1. GLFE Real-time TAMDAR Impact Experiments with the 20km RUC Stan Benjamin,Tracy Lorraine Smith, Bill Moninger, Brian Jamison NOAA Forecast Systems Laboratory Boulder, CO

2. Outline of talk Part 1: General description of RUC 1h cycle - Data assimilated - Spatial effects of assimilated data Recent enhancements to RUC assimilation Previous results from RUC data impact experiments Design of RUC parallel experiments – Dev / Dev2 Examples of Dev/Dev2 difference fields (Brian Jamison) Part 2: Actual statistical results

3. Provide high-frequency mesoscale analyses, short-range model forecasts Use all available observations Users: – aviation/transportation – severe weather forecasting – general public forecasting Focus on 1-12 hour forecast range Accuracy of surface fields very important Purpose for Rapid Update Cycle (RUC) model run operationally at NCEP

4. Vert Coord Hybrid sigma-isentropic Stable clouds, NCAR mixed-phase (cloud water, precipitation rain water, snow, graupel, ice, ice particle number. concen. Sub-grid Grell-Devenyi ensemble scheme precipitation (144 members, mean to model) Land-surfaceRUC LSM - 6-level soil/veg model, 2-layer snow model (Smirnova) Rapid Update Cycle Model Features Current operational RUC: 20-km Planned upgrade 2005:13-km

5. Updates 1-h cycle Analysis 3DVAR on  surfaces (     unb  lnQ) Details Balanced height (NMC method) Length-scales modified from OI Sfc Obs/ Use surface obs through PBL PBL Structure Lapse-rate checks Noise Adiabatic digital filter initialization Clouds/ Cloud analysis (GOES cloudtop pres, moisture radar reflectivity, METAR clouds) Cycling of cloud, hydrometeor, land-surface fields RUC Assimilation System

6. RUC Hourly Assimilation Cycle 11 12 13 14 15 16 Time (UTC) 1-hr fcst Background Fields Analysis Fields 1-hr fcst 3DVAR Obs 1-hr fcst 1-hr fcst 1-hr fcst 3DVAR Obs 3DVAR Obs 3DVAR Obs 1-hr fcst 3DVAR Obs

7. RUC Hourly Assimilation Cycle EC 11 12 13 14 15 16 Time (UTC) 3-hr fcst Background Fields Analysis Fields 3-hr fcst 3-hr fcst 3DVAR Obs 3-hr fcst 12-h fcst 12-h fcst 3DVAR Obs 3DVAR Obs 3DVAR Obs 3DVAR Obs

8. Observations used in RUC Data Type ~Number Freq. -------------------------------------------------- Rawinsonde 80 /12h NOAA profilers 30 / 1h VAD winds 110-130 / 1h Aircraft (V,temp) 1400-4500 / 1h Surface/METAR 1500-1700 / 1h Buoy/ship 100-150 / 1h GOES precip water 1500-3000 / 1h GOES cloud winds 1000-2500 / 1h GOES cloud-top pres ~10km res / 1h SSM/I precip water 1000-4000 / 6h -------------------------------------------------- GPS precip water ~300/ 1h Mesonet~5000/ 1h PBL – prof/RASS ~20/ 1h Radar refl / lightning 4 km res -------------------------------------------------- NCEP operational FSL experimental Cloud anx variables

9. Application of Digital Filter Initialization in RUC 45 min forward, 45 min backward – no physics Average over DFI period

10. RUC Analysis 3-d effect of observations dependent on statistically determined forecast error covariance vertical – dependent on  horizontal – smaller near surface, larger aloft,

11. RUC20 Wind forecast Accuracy -Sept-Dec 2002 Verification against rawinsonde data over RUC domain RMS vector difference (forecast vs. obs) RUC is able to use recent obs to improve forecast skill down to 1-h projection for winds 6 8 10 12 (kts) 1 3 69 12 Analysis ~ ‘truth’

12. Results from fall 2002 – better moisture results in RUC13 Potential for more improvement from TAMDAR – V, T, RH

13. Use of surface obs information throughout boundary layer in the RUC analysis Problem Information from surface observation not used through depth of PBL by RUC analysis Surface observation not retained in model forecast Temperature Original analysis Dewpoint Surface Observation *

14. Use of surface obs information throughout boundary layer in the RUC analysis Problem Information from surface observation not used through depth of PBL by RUC analysis Surface observation not retained in model forecast Solution Use METAR observation throughout PBL depth (from background field) Better model retention of surface observations Temperature Original analysis Dewpoint Analysis with use of PBL depth Surface Observation *

