NOCN02 CWAO 141930 GENOT FAX NO. 002 GENOT FAX NO.002 EST LA VERSION ANGLAISE DE GENOT FAX NO 003. IMPLEMENTATION OF A NEW FORECAST MODEL AT CMC... AS OF MONDAY FEBRUARY 24 1997 AT 1200 UTC, A NEW FORECAST MODEL WILL BE IMPLEMENTED AT THE CANADIAN METEOROLOGICAL CENTRE. THE GEM (GLOBAL ENVIRONMENTAL MULTISCALE) MODEL WILL REPLACE THE OPERATIONAL RFE MODEL THAT HAS BEEN USED TO PRODUCE SHORT RANGE FORECASTS AT CMC SINCE APRIL 22 1986. ALL THE PRODUCTS CURRENTLY PRODUCED FROM THE RFE MODEL WILL BE DERIVED USING THE NEW GEM MODEL OUTPUTS (CHARTS, GRIB AND BUFR DATA, STATISTICAL AND SCRIBE FORECASTS). WHILE THE CURRENT RFE MODEL IS INTEGRATED AFTER A 12-HOUR SPINUP CYCLE, THE NEW GEM MODEL WILL BE INITIALIZED FROM AN EARLY GLOBAL ANALYSIS (DATA CUTOFF TIME OF T+1:50) ON 16 PRESSURE LEVELS. THIS IS AN INTERIM MEASURE SINCE FOR VARIOUS REASONS IT IS NECESSARY TO WAIT FOR THE 3D-VAR ANALYSIS (CURRENTLY RUNNING IN PARALLEL MODE) TO BE USED JOINTLY WITH THE GEM MODEL IN A REGIONAL ASSIMILATION SYSTEM. THE DECISION TO IMPLEMENT USING THE ABOVE SETUP IS FULLY JUSTIFIED BY THE FACT THAT BOTH OBJECTIVE AND SUBJECTIVE EVALUATIONS OF GEM VS RFE FORECASTS OVER SEVERAL MONTHS PROVE THAT GEM FORECASTS DO COMPARE ADVANTAGEOUSLY TO THE ONES FROM THE RFE. HOWEVER, WE EXPECT TO TEST A REGIONAL ASSIMILATION SYSTEM USING THE GEM MODEL AND THE 3D-VAR ANALYSIS AS SOON AS POSSIBLE AND TO IMPLEMENT IT IF FORECASTS ARE THUS IMPROVED. THE MAIN DIFFERENCES BETWEEN THE GEM AND THE RFE MODELS ARE: RFE GEM DOMAIN HEMISPHERIC GLOBAL CENTRAL WINDOW NORTH AMERICA NORTH AMERICA COVERAGE RESOLUTION OVER 35 KM 0.33 DEGREE ( ABOUT 35 CENTRAL WINDOW KM) VERTICAL LEVELS 28 SIGMA LEVELS 28 ETA LEVELS SPATIAL 3D FINITE ELEMENTS 3D FINITE ELEMENTS DISCRETIZATION TIME SEMI-LAGRANGIAN SEMI-LAGRANGIAN DISCRETIZATION SEMI-IMPLICIT IMPLICIT 3-TIME LEVEL 2-TIME LEVEL TIMESTEP 600 SEC. 1350 SEC. INITIALIZATION ADIABATIC IMPLICIT DIABATIC DIGITAL NORMAL MODE FILTER REGIONAL SPINUP YES NO THE MAIN POINTS TO NOTE ARE: 1. THE GEM MODEL IS RUN OVER A GLOBAL DOMAIN COVERED BY A VARIABLE RESOLUTION GRID; THE UNIFORM RESOLUTION AREA COVERS THE SAME AREA AS THE RFE WITH A GRID SPACING OF 0.33 DEGREE (ABOUT 35 KM). RESOLUTION DEGRADES SMOOTHLY AWAY FROM THE UNIFORM RESOLUTION DOMAIN IN THE SAME MANNER AS FOR THE RFE MODEL. 2. THE GEM MODEL USES A HYBRID COORDINATE SYSTEM (ETA COORDINATES) WHERE ETA LEVELS=(P-PT)/(PS-PT), PT BEING THE PRESSURE AT THE TOP OF THE MODEL (10 HPA) AND PS BEING THE SURFACE PRESSURE. THESE COORDINATES FOLLOW CLOSELY THE TERRAIN NEAR THE BOTTOM OF THE MODEL (LIKE THE SIGMA COORDINATES USED IN THE RFE) AND PRESSURE LEVELS NEAR THE TOP OF THE MODEL; THIS REDUCES SOME OF THE PROBLEMS NEAR THE TOP OF THE MODEL. 3. THE 2-TIME-LEVEL TIME DISCRETIZATION PERMITS LONGER TIME STEPS (1350 SECONDS VERSUS 600 SECONDS) WITHOUT LOSS OF ACCURACY. 