Radiosonde O-B statistics in the stratosphere AOPC meeting in Exeter 29 March 2017 Bruce Ingleby ECMWF Bruce.Ingleby@ecmwf.int ECMWF April 3, 2017
Overview Introducing myself (1 slide) Radiosonde trends and data continuity Move to BUFR and high resolution Status of GUAN 2015-2016 O-B statistics by radiosonde type Overview Documentation Meta-data problems, including position Summary 2
My work Met Office (1984-2013): some work on climate and seasonal forecasting, mainly data assimilation for NWP 2007-2013 use of in situ observations, dust assimilation Ingleby et al (2013, JTech) Surface humidity reports Ingleby + Edwards (2015, ASL) HiRes radiosonde data ECMWF (2013-now): assimilation of in situ observations Use of aircraft humidity, buoy OSEs Move to BUFR data (radiosonde, SYNOP, marine) Study of different radiosonde types Using GRUAN radiosondes as reference data, GAIA-CLIM 3
Trends 1: general Dirksen et al (2014, AMT) paper on GRUAN processing for RS92 very useful Some users changing from RS92 to RS41 or other radiosondes, Vaisala will stop making RS92 at the end of 2017! Meisei bought out new, smaller radiosonde ims-100 Ongoing move to radiosondes without pressure sensors for most purposes US NWS starting to move away from use of 1680 MHz carrier frequency to 403 MHz (used elsewhere) Use of GPS for surveying station locations: OK but surveyors need to make sure that heights are relative to geoid (sea level) rather than reference ellipsoid (GPS default)! One 40 m error found. WMO guidance should mention this issue. 4
Trends 2: move to BUFR data (and high resolution) Migration supposed to be complete by November 2010 2014 still going on ECMWF set up https://software.ecmwf.int/wiki/display/tcbuf/ to help Radiosonde more complicated than surface data: change structure (no parts A/B/C/D); allow high resolution reporting plus time/lat/lon at each level; allow extra metadata (eg. software version, radiosonde serial number). Launch position in BUFR (some diffs from Pub9A & OSCAR/surface) Currently 20% of radiosonde stations send high resolution BUFR, 8% send low resolution native BUFR, 44% send reformatted TEMP, 28% don t send BUFR Most reformatted TEMP have parts converted separately doesn t meet BUFR coding standard Paper Progress towards high-resolution, real-time radiosonde reports, Ingleby et al (2016, BAMS) BUFR data not currently in an open archive : discussions with NCEI regarding addition to IGRA 5
Trends 3: Data precision/rounding Comparison between TEMP and BUFR (Ingleby and Edwards, 2015, ASL) showed up some issues with TEMP coding/decoding, last bit used to indicate + or - ºC so TEMP precision is 0.2 degrees. Temperature offsets look at one decimal place (1DP) case first: TEMP coding TEMP decoding +13.4 C +13.4 C (+273.1) MO 134 +13.5 C +13.45 C (+273.15) EC RS92 with DigiCORA III coding (raw data to 2DP): 13.40 13.59 C coded as 134, average ~13.5 C. The values as decoded by ECMWF are about 0.05 low (~ 0.15 for MO) MW41 (some RS92, ~all RS41) coding seems to have changed (so ECMWF decoded values are about 0.05 high? TBC) Modem M10 TEMP reports seem OK comparing TEMP & BUFR at ECMWF (TEMP reports reformatted to BUFR not necessarily consistent) 6
Trends 4: austerity measures Jan 2015 Russia cut back from 2 ascents/day to 1 ECMWF performed OSEs using 2013/14 data cut back worsens forecasts in winter (EC newsletter 149) ECMWF DG made representations via WMO April 2015 Russia reversed the cutback 2016: Mexico and Brazil reduced reporting at 00 UTC (dashed lines) African numbers fluctuate (comms problems?) India and Indonesia have increased radiosonde ascents Quality of Indian reports has improved somewhat (but their JinYang reports worse than Korean ones) 7
Impact of 1 vs 2 Russian reports per day (EC newsletter 149) Short range impact mainly at low levels over Russia (plot from Mark Rodwell) Spreads downstream and upwards (~6 months NWP improvements to NH at D+5) This is in Winter (no tropospheric sat. radiances over snow) less impact in Summer No wind profilers, very few aircraft ascents/descents over Russia Russian radiosonde quality worse than average in UTLS (less difference in LT) 8
15 GUAN stations not reporting in December, esp island/african stns Some are temporary, but this is typical GUAN not distinguished in NWP GUAN and other radiosonde subsets GRUAN dominated by (now restricted to) northern extratropics some Australian stns in the pipeline 9
GUAN 2 Mean and rms O-B for two years by latitude band GUAN is not homogenous in quality: Russian radiosondes worse. Select group RS41, LMS6, M10, Meisei, Shanghai similar T stats to RS92 (some worse for UT humidity) Mean O-B very similar except at top and bottom ECMWF B too cold between 100 and 20 hpa 10
