Gravity-Wave Drag Parameterization in a High-Altitude ...

16
Slide 1 Gravity-Wave Drag Parameterization in a High-Altitude Prototype Global Numerical Weather Prediction System Marine Meteorology Division (Code 7500) NRLDC lower atmosphere middle atmosphere Remote Sensing Division (Code 7200) Space Science Division (Code 7600) N. Baker T. Hogan M. Peng C. Reynolds B. Ruston … L. Coy S. Eckermann A. Kochenash J. McCormack F. Sassi D. Allen K. Hoppel D. Kuhl G. Nedoluha

Transcript of Gravity-Wave Drag Parameterization in a High-Altitude ...

Slide 1

Gravity-Wave Drag Parameterization in a High-Altitude Prototype Global Numerical Weather Prediction System

Marine Meteorology Division (Code 7500)

NRLDC

lower atmosphere

middle atmosphere

Remote Sensing Division (Code 7200)

Space Science Division (Code 7600)

N. BakerT. HoganM. PengC. ReynoldsB. Ruston …

L. CoyS. EckermannA. KochenashJ. McCormackF. Sassi

D. AllenK. HoppelD. KuhlG. Nedoluha

Slide 2

Global SpectralForecast Model

Data Assimilation SystemNAVDAS/NAVDAS-AR

Global 0-100 km observationsover next 0-6 hours

Xb

0-10 Day Forecasts

0-9 Hour Forecasts

y

Xa

6 hourly global0-100 km analysis fields

NOGAPS-ALPHAhttp://uap-www.nrl.navy.mil/dynamics/html/nogaps.html

6-hourly update cycle

Slide 3

NOGAPS-ALPHA

~1 hPaTop Data Insertion

Operationally

Navy Operational Global Atmospheric Prediction System (DoD NWP System)

Advanced‐Level Physics & High Altitude NWP Prototype

0.002 hPaTop Data Insertion

NAVDAS: NRL Atmospheric Variational Data Assimilation System (3DVAR)

T79L68

Slide 4

Gravity-Wave Drag Parameterizations

• Orographic gravity-wave drag (OGWD) schemes– Lindzen type scheme (Palmer et al. 1986)– New Met Office scheme (Webster et al. 2003)– Developmental NRL scheme (Kim and Doyle 2005)

• Nonorographic gravity-wave drag (NGWD) schemes– Rayleigh friction– Alexander and Dunkerton (1999) multiwave scheme– Kim-Chun convective GWD schemes– Hines (1991) Doppler-spread scheme– WACCM 3.0 65-wave Lindzen scheme (Garcia et al. 2007)

NOGAPS-ALPHA T79L68 “Frozen” Production Configuration (Eckermann et al. JASTP 2009)

Eckermann, S. D., K. W. Hoppel, L. Coy, J. P. McCormack, D. E. Siskind, K. Nielsen, A. Kochenash, M. H. Stevens, C. R. Englert, and M. Hervig, High-altitude data assimilation system experiments for the northern summer mesosphere season of 2007, J. Atmos. Sol.-Terr. Phys., 71, 531-551, 2009

Slide 5

Gravity Wave Drag (GWD) Parameterization Issues…The GWD schemes must be tuned to improve both the forecasts xb

and the analyses xa (to eliminate obs-forecast [O-F] biases)

1. Computational Speed• WACCM 3.0 NGWD schemes consumes ~20-40% of the total run time of the

forecast model

• To be competitive for transition to operational NWP centers, this is (at least) an order of magnitude too expensive

Question 1: how can these schemes be made much more efficient?

2. Extensive (endless?) tuning • many forecast model runs and iterative tuning are needed to reproduce temperature

fields from SABER and MLS and long-term climatologies

• essential to reduce biases between observations and background (O-Fs) that affect the quality of the final analysis (A) fields

Question 2: Can the analysis fields xa provide an observations-based method of objectively tuning the GWD parameterizations?

