RUC/Rapid Refresh Development and Testing Including TAMDAR and Radar Reflectivity Impact Studies NOAA Earth System Research Lab (ESRL) Global Systems Division (GSD) Boulder, CO NOAA/ESRL/GSD Stan Benjamin Steve Weygandt John M. Brown Tanya Smirnova Dezso Devenyi Kevin Brundage Georg Grell Steven Peckham Tom Schlatter Tracy Lorraine Smith NCEP/EMC – Geoff Manikin Major transitions: RUC13 change package – ~Jan 2008 – radar reflectivity assimilation, TAMDAR, current status, testing RUC changes in current testing Rapid Refresh WRF testing in planned testing in Current impl plan – FY09 – 4Q

RUC Hourly Assimilation Cycle Time (UTC) 1-hr fcst Background Fields Analysis Fields 1-hr fcst RUC 3dvar Obs 1-hr fcst 3dvar Obs Cycle hydrometeor, soil temp/moisture/snow plus atmosphere state variables Hourly obs in 2008 RUC Data Type ~Number Rawinsonde (12h) 80 NOAA profilers 30 VAD winds PBL – prof/RASS ~25 Aircraft (V,temp) TAMDAR (V,T,RH) Surface/METAR Buoy/ship GOES cloud winds GOES cloud-top pres 10 km res GPS precip water ~300 Mesonet (temp, dpt) ~7000 Mesonet (wind) ~ 600 METAR-cloud-vis-wx ~1500 Radar / lightning 2km

RUC Rapid Refresh (2009) Hourly NWP Update for: - CONUS - AK/Can - Pac/Atl - Caribbean Current RUC CONUS domain Planned approx Rapid Refresh domain NWP updated hourly w/ latest obs Aviation / transportation Severe weather Decision support tools

RUC History – NCEP (NMC) implementations First operational implementation of RUC - 60km resolution, 3-h cycle 1998 – 40km resolution, 1-h cycle, - cloud physics, land-sfc model 2002 – 20km resolution - addition of GOES cloud data in assimilation 2005 – 13km resolution, new obs (METAR clouds, GPS moisture), new model physics 2008 (Jan) – Assimilation of radar reflectivity, improved thunderstorm forecasting 2009 – WRF-based Rapid Refresh to replace RUC NOAA/GSD testing – precedes NCEP ops by years

Jan 2008 Changes for oper RUC upgrade Assimilation Use of radar reflectivity in RUC diabatic digital filter initialization in RUC model Mesonet winds using mesonet provider uselist TAMDAR aircraft observations (TAMDAR impact para RUC tests at GSD) Model physics RRTM longwave radiation Mod to Grell-Devenyi – decrease areal coverage Mod to RUC land-sfc model – fresh snow density Post-processing – add reflectivity fields

RUC change package – Jan continued Model changes  RRTM longwave radiation, replacing Dudhia LW (largely eliminates warm bias in RUC) Mod to Grell-Devenyi convective scheme (reduce excessive areal coverage for light precip)  Mod to snow component of RUC land-sfc model for snow density (decrease excessively cold 2m air temps over fresh snow cover at night)

RRTM Longwave Radiation in RUC Upgrade Effect on 2-m temperature forecasts Much decreased warm bias, esp. at nighttime 1-month comparison 14 May –13 June 07 Eastern US only RUC para – RRTM LW RUC oper – Dudhia LW 2-m temp bias (obs – forecast) COLD WARM 12h 3h UTC

9h fcst - Oper RUC9h fcst – para RUC Anx valid 09z Better 2m temp forecast From para RUC w/ RRTM LW

9h fcst - Oper RUC9h fcst – para RUC Anx valid 09z Better 2m temp forecast From para RUC w/ RRTM LW Today

Data type used in the RUC model Number (resolution) of observations Frequency of data insertion Rawinsonde80/12h NOAA/PBL wind profiler /1h VAD winds /1h Aircraft (V, T) /1h Surface / METAR /1h Buoy / ship /1h GOES precipitable water /1h GOES cloud drift winds /1h GOES cloud-top pressure(10-km resolution)/1h SSM/I precipitable water /6h GPS precipitable water~300/1h Mesonetwork data~6000/1h METAR cloud visibility~1500/1h Radar / lightning(4-km resolution)/1h GOES / POES radiancesWRF-Rapid Refresh + GSI/ 1h

