on the accuracy of different radiosonde types · • thanks to aki lilja (vaisala), lars isaksen...

14
© ECMWF October 17, 2016 On the accuracy of different radiosonde types TECO-2016 Madrid, 30 September 2016 Bruce Ingleby ECMWF [email protected]

Upload: others

Post on 02-Oct-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

© ECMWF October 17, 2016

On the accuracy of different radiosonde typesTECO-2016 Madrid, 30 September 2016

Bruce [email protected]

Page 2: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

Radiosonde height reports• Short-range Z500 verification:

• Vs analyses shows improvements in recent years

• Vs radiosondes – insensitive to forecast improvement (large obs error?)

• Reasons:

• Rounding to 10 m in TEMP code (native BUFR improvement)

• Wrong station height (eg some Saudi stations after upgrade to RS41, confusion between different heights or simple typing error?)

• Use of GPS height without adjustment to the geoid (eg Greenland summit)

• Ideally fix height errors at source (could apply correction at NWP centres)

• ‘Irreducible’ measurement error?

• Poster (Ingleby, Dahoui and Lehmuskero) being prepared for TECO 2016

2EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Page 3: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

Overview• Introduction

• Results for Jan-July 2016 (2015 similar)

• Migration to BUFR

• Summary

• Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions

• This study was supported by Vaisala Oyj (other radiosonde types are also available)

3EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Page 4: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

Intoduction• Major radiosonde intercomparisons (eg Nash et al, 2011) are infrequent and expensive, and can’t look at all radiosonde types

• They don’t sample some ‘real world’ problems (eg wrong station height)

• Numerical Weather Prediction (NWP) Data Assimilation systems routinely generate comparisons between observation (O) and background (B – 12 hour ECMWF forecast) for all reports on the GTS

• Issues:

• different geographical sampling for different radiosonde types (present results by latitude zone to minimise this)

• Seasonal and diurnal variations (average over these)

• Results presented for TEMP standard levels

• One motivation is to introduce uncertainty estimates that vary by radiosonde type – try to improve analyses and forecasts

4EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Page 5: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

Coverage• hi

5EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTSVaisala (RS92-NGP used in US) M10 Various +others 4 types inc Paza (UKR) InterMet (ZAF)

Page 6: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

Temperature statistics• Dashed lines O-B bias, solid rms

• Key: TYPE NNN/100

• 50-90N: large diffs at 1000 hPa –B not good for inversions (Winter), RS92/RS41/LMS6/M10 good, Russian – less good fit; +ve O-B in stratosphere

• Tropics: +ve O-B at low levels (B), large rms in stratosphere (gravity waves), most types good ex Graw

• 20-50N (split): PAZA and Nanjing have problems; all have +ve O-B above 100 hPa

6EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Page 7: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

Relative Humidity statistics• Very variable in UT, most types ECMWF doesn’t use for T<-40, P<300

• Vaisala types (RS92/RS41/NGP) generally perform well, also LMS6, M10 and Graw (100 hPa peaks in tropics), Meisei OK but stops at 250 hPa, Chinese and Jin Yang poor in UT, Russian sondes often saturated after cloud – huge UT biases

• Low/mid troposphere biases not bad, large tropical UT bias partly use of Buck svp formula in ECMWF processing, being replaced by Sonntag

7EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Page 8: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

Wind statistics• Speed bias and vector wind rms

• Less type dependence than for T and RH, small speed bias (B problem?)

• PAZA (and Jin Yang) problems

• Meisei iMS-100 (Me35; used in Turkey): over-smoothed?

8EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Page 9: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

Height statistics• At low latitudes, low level B T bias gives Z bias at 500 hPa and above

• Also problems from station height errors at 30+ stations (various types)

• Also geometric height conversion problems(?) at some ** stations in tropics

• BUFR gives better precision of 1 m for P<700 hPa – improves rmsO-B

• See poster for more details

9EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Page 10: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

ECMWF applies bias corrections to T and RH

• Based on Agusti-Panareda et al (2009, QJ), updated monthly

• Function of sonde type, pressure and solar elevation angle

• Use nighttime RS92 as reference and O-B – assumes B biases are homogeneous

• T corrections mainly small, biggest for Graw out of common types

• We would like to reduce or remove RH bias corrections – trials being performed

10EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Page 11: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

Migration to BUFR

• Migration slow, most reformatted TEMP reports (orange) are unusable

• https://software.ecmwf.int/wiki/display/TCBUF Ingleby et al (2016, BAMS)

11EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Page 12: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

A curious feature: low level dry slot• Very low near-surface Td values seen various times from 16429, occasionally from other stations

• Hypothesis: real but not natural –industrial plume?

• Any other ideas?

12EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Page 13: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

Summary and future work• Results from mean/rms O-B and rejection rates (not shown)

• Some radiosonde types have more quality problems than average: Paza, Graw, InterMet, Nanjing and Jin Yang1 (1 at Indian stations, OK at Korean stations). Some Russian stations OK, others less good.

• Vaisala types generally good, also LMS6 and Modem M10 (slightly less good at UT humidity in tropics) and Meisei and Shanghai (less good at UT humidity, Me35 wind smoothing?)

• Happy to talk to manufacturers and NMSs to provide feedback and understand issues better.

• Write up study, introduce type dependent error estimates and reduce bias corrections applied at ECMWF.

• TEMP to BUFR migration is ongoing – please get your BUFR reports checked by other (NWP) centres before switching off TEMP

13EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Page 14: On the accuracy of different radiosonde types · • Thanks to Aki Lilja (Vaisala), Lars Isaksen (ECMWF) and Alexander Kats (Roshydromet) for various discussions • This study was

RS92 O-B statistics• Stratospheric T and Z (and wind?) slightly better at night –consistent with solar radiation causing greater uncertainty (egDirksen et al, 2014)

• Mid-tropospheric RH worse at night?

• Diurnal variation mainly due to obs error (rather than model?)

• Min rms O-B for T <0.7°(depends partly on latitude)

14EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS