Production suite overview

Operational runs

Table of contents

1. Structural Design Constraints

2. Structure of Production System

3. Interdependencies of the Operational Runs

   • 3.1 The Objective Analysis
   • 3.2 The Global Data Assimilation Cycle
   • 3.3 The Regional Analysis Spin-up
   • 3.4 Surface Analysis Dependencies

4. Operational Runs Description

   • 4.1 G3 [00,06,12,18] Early Surface Analyses Run
   • 4.2 R1 [00,12] Regional Model Forecast Run
   • 4.3 G5 [00,06,12,18] Complete Surface Analysis Run
   • 4.4.G1 [00,12] Global Forecast Run
   • 4.5 G6 [00,06,12,18] Final Global Surface Run
   • 4.6 R6 [00,06,12,18] Regional Surface Analyses Runs
   • 4.7 R2 [00,06,12,18] Regional Data Assimilation System Runs
   • 4.8 R3 [00,12] Regional Final Analyses Runs
   • 4.9 R1 [06,18] Regional Early Upper Air Analysis Run
   • 4.10 G1 [06,18] Global Early Upper Air Analysis Run
   • 4.11 RW06 and RE06 HIMAP high resolution Run
   • 4.12 E2 [00,06,12,18] and E100 Ensemble prediction runs
   • 4.13 GV [00,06,12,18] Global Verification Run
   • 4.14 EOW/EOM: End-of-week and End-of-month Runs
   • 4.15 AJ: Automatic Job Initiation Run
   • 4.16 Parallel Runs

5. Table 1 CMC model and upper air objective analysis run schedule

6. Figure 1 Data Assimilation System

CMC's fully automated production system ingests large amounts of observational data, analyzes them, makes a forecast of the future state of the atmosphere and disseminates the meteorological information to users within and outside the Meteorological Service of Canada (MSC) through a variety of means. The automated production system consists of a series of operational runs, which are sequences of computer jobs executed in batch on the various CMC computers.

The main runs are R100 and R112 in which the Regional model is executed, and G100 and G112 in which the Global model is executed.

Presently, the forecast models and objective analyses execute on the backend supercomputer (IBM pSeries). The frontend computer (SGI ORIGIN 3000) is used for observation decoding and for the production of charts, GRIB and bulletins.

1 Structural Design Constraints

The structure of the CMC production system and the operational runs is dependent on a number of factors. Much of it has evolved from historical requirements, which are still in force today.

The first major constraint is the operational timeliness of the production system's products. Certain numerical weather products are required with short lead times and with a minimum of two issues per day. This necessitates a forecast cycle that is repeated twice daily with an early data cut-off time.

Secondly, the geographical area of coverage is an important consideration in design. There is a natural association between the range of forecast and the area of coverage. Medium or long range forecasts require a global coverage in data assimilation to satisfy the forecast time scales involved. Short-range forecasts requirements can be satisfied through a more limited area coverage.

Thirdly, in an operational run there is a set of basic processes which must run sequentially (a forecast runs after an objective analysis). There is also the inter-dependency of the different operational runs, for example, one run may provide the trial field for the next run's objective analysis and so on. This linking of various operational runs becomes a vital consideration in the overall design.

2 Structure of Production System

The production system is comprised of a series of operational runs. They can be classified into four categories: regional surface, regional upper air runs, global surface, and global upper air runs.

There is a basic 24-hour cycle in the runs when all runs and products repeat themselves. For all practical purposes there is also a 12-hour cycle in which the runs basically repeat themselves. There are, however, significant differences in the global runs at night based on 00 UTC data and the run during the day based on 12 UTC data.

Although the contents of each run differ they share a similar set of fundamental processes. They are data extraction, data assimilation, forecasting the future state of the atmosphere (for the main runs) and a host of post-processing activities. These post-processing activities include complex graphical calculations, statistical and diagnostic forecasts, bulletin preparation, verifications, archiving of observations, analyses and forecasts. Except for the models most of this post-processing work is done on the frontend computer. Model output files are downloaded in data bases on the frontend and then post-processing activities using these files are initiated.

