GSD – Assimilation and Modeling Branch

GSD – Assimilation and Modeling Branch

35th Conference on Radar Meteorology30 September 2011

Analysis of forecast improvement from radar-data assimilation

within RUC, Rapid Refresh and HRRR models

Steve Weygandt, Curtis Alexander, Ming Hu, Patrick Hofmann,

Eric James, Haidao Lin, Tanya Smirnova, David Dowell,

Stan Benjamin, John Brown

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 Domain Grid

PointsGrid

SpacingVertical Levels

Vertical Coordinate

Height Lowest Level

Pressure Top

RUC CONUS 451 x 337 13 km 50 Sigma/ Isentropic 5 m ~50 mb

RR North America 758 x 567 13 km 50 Sigma 8 m 10 mb

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

Model Run at: Time-Step Forecast Length Initialized Boundary

ConditionsRun Time

# of CPUs

RUCNCEPoper

18 s 18 hrs Hourly (cycled) NAM ~25 min 32

RR GSD, EMC 60 s 18 hrs Hourly (cycled) GFS ~25 min 160

HRRR GSD 15-20s 15 hrs Hourly (no-cycle)

RUC(RR --Apr 11) ~50 min ~1000

Models at NOAA/ESRL/GSD/AMB

Model Assimilation DFI Microphysics Radiation Convection PBL LSM

RUC RUC-3DVAR Yes w/radar Thompson RRTM/Dudhia G3 MYJ RUC

RR GSI Yes w/radar Thompson RRTM/Goddard G3 MYJ RUC

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

RR physics behavior similar to RUC – good for aviation, convective environment , surface and

sensible weatherNCEP upgrade from RUC to RR will provides moreadvanced model and analysis components

- WRF-ARW: advanced numerics, non-hydrostatic- GSI: advanced satellite DA, 4DVAR/EnKF development- Both community-based, ongoing code contributions

• “Radar-DFI” assimilation procedure in NCEP RUC since Nov. 2008 improved fcsts precip systems

• Radar-DFI in GSD RR since 2008, RR to replaceRUC at NCEP late 2011

• Hourly CONUS HRRR nest in RUC since 2008 HRRR benefitted from RUC radar-DFI

• HRRR switched to run as nest within RR (with RR radar-DFI) in April 2011.

• HRRR run in real-time to support CoSPA, SPC, NWS regions, renewable energy

Quantify radar DFI impact on RUC, RRand HRRR nests for summer-time convection

Establish baseline for more sophisticated radar assimilation efforts

Background / Goals

Rawinsonde (balloons)Wind Profilers (405 MHz, 915 MHz)

RASS virtual temperaturesVAD winds (WSR-88D radars)

Aircraft (ACARS, TAMDAR) Surface (METAR , Buoy,

Mesonet )Precipitable water (GPS, GOES,

SSM/I)GOES cloud-drift winds

GOES cloud-top pressure/tempRadar reflectivity, lightning

Ship reports/dropsondesSatellite radiances (RR)

Rapid Refresh Hourly Cycle / Observations

1-hrfcst

1-hrfcst

1-hrfcst

11 12 13Time (UTC)

AnalysisFields

3DVAR

Obs

3DVAR

Obs

Back-groundFields

Hourly updating using all available observations

Data types used

Spring 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

3DVAR

Obs

3DVAR

Obs

Back-groundFields

18-h fcst 18-h fcst

1-hr

fcst

DDFI DDFI

1-hr

fcst

18-h fcst

1-hr

fcst

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

Digital Filter-based reflectivity assimilationinitializes ongoing precipitation regions

Radar reflectivity assimilation

Forward integration,full physics with radar-based latent heating

-20 min -10 min Initial +10 min + 20 min

RUC / RR HRRR model forecast

Backwards integration, no physics

Initial fields with improved balance, storm-scale circulation

Digital Filter-based reflectivity assimilationinitializes ongoing precipitation regions

Radar reflectivity assimilation

Forward integration,full physics with radar-based latent heating

-20 min -10 min Initial +10 min + 20 min

RUC / RR HRRR model forecast

Backwards integration, no physics

Initial fields with improved balance, storm-scale circulation

+ RUC/RR Convection suppression

No precipitation hydrometeor specification*

psfct

Diabatic Digital Filter InitializationReduce noise in RUC and Rapid Refresh

No

ise

pa

ram

ete

r

Rapid Refresh (GSI + ARW) reflectivity assimilation example

Low-levelConvergence

Upper-levelDivergence

K=4 U-comp. diff (radar - norad)

