Hourly Updated NOAA NWP Models

35th Conference on Radar Meteorology30 September 2011

Development and Testing of Improved Radar-Data Assimilation Techniques for the Rapid Refresh and High-Resolution

Rapid Refresh (HRRR)

NOAA/ESRL/GSDCurtis Alexander, David Dowell,

Ming Hu, Steve Weygandt, Tanya Smirnova,

Stan Benjamin, John Brown, Eric James and Patrick Hofmann

Hourly Updated NOAA NWP Models

13km Rapid Refresh

(mesoscale)

13km RUC (mesoscale)

3km HRRR (storm-scale)

RUC – current oper Model, new 18h fcst every hour

High-Resolution Rapid Refresh Experimental 3km nest inside RR, new 15-h fcst every hour

Rapid Refresh (RR) replaces RUC at NCEP in 2011 WRF, GSI with RUC features

NOAA/ESRL/GSD/AMB ModelsModel Run at: Domain Grid

PointsGrid

SpacingVertical Levels

Vertical Coordinate

Lowest Level

Pressure Top

RUC GSD, NCO CONUS 451 x

337 13 km 50 Sigma/ Isentropic 5 m ~50 mb

RR GSD,EMC

North America

758 x 567 13 km 50 Sigma 8 m 10 mb

HRRR GSD CONUS 1799 x 1059 3 km 50 Sigma 8 m 85 mb

Model Version Time-Step Forecast Length Initialized Boundary

ConditionsRun Time

# of CPUs

RUC N/A 18 s 18 hrs Hourly (cycled) NAM ~25 min 36

RR WRF-ARW v3.2+ 60 s 18 hrs Hourly

(cycled) GFS ~25 min 160

HRRR WRF-ARW v3.2+ 15-20 s 15 hrs Hourly

(no-cycle) RUC ~50 min 1000

NOAA/ESRL/GSD/AMB Models

Model Version Time-Step Forecast Length Initialized Boundary

ConditionsRun Time

# of CPUs

RUC N/A 18 s 18 hrs Hourly (cycled) NAM ~25 min 36

RR WRF-ARW v3.2+ 60 s 18 hrs Hourly

(cycled) GFS ~25 min 160

HRRR WRF-ARWv3.2+ 18-23 s 15 hrs Hourly

(no-cycle) RR ~50 min 1064

Model Run at: Domain Grid Points

Grid Spacing

Vertical Levels

Vertical Coordinate

Lowest Level

Pressure Top

RUC GSD, NCO CONUS 451 x

337 13 km 50 Sigma/ Isentropic 5 m ~50 mb

RR GSD,EMC

North America

758 x 567 13 km 50 Sigma 8 m 10 mb

HRRR GSD CONUS 1799 x 1059 3 km 50 Sigma 8 m 20 mb

Model Assimilation Radar DFI Microphysics Radiation Cum

Param PBL LSM

RUC RUC-3DVAR Yes - strong Thompson RRTM/Dudhia Grell-

DevenyiBurk-

Thompson

RUC

RR GSI Yes -strong Thompson RRTM/Goddard G3 +

Shallow MYJ RUC

HRRR None: RUC I.C. No Thompson RRTM/Goddard None MYJ RUC


Model Assimilation Radar DFI Microphysics Radiation Cum

Param PBL LSM

RUC RUC-3DVAR Yes - strong Thompson RRTM/Dudhia Grell-

DevenyiBurk-

Thompson

RUC

RR GSI Yes -strong Thompson RRTM/Goddard G3 +

Shallow MYJ RUC

HRRR None: RR I.C. No Thompson RRTM/Goddard None MYJ RUC


Summary of 2011Changes & Evaluation Activities

(1) RR initialization (parent model/assimilation)(2) Latent heating strength (radar data assimilation) in RR(3) Reduced diffusion (no 6th order diffusion)(4) Raised pressure top(5) Increased min/max time step(6) Higher resolution SST data (0.083°)

