1. Application of the WRF-LETKF System over Southern South America: Sensitivity to model physics.
María E. Dillon, Yanina Garcia Skabar, Juan Ruiz,
Eugenia Kalnay, Estela A. Collini,
Pablo Echevarría, Marcos Saucedo,
Takemasa Miyoshi

2. Different approaches for including model error in ensemble forecasts
Develop a state-of-the-art regional data assimilation system, that can be implemented operationally at the National Weather Service of Argentina and provide better forecasts.
Goal
Produces an ensemble of analyses
Is independent from the model
Computationaly efficient
Has tuneable parameters
Tested over different regions and scales (Miyoshi et al., 2007; Yang et al., 2009; Seko et al., 2011; Miyoshi y Kunii, 2012)
Why LETKF?
Why Multimodel?
Including model error in an ensemble can lead for a more realistic spread of the forecast solution (Houtekamer et al., 1996)
Consider different physical parameterization schemes (Stensrud et al., 2000; Meng and Zhang, 2007)
Different approaches for including model error in ensemble forecasts

3. Methodology WRF-LETKF System developed at the University of Maryland.
(Hunt et al. 2007)
Test period: 01 Nov – 31 Dec 2012
Terrain heigth (m)
40 Ensemble Members
Spatial Localization
H= 400 km
V~4 km
Horizontal resolution: 40 km
Vertical resolution: 30 levels
B. C.: 3hs GFS 0.5° deterministic forecasts (not perturbed for each member)
I. C.: 01 Nov 00 UTC. The GFS Analysis was perturbed using differences between consecutive atmospheric states. To generate the 40 perturbations, analysis from October and November of 2010 were used.

4. Lack of thermodynamic observations!
3 hours Analysis with a backward assimilation window of 3 hs 00 03 06 09 12
obs
obs
obs
obs
Observations from the NCEP prepbufr were assimilated
Lack of thermodynamic observations!

5. Example of data assimilated for 12 UTC:
ADPSFC - SFCSHP
ADPUPA - AIRCFT
SATWND (GOES)
A superobbing
technique was
applied to
ASCAT winds

6. Physical parameterizatios used in the WRF model (v3.3.1)
Cumulus: Kain-Fristch (Kain, 2004)
Planetary Boundary Layer: YSU (Hong, Noh and Dudhia, 2006)
Microphysics: WSM6 (Hong and Lim, 2006)
SW radiation: Dudhia (Dudhia, 1989)
LW radiation: RRTM (Mlawer, 1997)
Soil model: Noah (Chen and Dudhia, 2001)
Control
Multimodel
Combination between different options of Cumulus and PBL parameterizations:
Kain-Fristch (Kain, 2004)
BMJ (Janjic, 1994, 2000)
Grell (Grell and Devenyi, 2002)
YSU (Hong, Noh and Dudhia, 2006)
5 members
MyJ (Janjic, 1994)
4 members
Quasinormal (Sukoriansky, Galperin and Perov, 2005)

7. An adaptive inflation was used (Miyoshi, 2011)
The multiplicative inflation parameters are estimated adaptively
Control
They are computed simultaneously with the ensemble transform matrix at each grid point
1st run of Nov-Dec 2012, starting with constant inflation 1.1
2nd run of Nov-Dec 2012, starting with the inflation obtained before
Spatially and temporally varying adaptive inlfation
Multimodel
Convergence of the inflation parameter
(Period that we used to analyze the results)

8. Bigger errors for 18 and 21 UTC
Results
Mean of the RMSD for each DA cycle
Analyses
Little enhancement of the spread with the multimodel exp
Bigger errors for 18 and 21 UTC
3hs fcst
RMSD of Multimodel scheme is less than or equal to RMSD of Control run.
We are happy that the RMSD in our first trial are comparable to GFS based forecasts!
6hs fcst

9. Case Study: 6-7 December 2012
3B42-V7 estimation
A mesoscale convective system developed ahead of a cold front. Strong vertical shear, high values of CAPE.
Warm and moisture advection at 850 hPa
07 Dec 00 UTC
Deterministic WRF I.C. GFS
Ens mean I.C. Control
Ens mean I.C. Multimodel
Accum precip 18 UTC 06dec – 06 UTC 07dec