RUC cloud analysis since 2005 – Use of cloud / hydrometeor observations to modify cycled cloud / hydrometeor fields Background Cloud water + cloud ice Cloud assessment (YES/NO/UNKNOWN) from observations (METAR/sat/radar)

RUC change package – 2007 – one more thing  Assimilate mesonet winds from accepted provider uselist Allows about 600 additional wind observations to be used hourly in RUC Primary mesonet providers (e.g., Citizens Weather, AWS) have common siting problems Mesonet winds turned off for RUC13 implementation in 2005 New RUC treats each mesonet provider separately Station-by-station reject list will allow use of many Citizens-Wx, AWS stations in future Poster – Mesonet QC monitoring web site – Benj et al

–Most significant weather problem for aviation operations. –Requirement for improved hourly-updated 2-8h forecast for air traffic management. –Current situation Large uncertainty even for short lead time Very small-scale, deterministic forecasts difficult – Severe convective weather -- significant impact on human activities The Thunderstorm Prediction Problem for Aviation and Severe Weather

Two separate, but related problems: – Convective initiation – Convective evolution (including decay) The Thunderstorm Prediction Problem

Two separate, but related problems: – Convective Initiation – Convective evolution The Thunderstorm Prediction Problem Need very accurate forecast of mesoscale environment

Two separate, but related problems: – Convective Initiation – Convective evolution –Excellent near-surface physics –High frequency assimilation of asynoptic observations (surface METAR and mesonet, wind profilers, aircraft) The Thunderstorm Prediction Problem Need very accurate forecast of mesoscale environment

Two separate, but related problems: – Convective Initiation – Convective evolution –Excellent near-surface physics –High frequency assimilation of asynoptic observations (surface METAR and mesonet, wind profilers, aircraft) The Thunderstorm Prediction Problem Need very accurate forecast of mesoscale environment Nighttime convection often elevated – even harder

WSR-88D provides invaluable observations Much improved operational warning, nowcasting Operational NWP community slow to utilize radar reflectivity data in models – Real-time Level II (3-d) data access issues – Difficulties assimilating reflectivity Once convection is ongoing…

Forward integration, full physics Diabatic Digital Filter Initialization (DDFI) -30 min -15 min Init +15 min RUC model forecast Backwards integration, no physics Obtain initial fields with improved balance

Forward integration, full physics Apply latent heating from radar reflectivity, lightning data Diabatic Digital Filter Initialization (DDFI) add assimilation of radar data -30 min -15 min Init +15 min RUC model forecast Backwards integration, no physics Obtain initial fields with improved balance, vertical circulations associated with ongoing convection

1.Force precipitation in areas with radar echoes –Specify 3D latent heat from reflectivity –Apply latent heat within forward integration of diabatic digital filter initialization (DDFI) –Replace temperature tendency from cumulus parameterization and explicit microphysics –Moisten echo region, do not add hydrometeors 2.Suppress convection in echo-free areas –Create convective suppression mask (> 300 mb deep layer > 100 km from existing convection –Inhibit cumulus parameterization during DDFI and 1 st 30 min. of model integration RUC reflectivity assimilation procedure

1.Minimal shock to model –Coherent wind, temperature and moisture fields evolve in response to heating within DDFI 2.Very little additional computer cost –DDFI already used to control noise 3.Independent of model or physics packages –Can be applied to Rapid Refresh WRF, other models Advantages of radar assimilation procedure

RUC radar assimilation test case NSSL radar reflectivity mosaic Test case 00z 8 Jan 2007 Experiment (EXP) –LH temperature tendency in DDFI (no moistening, no suppression yet) Control (CNTL) –Standard initialization (no radar assimilation)

NSSL 3-km radar reflectivity (dbz) K=15 LH temp. tend. (K / 15 min) Contour interval = 0.5 K00z 8 Jan 2007

Reflectivity experiments within RUC -CNTL standard initialization – no radar init -EXP LH nudging within DDFI no adding hydrometeors from reflectivity no moistening in reflectivity region

Low-level Convergence K=15 U-comp. diff (EXP - CNTL) K=35 U-comp. diff (EXP - CNTL) Upper-level Divergence Contour interval = 0.2 m/s

1700 UTC 27 Jan 2004 NSSL mosaic Z = 5 kft NSSL mosaic Z = 10 kft NSSL mosaic Z = 20 kft “Pyramids of silence” complementing “cones of radar data” Latent heating in diabatic forward DFI step specified only where 3-d radar data available

cint = 0.5 mm EXP min. prec. 00z 3-km refl (dBZ) NSSL mosaic

CNTL min. prec. cint = 0.5 mm EXP min. prec.