HRRR Milestones• Inception over northeastern US

Sept 2007• Integration into CoSPA: Aviation Users

Spring 2008• Domain expansion to eastern US

Mar 2009• HCPF time-lagged ensemble inception May

2009• HRRR WRF-ARW updated to v3.1.1

Oct 2009• Domain expansion to CONUS

Oct 2009• HRRR WRF-ARW updated to v3.2

Apr 2010• Forecast period extended to 15 hrs

Apr 2010• Real-time multi-scale reflect. verification June

2010• Parallel (shadow) retrospective system Sept

2010• Attained ~95% reliability

Jun 2010• Reduced latency to ~2 hrs

Dec 2010

HRRR (and RR) Future Milestones• Conversion of all output to GRIB2 format

Apr 2011• Transition from RUC to RR parent model

Apr 2011• DOE-funded HRRR FTP site for energy industry May

2011• Update to WRF-ARW v3.3.1

Nov 2011• Reflectivity data assimilation at 3 km scale 2012• Assimilate Radial Velocity at 3 km scale

2012• Incorporate SatCast products at 3 km scale 2012• HRRR demo @ESRL improves

2012-2014• Rapid Refresh operational at NCEP

2011• Ensemble Rapid Refresh (NARRE) at NCEP

2014• HRRR operational at NCEP

2015?• Ensemble HRRR (HRRRE) at NCEP

2016?

HRRR (and RR) Future Milestones• Conversion of all output to GRIB2 format

Apr 2011• Transition from RUC to RR parent model

Apr 2011• DOE-funded HRRR FTP site for energy industry May

2011• Update to WRF-ARW v3.3.1

Nov 2011• Reflectivity data assimilation at 3 km scale 2012• Assimilate Radial Velocity at 3 km scale

2012• Incorporate SatCast products at 3 km scale 2012• HRRR demo @ESRL improves

2012-2014• Rapid Refresh operational at NCEP

2011• Ensemble Rapid Refresh (NARRE) at NCEP

2014• HRRR operational at NCEP

2015?• Ensemble HRRR (HRRRE) at NCEP

2016?

Radar Reflectivity Data Assimilation

Model Forecasts (WRF-ARW)

Reflectivity = Diagnostic variable ESTIMATE

Hydrometeors are prognostic variables in microphysics scheme(s)

Radar Observations (WSR-88D)

Reflectivity = SUM(Rain, Snow, Hail, etc…)

Not a direct observation of individual hydrometeors and distributions

Reflectivity observation type does not correspond to a state variable

Options:A. Specification of hydrometeors using some assumptionsB. Use implicit condensate formation as a model forcing functionC. Combination of A and BD. Others

Radar Reflectivity Data AssimilationB. Use implicit condensate formation from > 0 dBZ 3-D reflectivity as aforcing function in the model: Latent heating in microphysics (MP)

(1) Latent heating rate in the absence of reflectivity observationsOutside of radar coverage => Use model MPBelow lowest elevation and above PBL top => Extrapolate obs down

Testing:Rely upon other observation types (lightning, satellite) as proxy

(2) Latent heating rate where 3-D reflectivity observation < 0 dBZConvective/precipitation suppression => Set MP latent heating = 0

Testing:If model environment supports convection this is not enough,Need to force weak subsidence to avoid spurious convection

Radar Reflectivity Data AssimilationB. Use implicit condensate formation from > 0 dBZ 3-D reflectivity as aforcing function in the model: Latent heating in microphysics (MP)

(3) Latent heating rate for a given 3-D reflectivity observation > 0 dBZApplication in planetary boundary layer (PBL)? => No, use MP value

Function of stability? => NoTesting: Reduce heating for more unstable environments

Vertical/horizontal location of heating? => Currently co-locatedTesting: Vertical displacement (i.e. reflectivity trailing indicator)

Time-scale of condensation formation? => Currently 10 minTesting: 5 min (2X), 20 min (1/2 X), 30 min (1/3 X)

Length of time to apply heating? => Currently constant for 20 minTesting: Cycle in four 15-min periods with different values

Adjust microphysics heating rate? => Currently replace (specify)Testing: Averaging or adding obs-based and MP heating