Slide 6

Multiwave Deterministic Nonorographic Gravity Wave Drag Parameterization

1. Source Level Momentum Flux τ(c)

2. Discretize τ(c) among ngw=2nc+1=65 individual gravity waves & propagate each wave upwards

3. Add up flux deposition (drag) wave-by-wave to compute the total mean-flow acceleration

Slide 7

New Stochastic Version of the Nonorographic Gravity Wave Drag Scheme

Deterministic source• discretized identically in every

grid box using ngw=65 different phase-speed waves

Stochastic Analogue• sample the source spectrum

stochastically using nsgw=1 wave per grid box based on uniform random number generator

• New scheme is explicitlystochastic and intermittent

• In theory, 65 times faster!

Slide 8

Single Column Tests

• Same time-mean drag (and diffusion and heating rates), but accompanied by explicit stochastic variability about the time mean

Slide 9

Tests in NOGAPS-ALPHA: 3 Year Nature RunZonal-Mean Zonal Winds for July

Slide 10

Tests in NOGAPS-ALPHA: 3 Year Nature RunZonal-Mean GW-induced Acceleration and Heating Rates for July

Stochastic 1-wave scheme produces identical climate to deterministic 65-wave scheme but it is between 10-20 times faster computationally in the model.

[Eckermann, S. D., Explicitly stochastic parameterization of nonorographic gravity-wave drag, J. Atmos. Sci., in press, 2011]

Slide 11

Tests in NOGAPS-ALPHA: 3 Year Nature RunMean Spectral Variability for June

Slide 12

Extension to Autoregressive (AR) Stochastic Model

Existing StochasticGaussian flux spectrum

Uniform random c’s

Momentum Flux

Phase Speeds c

AR StochasticUniform flux spectrum

Normal random c’s

1. Normal rather than uniform distribution of random phase speeds2. Phase speed now has a “time history” (temporal autocorrelation)3. Autocorrelation values (0<AC<1) control degree of temporal scale of

the random walk (AC 0.0 short, AC 1.0 long)

Slide 13

Gravity Wave Drag (GWD) Parameterization Issues…The GWD schemes must be tuned to improve both the forecasts xb

and the analyses xa (to eliminate obs-forecast [O-F] biases)

1. Computational Speed• WACCM 3.0 NGWD schemes consumes ~20-40% of the total run time of the

forecast model

• To be competitive for transition to operational NWP centers, this is (at least) an order of magnitude too expensive

Question 1: how can these schemes be made more efficient?

2. Extensive (endless?) tuning • many forecast model runs and iterative tuning are needed to reproduce temperature

fields from SABER and MLS and long-term climatologies

• essential to reduce biases between observations and background (O-Fs) that affect the quality of the final analysis (A) fields

Question 2: Can the analysis fields xa provide an observations-based method of objectively tuning the GWD parameterizations?

Slide 14

Drag in the Analyses xa – How Much from Observations, and How Much from Model Parameterizations?

xa

Slide 15

Mesospheric Temperature Corrections (xa-xb) from SABER and MLS for one 6-hour cycle

Preliminary Hypothesis:• Sparse data

assimilated in upper mesosphere constrain only zonal wavenumbers ~1-8

• Parameterized OGWD has large signal in analyses since it is applied at zonal wavenumbers>8

• Parameterized NWD applied gravest zonal wavenumbers, so can be corrected by observations

Slide 16

Summary

GWD parameterizations presents somewhat different (but similarly thorny) challenges for operational NWP systems (compared to climate models)

• must be very cheap computationally (cheaper than climate model requirements

• have important complex influences that manifest both in the forecasts and in the analyses

A new explicitly stochastic analogue of an existing NGWD scheme yields

• An order of magnitude improvement in computational speed

• Essentially identical long-term climate in nature runs

• No excessive increases in variability at smallest space-time GCM scales

• Can add greater and more realistic spread for ensemble forecasting applications

Subgridscale zonal drag residuals in the analyses

• Show signals of parameterized OGWD from forecast model

• Suggests NGWD rather than OGWD might be more effectively constrained by sparse data assimilated in mesosphere