CNTL min. prec. cint = 0.5 mm EXP min. prec.

CNTL min. prec. cint = 0.5 mm EXP min. prec.

cint = 0.5 mm EXP min. prec. 02z 3-km refl (dBZ) NSSL mosaic

NSSL reflectivity min. accum. precipitation CNTLEXP Contour interval = 0.5 mm 02z 8 Jan 2007

Radar obs – 00z 25 Mar 2007 With radar assimilation Assimilation improvement most clear in reflectivity field (smeared in precipitation field) Real-time parallel testing at GSD (started February 2007) Added simulated RUC radar reflectivity field RUC 3-h forecasts valid 00z 25 Mar 2007 No radar assimilation

03z 1 Mar 2007 Obs. reflectivity 03z 1 Mar 2007 Analyzed 700 mb vert. vel. from RUC with and without radar assimilation Real-time case from 03z 1 March 2007 NSSL Reflectivity mosaic 0-h vert. vel. 03z 1 Mar h vert. vel. No radar assim Radar assim

Obs. reflectivity 06z 1 Mar 2007 Radar assim No radar assim 06z 1 Mar h fcst accum. precip. from RUC with and without radar assimilation Real-time case from 03z 1 March z 1 Mar h precip. NSSL Reflectivity mosaic

Radar reflectivity assimilation Part 2 – convection suppression Define suppression areas as follows: No reflectivity > 20 dbZ within 100 km Depth of radar data > 300 hPa Complemented by GOES fully clear areas No coverage Allow convection Suppress convection Accomplish in RUC model: Specify minimum cap depth as 0 hPa in DFI step and first 30 min in actual forecast

No coverage Allow convection Suppress convection Convective suppression example Real-time case from 12z 7 June 2007 NSSL reflectivitySuppression mask

NSSL 3-h precipitation Radar Assimilation Control convective suppression - How does it work? – Reduces latent heating, vert. motion in erroneous conv areas Real-time 3-h forecasts valid 15z 7 June 2007 Valid 15z 7 June 2007 Convective suppression example

0900z NSSL Q2 composite refl RUC “analysis” composite refl (actually 1h fcst) 0900z Tues 17 July 2007

0900z NSSL Q2 composite refl RUC “analysis” composite refl (actually 1h fcst) 0900z Tues 17 July 2007

Valid 1200z Tues 17 July 2007 RUC 3h forecast composite refl NSSL Q2 composite refl 1200z

Precip – 3h –09-12z Tues 17 July 2007 NSSL Q2 QPE RUC 3h forecast with RADAR ASSIM

Precip – 3h –09-12z Tues 17 July 2007 NSSL Q2 QPE RUC 3h forecast with RADAR ASSIM

Precip – 3h –09-12z Tues 17 July 2007 NSSL Q2 QPE RUC 3h forecast without RADAR ASSIM

Precip – 3h –09-12z Tues 17 July 2007 NSSL Q2 QPE without RADAR ASSIM RUC 3h forecasts with RADAR ASSIM

NSSL Q2 composite refl RUC “analysis” composite refl (actually 1h fcst) 1200z Tues 17 July z

NSSL Q2 composite refl 1800z Tues 17 July 2007 RUC 3h forecast with RADAR ASSIM 1500z

NSSL Q2 composite refl 1500z Tues 17 July 2007 RUC 6h forecast with RADAR ASSIM 1800z RUC forecast unable to hold onto squall line - result of Grell-Dev limitations, not assim

Precip – 3h –12  15z Tues 17 July 2007 without RADAR ASSIM RUC 3h forecasts with RADAR ASSIM 12-15z NSSL Q2 QPE

Precip – 3h –15->18z Tues 17 July 2007 NSSL Q2 QPE without RADAR ASSIM RUC 6h forecasts with RADAR ASSIM 1800z

Sfc Temp – 21z Tues 17 July 2007 without RADAR ASSIM RUC 9h forecasts with RADAR ASSIM 2100z Evaporative cooling - improved cold pool

Radar assim. 3-h fcst. precip. 12z 12 Feb 2007 No radar assim. Obs. reflectivity 11z 12 Feb h fcst. precip. 12z 12 Feb h fcst accum. precip. from RUC with and without radar assimilation Real-time case from 09z 12 Feb 2007 RUC Radar refl assim w/ DDFI effective on winter and summer events no added CPU

