NOCN02 CWAO 131830
GENOT FAX NO. 009

GENOT NO. 009 EST LA VERSION ANGLAISE DE GENOT NO. 010

IMPLEMENTATION OF A NEW DATA ASSIMILATION SYSTEM AT CMC

AS OF WEDNESDAY JUNE 18 1997 AT 1200 UTC, A NEW DATA
ASSIMILATION SYSTEM WILL BE IMPLEMENTED AT THE
CANADIAN METEOROLOGICAL CENTRE. THE 3D-VAR (3-
DIMENSIONAL VARIATIONAL) DATA ASSIMILATION SYSTEM WILL
REPLACE THE OPERATIONAL OI (OPTIMUM INTERPOLATION)
SCHEME THAT HAS BEEN USED FOR DATA ASSIMILATION AT CMC
SINCE THE MID SEVENTIES. ALL THE UPPER AIR ANALYSES
WILL BE PRODUCED USING THE 3D-VAR SCHEME. HOWEVER, ALL
SURFACE ANALYSES WILL CONTINUE TO BE PRODUCED USING
THE OI SCHEME. AT THE SAME TIME, A CHANGE IN THE
SELECTION METHOD OF HUMSAT DATA USED BY THE ANALYSIS
WILL ALSO BE IMPLEMENTED. THE NEW ANALYSIS WILL BE
USED AS INPUT TO BOTH THE GLOBAL SEF MEDIUM RANGE AND
THE GEM REGIONAL FORECAST MODELS.

DATA ASSIMILATION IS AN ESSENTIAL FEATURE OF ANY
NUMERICAL WEATHER PREDICTION SYSTEM AND ITS PURPOSE IS
TO BLEND INFORMATION CONTAINED IN OBSERVATIONS FROM
DIFFERENT SOURCES WITH INFORMATION CONTAINED IN A
PRIOR ESTIMATE OF THE STATE OF THE ATMOSPHERE REFERRED
TO AS THE BACKGROUND STATE (TRIAL FIELDS). IN
PRACTICE, CURRENT OPERATIONAL ASSIMILATION SYSTEMS
TAKE THE BACKGROUND STATE AS A SHORT TERM FORECAST
FROM A NWP MODEL WHICH CARRIES FORWARD IN TIME THE
INFORMATION GAINED FROM PAST OBSERVATIONS. TO PROPERLY
ASSIMILATE THE DATA, ONE NEEDS FORWARD MODELS FOR EACH
TYPE OF OBSERVATION TO DESCRIBE THE LINK BETWEEN THE
MODEL STATE AND THE OBSERVATIONS. IT IS IMPORTANT TO
HAVE A GOOD ESTIMATE OF THE ACCURACY OF BOTH THE
OBSERVATIONS AND THE BACKGROUND STATE IN ORDER TO
OBTAIN AN ANALYSIS THAT TRULY MINIMIZES THE ANALYSIS
ERROR. THE QUALITY OF THE ANALYSIS IS CONTROLLED
LARGELY BY THE NATURE OF THE COVARIANCES OF BOTH THE
FORECAST AND OBSERVATIONAL ERRORS. ALTHOUGH BOTH THE
OI AND 3D-VAR SYSTEMS ARE BASED ON THIS SAME
STATISTICAL PRINCIPLE, THE 3D-VAR ANALYSIS DOES NOT
HAVE MANY OF THE CONSTRAINTS THAT THE OI SYSTEM HAS:

1.   DATA SELECTION
TO PRODUCE AN ANALYZED VALUE AT A PARTICULAR GRID
POINT, THE OI HAS TO RESTRICT ITSELF TO A MAXIMUM
NUMBER OF OBSERVATIONS: THIS IS REFERRED TO AS DATA
SELECTION WHICH WILL NOT BE REQUIRED ANYMORE.  IN 3D-
VAR, ALL AVAILABLE DATA ARE CONSIDERED.

2.   DIRECT ASSIMILATION OF OBSERVATIONS
THE OI SYSTEM DOES NOT ALLOW FOR COMPLEX FORWARD
MODELS AND ALL OBSERVATIONS MUST BE TRANSFORMED INTO
MODEL VARIABLES (E.G., GEOPOTENTIAL, WIND SPEED AND
DIRECTION, ETC.). ONE OF THE KEY ADVANTAGE OF THE
VARIATIONAL FORMULATION IS THAT IT IS A MORE SUITABLE
FRAMEWORK TO ASSIMILATE OBSERVATIONS INDIRECTLY (EVEN
NON LINEARLY) RELATED TO MODEL VARIABLES WHICH IS
OFTEN THE CASE FOR SATELLITE DATA.

3.   IMPROVEMENT IN THE REALISM OF BACKGROUND ERROR STATISTICS
SIGNIFICANT IMPROVEMENTS TO THE ANALYSIS REQUIRE MORE
REALISM IN THE REPRESENTATION OF THESE STATISTICS. THE
3D-VAR SYSTEM IS OPENING NEW AVENUES FOR SUCH
IMPROVEMENTS. CURRENTLY, IN ORDER TO PROPERLY VALIDATE
THE SYSTEM, THE SAME STATISTICS HAVE BEEN INTRODUCED
AND THE EXTENSIVE TESTING DONE IN THE COURSE OF THE
DEVELOPMENT HAS SHOWN THAT WHATEVER OI CAN DO, IT IS
POSSIBLE TO REPRODUCE IT TO A GREAT EXTENT IN THE 3D-
VAR.