2011 Hourly HRRR Initialization from RR

HourlyRR

LateralBoundaryConditions

Interp to 3 km grid

HourlyHRRR

15-h fcst

Initial Condition

Fields

11 z 12 z 13 z

Time (UTC)

AnalysisFields

3DVARObs

3DVARObs

Back-groundFields

18-h fcst 18-h fcst

1-hrfcst

DDFI DDFI

1-hrfcst

18-h fcst

1-hrfcst

Interp to 3 km grid

15-h fcst

Use 1-h old LBC

to reducelatency

Use most recent IC (post-DFI)

to get latest radar info

Reduced Latency:

~2h for 2011

Forward integration, full physicsApply latent heating from radar reflectivity, lightning data

Diabatic Digital Filter Initialization (DDFI) -20 min -10 min Init +10 min

RR model forecast

Backward integration,no physics

Obtain initial fields with improved balance, vertical circulations associated withongoing convection

• The model microphysics temperature tendency is replaced with a reflectivity-based temperature tendency.– Dynamics and microphysics respond to thermodynamic forcing.

• Analysis noise is reduced by digital filtering.

00 hr HRRR forecasts

RR – No radar data RR – Radar data

Obs00z 12 August 2011


ObsObs


01z 12 August 2011


Reflectivity00z 12 August 201100 hr forecasts

Convergence Cross-Section



Reflectivity01z 12 August 201101 hr forecasts



Obs20z 14 August 2011



21z 14 August 2011 ObsObs


20z 14 August 201100 hr forecasts

Reflectivity




21z 14 August 201101 hr forecasts

Reflectivity

HRRR Reflectivity VerificationDDFI in 13-km RUC/RR (parent model) but not in 3-km HRRR

Eastern US, Reflectivity > 25 dBZ11-20 August 2011

Increase in forecast skill when reflectivity assimilated in parent model

Neighborhood skill apparent on coarser verification grid (right)

CSI 03 km CSI 40 km

RUC + RadarRR + Radar

RUC + No RadarRR + No Radar

RUC + RadarRR + Radar

RUC + No RadarRR + No Radar

HRRR Reflectivity VerificationDDFI in 13-km RUC/RR (parent model) but not in 3-km HRRR

Eastern US, Reflectivity at 03 km11-20 August 2011

Higher bias when assimilating radar data, especially with the RR parent

RR + RadarRR + No RadarRUC + RadarRUC + No Radar

BIAS 25 dBZ BIAS 40 dBZ

RR + RadarRR + No RadarRUC + RadarRUC + No Radar

OptimalOptimal

HRRR Reflectivity VerificationDDFI in 13-km RR (parent model) but not in 3-km HRRR

Eastern US, Reflectivity > 25 dBZ14-24 July 2010

Increased latent heating forcing from parent increases skill and biasIncreased skill not sustained at longer forecast times (> 4 hrs)

CSI 40 km BIAS 03 km

Optimal

2x Latent heating rate in RR1x Latent heating rate in RR1/3x Latent heating rate in RR

2x Latent heating rate in RR1x Latent heating rate in RR1/3x Latent heating rate in RR


HourlyRR


Interp to 3 km grid

HourlyHRRR

15-h fcst

Initial Condition

Fields

11 z 12 z 13 z

Time (UTC)

AnalysisFields

3DVARObs

3DVARObs

Back-groundFields

18-h fcst 18-h fcst

1-hrfcst

DDFI DDFI

1-hrfcst

18-h fcst

1-hrfcst

Interp to 3 km grid

15-h fcst

Use 1-h old LBC

to reducelatency



Reduced Latency:

~2h for 2011

Attempt application of 13-km DDFI again at 3-km

HourlyRR


Interp to 3 km grid

HourlyHRRR

15-h fcst

Initial Condition

Fields

11 z 12 z 13 z

Time (UTC)

AnalysisFields

3DVARObs

3DVARObs

Back-groundFields

18-h fcst 18-h fcst

1-hrfcst

DDFI DDFI

1-hrfcst

18-h fcst

1-hrfcst

Interp to 3 km grid

15-h fcst

Use 1-h old LBC

to reducelatency



Reduced Latency:

