Canadian Meteorological Centre
2121 Trans-Canada Highway
Dorval, Québec H9P 1J3

Major modifications were incorporated into the global data assimilation system. The analyses are now performed on the GEM model eta levels, and new forecast error statistics were implemented in June 2000. Also, TOVS/ATOVS radiances from the NOAA-14 and NOAA-15 satellites, as well as wind data from ACARS/AMDAR were introduced in September 2000. The global model was also changed in June 2000 with modifications aimed at increasing the level of activity in the model.

The mesoscale short range model (HIMAP), running at 10-km resolution, had its convection parameterization scheme upgraded to the Kain-Fritsch scheme in July 2000.

The Updateable Model Output Statistics (UMOS) postprocessing system was connected to the regional model outputs in September 2000.

The following types of observations are presently used at the Centre. For these types, we use all observations that are available from the GTS, on the global scale. The numbers indicate typical amounts received during a 24-hour period :

- SYNOP/SHIP 20,300

- TEMP (500 hPa GZ) 1,150

- TEMP/PILOT (300 hPa UV) 1,250

- DRIFTER 2,300

- AIREP/ADS 3,700

- SATEM 5,500 *

- SATOB 15,000

- SATOB-SST 2,000

- SA/METAR 172,000

- AMDAR/ACARS 85,000

- PIREP 900 *

- PROFILER 635 *

- HUMSAT 12,000 **

- TOVS/ATOVS 95,000

- SSM/I 1,000,000 ***

- ERS 280,000 *

* Not assimilated yet, or no longer assimilated.

** Locally produced GOES moisture profiles.

*** A third of these are used for ice analyses.

b) Products

GRIB ECMF

GRIB KWBC

GRIB EGRR

FDCN KWBC

FDUS KWBC

U.S. Difax products

Significant weather forecasts

Winds/Temperature forecasts for various flight levels

Fully automated.

Various real-time quality control checks are performed for each observation received from the GTS. In particular :

- all reports are checked for gross errors;

- values for main items, such as height, pressure, temperature, dew point and wind are checked to be inside physical and climatological limits;

- temperature profile check;

- hydrostatic check;

- horizontal check (spatial consistency with neighbours and first guess fields)

These checks are done at, or after the decoding phase of the bulletins. Canadian observations are put on the GTS before such quality control is performed. However, Canadian observations are subject to quality control at the observing site, before transmission to the national centre.

The information generated by the quality control system inside the objective analysis is fed back into the observations database in order for non real-time monitoring and quality control activities to be performed. This monitoring is done on the global scale. Nationally, we also monitor the bursting altitude of upper-air soundings and results are distributed to data producers on a daily basis and monthly reports are distributed.

Each Canadian synoptic report (manned stations only) is also monitored in real time for completeness and timeliness. Requests to individual stations are made if certain criteria are met. Observing stations send corrections if time permits. These corrections are sent to the GTS for transmission. A monthly summary of errors is produced and distributed to data producers.

Monitoring the availability of observations on the global scale is an inherent portion of operations at the CMC. Information on the current content of the observational databases is available in real time, by observation types and by geographical areas. A chart showing the geographical distribution of observations, by types, used to initialize the numerical models is distributed to forecast centres across the country in real-time. A monthly report describing the availability of upper air observations is produced and distributed to data producers.

The information on the availability and quality of observations available for use in the final global analyses is assembled each month into the "CMC Global Data Monitoring Report". The statistics presented in the reports are prepared in accordance to the WMO/CBS approved procedures. The reports are sent to the WMO Secretariat as well as other major GDPS centres.

In 1993, CMC was designated by CBS as the lead centre for the monitoring of the quality of land surface observations in WMO RA-IV (North and Central America). In 1994, the CMC began to fulfil its role and since then has regularly produced its 6-monthly reports entitled " Report on the Quality of Land Surface Observations in Region IV". Two such reports were distributed in 2000. Monitoring results are distributed directly to national focal points for most countries within RA-IV.

7.1 System run schedule

The following table summarizes the operational runs at CMC. The core of the operational runs executes in batch on the NEC SX-4/32. Most of the postprocessing jobs, including CMC products, execute on the front end computer (SGI Origin 2000).

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

7.2.1 Data assimilation and objective analysis

7.2.3.1 Analysis

A series of classic analysis products are available in electronic or chart form ( i.e. surface analysis of snow and cover, sea surface temperature, surface MSLP and fronts, upper-air geopotential, winds and temperature at 1000, 850, 700, 500, 250 hPa, etc.).

7.2.3.2 Forecasts

A series of classic forecast products are available in electronic or chart form ( i.e. MSLP and 1000-500 hPa thickness, 500 hPa geopotential height and absolute vorticity, cumulative precipitation and vertical velocity, 700 hPa geopotential height and relative humidity). A wide range of bulletins containing spot forecasts for many locations are produced . As well, other specialized products such as precipitation and probability of precipitation forecasts, temperature and temperature anomaly forecasts, etc., are produced.

5. Ensemble Prediction System

The 16 member ensemble prediction system (EPS) runs once a day up to 10 days (Houtekamer et al., 1996; Lefaivre et al., 1997; Plante et al. 1999). Eight perturbed analyses are obtained by running independent assimilation cycles that use perturbed sets of observations and are driven by eight different versions of the spectral global model (SEF model T95, Ritchie, 1991). The number of perturbed analyses is doubled as follows: the mean of the analyses is subtracted to the operational analysis and a fraction of this difference is added to the original perturbed analyses. Every day, at 00 UTC, two separate models are used to produce the 10-day forecasts: the SEF model and the GEM model (resolution of 1.875°, Côté et al., 1998a and 1998b). Each model uses different versions of their physical parameterizations.

Ensemble outputs of the following products are available on the web (http://www.cmc.ec.gc.ca/cmc/htmls/forecasts.html, then select ensemble forecast system): spaghetti plots of the 500 hPa heights; composite MSLP highs and lows; cumulative precipitation amounts; forecast charts of precipitation amounts probability for various thresholds. Ensemble outputs are also used to feed the Perfect Prog statistical package to forecast daily maximum/minimum temperatures.

7.3.3.1 Analysis

A series of classic analysis products are available in electronic or chart form ( i.e. surface analysis of snow and cover, sea surface temperature, surface MSLP and fronts, upper-air geopotential, winds and temperature at 1000, 850, 700, 500, 250 hPa, etc.).

7.3.3.2 Forecasts

A wide variety of forecast products are available in electronic or chart form. These include the classic charts such as MSLP and 1000-500 hPa thickness, 500 hPa geopotential height and absolute vorticity, cumulative precipitation and vertical velocity, 700 hPa geopotential height and relative humidity. A myriad of special charts are produced in the context of the summer or winter severe weather (tropopause, stability indices, wind shear, helicity, wind chill, liquid water content, streamlines, low-level maximum wind, vertical motion, etc.) or in the specific support for aviation forecasting (icing, freezing level, height of cloud ceiling, momentum flux, turbulence, etc.). A wide range of bulletins containing spot forecasts are produced for many locations over North America.

 7.3.4 Operational techniques for application of NWP products

5. High resolution model for short range forecast (HIMAP)

A high resolution model is run once a day for 24 hours over two sub-areas of Canada: western Canada and upstream waters; Great Lakes and eastern Canada. This strategy was given the name of High Resolution Model Applications Project (HIMAP, Pellerin et al., 1998). The model used is the same unified model described in section 7.3.2, except for the following differences:

• the window of uniform grid has a resolution of .09 degrees (~10 km);
• the number of vertical levels is 35 with top of the model at 10 hPa;
• the convection scheme is the Kain-Fritsch scheme (Kain and Fritsch, 1990, Kain and Fritsch, 1993).

The model is started from the 6-hour forecast of the regional model following the 00 UTC run. Outputs of surface fields covering the uniform grid area are transmitted in GRIB formats to Canadian Regions. Series of coloured images (including animation) are also made available through the internal Web.

The CMC is able to provide in real-time air concentrations and surface deposition estimates of airborne pollutants. These fields are obtained from a 3-D long range atmospheric transport/dispersion/deposition model, named the "Canadian Emergency Response Model" or "CANERM". The main applications for this model have been for estimating concentrations of radionuclides and volcanic ash. Based on this operational capability, the CMC has been designed by the WMO as a Regional Specialised Meteorological Centre (RSMC) with specialization in Atmospheric Transport Modelling Products for Environmental Emergency Response. In addition, CMC has been designed by the ICAO as a Volcanic Ash Advisory Centre (VAAC).

Fields of wind, moisture, temperature and geopotential heights must be provided to CANERM. These are obtained either from the Global or the Regional forecast and objective analysis systems. Please refer to the above section 7.2 for more information on these NWP products.

Latitude, longitude and time of the release are necessary input parameters for CANERM. Estimates of intensity and duration of the release are also required. In the absence of actual source data, the standard default values adopted at the WMO's First International Workshop on Users' Requirements for the Provision of Atmospheric Transport Model Products for Environmental Emergency Response (September 1993) would be used. These are :

- uniform vertical distribution up to 500 m above the ground;

- uniform emission rate during the first 6 hours;

- total pollutant release of 1 arbitrary unit;

- type of radionuclide is Caesium 137.

CANERM was developed by Janusz Pudykiewicz of Environment Canada and is described in Pudykiewicz, 1989. The horizontal and vertical advection in the model are performed using the semi-Lagrangian algorithm of Ritchie, 1987. Diffusion is modelled according to K-theory. The diffusivities are constant in the free atmosphere but have a vertical profile in the boundary layer which is dependent on the state of the surface layer; the vertical diffusivity within the surface layer is approximated using the relations provided by the analytical theory of the surface layer. CANERM simulates wet and dry scavenging, wet and dry deposition and radioactive decay for selected tracers. Wet scavenging is modelled by a simplified statistical parameterization based on the relative humidity. The source term is modelled by a narrow gaussian distribution to simulate both the release and subgrid scale mixing.

CANERM can be executed in forecast mode up to day 10, using the operational Global forecast model, and up to 2 days using the operational Regional forecast model. CANERM can also be executed in hindcast mode using Global or Regional objective analyses. Presently, three horizontal resolutions are available : a resolution of 150 km on a quasi-hemispheric domain, a movable continental domain with a resolution of 50 km and a mesoscale domain with a resolution of 25 km. CANERM can be executed in both the Northern and the Southern Hemispheres.

(atmospheric transport/dispersion/deposition) products

Upon request from the appropriate WMO Member Countries Delegated Authorities, the CMC will provide the following standard set of basic products :

The standard set of products was agreed upon at the First International Workshop on Users' Requirements for the Provision of Atmospheric Transport Model Products for Environmental Emergency Response. The CMC can also provide charts of air concentration estimates for the surface, 850, 700, 500, 300 and 250 hPa levels as well as total surface deposition estimates, at 3 or 6-hour intervals, if required. All the products can be transmitted by facsimile, in real time, during environmental emergencies. In addition, CMC is designed by the ICAO as a Volcanic Ash Advisory Centre (VAAC).

The Canadian Global model is used to prepare ozone and UV Index forecast at the 18-hour projection time based on 00 UTC data and at the 30-hour projection time based on 12 UTC data (Burrows et al., 1994). A Perfect Prog statistical method is used for forecasting total ozone, which is then supplemented with an error-feedback procedure. UV Index is calculated from the corrected ozone forecast. Charts of the total ozone forecast and of the UV Index forecast are prepared and transmitted to the Regional Offices. Bulletins giving the forecast UV Index at an ensemble of stations across Canada are also generated.

Sea-state forecasts of 48 hours over the Eastern Pacific and Western Atlantic are generated twice a day (00 UTC and 12 UTC) by the WAM (WAve Modeling) model. The Pacific version of the wave model uses the surface level winds from the global model while the Atlantic version uses the regional model wind outputs. Various parameters are plotted on the wave forecast chart (wave height, swell period, swell height and direction, etc.).

The Canadian Meteorological Centre produces low level ozone concentration forecasts in support for the regional air quality program (Burrows et al, 1997). Dynamic-statistical forecast models provide a low cost solution to the problem of producing forecast of environmental variables such as low level ozone concentration, and can give a good trade-off between significance and precision in return for substantially lower computer resource requirements. Most environmental variables have a non-linear dependencies on predictors in part or all of their distribution. recent non-linear regression techniques allow for a significant increase in accuracy of dynamic-statistical models compared to traditional linear regression methods. One of these is Classification and regression Trees (CART) (Brieman et al, 1984), another is the Neuro-Fuzzy Inference System (NFIS) (Chiu, 1994). Both CART and NFIS are capable if modelling en entire predictand data distribution, including the tails, with better accuracy than linear regression and with fewer predictors. Given a learning data base of matched predictand and predictors, CART regression produces a non-parametric, non-linear, tree-based, piece-wise continuous model of the predictand data. It uses a variance minimizing algorithm which optimizes the task of predictor selection, often greatly reducing initial data dimensionality. NFIS first reduces initial data dimensionality by a procedure known as subtractive clustering which finds vector cluster centers, but it does not of itself eliminate predictors. Overlapping coverage in predictor space is enhanced by allowing a Gaussian membership function to apply for each component of clusters. Coefficients for a continuous predictand response model bases on the fuzzified cluster centers are then obtained by least-squares estimation procedure.

CANFIS is a two-stage data modelling technique combining the strength of CART to optimize selection of predictors from a large pool of potential predictors, with the modelling strength of NFIS. A CANFIS model requires minimal computer time to run. It will be used to produce models for ground level ozone, particulates, and other pollutants at more than 100 sites in Canada. The models will be used in a Perfect Prog forecast mode twice daily. Upstream and local Lagrangian meteorological and emissions predictors will be produced from 72-hour back-trajectories calculated at 925 hPa with the operational Global model.

The primary contributors to occurrences of high levels of ground level ozone are photochemical reaction involving ozone and NOx/VOC precursors of ozone. Ozone and its precursors can be produced locally and also transported to a site from upwind locations. Knowledge of airmass properties and pollution emissions locally and upwind from a site are essential for prediction of ground level ozone. Predictors for a dynamic-statistical model should involve both local and upwind meteorology and NOx/VOC emissions. Many potential predictors should be included, then the most relevant for a site are selected by the CANFIS procedure.

A database of hourly ground level ozone observations matched with a large pool of potential predictors was gathered for the years 1980-1994 for more than 100 sites in Canada to develop the dynamic-statistical model. Three predictands are forecast for each 6-hour period ending at 00, 06, 12 and 18 UTC. These are maximum 3-hour running average ground level ozone concentration in each 6-hour period, maximum ozone and average ozone concentration.

To build the predictors, 1988 emissions of NOx/VOC for point, area, and mobile sources in eastern North America were summed on a 20 km grid, stratified by four seasons, and by day of week. A three-dimensional 72-hour back trajectory starting from 925 hPa was run from each site at 00, 06, 12 and 18 UTC every day to calculate upwind predictors. Emissions at hourly trajectory locations are assumed to be injected into the air and are summed every 6 hours along the trajectory, with a grand sum over the entire trajectory. The combination of maximum temperatures and emissions predictors for the precious two days along the trajectories serve as surrogate for the process of injecting a load of pollution and ozone into an airmass upwind of a site and transporting it to the site. There are two predictors that are themselves estimated along the trajectories: maximum temperature and occurrence of precipitation. These predictors are estimated for two reasons: the trajectory locations may often not be near a meteorological observation site, so an observed value may not be available; they must be forecast for the day of validity at the site in the national operational model, either from 00 or 12 UTC runs the current day.

The system is driven by the Canadian operational Global model.

5. Extended range forecasts (10-30 days)

Ten-day temperature anomaly forecasts (Verret et al. 1998) are generated once a day and fifteen-day temperature anomaly forecasts once a week using a perfect prog approach from the medium-range model described at section 7.2.2.

Monthly temperature forecasts based on numerical weather prediction techniques, are issued at the beginning and mid-month of every month. An ensemble of 5 runs, obtained from 24-hour time lag, is produced. The model used is very similar to the former operational spectral global model (Ritchie, 1991), except it has lower horizontal resolution (T63 L23) and has evolving geophysical forcing: the anomalies (analysis-climatology) of sea surface temperature (SST) and snow, observed during the previous 30 days, are added to the daily climatology during the integration. Direct model surface temperature outputs ensemble means are averaged over the 30-day period and subtracted from model climatology obtained from a 26-year hindcast period (see section 7.6). These temperature anomalies are then normalised by the model standard deviation multiplied by .43 (to get equiprobable classes) and categorised in above, below and normal classes. Charts are produced, showing above normal, below normal and near normal temperature categories. Monthly forecast products are on the Web (address http://www.cmc.ec.gc.ca/climate/htmletc/m_fcst-e.html).

6. Long-range forecasts (seasonal forecasts)

Seasonal forecasts are issued 4 times a year (at the beginning of March, June, September and December). Seasonal products are distributed internationally and nationally through Internet (address http://www.cmc.ec.gc.ca/~cmcdev/saisons/seasons.html on the Web). They are also distributed nationally on the National Telecommunications System and to selected users by facsimile and made available on electronic bulletin boards. The charts are accompanied by a verification chart giving the performance of the forecast over the hindcast period. Also, verification charts, showing the previous season's prediction and a preliminary analysis of the observed anomaly, are provided.

1. Season 1 forecasts (zero lead time)

Season 1 forecasts are produced using a numerical approach (Derome et al., 2000). Two ensembles of 6 runs, obtained from 24-hour time lag, are produced: 6 from the T63 L23 model described in section 7.5, 6 from a general circulation model (GCM) (McFarlane et al., 1992) (T32 L10). Both models use the same initial operational analyses. SST anomalies, that have been observed over the previous 30 days, are added to climatological values over the period; snow is relaxed towards climatology at the end of the first month, except for the GCM, where it is a prognostic variable. A simple statistical linear regression equations relates the 1000-500 hPa thickness anomalies (forecast minus model climatology) to surface temperature anomalies, using regression coefficients for 90-day forecasts. Maps are similar to monthly ones: 3 classes, separated using the .43 standard deviation of observed climatology.

The precipitation forecast is produced using a more direct approach: the two ensemble means of forecast precipitation are subtracted from their respective models’ climatologies, and normalised by models’ standard deviations. These normalised forecasts are then added, divided by two and used to produce a map, categorised in 3 classes, using the .43 value for separation.

Skill maps of temperature and precipitation, as obtained over the 26 years of historical runs, are shown for each of the 4 seasonal forecasts periods.

2. Season 2, 3 and 4 forecasts

Seasonal forecasts at lead time of 3, 6 and 9 months are produced, using a Canonical Correlation Analysis technique (Shabbar and Barnston, 1996). The technique uses the SST anomalies observed over the last year to predict temperature and precipitation anomalies at Canadian stations (51 for temperatures, 69 for precipitation) for the following 3 seasons. Maps of above, normal and below temperature and precipitation are produced. These are accompanied by skill maps, as obtained from cross-validation over a 40-year period.

8. Verifications of prognostic products

The objective verification of the operational numerical models is done on a continuing basis. S1 skill scores, bias and root mean square error scores are produced for the Canadian verification area. A monthly verification summary is produced and distributed to our clients.

A verification system following the WMO/CBS recommendations has been implemented in 1987. Results are regularly exchanged with other participating centres. The table on the following page is a summary of the verification scores for year 2000 along the recommended format. Since 1994, CMC has proceeded with an electronic exchange of these verification scores with other NWP centres. This electronic exchange proved to be very useful and results from various operational NWP models can now be compared much more easily and thoroughly.

The regional data assimilation system will be modified in January 2001 to incorporate radiances from TOVS/ATOVS and ACARS/AMDAR winds, while moving to the analysis on model levels (as is done in the global system).

Improvements will be made to the regional and global 3D-variational systems: use of significant level temperature data and surface pressure instead of geopotential as well as temperature data from aircrafts. There will be a complete revision of the data quality control scheme, including a background check and variational quality control in the 3D-var. We will also incorporate the assimilation of additional type of observations (TOVS/ATOVS level 1b, high density SATOBS, radiances for IR channels (HIRS, GOES, AIRS) and surface wind data (SSM/I, QuikScat).

The unified GEM model will be converted to MPI and will be able to run on multiple nodes of the SX-5 machines.

The regional model resolution will be increased, with improved surface and condensation schemes.

Ensemble prediction system will see its resolution increased and will be used to issue probabilistic products.

The HIMAP runs will be executed twice a day.

Air quality forecasts will become fully operational with the implementation of a dynamically based ozone forecast model.

Instead of issuing 4 "seasonal" forecasts per year, 90-day forecasts are to be issued every month, blending dynamical and empirical techniques.