1. Item 4.6 - Taking into account radiosonde position in verification
ET-PWFPS
Beijing, China
Tom Robinson
Canadian Meteorological Centre
March 12-16, 2018

2. Changes to radiosonde observations
New selection scheme for TAC vs BUFR profiles
Use of high-precision of temperature and humidity observations
ECMWF rejection criteria for humidity
Revised saturation vapor pressure formula (AERK)
Revised humidity limits in the analysis
7 February 2018, 1200 UTC

3. Benefits of using radiosonde data in BUFR
Better usage of data because of accurate time and position information for all levels
More data because of more levels
e.g. every 2 seconds giving ~3500 levels
Higher precision of observations
Temperature: 0.01 K
Dew point: 0.01 K
Wind direction: 1 degree
Geopotential height: 1 m

4. Radiosonde data processing
Selection process :
verifies BUFR profile completeness
QC of native BUFR profiles compares native drift and estimated drift based on the native time displacement and the averaged wind between 2 pressure levels.
a dry energy norm of short-range forecast departures (O-B) is calculated for both TAC and BUFR profiles.
BUFR profile selected only if:
the number of observations in the BUFR profile is greater than in the TAC profiles
less than 10% of the profile contains suspicious drift positions
the dry energy norm of the BUFR profile is less than 1.3 the norm of the TAC profile
GTS
~95%
42%
28%
Reformatted BUFR
No drift positions
Native BUFR
Drift positions
TAC
Selection
scheme
TAC 83%
BUFR 17%
Data Assimilation
Systems

5. Radiosonde data processing
Radiosonde profile Schleswig, Germany
16 December Z after thinning
Radiosonde profiles are thinned in such a way that one set of observations (temperature, wind and dewpoint) are selected per model levels (dashed green lines). The figure shows an example of data selection for the Schleswig station, which transmitted a high-resolution BUFR profile of 2468 levels and a TAC profile of 80 levels at 0000 UTC, 16 December 2015.
After thinning, 68 levels were selected from the BUFR profile, corresponding to all model levels up to 8 hPa, while 29 levels for the temperature and 43 levels for the wind were selected from the TAC profile. The number of levels for temperature and wind are not the same for TAC because significant levels for temperature and wind are not the same.
BURF
TAC

6. New rejection criteria for humidity observations
Currently all radiosonde humidity observations are assimilated from surface to 70 hPa

7. New rejection criteria for humidity observations
Currently all radiosonde humidity observations are assimilated from surface to 70 hPa
New rejection criteria for humidity (based on ECMWF implementation):
Rejected if above 100 hPa for Vaisala 92 and 41 series
Rejected if above 300 hPa for all other radiosonde types
Rejected if T < -60 C for Vaisala series 92 et 41
Rejected if T < -40 C for all other radiosonde types

8. Change in the number of observations assimilated per day in the GDPS
Average number of observations assimilated (in million)
during the winter 2017 period
GDPS 6.0.0
GDPS 6.1.0
Variation
Radiosonde
0.165
0.178
+7.6%
AMV
0.278
0.271
-2.2%
Radiances
10.857
11.077
+2.0% (CSR +0.6%)
Total
12.585
12.794
+1.7%

9. Verification Scores
Verification scores against radiosondes were made for the two following period:
Summer 2016: 16 June 2016 – 31 August 2016
Winter 2017: 15 December 2016 – 28 February 2017
The final cycles that use the operational GDPS components is the control (in blue)
The cycles that includes all the changes is the experiment (in red)

10. O-A, O-B over Northern Hemisphere (Winter 2017)

11. Verification Scores against Radiosondes over Northern Hemisphere (Winter 2017)

12. Verification Scores against Radiosondes over Southern Hemisphere (Winter 2017)

13. Verification Scores against Radiosondes over Northern Hemisphere (Summer 2016)

14. Verification Scores against Radiosondes over Southern Hemisphere (Summer 2016)

15. Contribution of each component of Phase 1 to the GDPS forecast improvements in the troposphere
Northern Hemisphere
Summer 2016
Winter 2017
Short-range
Medium-range
4D-EnVar
RTTOV-12 o
AMVs
Radiosondes (TAC vs BUFR)
Radiosondes (ES screening)
CSR (WV channels)
GB-GPS
Aircraft + + + + + + + _ + + + + _ _ + + _ + + +
Southern Hemisphere
Summer 2016
Winter 2017
Short-range
Medium-range
4D-EnVar
RTTOV-12
AMVs o
Radiosondes (TAC vs BUFR)
Radiosondes (ES screening)
CSR (WV channels)
GB-GPS
Aircraft + + + + + + + + + + + _ + + + +

16. Mean Drift Distance and Position Error
The position errors are calculated by comparing the retrieved positions with the actual positions from the SPARC dataset, which are obtained from the Global Navigation Satellite System (GNSS).
The mean horizontal drift distance steadily increases with height and reaches 45 km at 10 hPa for July 2008 and close to 100 km for December 2008.
Overall, the mean retrieved position error for both methods is about 10% of the mean balloon drift distance.

17. Estimation of the Horizontal Position and Time of Radiosonde Data
1200 UTC 15 December 2008

18. Verification scores against radiosonde data with and without balloon drift
Verification scores made with forecasts at synoptic time (0000 and 1200 UTC)

19. Verification scores against radiosonde data with and without balloon drift
Verification scores made with forecasts at synoptic time (0000 and 1200 UTC)

20. Verification scores against radiosonde data with and without balloon drift
Verification scores made with forecasts at synoptic time (0000 and 1200 UTC)

21. Verification scores against radiosonde data with and without balloon drift
Verification scores made with forecasts at synoptic time (0000 and 1200 UTC)

22. Conclusions
Radiosonde drift position has an effect on short-range forecast verification scores above 500 hPa
The effects of using both time and position have not been studied
The issue is not critical yet as only 30 % BUFR compliance, but needs to be addressed
It is proposed that a plan of action be put in place for further study of the issue and for developing a proposal for possible implementation of radiosonde position and time in the CBS verification exchange for the next meeting of the ET-OWFPS