NOCN03 CWAO 052030 GENOT NO. 002 GENOT NO. 001 EST LA VERSION FRANCAISE DU GENOT NO. 002. ON WEDNESDAY JANUARY 12, 2005, AT 12 UTC, THE CANADIAN METEOROLOGICAL CENTRE WILL IMPLEMENT A MAJOR CHANGE TO THE ENSEMBLE PREDICTION SYSTEM. THE CURRENT ANALYSIS CYCLE USING THE OPTIMAL INTERPOLATION TECHNIQUE WILL BE REPLACED BY AN ANALYSIS CYCLE BASED ON THE ENSEMBLE KALMAN FILTER. THE 16 MODELS USED TO PRODUCE THE 10 DAY FORECASTS WILL NOT BE MODIFIED. MOTIVATION THE ASSIMILATION SYSTEM OF THE ENSEMBLE PREDICTION SYSTEM (EPS) IS DESIGNED TO PROVIDE AN ENSEMBLE OF ANALYSES THAT REFLECTS THE ERROR IN THE ENSEMBLE MEAN AND THAT CAN BE USED TO INITIALIZE A MEDIUM RANGE EPS. THE ENSEMBLE KALMAN FILTER (ENKF) USES AN ENSEMBLE OF TRIAL FIELDS TO PROVIDE FOUR- DIMENSIONAL FLOW-DEPENDENT STATISTICS OF THE TRIAL-FIELD ERROR TO THE ASSIMILATION PROCEDURE. IT APPROXIMATES THE KALMAN FILTER WHICH IS THE OPTIMAL DATA ASSIMILATION METHOD UNDER CERTAIN CONDITIONS SUCH AS LINEARITY OF ERROR EVOLUTION. THE ENKF TECHNIQUE WAS DEVELOPED BY THE DATA ASSIMILATION AND SATELLITE METEOROLOGY DIVISION OF THE METEOROLOGICAL RESEARCH BRANCH, IN COOPERATION WITH THE DEVELOPMENT BRANCH OF THE CANADIAN METEOROLOGICAL CENTRE. ASSIMILATION THE PRODUCTION OF THE 16 INITIAL CONDITIONS NEEDED TO LAUNCH THE 16 MEDIUM RANGE INTEGRATIONS OF THE CURRENT OPERATIONAL EPS HAS THE FOLLOWING CHARACTERISTICS THERE ARE 8 INDEPENDENT ASSIMILATION CYCLES THE ANALYSES ARE PERFORMED USING OPTIMAL INTERPOLATION (OI) THE TRIAL FIELDS ARE PRODUCED BY THE SEF MODEL 8 EXTRA ANALYSES ARE PRODUCED USING A CORRECTION TOWARDS THE DETERMINISTIC OPERATIONAL GLOBAL ANALYSIS. WITH THE NEW EPS, THE 16 INITIAL CONDITIONS ARE OBTAINED IN THE FOLLOWING MANNER THE ENKF CONSISTS OF 96 QUASI-INDEPENDENT ASSIMILATION CYCLES THE TRIAL FIELDS ARE PRODUCED BY THE GEM MODEL AT A HORIZONTAL RESOLUTION OF 1.2 DEGREE CONFIGURED SIMILARLY TO THE CURRENT DETERMINISTIC OPERATIONAL GLOBAL MODEL 16 REPRESENTATIVE MEMBERS ARE CHOSEN AMONG THE 96 ANALYSES THE ENSEMBLE SPREAD IS INFLATED BY A FACTOR OF 1.5 TO ARRIVE AT SUFFICIENT SPREAD IN THE MEDIUM RANGE. FORECAST THE 16 MEDIUM-RANGE FORECASTS RUNNING TO 10 DAYS ARE LAUNCHED DAILY AT 00 UTC. THIS COMPONENT OF THE EPS REMAINS UNCHANGED. DATA THE ENKF WILL BENEFIT FROM THE DATA QUALITY CONTROL PROVIDED BY THE CURRENT 3DVAR OPERATIONAL GLOBAL DATA ASSIMILATION SYSTEM (GDAS), WHEREAS THE OI SYSTEM HAD AN INDEPENDENT DATA STREAM FOR THE OBSERVATIONS. THE NEW TECHNIQUE ALLOWS THE USE OF NEW DATA SETS INCLUDING SATELLITE RADIANCE DATA FROM THE AMSU/A AND AMSU/B INSTRUMENTS, WHICH WAS NOT POSSIBLE WITH THE PREVIOUS TECHNIQUE. INDEED ALMOST ALL GDAS DATA ARE ASSIMILATED IN THE ENKF SYSTEM. SUMMARY OF THE OBJECTIVE EVALUATIONS EXTENSIVE PRELIMINARY TESTING OF THE