K=17 U-comp. diff

(radar - norad)

NSSL radar reflectivity

(dBZ)

14z 22 Oct 2008Z = 3 km

• No cycling on the 3-km HRRR grid, must “spin-up storms within each run followed by accelerated storm evolution

“match” with real-world evolution depends on stage of actual storm evolution

• How effective is cycled “radar-DFI” procedure applied on mesoscale grid?-- for mesoscale “parent” grid?-- for storm-scale “child” grid?

• How does parameterized convection on parent grid impact evolution of storm-scale model?

• Similarities / difference RUC-HRRR vs. RR-HRRR?

Issues / Questions

• To what degree is storm evolution controlled by mesoscale vs. storm-scale forcing?

Highly variable based on regime, but mesoscale forcing is very significant

-- We don’t accurately observe or predict mesoscale

mesoscale forcing == high predictability

-- Strong mesoscale forcing does imply reduced impact from radar data assimilation, unlessassimilated radar data improves mesoscale fields

(cold pools, low-level wind fields, etc.)

Issues / Questions

R/T-Retro Experiment Weather

11 12 13 14 15 16 17 18 19 20 21

RRAug 2011

RUCRR -- HRRR

RUC -- HRRR NO radarassimilation

YES radarassimilation

REAL-TIME

RETROSPECTIVE

R/T-Retro Experiment Weather

11 12 13 14 15 16 17 18 19 20 21

RRAug 2011

RUCRR -- HRRR

RUC -- HRRR NO radarassimilation

YES radarassimilation

REAL-TIME

RETROSPECTIVE

R/T-Retro Experiment Weather

11 12 13 14 15 16 17 18 19 20 21

RR -- HRRRAug 2011

RUC -- HRRRRR -- HRRR

RUC -- HRRR NO radarassimilation

YES radarassimilation

REAL-TIMEEvery 2 hoursEvery 1 hour

RETROSPECTIVE

R/T-Retro Experiment Weather

11 12 13 14 15 16 17 18 19 20 21

RR -- HRRRAug 2011

RUC -- HRRRRR

RUC NO radarassimilation

YES radarassimilation

REAL-TIMEEvery 2 hoursEvery 1 hour

RETROSPECTIVE

R/T-Retro Experiment Weather

11 12 13 14 15 16 17 18 19 20 21

RR -- HRRRAug 2011

RUC -- HRRRRR -- HRRR

RUC -- HRRR NO radarassimilation

YES radarassimilation

REAL-TIMEEvery 2 hoursEvery 1 hour

RETROSPECTIVE

R/T-Retro Experiment Weather

11 12 13 14 15 16 17 18 19 20 21

RR -- HRRRAug 2011

RUC -- HRRRRR -- HRRR

RUC -- HRRR NO radarassimilation

YES radarassimilation

REAL-TIMEEvery 2 hoursEvery 1 hour

00z 13 Aug

02z 12 Aug

RETROSPECTIVE

R/T-Retro Experiment Weather

11 12 13 14 15 16 17 18 19 20 21

RR -- HRRRAug 2011

RUC -- HRRRRR -- HRRR

RUC -- HRRR NO radarassimilation

YES radarassimilation

REAL-TIMEEvery 2 hoursEvery 1 hour

00z 13 Aug

02z 12 Aug

22z 13 Aug

RETROSPECTIVE

R/T-Retro Experiment Weather

11 12 13 14 15 16 17 18 19 20 21

RR -- HRRRAug 2011

RUC -- HRRRRR -- HRRR

RUC -- HRRR NO radarassimilation

YES radarassimilation

REAL-TIMEEvery 2 hoursEvery 1 hour

00z 13 Aug

02z 12 Aug

22z 13 Aug

14z 14 Aug

00z 15 Aug

RETROSPECTIVE

R/T-Retro Experiment Weather

11 12 13 14 15 16 17 18 19 20 21

RR -- HRRRAug 2011

RUC -- HRRRRR -- HRRR

RUC -- HRRR NO radarassimilation

YES radarassimilation

REAL-TIMEEvery 2 hoursEvery 1 hour

00z 13 Aug

02z 12 Aug

22z 13 Aug

14z 14 Aug

00z 15 Aug

04z 19 Aug

RETROSPECTIVE

20z 19 Aug

80-km

valid (GMT)valid (EDT)

