, please contact us.
Announcements from the dd_info mailing list
Announcements related to this dataset are available in the dd_info list.