Radar assimilation impact on 3-h precipitation skill scores Significant improvement in ETS and bias Spring - daytime

Radar assimilation impact on 3-h precipitation skill scores Summer - overnight

Radar assimilation impact on 3-h precipitation skill scores 4 x 0-3h vs. 1x0-12h Summer - daytime

WSI Nowrad 0500 UTC 23 Mar 2005 Radar reflectivity (dbz)

NLDN 0500 UTC 23 Mar 2005 (strikes/40 min/gridbox) Lightning data

Radar data Lightning data Rainwater from radar/lightning data UTC 23 Mar 2005

---- one more time Changes for oper RUC upgrade Assimilation Use of radar reflectivity in RUC diabatic digital filter initialization in RUC model Mesonet winds using mesonet provider uselist TAMDAR aircraft observations (TAMDAR impact para RUC tests at GSD) Model physics RRTM longwave radiation Mod to Grell-Devenyi – decrease areal coverage Mod to RUC land-sfc model – fresh snow density Post-processing – add reflectivity fields

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

AMDAR and TAMDAR “AMDAR” (Automated Meteorological Data and Recording) – are automatically sent from commercial aircraft, mostly large jets “TAMDAR” (Tropospheric AMDAR) – automatic reports from (currently) ~50 turboprops flying regionally in the US Midwest –Provided by AirDat LLC –Agreement between Northwest Airlines (Mesaba – regional subsidiary) and AirDat LLC –New agreement between NWS/FAA and AirDat for use of TAMDAR

Coverage is limited to major hubs below 20 Kft, (without TAMDAR)

Below 20 Kft, with TAMDAR – better regional coverage in the Midwest

TAMDAR Variables TAMDAR measures temperature and winds aloft, as does the rest of the AMDAR fleet In addition, TAMDAR measures –water vapor –turbulence (not discussed) –icing (not discussed)

Parallel real-time RUC cycles to monitor TAMDAR impact since 2005 “dev2” – includes TAMDAR and all other data typically assimilated by the RUC “dev” – lacks only TAMDAR data Both cycles use NAM boundary conditions Both run at 20-km, but are otherwise similar to the operational 13-km runs Background fields are set equal every 48 h

Great Lakes Region includes 13 RAOBs Eastern US Region includes 38 RAOBs National Region is the RUC domain (CONUS and adjacent) Results today are from the Great Lakes region, where most current TAMDAR- equipped aircraft fly

TAMDAR – regional aircraft with V/T/RH obs GSD impact study with RUC parallel cycles (ongoing) 10-30% reduction in RH, temperature, wind fcst error w/ TAMDAR assimilation 3h Fcst errors – RUCdev (no TAMDAR), RUCdev2 (w/ TAMDAR) noTAM wTAM Temp RH Wind wTAM noTAM

Temperature RMS error time series, 3-h forecasts, Great Lakes Region, surface to 500 mb TAMDAR impact up to 0.2 K Red: dev RMS error Blue: dev2 RMS error Black: difference

Temperature RMS error profile, 3-h forecasts, Great Lakes Region TAMDAR impact max 0.4K at 900 mb -- Inversion level, cloud ceiling For 4 months in Jan-May 2007 Thus TAMDAR impact represents about 35% of the maximum expected 3-h T forecast error reduction at 900 mb.

Wind RMS error profile, 3-h forecasts, Great Lakes Region TAMDAR impact: 0.25 m/s at 700 mb Analysis fit to RAOBs is ~2.2 m/s Thus, TAMDAR impact on 3-h wind forecasts represents a 15% reduction in 3-h wind forecast error at 700 mb

RH error time series, 3-h forecasts, Great Lakes Region, surface to 500 mb TAMDAR impact up to 2% RH

RH RMS error profile, 3-h forecasts, Great Lakes Region TAMDAR impact ~2 %RH below 550 mb

A look ahead (2) AirDat will install TAMDAR on additional fleets over the next several months Covering Alaska and the Western US These fleets include some jet aircraft –higher altitudes, speeds (implications??) –better heading => reduced wind errors GSD will evaluate the quality and impact of these data (with FAA funding) (Unfortunately, these new data will not be available beyond GSD, per AirDat)

