Vortex-Scale Hurricane Data Assimilation:

Overview of NOAA/AOML/HRD’sHurricane Ensemble Data Assimilation System (HEDAS) and Its

Performance for the 2010 Hurricane Season

Altuğ Aksoy1,2, Sim Aberson1, Tomislava Vukicevic1,Kathryn Sellwood1,2, Gopal1, & Lisa R. Bucci1,3

1NOAA/AOML Hurricane Research Division2U. Miami/RSMAS Cooperative Institute for Marine & Atmospheric Studies

3Science Applications International Corporation

NOAA Operates Three Aircraft to ObserveHurricane Environment and Vortex Structure

Flight-Level Wind, T, Q

Gust probeC-Band Weather Radar

Chemistry inlet

RadiometerLower-fuselage Weather Radar

Launch tubes and chutes:

Pylons

Radar altimeterGPS

(Stepped-frequency Microwave Radiometer, cloud physics, aerosol)

Dropwindsonde,

AXBT, AXCP, AXCTD

Tail Doppler Radar2 P-3 Aircraft – NOAA 42, 43(hurricane core penetrations)

1 G-IV Aircraft – NOAA 49(hurricane environment)

HRD HEDAS Effort Primarily Focuses onDropsonde, Doppler Radar, and Flight-Level Observations

C-Band Weather Radar

GPS navigation

Stepped-frequency Microwave Radiometer

Tail Doppler Radar (new)

Dropwindsonde chute

Data Assimilation: An Overview of the Ensemble Kalman Filter (EnKF)

• Data assimilation combines model background with observations to obtaina best estimate of atmospheric conditions to initialize a model run

• The EnKF is an advanced data assimilation technique (Evensen 1994)• An ensemble of short-range forecasts provides the flow-dependent background

covariance information

t = t0 t = t0+Δt t = t0+2Δt

For the assimilation of obs, use covariances

sampled from the ensemble of forecastsAnalysis uncertainty becomes the initial

condition uncertainty for the new forecast cycle

Subsequent forecast cycle is

initialized from the previous

analysis ensemble… Start with an ensemble of states (ensemble members)

that better represent theinitial uncertainty about the

“mean” state

Instead of a single state that represents the initial state of

the atmosphere …

EnKF

Observations

HRD’s Hurricane Ensemble Data Assimilation System (HEDAS)

• Forecast model:– HRD’s Experimental HWRF (HWRF-X)– 2 nested domains (9/3 km horizontal resolution, 42 vert. levels)– Static inner nest to accommodate covariance computations

Inner nest size: ~10x10 degrees– Ferrier microphysics, explicit convection on inner nest

• Ensemble system:– Initialized from semi-operational GFS-EnKF (NOAA/ESRL) ensemble– Initial ensemble is spun up for 3-4 hours before assimilation begins– 30 ensemble members

• Data assimilation:– Square-root ensemble Kalman filter, EnKF (Whitaker and Hamill 2002)– Assimilates inner-core aircraft data on the inner nest

NOAA P-3, NOAA G-IV, USAF, PREDICT G-V– Covariance localization (Gaspari and Cohn 1999)

2010 HEDAS Semi-Real-Time Runs for HFIP:DA Cycling Workflow

• Only ran when Doppler radar wind data was available from NOAA P-3 flights( 19 cases)

• Used 1452 processors on NOAA’s tJet cluster (supported by HFIP)T T + 126 hT – 6 h