00 02 04 06 08 10 12 14 16 18 20 22 8p 10p 12 2a 4a 6a 8a 10a 12 2p 4p 6p

“neighborhood” verification of 6-hforecasts from 3-km HRRR

verification:10 June – 26 Sept 2010

40-km

20-km

25 dBZ6-h fcst

3-km

80-km

valid (GMT)valid (EDT)

00 02 04 06 08 10 12 14 16 18 20 22 8p 10p 12 2a 4a 6a 8a 10a 12 2p 4p 6p

“neighborhood” verification of 6-hforecasts from 3-km HRRR

verification:10 June – 26 Sept 2010

ConvectiveInitiation

period

40-km

20-km

25 dBZ6-h fcst

3-km

80-km

valid (GMT)valid (EDT)

00 02 04 06 08 10 12 14 16 18 20 22 8p 10p 12 2a 4a 6a 8a 10a 12 2p 4p 6p

“neighborhood” verification of 6-hforecasts from 3-km HRRR

verification:10 June – 26 Sept 2010

ConvectiveInitiation

period

40-km

20-km

25 dBZ6-h fcst

3-km

ConvectiveDecayperiod

80-km

valid (GMT)valid (EDT)

00 02 04 06 08 10 12 14 16 18 20 22 8p 10p 12 2a 4a 6a 8a 10a 12 2p 4p 6p

“neighborhood” verification of 6-hforecasts from 3-km HRRR

verification:10 June – 26 Sept 2010

ConvectiveInitiation

period

40-km

20-km

25 dBZ6-h fcst

3-km

ConvectiveDecayperiod

R/T-Retro Experiment Configuration

Mostly compare RUC-HRRR vs. RR-HRRR (w/ and w/o radar)

Also compare RUC and RR (w/ and w/o radar assimilation)

Reflectivity and precipitation verification (also upper-air, surface)

RUC RR

NO radar assim

YES radar assim

also RUC RR HRRR HRRR

R/T-Retro Experiment Configuration

Mostly compare RUC-HRRR vs. RR-HRRR (w/ and w/o radar)

Also compare RUC and RR (w/ and w/o radar assimilation)

Reflectivity and precipitation verification (also upper-air, surface)

RUC RR

NO radar assim

YES radar assim

also RUC RR HRRR HRRR

Initial Hydrometeor specification:RUC – No precipitation hydrometeor specification0-h RUC-HRRR reflectivity from 1-h RUC explicit hydrometeorsRR – Specify snow hydrometeors only0-h RR-HRRR reflectivity from 1-h RR explicit HMs + radar snow

13-km Eastern USMatched comparison

2 X 12 hr fcstvs. CPC 24-h analysis

11 – 21 Aug. 2011All init times

Rapid Refreshvs. RUC

PrecipitationVerification

RUC radar

| | | | | | | 0.01 0.10 0.25 0.50 1.00 1.50 2.00 in.

CSI(x 100)

RUC no radar

RR radar

RR no radar

| | | | | | | 0.01 0.10 0.25 0.50 1.00 1.50 2.00 in.

bias(x 100)

100(1.0)RUC

radarRUC

no radar

RR radarRR

no radarRadar assim

improves precip CSIs scores,

but also increases bias

Threshold

| | | | | | | 0-h 2-h 4-h 6-h 8-h 10-h 12-h

25 dBZ 13-kmEastern USMatched Comparison12,13,14,19 Aug. 2011All init times

“parent” vs. “child”ReflectivityVerification

| | | | | | | 0-h 2-h 4-h 6-h 8-h 10-h 12-h

CSI(x 100)

RR-HRRR radar

RR-HRRR no radar

3-km fcsts improve upon parent 13-km

forecasts radar assim

adds skill at both 13-km and 3-km

RR radar RR

no radar

Forecast Lead Time

RUC-HRRR radar

RUC-HRRR no radar

RUC radar RUC

no radar

CSI(x 100)

RUC-HRRRvs. RR-HRRR ReflectivityVerification

RUC-HRRR radar

CSI(x 100)

RUC-HRRR no radar

RR-HRRR radar

RR-HRRR no radar

bias(x 100)

100(1.0)

Radar assim improves precip CSIs scores,

but also increases bias

| | | | | | | 0-h 2-h 4-h 6-h 8-h 10-h 12-h

Forecast Lead Time

| | | | | | | 0-h 2-h 4-h 6-h 8-h 10-h 12-h

Eastern US, 25 dBZ 13-km CSI3-km biasMatched Comparison12,13,14,19 Aug. 2011All init times

RUC-HRRR radar

RUC-HRRR no radar

RR-HRRR radar

RR-HRRR no radar

| | | | | | | 0-h 2-h 4-h 6-h 8-h 10-h 12-h

RUC-HRRR radar

RUC-HRRR no radar

RR-HRRR radar

RR-HRRR no radar

| | | | | | | 0-h 2-h 4-h 6-h 8-h 10-h 12-h Forecast Lead TimeForecast Lead Time

| | | | | | | 0-h 2-h 4-h 6-h 8-h 10-h 12-h

| | | | | | | 0-h 2-h 4-h 6-h 8-h 10-h 12-h

18z inits 00z inits

06z inits 12z inits

CSI for different Initialization times

00z 06z 06z 12z

12z 18z 18z 00z

18z

06z

00z

12z

RRHRRR

RADAR

NSSL mosaic

RRHRRR

no radar

RUCHRRR

RADAR

RUCHRRR

no radar

01z 13 Aug 201100z

init

RRHRRR

RADAR

NSSL mosaic

RRHRRR

no radar

RUCHRRR

RADAR

RUCHRRR

no radar

01z 13 Aug 201100z

init0h 2h 4h 6h 8h

CSI scores

01z 13 Aug 2011

RRHRRR

RADAR

NSSL mosaic

RRHRRR

no radar

RUCHRRR

RADAR

RUCHRRR

no radar

00z + 1 hr fcsts

00z 13 Aug 2011

NSSL mosaic

RUCHRRR

RADAR

00z initialization

00z 13 Aug 2011

NSSL mosaic

RUCHRRR

RADAR

00z initialization

RRHRRR

no radar

RUCHRRR

no radar

RRHRRR

RADAR

00z 13 Aug 2011

NSSL mosaic

RUCHRRR

RADAR

RRHRRR

no radar

RUCHRRR

no radar

RRHRRR

RADAR

00z initialization

00z 13 Aug 2011

NSSL mosaic

RUCHRRR

RADAR

RRHRRR

RADAR

RUCHRRR

no radar

RRHRRR

no radar

00z initialization

01z 13 Aug 2011

RRHRRR

RADAR

NSSL mosaic

RRHRRR

no radar

RUCHRRR

RADAR

RUCHRRR

no radar

00z + 1 hr fcsts

02z 13 Aug 2011

RRHRRR

RADAR

NSSL mosaic

RRHRRR

no radar

RUCHRRR

RADAR

RUCHRRR

no radar

00z + 2 hr fcsts

04z 13 Aug 2011

RRHRRR

RADAR

RRHRRR

no radar

RUCHRRR

RADAR

RUCHRRR

no radar

00z + 4 hr fcsts

NSSL mosaic

06z 13 Aug 2011

RRHRRR

RADAR

RRHRRR

no radar

RUCHRRR

RADAR

RUCHRRR

no radar

00z + 6 hr fcsts

NSSL mosaic

• “Radar-DFI” in mesoscale model improves mesoscale model precipitation and nested storm- scale reflectivity forecasts

• Results very similar for RUC-HRRR, RR-HRRR

• Radar-DFI focuses mesocale convective areas, suggestion of indirect convective suppression

• Radar data impact quite case dependent,strong diurnal aspect with little retention across mid-day convective minimum

• RR-HRRR convects more easily than RUC-HRRR, HRRR has difficulty propagating leading convective lines for both RR and RUC fields, especially for strongly capped environments

• Limitations to non-cycled 3-km forecasts

Conclusions and future work