The general run naming convention is:

[R,G][code][HH]

where HH is the origin UTC time

R = Regional; G = Global

code: 1 model run (usually, exception is the HIMAP high resolution run « E » for East and « W » for West)

2 assimilation system

5 surface analyses

6 final surface analysis

As will be seen in more details in the following sections the major upper air runs are:

• R1 [00,12] the Regional (GEM : Global Environmental Multi-Scale) model runs
• G1 [00,12] the Global (GEM : uniform grid version) model runs
• R2 [00,12] start of the spin-up of regional assimilation system
• R2 [06,18] regional assimilation system
• G2 [00,06,12,18] the global assimilation cycle

Other « special » runs include :

• RE06 and RW06 : HIMAP high resolution regional runs (Eastern and Western windows)
• E1 and E2 [00,06,12,18] : ensemble prediction forecast system
• M100 : monthly and seasonal forecasts
• c100: CHRONOS model forecasts (air quality model)

Of lesser importance are the R1 [06,18] which contain analyses based on 06 and 18 UTC data with an early data cut-off time and the R3 [00,12] which contain final regional upper air analyses. In addition there is an early global analysis in G1 [06,18].

The surface runs are:

• G3 [00,06,12,18] which are the early global surface runs;
• G5 [00,06,12,18] which are the complete global surface analysis runs;
• G6 [00,06,12,18] which are the final surface analysis runs and are used for the assimilation cycle.

Table 1 gives the present configuration of the runs and their respective data cut-off times. The CMC operational meteorologist on duty must also decide whether sufficient data has been received to effectively begin the run. On occasions when certain difficulties arise (hardware, software, or telecommunications problems) the runs may be delayed by the CMC meteorologist. The operational production system works on a tight schedule and delays in the primary runs will cause delays in products in that run and even delays in subsequent runs. In addition, consideration must be given to the interdependencies of the runs.

3 Interdependencies of the Operational Runs

3.1 The Objective Analysis

The upper air objective analysis uses as input the 6-hour global or regional model forecast on pressure levels as well as meteorological observations contained in what is called a derivate file; the latter is a snapshot of observations valid in a window comprised between T-3 hours and T+3 hours, where T is the analysis time. The residual (i.e. the differences between the observation and the 6-hour forecast field interpolated at the position of the station) is calculated at each point of observation. An analysis of these residuals is performed on pressure coordinates using 3-dimensional variational analysis (3D-VAR). The analysis on pressure levels is then obtained by combining the analyzed residuals and the trial field. These residuals are then interpolated from pressure levels to the model's eta levels and combined with the eta trial field to obtain the analysis on eta coordinates.

When too large differences exist between an observed value and the trial field value for the same point, then the observed data will be rejected. The rejected data flags and the observations as "seen" by the analysis are written into another file and saved as a separate file called a "post-derivate". Both the derivate and post-derivate files are part of the "ADE" data base.

3.2 The Global Data Assimilation Cycle

The global data assimilation cycle (see Figure 1) comprises an uninterrupted series of final global analyses at main synoptic hours (with a data cut-off time of T+6 hours for the 06 and 18 UTC analyses, T+9 hours for the 00, 12 UTC analyses), each of which is followed by a 6-hour global model forecast that feeds into the next final analysis.

The eta levels of the global model used for the 6-hour forecast are the same as those of the analysis. This has the advantage that in data sparse areas the analysis will preserve many of the features of the forecast; data at T-6 will then have an effect at time T because of the trial field. Since there are many eta levels in the vertical, the vertical structures will be preserved - especially those of the boundary layer.

To help understand the concepts explained above, let us assume a data assimilation cycle beginning with the 06 UTC analysis (see Figure 1). The following is a brief outline of the steps.

a) Near 12 UTC (the data cut-off time of T+6 hours), a final derivate file is created with data valid from 03 to 09 UTC. Together with the trial field valid for 06 UTC, this file is fed to a global objective analysis program which produces a final global analysis (valid for 06 UTC) of meteorological variables such as virtual temperature, specific humidity, and the u and v component of the wind on eta levels, as well as temperature, dew point depression, and u and v components of the winds on pressure surfaces. The eta level analysis is fed into the global forecast model which produces a global trial field valid at 12 UTC.

b) Near 21 UTC (the data cut-off time of T+9 hours) a final derivate file is created with data valid from 09 to 15 UTC. With this derivate file and the trial field valid for 12 UTC, a final global analysis valid for 12 UTC is made. With the final analysis on eta levels, a 6-hour forecast (valid 18 UTC) of the global model is made.