Ed Szoke (GSD) TAMDAR evals - Overview TAMDAR soundings have been shown to be useful for forecasting Talks at the last SLS Conference and previous Annual Meetings WFO Green Bay helps maintain the official NOAA TAMDAR web page at In this talk we focus on the impact on NWP: Evaluation of RUC precipitation forecasts for runs with and without TAMDAR for significant weather events –Mostly a subjective evaluation, but objective scoring for 2007 cases

Still one of the better cases for TAMDAR impact Oct 2005: heavy precip in the Upper Midwest. Flooding reported in Minnesota to northern Wisconsin. Case 1: 4 October 2005 – 2100 UTC Surface analyses and reflectivity

Very sharp cut off to the precip in WI and a huge gradient with a 2-3” max. NPVU estimated precipitation for 6-h ending 0000 UTC 5 October 2005

Both runs forecast too much precip in southern half of Wisconsin, but the RUC run with TAMDAR correctly forecasts more precip (small spots of >1.00”) across the northern half of the state. RUC forecasts from the 4 October UTC runs 6-h total precipitation ending 0000 UTC 5 October No TAMDAR With TAMDAR

Sounding comparison: RUC 6-h forecasts with (labeled dev2) and without (labeled dev1, in black) TAMDAR, compared to the RAOB for Detroit (green) at 0000 UTC 5 Oct 05. Incorrect dry layer in the dev1 (noTAM) forecast.

Same comparison but for Peoria, Illinois. The RUC run with TAMDAR is closer to the RAOB especially at and below 700 mb.

Heavy precip continues in the same areas Case 1/part 2: 5 October 2005 – 0300 UTC Surface analyses and reflectivity

NPVU estimated precipitation for 6-h ending 0600 UTC 5 October 2005

For this period the RUC run that used the TAMDAR data is a much better forecast with a very sharp cut off to the precipitation in Wisconsin and a better location for the heavy precip. RUC forecasts from the 5 October UTC runs 6-h total precipitation ending 0600 UTC 5 October No TAMDARWith TAMDAR No TAMDAR With TAMDAR

Case 4: 22 March 2007 – 0000 UTC Surface analyses and reflectivity Strong spring storm with lots of severe weather

22 March 2007 – 0300 UTC Surface analyses and reflectivity

SPC severe reports for 24-h ending 1200 UTC/22 March 2007

Some differences are seen – these are outlined in the forecasts The RUC forecast that uses TAMDAR is generally better except within the orange oval area, where no precipitation fell. RUC forecasts from the 22 March UTC runs 6-h total precipitation ending 0600 UTC 22 March No TAMDARWith TAMDAR No TAMDAR

Case 5: 21 June 2007 – 2100 UTC Surface analyses and reflectivity Strong convection with many reports of severe weather

22 June 2007 – 0000 UTC Surface analyses and reflectivity

SPC severe reports for 24-h ending 1200 UTC/22 June 2007

Main difference is the precipitation in IL and IN predicted by the RUC run without TAMDAR compared to almost nothing in the run with TAMDAR. Verification showed that no precipitation fell in the IL/IN area. RUC forecasts from the 21 June UTC runs 6-h total precipitation ending 0000 UTC 22 June No TAMDAR With TAMDAR

Sounding comparison for 6-h forecasts for RUC with TAMDAR (dev2) vs RUC without TAMDAR (dev) compared to the DVN RAOB at 0000 UTC 22 June 2007

Sounding comparison for 6-h forecasts for RUC with TAMDAR (dev2) vs RUC without TAMDAR (dev) compared to the ILX RAOB at 0000 UTC 22 June 2007

Summary When we began to examine precipitation forecasts in late 2005 were impressed by the 4-5 October 2005 case with significantly better forecasts by the RUC run that used TAMDAR –But that remains our best case More typically, we see much smaller impacts These tend to favor the RUC run that uses TAMDAR, but not always –And sometimes mixed...forecast better in some spots but not in others Objective scoring of the precipitation forecasts that began in 2007 agrees with our overall subjective impression –Longer-term statistics show relatively small differences generally favoring the RUC run that uses TAMDAR –But on a case by case basis can see greater differences in the scores

Decreased vertical resolution decreases TAMDAR impact on 3-h T forecasts by ~30% at 750 mb ~10% at 900 mb Effect of vertical resolution on TAMDAR 3-h Temperature forecast impact Each curve shows the amount that TAMDAR reduces the RMS error. Low-res reduces the error less => less TAMDAR impact.