~2h for 2011

DDFI DDFI

HRRR Reflectivity VerificationDDFI in 13-km RR (parent model) AND in 3-km HRRR

Eastern US, Reflectivity > 25 dBZ14-24 July 2010

Application of DDFI at both scales results in spurious convectionSignificant loss of skill

2x Latent heating rate in RR1x Latent heating rate in RR1/3x Latent heating rate in RR1x Latent heating in RR and HRRR


Optimal



HourlyRR


Interp to 3 km grid

HourlyHRRR

15-h fcst

Initial Condition

Fields

11 z 12 z 13 z

Time (UTC)

AnalysisFields

3DVARObs

3DVARObs

Back-groundFields

18-h fcst 18-h fcst

1-hrfcst

DDFI DDFI

1-hrfcst

18-h fcst

1-hrfcst

Interp to 3 km grid

15-h fcst

Use 1-h old LBC

to reducelatency



Reduced Latency:

~2h for 2011

13-km DDFI and sub-hourly cycling at 3-km

HourlyRR


Interp to 3 km grid

HourlyHRRR

Initial Condition

Fields

11 z 12 z 13 z

Time (UTC)

AnalysisFields

3DVARObs

3DVARObs

Back-groundFields

18-h fcst 18-h fcst

1-hrfcst

DDFI DDFI

1-hrfcst

18-h fcst

1-hrfcst

Interp to 3 km grid

Reduced Latency:

~1h for 2012?

Use 1-h old IC

and 2-h old LBC

15-h fcst 15-h fcst

Reflectivity DA on 3-km (HRRR) Grid

interpolation from RR,hydrometeor specification

t060 min t045 min t030 min t015 min t0

reflectivity-based temperature tendency

NSSL3D

mosaic

HRRR composite reflectivity

t045 min

t030 min t015 min t0

HRRR (3-km) grid produces convective storms explicitlyReflectivity-based temp. tendencies are applied during sub-

hourly cycling (forward model integration only, no digital filtering)

Initial Temperature atLowest Model Level

(1) HRRR initializedwithout 3-km radar DA

(2) HRRR initializedwith 3-km radar DA

1000

km

2000 UTC 11 May 2011

CompositeReflectivity2100 UTC

11 May 2011

1-h fcstwith 3-km

radar cycling

obsconvection develops quickly (RR cycling,

DDFI)

1-h fcstwithout 3-kmradar cycling

1-h fcstwith 3-km

radar cycling

more accuraterepresentation ofsystem maturity

CompositeReflectivity0200 UTC

11 May 2011

6-h fcstwith 3-km

radar cycling

more accurateforecast of

convective system propagation

6-h fcstwithout 3-kmradar cycling

obs

HRRR Reflectivity VerificationDDFI in 13-km RR (parent model) AND 3-km HRRR 15-min cycling

Eastern US, Reflectivity > 25 dBZSelect Cases May-July 2011

Application of DDFI at 13-km and cycling at 3-km results improves skillRetention of convective-scale forcing/evolution remains a challengeMesoscale environment often dominates storm-scale forcing

1x Latent heating rate in RR and HRRR1x Latent heating rate in RR


Optimal


CSI 40 km1x Latent heating rate in RR and HRRR1x Latent heating rate in RR

HRRR Forecast Behavior2012 Targets

(1) Lower peak bias in convection over eastern US

(2) Less difficulty propagating/maintaining MCSs

(3) Improve timing convective

initiation (early AM runs)

(4) Fewer false alarm cases

RR/HRRRModelDevelopment andEvaluation

“Simplistic” 13-km latent heatingNo 3-km data assimilation

“Smarter” 13-km latent heating3-km radar data assimilation

2011(1) Higher bias in convection over eastern US

(2) Difficulty propagating/maintaining MCSs

(3) Lead in convective

initiation (early AM runs)

(4) False alarm cases

Implement

Documents

Hourly Updated NOAA NWP Models