c) Near 00 UTC (the data cut-off time of T+6 hours), a final derivate file is created with data valid from 15 to 21 UTC. With the derivate file and the trial field valid for 18 UTC, a final global analysis valid for 18 UTC is made. With the final analysis on eta levels, a 6-hour forecast (valid 00 UTC) of the global model is made.

d) Near 09 UTC (the data cut-off time of T+9) a final derivate file is created with data valid from 21 to 03 UTC. With the derivate file and the trial field valid for 00 UTC, a final global analysis valid for 00 UTC is made. With the final analysis on eta levels, a 6-hour forecast (valid for 06 UTC) of the global model is made.

This is the last step of this 24-hour cycle; the cycle repeats itself afterwards starting with step a).

3.3 The Regional Analysis Spin-up

The regional analysis spin-up is connected to the global data assimilation cycle once every 12 hours (see Figure 1).

A completely decoupled regional analysis cycle could have been designed. The reason for the coupling is that the regional analysis will benefit from the features in the global system in areas where the resolution of the variable grid of the regional model is much coarser than the corresponding resolution of the global model (over the Western Pacific for example).

The reason for the spin-up and the analysis on the same eta levels and on the same variable grid as the ones used by the regional model is to provide an analysis consistent with the regional model's levels and grid, therefore eliminating the shock resulting from the interpolation process (vertical and horizontal) from one grid to the other. Both horizontal and vertical features from the trial fields will be better carried in the analysis; the structure of the boundary layer will be better preserved, and the regional analysis over North America and adjacent waters will be of higher resolution than the one carried in the present global analysis.

Once again, to help understand the concept explained above, we will look at how the 00 UTC regional analysis spin-up is performed; the 00 UTC regional analysis spin up is defined as the series of analyses and trial fields on the regional model grid that are required to produce the 00 UTC regional analysis that will be fed to the CMC regional model for the production of forecasts out to 48 hours.

a) the spin-up begins by interpolating on the regional model eta vertical levels a global model trial field in eta co-ordinate valid for 12 UTC. This trial field and the appropriate derivate file (i.e. observations) are combined to produce a final 12 UTC analysis on the regional grid, which is then fed to the regional model to produce the 6-hour trial field valid for 18 UTC.

b) This trial field and the appropriate derivate file are then used to produce a final regional analysis valid at 18 UTC, which is then fed to the regional model to produce the 6-hour trial field valid for 00 UTC.

c) This trial field and the appropriate derivate file (this time with an early data cut-off time) are then used to produce an early analysis valid for 00 UTC, in time to be used as input to the regional model's 48 hour integration. This run is the R100 run.

This then completes the 00 UTC regional analysis spin-up. An equivalent scenario is used for the 12 UTC spin-up as well.

3.4 Surface Analysis Dependencies

In every run where a forecast model is run (global or regional, trial or 'model' run), the latest surface based fields are added to the upper air analysis and offered to the model.

To the upper air analysis are added the latest surface temperature, deep soil temperature, albedo, snow, ice and sea surface temperature analysis. In the case of the regional analysis, a soil moisture analysis is also added.

All these fields are used by the models. If for any reason they have not been updated recently, the models will still run, using the last surface analysis available

4 Operational Runs Description

4.1 G3 [00,06,12,18] Early Surface Analyses Run

This is the first run of the forecast cycle after the main synoptic times. It has a data cut-off time of about T+45 minutes. Its purpose is to produce an early analysis of mean sea level pressure and surface temperature using the 6-hour trial fields from the global cycle; the surface temperature analysis produced in this run is offered to the regional model that will execute shortly after (R1 run).

Some processing activities (on the frontend computer) are also part of this run. They include the production of objectively analyzed surface charts that are enhanced by adding fronts and that can be modified subjectively by the CMC meteorologist to provide a more accurate representation of surface lows, isobars, etc. Plotted charts of surface observations are also transmitted from this run.

4.2 R1 [00,12] Regional Model Forecast Run

This run is the main run where the Regional model is integrated to 48 hours. Its main purpose is to provide high resolution short range forecast products over Canada and adjacent oceans.