Current RUC CONUS domain RUC Rapid- Refresh ( ) Continental situational awareness model Hourly NWP Update for: - CONUS - AK/Can - Pac/Atl - Caribbean Planned Rapid Refresh domain

RUC to Rapid Refresh North American domain (13km) GSI (Gridpoint Statistical Interpolation) WRF model (ARW dynamic core almost certainly) CONUS domain (13km) RUC 3dvar RUC model

Rapid Refresh Hourly Assimilation Cycle Time (UTC) 1-hr fcst Background Fields Analysis Fields 1-hr fcst GSI Obs 1-hr fcst GSI Obs Cycle hydrometeor, soil temp/moisture/snow plus atmosphere state variables Hourly obs used in RR Data Type ~Number Rawinsonde (12h) 80 NOAA profilers 30 VAD winds PBL – prof/RASS ~25 Aircraft (V,temp) TAMDAR (V,T,RH) Surface/METAR Buoy/ship GOES cloud winds GOES cloud-top pres 10 km res GPS precip water ~300 Mesonet (temp, dpt) ~7000 Mesonet (wind) ~ 600 METAR-cloud-vis-wx ~1500 Radar / lightning 2km Sat radiances – AMSU-A/B, GOES QuikSCAT

WRF physics options -- All available with both ARW and NMM cores w/ WRFv All combinations tested with WRF-RR core-test Phase 1 - Default NMM physics Phase 2 - RUC-like physics NAM-NMMRapidRefresh Explicit cloudsFerrierThompson-NCAR Sub-grid convection Betts-Miller- Janjic Grell-Devenyi Land-surfaceF77 version of Noah (“99” LSM) RUC-Smirnova or Noah Turbulent mixingMellor-Yamada- Janjic

1-h cycling of atmospheric (including hydrometeor) and land surface model fields Update cycled fields with all available hourly observations Utilize GSI satellite radiance assimilation scheme Build in “RR-specific” components: 1) ‘pre-forecast’ diabatic digital filter initiation (DDFI) 2) cloud analysis (satellite, METAR, radar, LTG obs) 3) surface obs assimilation (BL depth, coast-lines) 4) Force convection from radar, lightning data in model DDFI after GSI pre-processing RR application of GSI assimilation

Example from the preliminary Rapid Refresh real time cycle Analysis at 1200 UTC 29 May 2007 over RR domain. Wind field at model level 15 colored according to potential temperature. Each color represents a 5-K interval of potential temperature, with purple representing from K (north) and red representing from K (south).

12h forecast - 2m temperature 12z 12 May 2007 Rapid Refresh RUC More preliminary RR results

Rapid Refresh Data Assimilation timeline Fall 2006 – summer 2007 – Cycle WRF using GSI over RR domain - Use of WRF version 1.2 (WRFSI) - Testing of new surface assimilation and cloud analysis modules Fall 2007 Cycling over NAmerica with all observations - Update to new GSI, WRF versions; new computer - Use of NCEP prepbufr files, increase cycling frequency - Comparison with other systems, continue refinement Fall/Winter 2007/08 Full system with RR modifications - DDFI in place for chosen WRF model core - Detailed examination of cold season near surface aspects - Refinement of system toward operation skill

Spring/Summer 2008 Real-time and retro cycles - Testing of DDFI radar data assimilation - Focus on performance for convective situations - Transfer code to NCEP Winter Testing of complete RR at NCEP and GSD Summer/Fall 2009 Operational implementation at NCEP Rapid Refresh Planned Timeline

Transitioning to operations (RUC, RR) -Must run at NCEP -Must run within available computer resources and time constraints (5 min – assim, 17 min- 12h fcst) -Must be built into existing code infrastructure (e.g.: Build assimilation capability in GSI, develop probabilistic products within SREF framework)

Upcoming GSD tasks  Develop Rapid Refresh – North American 1h update 4-d assim/model – toward NextGen 4-d database  Tied with EMC more than before  Test and recommend physics options  Developed diabatic DFI (digital filter initialization) in WRF-ARW to allow RR 1h cycle  GSI assimilation with RUC-specific enhancements Work with EMC, NCEP centers, NWS, other RUC/FAA/AWRP users, DTC, WRF community on forecast evaluation and improvement Major transition: Rapid Refresh planned implementation 2009 Evaluation at NCEP This is where you folks come in!

