Real-Time Impact Experiments of RUC Model - Detailed Analysis

1. Stan Benjamin, Bill Moninger, Tracy Lorraine Smith, Barry Schwartz, Brian JamisonNOAA Forecast Systems LaboratoryBoulder, CO GLFE Real-time TAMDAR Impact Experiments with the 20km RUC - August 2005 report

2. Outline of talk Part 1: General description of RUC 1h cycle - Data assimilated - Spatial effects of assimilated data Design of RUC parallel experiments – Dev / Dev2 New example of Dev/Dev2 difference fields (Brian Jamison) Part 2: Actual statistical results

3. Purpose for Rapid Update Cycle (RUC) model run operationally at NCEP • 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

4. RUC Assimilation System Updates 1-h cycle Analysis 3DVAR on q/s surfaces (y/c/Zunb, q, 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 METAR clouds, 1-h forecast background) Cycling of cloud, hydrometeor, land-surface fields

5. 3DVAR 3DVAR 3DVAR 3DVAR 3DVAR Obs Obs Obs Obs Obs RUC Hourly Assimilation Cycle 1-hr fcst 1-hr fcst 1-hr fcst 1-hr fcst 1-hr fcst 1-hr fcst Background Fields Analysis Fields Time (UTC) 11 12 13 14 15 16

6. 12-h fcst 12-h fcst 3DVAR 3DVAR 3DVAR 3DVAR 3DVAR Obs Obs Obs Obs Obs RUC Hourly Assimilation Cycle 9-hr fcst 9-hr fcst 9-hr fcst 9-hr fcst Background Fields Analysis Fields EC Time (UTC) 11 12 13 14 15 16

7. Cloud anx variables 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/METAR1500-1700 / 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

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

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

10. RUC20 • Wind forecast • Accuracy • Sept-Dec • 2002 6 9 1 3 12 Analysis ~ ‘truth’ 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

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

12. Upgrades to RUC model 28 June 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 All upgrades used in 2005 FSL 20-km dev/dev2 RUC versions

13. 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

14. 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 it does not allow accumulation effects of TAMDAR data beyond 48h)

16. Dev-Dev2 difference – 0h analysis Init 1800z 24 Aug 2005 – 500 mb 1000-1800z

22. Part 2 – Statistical results

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

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

25. TAMDAR evaluation phases • Phase 1 – 9 Feb – 21 April 2005 • Winter/early spring – lower vertical resolution • Phase 2 – 22 April – 1 June • Spring – higher vertical resolution • Phase 3 – 2 June – 22 July • Summer – higher vertical resolution • Phase 4 – 23 July – 24 August • Summer – lower vertical resolution

26. TAMDAR evaluation w/ RUC parallel cycles • Summary evaluation over each of 4 phases • Screened any dates with questionable results • If logs showed missing 1h RUC runs for either dev or dev2 cycles • If dev/dev2 verification stat differences over full national domain for winds > 0.2 m/s averaged over 8 mandatory levels • Results only shown for 00z verification, GtLakes region. Less impact for 12z verif and/or E.US reg.

27. 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?)

28. 850 mb temp - clear improvement in the small (Gt Lakes) region Phase 1 – Jan-Apr 05

29. Phase 3-4 – Jun-Aug 05

30. Dev – Dev2 difference in wind error Positive  positive impact from TAMDAR assim -- 30% Reduction of 3h fcst err TEMP – 850 mb

31. Temperature: some improvement for 700 mb, 3-h forecast Phase 1 – Jan-Apr 05

32. TEMP – 700 mb Phase 3-4 – Jun-Aug 05

33. Winds: not much difference Phase 1 – Jan-Apr 05

34. 850 mb 700 mb Phase 3-4 – Jun-Aug 05

35. WIND – averaged over 850-500 mb -- 10% Reduction of 3h fcst err

36. Relative Humidity: not much difference Phase 1 – Jan-Apr 05

37. 850 mb 700 mb Phase 3-4 – Jun-Aug 05

38. RH -- 12% Reduction of 3h fcst err

39. FSL-RUC TAMDAR impact experiment results • (as of 25 August 2005) • Results (TAMDAR impact) have improved during continued TAMDAR shakedown phase • Recent results – June-Aug • Temperature impact strongest at 850 mb • ~15-20% reduction of 3h forecast error • Less positive impact at 700-500 mb • RH impact at 850 mb • ~12% reduction of 3h forecast error • Negative impact at 700-500 mb • Better Temp/RH impact w/ higher vert resolution • Wind impact – averages 10% reduction of 3h fcst errors in 850-700-500mb layer • Diurnal variations – more 3h impact at 00z than 12z • Results should improve further with: • Further improvements in TAMDAR data accuracy • ascent/descent temp biases, wind errs?