IN ADDITION, THE NEW 3D-VAR ANALYSIS WILL ALSO BENEFIT
FROM A BETTER SELECTION OF THE HUMSAT DATA USED BY THE
MOISTURE ANALYSIS. WHILE THE CURRENT OI SCHEME USES
HUMSAT DATA AT A STANDARD RESOLUTION OF 200 KM, THE
NEW 3D-VAR ANALYSIS WILL RATHER USE THESE DATA AT 100
KM RESOLUTION WHERE STRONG MOISTURE GRADIENTS HAVE
BEEN DETECTED, AND AT 300 KM RESOLUTION WHERE
HOMOGENEOUS MOISTURE CONDITIONS EXIST. THIS WILL
RESULT IN REFINED MOISTURE PATTERNS ASSOCIATED TO
SMALLER SCALE CLOUD FEATURES.

THIS IMPLEMENTATION OF 3D-VAR IS THE FIRST AND
ESSENTIAL STEP TOWARD 4 DIMENSIONAL DATA ASSIMILATION
(4DDA). BOTH THE OI AND 3D-VAR ARE STATIONARY SYSTEMS
IN THE SENSE THAT THEY ARE DESIGNED TO HAVE A MINIMUM
LEVEL OF ERROR WHEN AVERAGED OVER A PERIOD OF A MONTH
TO A SEASON. THIS MAY HOWEVER BE INSUFFICIENT FOR
PARTICULAR SYNOPTIC EVENTS THAT REQUIRE A SUBTLER
DESCRIPTION TO BE CAPABLE TO PROPERLY MODEL THE
ENSUING DEVELOPMENT. THIS CAN ONLY BE ACHIEVED THROUGH
4DDA ON WHICH INTENSIVE RESEARCH IS TAKING PLACE.

ALTHOUGH 3D-VAR WILL MAKE IT MUCH EASIER TO ASSIMILATE
NON-CONVENTIONAL OBSERVATIONS, IT MUST POINTED OUT
THAT AT THIS TIME, 3D-VAR WILL USE THE SAME SET OF
OBSERVATIONS AS IS CURRENTLY BEING USED IN THE
OPERATIONAL OI SCHEME. HOWEVER, IN 3D-VAR, AIREP AND
SYNOP OBSERVATIONS ARE NO LONGER SUPEROBBED. IN THE
FUTURE MONTHS, BACKGROUND ERROR STATISTICS WILL BE
IMPROVED BY USING SOME OF THE NEW CAPABILITIES THAT
THE 3D-VAR HAS. NEW TYPES OF DATA WILL BE
PROGRESSIVELY INTRODUCED AS WELL.

THIS NEW 3D-VAR DATA ASSIMILATION SYSTEM HAS BEEN RUN
IN PARALLEL FOR THE LAST 4 MONTHS. ANALYSES, AND
FORECASTS FROM THE GLOBAL MODEL INITIALIZED FROM THE
3D-VAR ANALYSIS HAVE BEEN VERIFIED AGAINST RADIOSONDE
DATA OVER SEVERAL NETWORKS, EVALUATED BY THE CMC
OPERATIONAL METEOROLOGISTS, AND COMPARED TO THEIR
OPERATIONAL COUNTERPARTS.

OVERALL, BOTH THE OPERATIONAL AND PARALLEL SYSTEMS
HAVE SHOWN SIMILAR LEVELS OF PERFORMANCE FOR ANALYSES
AND FORECASTS GENERATED BY THE SEF MODEL. THIS WAS TO
BE EXPECTED SINCE BOTH ANALYSES HAVE USED THE SAME
DATA (EXCEPT FOR HUMSAT DATA) SO AS TO VALIDATE THE 3D-
VAR SCHEME AGAINST THE OI.

THE OPERATIONAL ANALYSES OF THE MASS FIELDS (E.G.
GEOPOTENTIAL, TEMPERATURE, WINDS) ARE VERY SIMILAR TO
THE ONES PRODUCED IN THE PARALLEL 3D-VAR SYSTEM,
EXCEPT IN DATA SPARSE AREAS, WHERE SOME SIGNIFICANT
DIFFERENCES BETWEEN THE TWO ANALYSES HAVE BEEN
OBSERVED. THESE OCCUR MOSTLY OVER OCEANIC AREAS, AND
MORE ESPECIALLY OVER THE SOUTHERN HEMISPHERE, WHERE
DATA DENSITY IS LESS. THE LARGEST DIFFERENCES ARE SEEN
AT THE HIGHEST ANALYSIS LEVELS WHERE THE FIT TO
GEOPOTENTIAL DATA IS NOT AS CLOSE TO THE RADIOSONDE
OBSERVATIONS IN 3D-VAR AS IN THE OI. HOWEVER, THIS
SHOULD NOT BE INTERPRETED AS A DEFICIENCY OF THE 3D-
VAR BUT RATHER A SUBTLE SIDE EFFECT OF THE DATA
SELECTION PROCESS WHICH IS DONE DIFFERENTLY IN BOTH
SYSTEMS..

MOISTURE ANALYSES HAVE BEEN ASSESSED AS SLIGHTLY
BETTER IN THE PARALLEL SYSTEM, LIKELY DUE TO THE
IMPROVED SELECTION PROCESS OF HUMSAT DATA. THIS
RESULTS IN BETTER DEFINITION OF SMALLER SCALE CLOUD
SYSTEMS, AND IN THEIR MORE ADEQUATE REPRESENTATION IN
THE MOISTURE FIELD PRODUCED BY THE ANALYSIS.

MCBEAN/ADMA/TORONTO