2015-2016 temperature O-B Statistics on standard levels, split by radiosonde type (colours) and latitude band (plots). Bias dashed, rms solid. Lower stratospheric (100-20 hpa) cold bias in B, largest N of 50 N, smaller in tropics. Large near surface differences N of 50 N (esp in Winter), B not good at inversions; inversions too cold in Russian radiosonde data? Stratospheric rms larger in tropics than in extratropics probably due to more gravity waves in tropics (Alexander et al, 2002) 11
2015-2016 RH O-B Lower troposphere statistics comparable, but big diffs in UT (ECMWF only uses RS92/41 in UT) Huge UT biases in Russian data reduced by EC bias correction (would like to switch off EC biascorr for RH) Other sondes also suffer from cloud contamination to a lesser extent RS41 and RS92 best (esp in tropics) Tropical UT statistics improved by change to use Sonntag SVP eqn in late 2016 (not shown) 12
2015-2016 wind O-B Similar statistics for wind: bias of wind speed (dashed), vector wind rms (solid) B wind speeds slightly low (0.2 m/s) in general, but larger bias (~ 0.8 m/s vs RS92) in tropical UTLS Some Meisei ims-100 winds oversmoothed? Less difference between radiosonde types for wind than for other variables 13
2015-2016 height O-B Height not assimilated but used in verification (precision better in BUFR) Tropospheric bias problems in ~10% of stations (H/HP mix up at Saudi stns using RS41, linked to lack of P sensor) Beijing CF06 reporting geometric height not geopotential height! Low level background T bias in tropics gives height bias 14
Documentation of different radiosondes Type Papers GRUAN product? Vaisala RS92 Many Yes Vaisala RS41 1 In progress Meisei 2 Close for RS-11G Modem M10 - In progress Lockheed LMS6 - Chinese - (in Chinese?) Graw DFM-09 - In progress Intermet - Meteolabor SRS-C34 1 Close Russian - (1 preprint) US NWS implemented its Radiosonde Replacement System from 2005-13. various conference abstracts/presentations about data continuity study by R Brown et al, but no report. 15
Metadata Latitude, longitude, height Some problems in going to BUFR (eg 10 30 converted to 10.3 ) Occasional sign of latitude/longitude errors ECMWF resets BUFR position if significant difference from list (phase out?) Which is primary source of information BUFR or OSCAR/surface? Radiosonde type Sometimes missing (China, India, other) Sometimes wrong (CF06 type only allocated after it started being used operationally reports using no type in India, type 32 in Malysia) Software version and serial number 16
Some of the worst problems may be radar related? Example of PAZA type 15 radiosonde (uses radar, no P sensor, Ukraine) black, red line is ECMWF background Agreement OK in LT, dreadful in UT (too large in vertical extent to be model problem) looks like vertical offset, also seen in winds Radar misalignment? Other? Type 15 O-B stats very poor (4 stns) Some other types/ascents appear to show less severe forms of similar problem China uses radar but radiosondes have P sensor largely OK 17
Summary Various challenges including austerity measures Slow move to BUFR is both a challenge and an opportunity Shows up minor temperature offsets from TEMP (de)coding Global network, and GUAN, are not homogenous in quality Station height problems are reducing the value of verification against radiosonde heights (also geometric height reporting from CF06) Regular radiosonde monitoring by ECMWF and others but it is up to NMSs to note and fix problems EUMETNET more proactive within Europe generally successful Radiosondes still important for NWP (especially in NH) and climate reanalysis (especially pre-gpsro) Use as reference more difficult when swamped in numbers by satellite data (Eyre, 2016) Need reference data for different latitudes can we link operational RS92 to GRUAN uncertainties? 18
19
MetOffice ECMWF NWP- GRUAN The GRUAN Processor Temperature (B-O) for 2013 Mean total uncertainty LIN (NH) 1297 profiles MAN (Trop) 82 profiles LAU (SH) 32 profiles All sites* 3895 profiles This project has received funding from the European Union s Horizon 2020 research and innovation programme under grant agreement No 640276. www.gaia-clim.eu
Diurnal variation of temperature uncertainty Dirksen et al (2014, AMT) GRUAN (reference radiosonde) RS92 processing Their figure 10: --- estimated uncertainty. --- ΔT from dual launches Uncertainty increases with altitude, especially during daylight (top) radiation effects (orientation of the sensor) Diurnal variation can be seen (just) in (O-B) T statistics more in Z statistics GRUAN daytime radiation correction differs from Vaisala but net result similar (modified spike removal) Nighttime radiation correction small, GRUAN uses Vaisala correction 21
Winds 20 S-20 N at 100 and 70 hpa Hi 22
Winds 20 S-20 N at 30 and 10 hpa 23