15. RUC enhancements: 1.Use of METAR obs through boundary-layer depth (Sept 04) 2.Assimilation of GPS precipitable water observations (May 2005) CAPE impact from two RUC enhancements 3h fcst WITH enhancements 3h fcst OPERATIONAL 0000 UTC 21 Apr 2004 Severe reports NWS SPC Norman, OK

16. Use GOES CTP, radar reflectivity, lightning, METAR (clouds,wx,vis) to modify moisture fields Construct 3-d logical arrays (YES/NO/UNKOWN) for clouds and precipitation from all info Clear/build (change qc, qi, qv) with logical arrays Safeguards for pressure-level assignment problems (marine stratus, convective clouds) Use nationwide mosaic radar data to modify water vapor, hydrometeor fields Lightning data used as a proxy for radar reflectivity Feedback to cumulus parameterization scheme RUC Cloud Analysis

17. GOES cloud top pressureRadar/lightning data 100 200 300 400 500 600 700 800 900 999 PRES Qv Qc RH Cloudwater, water vapor and relative humidity before ( ) and after (----) GOES Cloud- top pressure adjustment Rainwater, snow, cloud ice and reflectivity before ( ) and after (----) GOES radar/lightning adjustment 100 200 300 400 500 600 700 800 900 999 PRES Qr Qs Qi dBZ

18. 3h 20km fcst WITH GOES cloud assim Cloud-top pressure (mb) NESDIS GOES Verification cloud-top prs 1200 UTC 9 Dec 2001 Sample 20-km RUC forecast impact from GOES cloud-top pres. assimilation 3h 40km fcst NO GOES cloud assim

19. - Nearest station up to 100 km distance - Maps info to 3-d cloud, precip. Y/N/U arrays - Change qc, qi, qv as follows: Build for BKN / OVC / Vertical Visibility - 40 mb thick layer (2+ model levels) - 150 mb thick for precip + GOES clouds - Can build multiple broken layers Clear - Up to cloud base (if needed) - to 12 kft for CLR report Assimilation of METAR cloud/wx/vis Better analysis, prediction of ceiling and visibility

20. analysis – with METAR cloud/ visibility obs RH,  Background – 1h fcst Cloud water mixing ratio Sample modification of cloud water (qc) from METAR cloud/weather/ Visibility obs 1700 UTC 27 Jan 2004 Cloud water mixing ratio

21. IFRLIFR VFR CLR MVFR Ceiling from RUC hydrometeors Aviation Flight Rules cloud ceiling height (meters) 1800 UTC 17 Nov 2003 Sample ceiling analysis impact Analysis WITH cld/wx/ vis obs Analysis NO cld/wx/ vis obs Observations

22. IFRLIFR VFR CLR MVFR Aviation Flight Rules cloud ceiling height (meters) Ceiling from RUC hydrometeors 2100 UTC 17 Nov 2003 Sample ceiling forecast impact 3h fcst WITH cld/wx/ vis obs 3h fcst NO cld/wx/ vis obs Observations

23. 2005  13-km operational at NCEP Assimilation of new observations - METAR cloud/vis/current weather - Mesonet - GPS precipitable water - Boundary layer profilers, RASS temperatures - Radar data (when available at NCEP) Model improvements – new versions of: - Mixed-phase cloud microphysics (NCAR-FSL) - Grell-Devenyi convective parameterization 2007  Planned operational implementation of WRF-based rapid-refresh Planned upgrades to RUC model

24. Convection - 2-12h forecast Ceiling/visibility Turbulence Terminal / surface 3-d RUC weather data updated hourly 20km x 50 vertical levels x 14 variables Better weather products require improved high- frequency high-resolution models with high-refresh data to feed them Icing Winds

25. Positive difference means CNTL experiment with profiler data had lower error than the EXP-P no-profiler experiment Wind forecast ‘errors’ - defined as rawinsonde vs. forecast difference Cntl = using all obs Exp = deny profiler obs Difference in errors between Cntl and Exp experiments w/ RUC - 4-17 Feb 2001 Anx Fit - ‘truth’

26. 3 h 6h 12 h Wind forecast impact – US National domain  Impact generally greatest for shorter forecast durations.  Decreases with projection except raobs  Raob impact largest at 12h – raob frequency is 12h.  Aircraft - largest overall impact at 3h, profiler next (much smaller)  Modest VAD and METAR impact aloft; (METARs improve low-level height field, which helps z  p mapping needed for VADs and profilers) 3h 6h 12h

27. Real-time TAMDAR impact experiment design Parallel 20km RUC 1-h cycles Dev cycle – all obs data but no TAMDAR Dev2 cycle – dev + TAMDAR data Lateral boundary conditions – same for Dev and Dev2 Control design Initialize Dev and Dev2 runs at exact same time Reset dev and dev-2 background field at 1000z every 48 h (even Julian dates) Ensure against any computer logistics differences Evolution of dev vs. dev2 is different Example – buddy check QC in each cycle may occasionally differ for non-TAMDAR data Slight difference in gravity waves Can propagate difference throughout domain Shows up in sfc temps, convection, esp. 925, 850 mb