FUTURE: High Resolution Rapid- Refresh (HRRR) Proposal for time frame Nested high-resolution domain (2-4 km) within RR Explicit depiction of convection (no cum. param.) Hourly or sub-hourly observation updating Need to initialize storm scale details… Use radial velocity (3DVAR, 4DVAR, retrieval techniques), reflectivity (polarimetric information), and lightning data to specify wind, temperature, and hydrometeors Storm building, adjustment, and removal

HRRR-ARW 6h fcst radRUC IC HRRR- ARW 6h fcst Oper RUC IC 06z obs refl NSSL Q2 product Results from sample real- time HRRR test Forecasts initialized 0000z 16 Aug 07 Radar-enhanced RUC critical for HRRR forecast

RUC/Rapid Refresh Development and Testing Including TAMDAR and Radar Reflectivity Impact Studies Major transitions: RUC13 change package – ~Jan 2008 – radar reflectivity assimilation - TAMDAR - Improved radiation, convection physics in RUC Rapid Refresh planned for FY09 WRF ARW, GSI, North America Ensemble Rapid Refresh proposed by 2012, to use ESMF framework High-Res Rapid Refresh (HRRR) – proposed to NCEP by km hourly updated 12h forecast In testing at GSD Covering NE Corridor

Planned implementation of TAMDAR sensors by AirDAT LLC into Horizon and PenAir Saab 340s

First Alaska TAMDAR data - 12 June 2007

AMDAR -24h Mar 07

SATELLITE DATA EXPERIMENTS FOR RAPID REFRESH -Both HIRS and MSU -Usage of CRTM -Preliminary experiments with NAM satellite files over CONUS domain using RUC background fields -Case of 11 April 2006, 1200 UTC. SATELLITE DATA EXPERIMENTS FOR RAPID REFRESH -Both HIRS and MSU -Usage of CRTM -Preliminary experiments with NAM satellite files over CONUS domain using RUC background fields -Case of 11 April 2006, 1200 UTC. SATELLITE DATA EXPERIMENTS FOR RAPID REFRESH Both HIRS and MSU Usage of CRTM Preliminary experiments with NAM satellite files over CONUS domain using RUC background fields Case of 11 April 2006, 1200 UTC. SATELLITE DATA EXPERIMENTS FOR RAPID REFRESH Both HIRS and MSU -- Use of Community Radiative Transfer Model (CRTM) Preliminary experiments with NAM satellite radiance and bias files over CONUS domain using RUC background fields. Case of 11 April 2006, 1200 UTC. Difference satellite minus no-satellite data; specific humidity. Model level=10.

Some Special Challenges – Aviation Forecasts with Expanded Domain Arctic low stratus – will be difficult to predict explicitly -- but we will try Sea ice – Leads, ponding, etc. Taiga in spring—snow cover and frozen ground under warm forest canopy Tundra in the land-surface model – prediction of surface conditions in the Far North Tropical cyclones—initializing, track prediction, intensity Sparse data over land – risk of “climate drift” in the model

Joint GSD-NCAR plans for HRRR model development High-Res Rapid Refresh (HRRR) 2-3 km, hourly updating, assimilation of hourly radar reflectivity Based off radar-assimilating Rapid Refresh (currently radar-assimilation RUC) Plans for FY08 1Q-2Q GSD, NCAR Rerun cases and test periods summer sources of external model grids GSD radar-assimilating RUC NCEP operational RUC (no radar assim) 3 variations of 3-km WRF-ARW ARW as is (already tested in real-time by GSD) ARW with FDDA (NCAR lead on development) ARW with radar-enhanced DFI (GSD lead)

Real-time HRRR for CoSPA for 3Q, 4Q FY08 Real-time HRRR forecasts over regional domain over NE Corridor area for May-August period –Forecasts out to 12h, reinitialized every 1-3h using radar-enhanced RUC or Rapid Refresh initial conditions –15-min output for selected 2-d fields including surface fields (2m temp/dewpoint, 10m winds), reflectivity, precipitation –Run at GSD, backup at NCAR Experimental HRRR output to MIT/LL, NCEP Storm Prediction Center, AWC (possible), others

Other FY08 activities toward HRRR Case studies from summer 2007 cases –NCAR - lead, GSD collaborating Case study testing of revised radar assimilation methods Development of simple forecast metrics for development retrospective tests and case studies –NCAR and GSD