1 / 20

Hybrid 4D-Var development at NRL: Results from two 3-month trials

Hybrid 4D-Var development at NRL: Results from two 3-month trials. David Kuhl 1 , Craig Bishop 2 , Tom Rosmond 3 , Elizabeth Satterfield 4 1 Naval Research Laboratory, Washington DC 2 Naval Research Laboratory, Monterey, CA 3 Science Application International Corp., Forks, WA.

naiya
Télécharger la présentation

Hybrid 4D-Var development at NRL: Results from two 3-month trials

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Hybrid 4D-Var development at NRL: Results from two 3-month trials David Kuhl1, Craig Bishop2, Tom Rosmond3, Elizabeth Satterfield4 1Naval Research Laboratory, Washington DC 2Naval Research Laboratory, Monterey, CA 3Science Application International Corp., Forks, WA. 4National Research Council/NRL, Monterey, CA * Xu et al. 2005 NAVDAS-AR 4DVAR system (Navy’s Operational Global Modeling System)

  2. Overview • Goal: Investigate impact of enhancing the conventional initial background error covariance of the NAVDAS-AR 4D-Var setup with an ensemble background error covariance. We refer to this as hybrid system. • Result: The hybrid system improves the 4D-Var DA.

  3. NAVDAS-AR • Where: • is the error covariance matrix for NAVDAS-AR specified at all time steps of the DA window • is the initial background error covariance matrix specified at the beginning of the DA window (3 hours after previous analysis time) • is the Tangent Linear Model (TLM) • is the adjoint model • is the model error covariance (opt. – not used) • TLM and adjoint are used to propagate initial background error covariance ( ) forward and backward through data assimilation window

  4. Hybrid Assimilation • Many groups (Buehner, Wang, Kleist) have found that hybrid assimilation results in improved analyses and forecasts • With hybrid assimilation we combine the of the conventional and ensemble methods in the linear fashion: • The resulting formulations incorporates aspects of both and • Bishop and Satterfield found theoretical justification for hybrid based on variances

  5. NAVDAS-AR Conventional • Variances ( ): • Geo-pot. height and temperature are in exact hydrostatic balance • Geo-pot. height and winds are approximately geostropically balanced in the extratropics and independent in tropics • Correlations ( ): • Isotropic correlation model based on balanced and unbalanced correlations separable in the vertical and horizontal (see Chapter 4 Daley and Barker 2001) • Strengths: • High rank • Preserves some aspects of geophysical balances • Weakness: • Not flow dependent • Horizontal length scale independent of height may not apply in both troposphere and stratosphere • Balance assumptions are incorrect in boundary layer and stratosphere

  6. Flow Dependent Ensemble • Where: • is the ensemble perturbation • is the number of ensemble members • is localization matrix • Ensembles: 9-banded Ensemble Transform (ET) (McLay et. al 2010) • Mean: 3-hour forecast of 4D-var analyses at high resolution • Covariances (balance of): • Operational 3D-Var analysis error variances • 3-hour forecast of ensemble members at low resolution • Strength: • Flow dependent errors of the day • Multivariate balances implied by the localized ensemble correlations • Weakness: • Localization damages geophysical balances • Cycled ensembles (generated in the manner of a Kalman filter) almost inevitably result in variances that are too small in some regions and too large in others. Getting this correct is a work in progress.

  7. Experimental Setup • 2 Experiments • Jun. 1, 2010 to Sep. 1 2010 • Jan. 1, 2011 to Apr. 1, 2011 • Discard 1st month of each analysis for Radiance Bias correction (VAR-BC) spin-up of ensemble • Model resolution (operational): • T319L42 outer (960x480 Gaus. Grid) • T119L42 inner (360x180 Gaus. Grid) • Ensemble resolution (same as inner): T119L42 • Number of Ensemble Members: 80 (size of operational ensemble) • Assimilating conventional observations and all operational radiances except Aqua and MHS

  8. Verification Metrics:Score Card • Anomaly Correlations: • Statistically significantly better with confidence level of 95% • All of the rest: • Statistically significantly better with confidence level of 95% • And error must be at least 5% less than the control

  9. Conventional vs. Hybrid Anomaly Height Correlation SH • Experiment comparison: • Blue is win for Conventional • Red is win for Hybrid • Percentage reduction/increase of anomaly height correlation relative to conventional • Anomaly height correlation is computed relative to self analysis at different forecast lead times 0-5 days • Forecasts were launched every 12 hours after one month spin-up for bias correction • Statistical significance of anomaly height correlation difference • Green Boxes show scorecard • Presented here 2 wins for Hybrid in SH Anomaly Height Correlation

  10. Anom. H.C. Conv.vs. Hybrid Jul-Aug 2010 Scorecard=1 Feb-Mar 2011 Scorecard=2

  11. Vect. Wind Conv.vs. Hybrid Jul-Aug 2010 Scorecard=0 Scorecard*=4 *=based on significance Feb-Mar 2011 Scorecard=0 Scorecard*=3

  12. Raobs.Conv.vs. Hybrid Jul-Aug 2010 Scorecard=0 Scorecard*=2 *=based on significance Feb-Mar 2011 Scorecard=0 Scorecard*=0

  13. Score Card Results • No statistical significance in either TC tracks or buoy wind speed verification • Classic Counting (RMS and Vect. Wind need 5% less than control) • Jul-Aug 2010: 1 points • Feb-Mar 2011: 2 points • Only 95% significance • Jul-Aug 2010: 7 points • Feb-Mar 2011: 5 points • Upgrade to 3D-Var: 1 point • Upgrade to 4D-Var (with other changes): 4 points

  14. Vect. Wind Conv.vs. Hybrid Jul-Aug 2010 Feb-Mar 2011

  15. Conclusions/Future Work • Conclusions: • Ensemble enhancement of improved analyses and forecasts with Hybrid alpha=0.5 • Full ensemble experiment ( ) shows marked improvement to tropical vector winds • Regional tuning of alpha may produce better impacts • Future Work: • Can large ensemble remove need for TLM and adjoint? • Climatological ensemble for the static . At low resolutions we saw very promising results in this direction • Tuning of the localization and alpha • Adaptive ensemble covariance localization • Bishop and Satterfield equation method for determining alpha

  16. Thank You.

  17. Anom. H.C. Conv.vs. Hybrid Jul-Aug 2010 Scorecard=1 Feb-Mar 2011 Scorecard=2

  18. Anom. H.C. Conv.vs. Hybrid Jul-Aug 2010 Feb-Mar 2011

  19. Raobs.Conv.vs. Hybrid Jul-Aug 2010 Feb-Mar 2011

  20. TC Tracks and Buoy • Tropical cyclone Track error • At lead time 4 days • Number of verification dates with storms in them: • Jul-Aug 2010: 42 • Feb-Mar 2011: 27 • No significant difference between either experiment • Global Buoy Surface Wind Speed Error • At lead time 3 days • No significant difference between either experiment

More Related