28. 0900z ---------  1200z Sunday 10 April 2005 Reset of dev-dev2 difference at 1000z by copying Dev RUC 1-h forecast from 0900z as background for Dev2 analysis at 1000z Reset is effective (although

29. Dev-Dev2 difference – 12h fcst Init 1200z 10 April 2005 – 500 mb

34. Dev-Dev2 difference – 12h fcst Init 1800z 10 April 2005 – 700 mb

39. Real-time TAMDAR impact experiment design Parallel 20km RUC 1-h cycles Dev cycle – all obs data but no TAMDAR Dev2 cycle – dev + TAMDAR data Lateral boundary conditions – same for Dev and Dev2 Control design Initialize Dev and Dev2 runs at exact same time Reset dev and dev-2 background field at 1000z every 48 h (even Julian dates) Ensure against any computer logistics differences Evolution of dev vs. dev2 is different Example – buddy check QC in each cycle may occasionally differ for non-TAMDAR data Slight difference in gravity waves Can propagate difference throughout domain Shows up in sfc temps, convection, esp. 925, 850 mb

40. Part 2 – Statistical results

41. Verification regions for FSL-RUC TAMDAR impact Large region (eastern half of US) -- 38 RAOB sites Small region (Great Lakes) includes 14 RAOBs

42. Positive difference means CNTL experiment with profiler data had lower error than the EXP-P no-profiler experiment Wind forecast ‘errors’ - defined as rawinsonde vs. forecast difference Cntl = using all obs Exp = deny profiler obs Difference in errors between Cntl and Exp experiments w/ RUC - 4-17 Feb 2001 Anx Fit - ‘truth’

43. Temperature shows notable improvement for 850 mb, 3-h forecast in large (E.US) region

44. Even clearer improvement in the small (Gt Lakes) region

45. Temperature bias: small improvement for 850 mb, 3-h forecast

46. Much improved temperature bias in small region

47. Temperature: some improvement for 700 mb, 3-h forecast

48. Winds: not much difference

49. Relative Humidity: not much difference

50. Only 00z times -------------- V m/s average diff by level pres 0h-an 12h 3h 6h 9h 1h 850 0.04 0.02 -0.07 -0.08 0.02 0.04 0.07 -0.01 0.09 700 -0.01 0.01 0.08 0.03 -0.05 -0.04 -0.02 0.03 0.06 500 -0.07 0.00 0.04 -0.04 -0.08 -0.04 -0.06 0.04 -0.08 400 -0.03 0.07 0.05 -0.06 -0.03 -0.02 0.03 -0.06 -0.04 300 0.00 0.06 0.04 0.09 -0.01 0.00 0.05 -0.02 0.00 250 0.00 -0.07 0.04 0.02 -0.05 0.03 0.02 0.00 0.05 200 0.01 0.07 -0.01 -0.04 -0.07 0.03 0.02 -0.01 0.04 150 0.02 0.01 0.07 -0.01 0.01 0.05 -0.01 -0.02 0.04 Results – wind – Great Lakes region only 1 March – 12 April 2005

51. Only 00z times -------------- T deg C average diff by level pres 0h-anx 12h 3h 6h 9h 1h 850 0.13 -0.02 0.23 0.11 -0.02 0.36 700 0.05 0.00 0.06 0.04 0.01 0.03 500 0.01 -0.01 0.04 0.01 -0.01 0.05 400 -0.01 0.00 -0.02 0.00 0.00 0.01 300 -0.02 0.00 -0.03 -0.02 -0.01 -0.01 250 -0.05 -0.01 -0.03 -0.02 0.00 -0.01 200 -0.03 0.01 -0.02 -0.04 -0.01 -0.03 150 0.03 0.03 -0.01 -0.02 0.04 0.01 Results – temperature – Gt Lakes region only 1 March – 12 April 2005

52. FSL-RUC TAMDAR impact experiment results (as of 12 April 2005) Impact experiments must be conducted such as to show value added to other existing observations RUC well-suited for this Real-time parallel cycles at FSL (Dev/Dev2) have provided well-controlled experiments and results Accelerated evaluation process Results are very preliminary and during TAMDAR shakedown phase Temperature impact strongest ~20% reduction of 3h forecast error RH impact less but positive No impact for wind Diurnal variations – more 3h impact at 00z than 12z Results will improve with: Improved TAMDAR data Future use of reject lists (updated weekly?)