Second GPS Radio Occultation Data Users Workshop, 22-24 August, 2005 National Conference Center, Lansdowne, Virginia Assimilation of GPS RO Data for Severe Weather Prediction In the Vicinity of Taiwan 1 Ching-Yuang Huang 1 Department of Atmospheric Sciences, National Central Universi ty, Taiwan Collaborators: 2 Ying-Hwa Kuo, 1 Shu-Ya Chen, 1 Jei-Zu Wang and 2 Yong-Run Guo 2 National Center for Atmospheric Research, Boulder, USA Sponsored by NSPO
27
Embed
Assimilation of GPS RO Data for Severe Weather Prediction In the Vicinity of Taiwan
Second GPS Radio Occultation Data Users Workshop, 22-24 August, 2005 National Conference Center, Lansdowne, Virginia. Assimilation of GPS RO Data for Severe Weather Prediction In the Vicinity of Taiwan. 1 Ching-Yuang Huang - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Second GPS Radio Occultation Data Users Workshop, 22-24 August, 2005
National Conference Center, Lansdowne, Virginia
Assimilation of GPS RO Data for Severe Weather PredictionIn the Vicinity of Taiwan
1Ching-Yuang Huang
1Department of Atmospheric Sciences, National Central University, Taiwan
Collaborators: 2Ying-Hwa Kuo, 1Shu-Ya Chen, 1Jei-Zu Wang and 2Yong-Run Guo
2National Center for Atmospheric Research, Boulder, USA
6 LEO satellites provide 2500 global daily measurements.
It will offer up to about 50~100 soundings in a regional model domain for a 6-h assimilation time window.
Dujuan Typhoon(2003/08/31/12)
CHAMP
09:00 (4.240 , 129.910)
10:29 (10.470 , 102.360)
10:34 (-2.400 , 133.310)
11:54 (49.050 , 83.740)
11:57 (42.390 , 104.010)
13:32 (27.040 , 86.150)
Cold front-1(2004/02/06/18)
CHAMP
17:01 (16.190 , 145.940)
18:28 (44.680 , 147.410)
19:59 (50.880 , 118.870)
20:05 (21.710 , 129.010)
20:10 (1.360 , 102.800)
Cold front-2(2004/02/07/06)
CHAMP
04:48 (39.490 , 147.130)
06:23 (44.550 , 151.430)
07:41 (-2.160 , 107.650)
07:49 (23.600 , 135.410)
Meiyu Front(2004/05/19/12)
CHAMP
09:57 (50.090, 103.420)
10:00 (45.490,114.790)
10:01 (33.450,139.760)
11:31 (50.560,100.360)
Data assimilated within initial time ± 3 hr
Conson Typhoon(2004/06/07/00)
CHAMP
06:04 (50.650,146.300)
06:09 (28.500,144.440)
06:15 (1.480,143.640)
07:37 (50.040,133.210)
Mindulle Typhoon(2004/06/29/06)
CHAMP
04:19 (9.100,141.050)
07:18 (35.680,97.080)
08:48 (48.790,80.240)
Aere Typhoon(2004/08/21/18)
CHAMP
15:00 (21.180,104.380)
16:32 (25.930,89.000)
16:37 (44.030,87.790)
Nanmadol Typhoon(2004/12/02/12)
CHAMP17:2218:5118:55
18:58
(35.50,112.720)
(47.270,87.850)
(24.320,93.90)
(18.180,92.590)
Cold front/Southwesterly(2004/12/03/12)
CHAMP
17:22
18:51
18:55
18:58
(35.500, 112.720)
(47.270,87.850)
(24.320,93.900)
(18.180,92.590)
Model Domain and Physics
The model simulations use three nested domains at 45-, 15- and 5-km resolutions.
All the simulations use MM5 version 3.5 or 3.6 with explicit treatments (Goddard’s scheme) for ice/graupel physics in the three domains (1, 2 and 3), Anthes Kuo’s scheme and Grell’s scheme for cumulus parameterization in domain 1 (largest) and domain 2, respectively, and the Blackadar scheme for PBL parameterization in all the domains.
3DVAR was performed for each domain with GPSrf.
MM5/WRF 3DVAR
Cost function defined by
)]}([)]([)(){(2
1 11 xhyOxhyxxBxxJ obsT
obsbT
b
x : N-component vector of the analysis variable ,
xb : N-component vector of the background variable (first guess)
yobs : M-component vector of the observations ,
B : Forecast-error covariance matrix ( N ×N ),
O : Observational covariance matrix ( M ×M ),
h : Transformation operator (converts analysis to observation),
N : Number of degrees of freedom in analysis,
M : Number of observations.
Covariance Matrix- O The GPS radio occultation observational covariance matrix is diagonal and thus has assumed no vertical correlations.
This assumption of vertical un-correlation is certainly not supportive of some existing dependence between observations, but in absence of statistical information on those correlations, the assumption insures that the data information is not underestimated in assimilation.