NEW EPS IN BOTH SUMMER AND WINTER CONDITIONS WAS PERFORMED BY THE RESEARCH AND DEVELOPMENT TEAM PRIOR TO OPERATIONAL PARALLEL TESTING. THE NEW EPS HAS BEEN RUNNING IN PARALLEL WITH THE OPERATIONAL EPS SINCE AUGUST 26, 2004, ALLOWING ADDITIONAL COMPARATIVE PERFORMANCE MEASUREMENTS BETWEEN THE TWO SYSTEMS. OBJECTIVE VERIFICATIONS OF AVERAGE 6-HOUR ENSEMBLE FORECASTS AGAINST UPPER AIR SOUNDING OBSERVATIONS FOR WIND, HEIGHT, TEMPERATURE AND DEW POINT DEPRESSION, FOR ALL TESTING PERIODS (PRELIMINARY AND PARALLEL), SHOWED POSITIVE AND STATISTICALLY SIGNIFICANT IMPROVEMENTS TO THE RMS ERROR. THE COMPARISON OF THE OPERATIONAL AND PARALLEL MEDIUM RANGE ENSEMBLE PREDICTIONS SHOWS THAT THE AVERAGE QUALITY OF THE ENSEMBLES IS IMPROVED DUE TO BETTER INITIAL CONDITIONS IN THE PARALLEL SYSTEM. WE NOTE IN PARTICULAR SMALLER ERRORS FOR THE ENSEMBLE MEAN A SMALLER AND MORE REALISTIC INITIAL SPREAD OF THE ENSEMBLE A FASTER GROWTH RATE FOR THE PERTURBATIONS, LEADING TO A BETTER OVERALL AGREEMENT BETWEEN THE ENSEMBLE SPREAD AND THE SIZE OF THE ENSEMBLE-MEAN ERROR. THE IMPROVEMENTS ARE PARTICULARLY SIGNIFICANT OVER THE SOUTHERN HEMISPHERE, WHERE THE ENKF SYSTEM IS ABLE TO DIRECTLY USE THE AMSU/A AND THE AMSU/B RADIANCES, WHEREAS THE OI SYSTEM USES SATELLITE DERIVED THICKNESS (SATEM) OBSERVATIONS. OBJECTIVE PRECIPITATION VERIFICATIONS AGAINST CANADIAN OBSERVATIONS INDICATE A SMALL IMPROVEMENT. SUMMARY OF THE SUBJECTIVE EVALUATIONS DURING THE LAST TWO MONTHS OF THE PARALLEL RUN, OPERATIONAL METEOROLOGISTS COMPARED THE TWO EPS OVER NORTH AMERICA FOR FORECAST TIMES UP TO 5 DAYS. WHEN COMPARING THE AVERAGE MASS FIELD OF THE 16 MEMBERS, IT WAS FOUND THAT THE TWO EPS WERE SIMILAR IN ABOUT 45 PERCENT OF CASES. FOR CASES SHOWING DIFFERENCES THE PARALLEL SYSTEM WAS PREFERRED MORE OFTEN THAN THE OPERATIONAL. IN THE WESTERN PORTIONS OF NORTH- AMERICA AND EASTERN PACIFIC, THE PARALLEL EPS WAS PREFERRED IN 66 PERCENT OF CASES SHOWING DIFFERENCES. IT WAS ALSO NOTED THAT THE PRECIPITATION PATTERNS WERE RATED EQUIVALENT IN 55 PERCENT OF CASES. OVERALL, THE NEW SYSTEM SHOWED SMALL IMPROVEMENTS WITH RESPECT TO PRECIPITATION FORECASTS. AGAIN, THE MORE SYSTEMATIC IMPROVEMENTS IN PRECIPITATION FORECAST WERE NOTED OVER WESTERN NORTH AMERICA. IT IS WORTH NOTING THAT THE NEW EPS SHOWED MUCH LESS SPREAD THAN THE OPERATIONAL SYSTEM AT LEAD TIME SMALLER THAN 5 DAYS, AND ESPECIALLY AT INITIAL TIME. HOWEVER, IN SOME CASES THE NEW SYSTEM DID SHOW LARGER SPREAD THAN THE OPERATIONAL SYSTEM ON DAYS 4 OR 5. OVERALL, THE SPREAD AMONG THE MEMBERS WAS PREFERRED IN THE PARALLEL SYSTEM. ADDITIONAL DETAILS ABOUT THIS IMPLEMENTATION WILL SOON APPEAR AT THE FOLLOWING WEB SITE: HTTP://WWW.SMC.EC.GC.CA/CMC/OP_SYSTEMS/RECENT_E.HTML (IN LOWER CASE). EVERELL / ADMA / SMA-SMC TORONTO