It has a rather early data cut-off time T+1:35, which allows most of North American and much off-shore data to be received and processed in time for the analysis fed to this model. Data coverage elsewhere in the Northern hemisphere is usually relatively low.

By finding deficiencies with the trial fields generated in both the regional spin-up and in the global assimilation cycle, the CMC meteorologist can insert corrective bogus data whenever necessary into the ADE data base so that it may be used in the regional OA, thereby providing a better analysis.

The model is integrated out to 48 hours with outputs at 3-hour intervals being processed and saved on the IBM supercomputer and downloaded to the frontend computer. Forecast fields on both pressure and eta levels are available.

The principal post-processing activities (all on the frontend computer) included in this run are:

• production of many graphic charts. These include the standard 4-panel charts together with specialized charts such as the summer, winter and aviation packages. These are transmitted to users via satellite broadcast or through the ECONET as well as on the CMC WEB page ;
• bulletins prepared in a multitude of formats including statistical forecasts, direct model output and diagnostic techniques;
• SCRIBE matrices;
• perfect prog statistical forecasts based on model outputs;
• GRIB and BUFR production: a variety of gridded forecasts in GRIB format and spot forecats in BUFR format are sent to the regional weather centers as well as to other users and clients ;
• the Atlantic wave model and wave forecast charts based on the Regional forecast winds;
• the statistical air quality forecast package as well as the chemical tracer model (CTM);
• some specialized graphical output for CMC operational workstations;
• amendments of the current upper air winds and temperature forecasts (FD's), production of the next valid set of FD's, and also of backup FD's (to be used later in case of computer failures in subsequent runs);
• file management of the grid point data base;

4.3 G5 [00,06,12,18] Complete Surface Analysis Run

This run is a global one as in G3 but a later data cut-off time (T+2:45) is used, thereby allowing much of the European and Asian synoptic observations to be received and processed in time for the analyses produced in this run.

Again as in G3 run the CMC operational meteorologist adds fronts and can, if necessary, subjectively modify the surface analyses produced in this run.

The surface temperature analysis produced in this run will be offered to the global model that will execute shortly after in G1 run.

4.4. G1 [00,12] Global Forecast Run

This run is a main run in which the GEM model in its global configuration is integrated using a global analysis with a data cut-off time of T+3:00. At this time the data coverage available for this run is much more complete than it is for the regional run.

The main purpose of the run is to provide medium range forecast products based on 00 UTC data; of course, it also provides global forecasts for the short range.

The global model is integrated to 72 hours based on 12 UTC data, and to 240 hours based on 00 UTC data.

Once a week, on early Sunday morning the model is run to 15 days.

The forecasts are saved at 3-hour intervals on both pressure and eta levels to 144 hours and then every 6 hours thereafter.

As in R1 runs, fairly intensive post-processing activities (on the frontend computer) are performed in this run:

• statistical forecasts based on various statistical techniques;
• many bulletins based on direct model output and/or statistical forecasts;
• production of many graphic analysis and forecast charts. These include the standard 4-panel charts together with specialized charts such as the summer, winter and aviation packages. These are transmitted to users via satellite broadcast or through the ECONET as well as on the CMC WEB page;
• the Pacific wave model and wave forecast charts based on global model forecast winds;
• forecasts related to ozone and UVB;
• SCRIBE matrices;
• SCRIBE international;
• GRIB and BUFR production: a variety of gridded forecasts in GRIB format and spot forecats in BUFR format are sent to the regional weather centers as well as to other users and clients;
• Special 6-15 day forecast products;
• specialized charts for the Department of National Defense, upon request;
• specialized graphical outputs for the CMC operational workstations;
• file management of the grid point database;

4.5 G6 [00,06,12,18] Final Global Surface Run

The late data cut-off time (T+5:15, T+8:15 for G600, G612) associated with this run allows the production of global surface analysis with a fairly complete global data coverage. As well, final surface analyses with fronts and other added features are produced by the CMC operational meteorologist from outputs of this run.

