Impact of Assimilating AMSU-A Radiances on forecasts of 2008 Atlantic TCs Initialized

with a limited-area EnKF

Zhiquan Liu, Craig Schwartz, Chris Snyder, and So-Young Ha

NCAR/NESL/MMMBoulder, Colorado, USA

05/22/2012 5th EnKF Workshop, Albany 1

NCAR is sponsored by the National Science Foundation

Outline

• Radiance DA in WRF/DART

• Results from 2008 Atlantic hurricane season

• Conclusions

05/22/2012 25th EnKF Workshop, Albany

DART/EAKF: vector-matrix formLiu et al., 2012, MWR

• a two-step square-root filter (Anderson, 2003)– adjustment step (shift+compact) for observation space analysis

– regression step from observation space to model space analysis increment

05/22/2012 5th EnKF Workshop, Albany 3

€

y ia = A y

1/ 2(HBHT )−1/ 2 y ib − yb

( ) + ya, i = 1,K ,N

€

x ia − x i

b = BHT(HBHT )−1 y ia − y i

b( )

05/22/2012 5th EnKF Workshop, Albany 4

Radiance DA in WRF/DART

• Make use of observation operators built in the WRFDA-3DVAR.– Radiance obs prior is calculated from WRFDA-3DVAR using CRTM

– Inflate the background before the radiance calculation

• Outlier check performed in DART

• Peak level of weighting function used for vertical localization

• Make use of bias correction utility in WRFDA-3DVAR

May June July Aug Sept.

Radiance Bias correction coefficient spin-up

Recent work indicates spinning up coeffs for O(months) is beneficial

05/22/2012 55th EnKF Workshop, Albany

€

J(β ) =1

2(β − β b )TB β

−1(β − β b ) +1

2[y − ˜ H (x r ,β )]TR −1[y − ˜ H (x r ,β )]

Reference field for coeffs training can be: from global analysis (NCEP GFS used in this study), or EnKF analysis, or other regional analysis

Experimental period

• 11 Aug-15 Sep 2008– 5 storms: Fay, Gustav, Hannah, Ike,

Josephine

05/22/2012 65th EnKF Workshop, Albany

Experimental Design• Two principal WRF/DART 6-hourly cycling experiments

– (1) Assimilate solely “conventional” (i.e., non-radiance) observations

– (2) Assimilate conventional observations and AMSU-A channels 5~7 radiances from NOAA-18 and METOP-2.

• WRF V3.2.1: 36km, 36 levels to 20hPa.– Deterministic 72-h forecasts from ensemble mean analyses at 00Z,

12Z

• DART: – 96 members, ±1.5 h time-window, LBCs from GFS– Adaptive inflation and localization, No surface obs except

altimeter

• Radiances:– Thinned to 72 km– Static bias correction coefficients from offline monitoring spun-

up for 3 months prior to experiment using GFS as the reference field

05/22/2012 75th EnKF Workshop, Albany

Obs coverage @ 00Z 17 Aug. 2008

05/22/2012 85th EnKF Workshop, Albany

Mean T analyses vs. ERA-Interim

(domain-averaged)

Time (x-axis)Model level (y-axis)

05/22/2012 5th EnKF Workshop, Albany 9

Single 3-d forecast before landfall

Fay Gustav

Hanna Ike

05/22/2012 105th EnKF Workshop, Albany

Track/intensity errors for all storms

05/22/2012 115th EnKF Workshop, Albany

Solid: no-radianceDashed: radiance

48-h forecasts vs. dropwindsondes obs

NOAA G-IV dropwindsondes sampled TC environment, not TC core.

208 dropsondes used in verification

05/22/2012 125th EnKF Workshop, Albany

Mean difference of EnKF analyses – ERA-Interim

05/22/2012 5th EnKF Workshop, Albany 13

Mean difference of Radiance minus non-radiance analyses

Radiances made the analyses colder over Atlantic, therefore reduced the warm bias and weakened steering flow.

05/22/2012 145th EnKF Workshop, Albany

Importance of simultaneouslyassimilating radiances and satellite winds

05/22/2012 155th EnKF Workshop, Albany

Conclusions

• Assimilating AMSU-A radiance improved TC track and intensity forecasts, particularly for forecast range beyond 36-h.

• Track improvement likely caused by improved environmental analysis, e.g, reduce T bias and improve the wind depiction.

• Simultaneous assimilation of radiances and satellite winds is important to maximize the benefit from both data sources.

05/22/2012 5th EnKF Workshop, Albany 16