10. One of the advantages of having an ensemble: to have an estimation of the uncertainty
Spread: shaded
Mean: contoured

11. Domain A
Domain B
Estimation 3B42-V7
Deterministic WRF
The ensemble schemes represent better the structure of the precipitation evolution, although the intensity is underestimated and there is a time lag.
The Multimodel scheme shows higher spread over both domains, and some members are closer to the estimation than the ensemble mean

12. Conclusions and Future work
These are the first experiments of DA of real observations in Argentina
Promising results for an operational application!!!
Method performance OK
3hs regional analyses
Ensemble forecasts
Multimodel scheme better than the Control
Wider spread
Less or equal RMSD
Better performance in a case study of intense precipitation
Problems?
Lack of thermodynamic observation
Little amount of observations for verification
More processing and storage capacity is needed

13. Thank you for listening!
Future work...
More exhaustive verification
Assimilation of thermodynamical observations such as vertical profiles from AIRS or ATOVS AMSU-A radiances
Implementation of no cost smoother for a “cheap” optimization of the analyses (Kalnay and Yang, 2010)
Study of more intense precipitation cases
Thank you for listening!

14. Acknowledgements References
We are grateful to Celeste Saulo for her sugestions. NCEP has very generously made available the GFS analyses and forecasts, as well as the prepbufr observations. Without this essential information, this work would not have been possible.
We also acknowledge the World Meteorological Organization (WMO), for the financial support for the assistance to this Conference.
The equipment used for this research is supported by PIDDEF 41/2010, PIDDEF 47/2010
We are grateful to the NWS of Argentina, the University of Buenos Aires and the CIMA, which are supporting this project.
References
Houtekamer P.L., and L. Lefaivre, Using ensemble forecasts for model validation. MWR, 125,
Hunt, B. R., E. J. Kostelich, and I. Szunyogh, Efficient data assimilation for spatiotemporal chaos: a local ensemble transform Kalman filter. Physica D, 77, 437–471.
Kalnay, E. and Yang, S., Accelerating the spin-up of Ensemble Kalman Filtering. QJRMS
Meng Z. and Zhang F., Tests of an ensemble kalman filter for mesoscale and regional-scale data assimilation. Part II: imperfect model experiments. MWR, 135, DOI: /MWR3352.1
Miyoshi, T., The Gaussian approach to adaptive covariance inflation and its implementation with the local ensemble transform Kalman filter. Mon. Wea. Rev., 139, 1519–1535.
Miyoshi T., S. Yamane, T. Enomoto, Localizaing the error covariance by physical distances within a local ensemble transform kalman filter (LETKF). SOLA, Vol 3, , doi: /sola
Miyoshi T. and Kunii M., The Local Ensemble Transform Kalman Filter with the Weather Research and Forecasting Model: Experiments with Real Observations. Pure Appl. Geophys. 169, DOI /s
Miyoshi T. and Kunii M., Using AIRS retrievals in the WRF-LETKF system to improve regional numerical weather prediction. Tellus, 64A, DOI /tellusa.v64i
Seko H., T. Miyoshi, Y. Shoji, K. Saito, Data assimilation experiments of precipitable water vapour using the LETKF system:intense rainfall event over Japan 28 July Tellus, 63A,
Stensrud D.J., J-W Bao, and T.T. Warner, Using initial condition and model physics perturbations in short-range ensemble simulations of mesoscale convective systems. MWR, 128,
Yang S-C, E. Kalnay, B. Hunt, N. Bowler, Weight interpolation for efficient data assimilation with the local ensemble transform kalman filter. QJRMS, 135, , doi: /qj.353

15. More Details ...
Vertical levels: not a regular distribution, with top at 50 hPa
Initial Conditions:
01 Nov 00 UTC. The GFS Analysis was perturbed using differences between consecutive atmospheric states.
To generate the 40 perturbations, analysis from October and November of 2010 were used.
Example I. C. Member 1: *
GFS Analysis from 16 Oct UTC minus GFS Analysis from 15 Oct UTC
GFS Analysis from 01 Nov UTC

16. Accum precip 18 UTC 06dec – 06 UTC 07dec
Case study: some particular members presented a better estimation of the precipitation than the mean of the ensemble, for both experiments.
3B42-V7 estimation
Ens mean I.C. Multimodel
Ens mean I.C. Control
Accum precip 18 UTC 06dec – 06 UTC 07dec
MEM 34 Multimodel
MEM 31 Multimodel
MEM 20 Control
Both with BMJ and YSU