EnsembleSpin-up

DA Cyclingwith EnKF

Deterministic HWRF-X Forecast from Ens. Mean

Real-TimeObservation

Pre-Processing

EnsembleInitialization

from T-6h GFS-EnKF

Mean of Final Analysis

Assumed Valid for T

1-hour Cycles

2010 HEDAS Semi-Real-Time Runs for HFIP:Summary of Cases

• A total of 19 cases are run (when NOAA P-3’s collected Doppler radar data):Ea

rl - 0

8/29

00Z

Earl

- 08/

29 1

2Z

Earl

- 08/

30 0

0Z

Earl

- 08/

30 1

2Z

Earl

- 08/

31 0

0Z

Earl

- 09/

01 1

2Z

Earl

- 09/

02 0

0Z

Earl

- 09/

02 1

2Z

Earl

- 09/

03 0

0Z

Earl

- 09/

03 1

2Z

Earl

- 09/

04 0

0Z

Inve

st92

- 09

/13

00Z

Inve

st92

- 09

/13

12Z

Inve

st92

- 09

/14

00Z

Karl

- 09/

16 1

8Z

Rich

ard

- 10/

23 1

2Z

Tom

as -

11/0

5 00

Z

Tom

as -

11/0

6 12

Z

Tom

as -

11/0

7 00

Z

0.1

1

10

100

1000

10000

0

5

10

15

20

25

# Cycles Run SFMR Dropsonde Flight Level Doppler Wind

Number of Observations Assimilated per Cycle

Num

ber o

f Obs

erva

tions

Num

ber of Cycles

2010 HEDAS Semi-Real-Time Runs for HFIP:Summary of Final Mean Analysis IntensityEa

rl - 0

8/29

00Z

Earl

- 08/

29 1

2Z

Earl

- 08/

30 0

0Z

Earl

- 08/

30 1

2Z

Earl

- 08/

31 0

0Z

Earl

- 09/

01 1

2Z

Earl

- 09/

02 0

0Z

Earl

- 09/

02 1

2Z

Earl

- 09/

03 0

0Z

Earl

- 09/

03 1

2Z

Earl

- 09/

04 0

0Z

Inve

st92

- 09

/13

00Z

Inve

st92

- 09

/13

12Z

Inve

st92

- 09

/14

00Z

Karl

- 09/

16 1

8Z

Rich

ard

- 10/

23 1

2Z

Tom

as -

11/0

5 00

Z

Tom

as -

11/0

6 12

Z

Tom

as -

11/0

7 00

Z

0

20

40

60

80

100

120

140

920

934

949

963

977

991

1006

1020

Intensity - Oper. Intensity - HEDAS Min. Pressure - Oper. Min. Pressure - HEDAS

Max

. 10-

m W

ind

Spee

d (k

nots

)M

in. Pressure (mbar)

2010 HEDAS Semi-Real-Time Runs for HFIP:Summary of Forecast Performance

Forecast Time (hour)

Intensity Error (kt) &Number of Cases

Track Error (km)

Frequency ofSuperior Performance (%)

HEDAS (in blue) performs better in intensity and comparably in track (to HWRF and HWRFx)

2010 HEDAS Semi-Real-Time Runs for HFIP:Comparison of Structures (Earl 08/30 12Z)

Wind Speed at 1 km (m/s)

Analysis Radar

Dist

ance

fro

m Ce

nter

(km

)

Distance from Center (km)

2010 HEDAS Semi-Real-Time Runs for HFIP:Comparison of Structures (Earl 08/30 12Z)

Frequency Histograms of Observed Quantities

SFMR

Wind Speed (m/s)

Freq

uenc

y

Flight Level Doppler Wind

2010 HEDAS Semi-Real-Time Runs for HFIP:Comparison of Structures (Earl 08/31 00Z)

Wind Speed at 1 km (m/s)

Analysis Radar

Dist

ance

fro

m Ce

nter

(km

)

Distance from Center (km)

2010 HEDAS Semi-Real-Time Runs for HFIP:Comparison of Structures (Earl 08/30 12Z)

Frequency Histograms of Observed Quantities

SFMR Flight Level Doppler Wind

Wind Speed (m/s)

Freq

uenc

y

2010 HEDAS Semi-Real-Time Runs for HFIP:Comparison of Forecasts (Earl 08/31 00Z)

Summary and Future Plans• Across the 19 cases for which HEDAS has run during the 2010 season:

– Intensity error was better than HWRFx (initialized with HWRF vortex) and operational HWRF

– Track error was comparable

• Issues remain with respect to initial adjustment during forecast when initialized from HEDAS vortex (this is not unique to HEDAS)

Research underway to address these issues

• Planned updates before the 2011 season:– Satellite data assimilation in the core (Vukicevic)– Representation of model error in surface and PBL physics (Aksoy and J. Zhang)– Improved parallelization

• All of HEDAS forecast case results can be found at the following link:https://storm.aoml.noaa.gov/realtime