The diagonal elements (variances) are prescribed as a profile exponentially decreasing from 3 N at 100 hPa to 10 N at 1000 hPa. The value of 10 N observational error near the surface is slightly than, but consistent with, the 3% refractivity difference between CHAMP radio-occultations and ECMWF analysis found at 1000 hPa, as reported by Kuo et al. (2004).
Threat Score (TS)
A: the number of the grids on which both
forecast and observation exceed the threshold, F: the number of the grids on which forecast exceeds
the threshold, and
O: the number of the grids on which observation exceeds the threshold.
> 1,500 verification grid points on the island.
AOF
ATS
Cases
Thresholds
NariGTS
NariGPSrf
NariGTS
NariGPSrf
NakriGTS
NakriGPSrf
NakriGTS
NakriGPSrf
0-24 h 0-24 h 24-48 h 24-48 h 0-24 h 0-24 h 24-48 h 24-48 h
0.25 mm 0.545 0.540 0.539 0.536 0.504 0.480 0.505 0.475
0.5 mm 0.544 0.539 0.538 0.535 0.482 0.464 0.498 0.448
1 mm 0.544 0.542 0.532 0.533 0.411 0.425 0.486 0.411
2 mm 0.530 0.524 0.527 0.531 0.376 0.363 0.463 0.378
5 mm 0.489 0.484 0.530 0.528 0.273 0.244 0.321 0.320
10 mm 0.489 0.492 0.527 0.530 0.182 0.174 0.277 0.278
15 mm 0.503 0.496 0.526 0.538 0.149 0.140 0.248 0.258
25 mm 0.524 0.496 0.543 0.531 0.126 0.127 0.233 0.235
50 mm 0.497 0.485 0.462 0.504 0.086 0.088 0.113 0.166
100 mm 0.473 0.444 0.277 0.393 0.009 0.000 0.000 0.129
TS is generally higher for the run with assimilated QuikSCAT data.
(Huang et al., 2005, Weather and Forecasting, August)
Table. Threat scores (TS) and root mean square errors (RMSE) (mm) of the accumulated rainfalls between 0-24 h and 24-48 h for different thresholds for the Lekima (2001) and Sinlaku (2002) simulations with and without GPS refractivity data. (Huang, 2003, NSPO project report)
Lekima Sinlaku
0-24h 24-48h 0-24h 24-48h
No GPS GPS No GPS GPS No GPS GPS No GPS GPS
0.25 mm 0.534 0.534 0.420 0.421 0.353 0.359 0.289 0.290
0.5 mm 0.531 0.531 0.401 0.404 0.350 0.348 0.273 0.269
1 mm 0.523 0.523 0.385 0.380 0.357 0.340 0.253 0.246
2 mm 0.504 0.504 0.361 0.359 0.340 0.328 0.231 0.228
5 mm 0.423 0.423 0.307 0.301 0.317 0.323 0.232 0.233
10 mm 0.373 0.373 0.226 0.233 0.308 0.308 0.240 0.244
15 mm 0.343 0.346 0.167 0.157 0.295 0.297 0.242 0.236
25 mm 0.276 0.282 0.080 0.075 0.254 0.272 0.219 0.207
50 mm 0.241 0.245 0.019 0.029 0.182 0.188 0.127 0.127
100 mm 0.218 0.207 0.000 0.000 0.018 0.032 0.005 0.004
Haitang Simulation ongoing with WRF…Episode : 07/17/2005-07/19/20052~4 GPSrf
Conclusions
The simulated results indicate the impact of several GPS refractivity data is only marginally positive for some severe weathers in the vicinity of Taiwan.
More positive impact may be procured by combinations with other unconventional data.
Much more data from COSMIC can be assimilated and might be very helpful for improving weather prediction in Taiwan.
topr
r
o rdrrr
yxdlyxrS
0
20
2
modmodmod
),(2),()(
Modeling of the phase along the straight line;Inverting the modeled phase into a virtual (modeled) refractivity
),(1),( modmod yxyxn
022
0
0modmod
/1)( dr
rr
drdSr
topr
r
2-D model refractive index:
The phase integrated along the straight line:
Inversion of the integrated phase:
rtopr 0r
0
dl(From Sokolovskiy et al. 2004)
Refractivity mapping
Nonlocal operator
local operator
WRF 3DVAR with local and nonlocal operators by 2006.
Comments on Nonlocal Refractivity Operator Assimilation
Errors increase considerably near upper boundary, due to the less cancellation effects in a shorter path. Thus, local refractivity operator may still be recommende
d above the tropopause.
The refractivity mapping normally has even larger errors compared to errors for excess phases, due to the application of Abel inversion in a finite domain with non-negligible local refractivity.
The straightline forward operator is very fast and the assi
milation will fit computationally into real-time operations.