4. THE PHYSICAL PARAMETERIZATION SCHEMES USED IN GEM ARE THE SAME AS THE ONES USED IN THE RFE. THE NEW GEM MODEL HAS BEEN RUN IN PARALLEL FOR SEVERAL MONTHS. FORECASTS FROM BOTH THE RFE AND THE GEM MODELS HAVE BEEN VERIFIED OBJECTIVELY AGAINST RAOBS OVER NORTH AMERICA, AND HAVE BEEN EVALUATED BY THE CMC OPERATIONAL METEOROLOGISTS. OVERALL, FORECASTS OF THE MASS FIELDS FROM THE GEM MODEL ARE GENERALLY SIMILAR TO THE ONES FROM THE RFE. UPPER AIR FEATURES WERE OFTEN SLIGHTLY BETTER HANDLED BY GEM, WHICH IS IN AGREEMENT WITH SMALLER RMS ERRORS OF THE GEOPOTENTIAL GEM FORECASTS WHEN VERIFIED AGAINST RAOBS. WE HAVE NOTICED THAT SOME WEATHER SYSTEMS WERE MOVED SLIGHTLY FASTER BY THE GEM MODEL, AND MORE OFTEN THAN OTHERWISE, THIS VERIFIED BETTER. SIGNIFICANT SURFACE LOWS WERE OFTEN FORECAST 2-6 HPA DEEPER BY THE GEM MODEL, AND THIS ONCE AGAIN VERIFIED BETTER ON MOST OCCASIONS. JETSTREAMS ARE OFTEN FORECAST SLIGHTLY STRONGER BY GEM THAN BY THE RFE, AND THESE SOMEWHAT STRONGER JETS VERIFY BETTER MOST OF THE TIME. PRECIPITATION FORECASTS PRODUCED BY BOTH MODELS ARE GENERALLY SIMILAR, EXCEPT IN THE FIRST HOURS OF INTEGRATION, WHERE THE ABSENCE OF A SPINUP FOR THE GEM MODEL GENERALLY RESULTS IN SLIGHTLY LESS PRECIPITATION. HOWEVER, AFTER THE FIRST SIX HOURS OF INTEGRATION, PRECIPITATION AMOUNTS ARE GENERALLY SIMILAR. DURING THE EVALUATION PERIOD, IN CASES WHERE DEEP CONVECTION WAS THE MAIN SOURCE OF PRECIPITATION (AS WITH SYSTEMS ORIGINATING OVER THE GULF OF MEXICO OR THE GULF STREAM), VERY OFTEN THE FORECAST AMOUNTS IN THE GEM MODEL OUTPUTS WERE LESS (BY 10-20%) THAN IN THE RFE; THIS IS LIKELY A DESIRABLE BEHAVIOR SINCE THE RFE MODEL HAS A TENDENCY TO OVERFORECAST PRECIPITATION IN THESE SITUATIONS. VERIFICATION OF THE STATISTICAL FORECASTS DERIVED FROM THE GEM OUTPUTS SHOWS THAT THEY ARE SLIGHTLY SUPERIOR TO THE ONES DERIVED FROM THE RFE, ESPECIALLY AT THE LONGER TIME RANGES. DUE TO THE ABSENCE OF SPINUP AND TO THE LOW VERTICAL RESOLUTION ANALYSIS USED TO FEED THE GEM MODEL, THE 0-HOUR FORECASTS MAY DEPART MORE FROM OBSERVATIONS THAN THE ONES FROM THE CURRENT RFE MODEL. THIS IS ESPECIALLY TRUE FOR FIELDS IN THE BOUNDARY LAYER (SUCH AS SURFACE TEMPERATURE) WHICH ARE EXTRAPOLATED FROM THE LOWEST PRESSURE ANALYSIS LEVEL. HOWEVER, AFTER A FEW HOURS OF INTEGRATION, THESE FIELDS HAVE ALREADY REACHED COMPARABLE VALUES TO THOSE PRODUCED IN THE RFE. USERS ARE THEREFORE ADVISED TO EXERCISE CAUTION IN THEIR INTERPRETATION OF THESE 0-HOUR FIELDS AND ON THE CONCLUSIONS THEY COULD DRAW REGARDING THE QUALITY OF THE GEM MODEL FORECASTS AT LONGER TIME RANGES. THE SLIGHTLY BETTER PERFORMANCE OF THE GEM MODEL IN SPITE OF THE ABSENCE OF SPINUP CAN BE ATTRIBUTED TO A REDUCTION IN HORIZONTAL DIFFUSION, THE USE OF HYBRID COORDINATES, THE ABSENCE OF A STRATOSPHERIC SPONGE NEAR THE TOP OF THE MODEL, AND THE ABSENCE OF AN EQUATORIAL BOUNDARY. A FULL ARTICLE FROM COTE ET AL. (1997) DESCRIBING THE DEVELOPMENT AND THE CHARACTERISTICS OF THE GEM MODEL, AS WELL AS A MORE DETAILED DISCUSSION OF THE PARALLEL RUN EVALUATION, WILL BE AVAILABLE AROUND FEBRUARY 20 ON THE CMC INTERNAL WEB PAGE, AT THE FOLLOWING ADDRESS: HTTP://IWEB.CMC.EC.GC.CA/CMC/INDEX.HTML/. FOR USERS NOT HAVING ACCESS TO ENVIRONMENT CANADA ECONET, THE INFORMATION IS ALSO LOCATED ON THE CMC PUBLIC WEB PAGE, AT HTTP://WWW.CMC.EC.GC.CA/CMC/INDEX.HTML MCBEAN/ADMA/TORONTO