In summary, here are the fields analyzed in the G6 run:

• At 00 UTC each day, a global ice cover analysis is produced, based mostly on SSMI data.
• At 00 UTC a global sea surface temperature analysis is produced; this analysis is based on the previous analysis and on sea surface temperature observations provided by ships, SATEM's, etc. A mean monthly sea surface temperature is also calculated.
• At all main synoptic times (00, 06, 12 and 18 UTC) the following global surface analysis are produced : surface temperature, snow depth, mean sea level pressure and surface dew-point temperature analysis. The surface temperature is used to calculate a deep soil temperature. The snow depth analysis is used in conjunction with the ice analysis, the vegetation field and the climatological albedo field to produce the albedo analysis. Also, the snow depth analysis produces other derived fields: the snow density, the snow age (time since last snowfall) and the snow cover which simply represents snow covered (snow depth greater than 1 cm) vs snow free surfaces.

The surface analyses are offered to the models by simply appending them to the upper air 3-dimensional analysis.

4.6 R6 [00,06,12,18] Regional Surface Analyses Runs

Currently only a regional soil moisture analysis is made at 18 UTC based on error feedback approach of the 6-h dew-point temperature forecast error.

4.7 R2 [00,06,12,18] Regional Data Assimilation System Runs

As explained in section 3, these runs constitute the regional assimilation system. The spin-up begins with R2 [00,12] which interpolates the global model 6-hour forecast to the eta levels of the regional model. Then an analysis and 6-hour forecast derived from the model are produced.

This 6-hour forecast is then used by the next part of the spin-up in R2 [06,18] which then produces an analysis, followed by a 6-hour forecast. This 6-hour forecast is subsequently used as the trial field in R1 [00,12] as well as in R3 [00,12] runs.

The data cut-off time used for these runs is about T+5:30.

Products from these runs include analysed charts as well as the charts of the trial field.

4.8 R3 [00,12] Regional Final Analyses Runs

The trial fields produced by the R2 [06,18] runs are used to produce "final" regional analyses with a data cut-off time of T+6:30. These analyses are of better quality than those of R1 [00,12] runs because of the slightly later data cut-off time.

4.9 R1 [06,18] Regional Early Upper Air Analysis Run

This is an early regional upper air run with a data cut-off time of T+1:40 as in R1 [00,12] except that no forecast model is run. The primary purpose of this run is to provide an early view of the analysis at both 06 and 18 UTC and also to generate the amendments for the FDCN2 bulletins, if required.

4.10 G1 [06,18] Global Early Upper Air Analysis Run

This is an early global upper analysis run with a data cut-off time of T+2:00 to provide the CMC meteorologists with an early view of the upper air analysis at 06 and 18 UTC.

4.11 RW06 and RE06 HIMAP high resolution Run

A higher resolution version of the regional GEM model is ran over 2 specific windows : the Western (RW) and Eastern (RE) window. These runs use the 6-h forecast of the R100 run as analysis. Gridded data outputs in GRIB format of the HIMAP runs are sent to the regional centers. As well, an extended set of images are posted on the CMC internal WEB page.

4.12 E2 [00,06,12,18] and E100 Ensemble prediction runs

These runs operate the ensemble prediction system which contains 8 members (i.e. 8 series of perturbed assimilation runs and 8 series of 10 day forecasts).

4.13 GV [00,06,12,18] Global Verification Run

These are runs that perform various verifications against the global analyses as well as against the radiosonde observations (WMO verifications).

4.14 EOW/EOM: End-of-week and End-of-month Runs

These are runs that execute once at the end of each week or month; they perform clean-up activities of various data files. They also provide summaries of scores, statistics, etc.

4.15 AJ: Automatic Job Initiation Run

These are jobs that are initiated at a specified time (as per the computer system clock) and are not included in any of the above runs. These runs produce the hourly weather summaries, the twice daily urban forecast bulletins, international temperature bulletins and other various bulletins (most of them being WBCN bulletins). Satellite data is usually processed through an AJ job. Several file clean-up jobs are also included in these runs.

4.16 Parallel Runs

Using a naming convention similar to the operational runs, many parallel runs are executed to test new analyses and/or forecasts systems. The data produced in these parallel runs is usually written in parallel directories and the parallel jobs run in non-operational computer classes as to not interfere with the operational runs.

Table 1

Note: There are also runs (not described here) that perform surface objective analyses and update geophysical fields; these are runs G3, G4, G5, G6 et R6.

Figure 1 Data Assimilation System

This figure describes the mechanics of the global and regional data assimilation system.

[View] (D)

The Green Lane^TM,
Environment Canada's World Wide Web Site.