Testing new generation radiosondes in the UK.
J. B. Elms and J. Nash
Met Office, Beaufort Park, Wokingham, Berkshire, RG40 3DN, UK
+44 (0)1344 855639 , +44 (0)1344 855897 (Fax),
1.Introduction
This paper presents results from tests performed at Camborne in Cornwall (WMO 03808) on new generation radiosondes by the UK Met Office in collaboration with various manufacturers. The radiosondes tested included the Graw DFM-97 (Germany) and Sippican Microsonde GPS (USA), Modem (formerly Geolink) GL98 (France) and various development variants of the Vaisala RS90 (Finland). One aim of the tests was to demonstrate that the Sippican and Graw radiosondes provide sufficiently accurate winds and heights to act as a working reference when testing other operational radiosonde systems, instead of using the old COSSOR tracking radar at Camborne. Various tests have been made using specialised variants of the RS90 radiosonde provided by Vaisala to resolve issues with humidity sensor contamination in cloud identified in the March 2000 test. The tests performed at Camborne are summarised in Table 1:
DATE / RADIOSONDES / OTHER SYSTEMS / REMARKSMarch 2000 / RS90 Acceptance test / Meteolabor Snow white
(Switzerland) / Poor calibration of RS90 humidity sensor
Humidity contamination above low cloud.
Day/night humidity differences.
May 2000 / Graw DFM97,
Sippican Microsonde II / SODAR (German),
COSSOR Radar / Comparison of radar and GPS winds.
December
2000 / MODEM GL98,
Graw + Sippican
Vaisala RS80/RS90 GPS / COSSOR Radar / Comparison of Radar and GPS winds
Preparing for WMO GPS Radiosonde
Comparison in Brazil.
June 2001 / RS90 capped and
uncapped humidity sensor / RS80 / Heat from upper humidity sensor (capped)
affected lower sensor measurements
Jan. 2002 / Graw +
Sippican [ chip thermistor]
RS90 GPS / COSSOR radar / Comparison of radar and GPS winds.
Feb/May
2002 / RS90 capped and
uncapped humidity sensor / RS80 / Testing of revised pulse heating cycle
for humidity sensors was successful
Table 1: Met Office Radiosonde Trials at Camborne 2000-2002
The December 2000 and January 2002 tests provided sufficient simultaneous wind and height measurements for significant statistical analysis. They also gave an opportunity to observe the temperature measurements from the faster response sensors, but there were insufficient measurements in separate day/night categories to provide reliable temperature comparison statistics.
- GPS WIND COMPARISONS
2.1 GPS Windfinding Problems.
A recent WMO report [1] has shown that the average percentage of missing data from measurements using Vaisala RS80G radiosondes is still about 12% world-wide. Table 2 summarises the amounts of missing wind data from the December 2000 test. Interference from the COSSOR radar before launch upset the synchronisation of the Vaisala GPS tracking so that no winds were reported on 3 RS80G ascents. Large gaps in wind messages were noted on another flight where the RS80G unwinder (dereeler) failed. The wind measurement failure rate from the UK ASAP is greater than 20 per cent. When a suitable data logging system was deployed with the ASAP system, the detailed archives showed that a significant number of failures could be attributed to interference from the navigation radars during preparations for launch. Recent tests have shown that the RS90G radiosonde wind measurements were not upset by radar interference in this manner. In the 2000 trial the Sippican GPS system took longest to prepare for launch, but had least amounts of missing wind and GPS height measurements. The Graw system was easiest to use, as GPS synchronisation could be obtained inside the operations room.
RADIOSONDE / No. ofFLIGHTS / FLIGHTS
With no
WINDS / FLIGHTS
With >
90% WINDS / REMARKS
RS80G
/ 16 / 3 / 15¹ / ¹ UNWINDER FAILURESIPPICAN / 18 / 0 / 17¹ / ¹Radiosonde transmitter failure after 51 minutes
MODEM / 13 / 0 / 11¹ / ¹ Faulty ground system antenna amplifier on early ascents
GRAW / 17 / 0 / 15¹ / ¹2GPS failure after about 35minute
RS90G / 4 / 0 / 4
Table 2: Summary of GPS Windfinding Problems during December 2000 GPs Trial.
2.2Simultaneous Windfinding Statistics
During the December 2001 test, upper winds were very strong and the radar was not able to track accurately enough to provide a reference. Thus, the average value of the Graw, Sippican and Vaisala RS80 winds was used as a reference. Figures 1 and show the RMS differences from this reference as a function of height. The statistics are from the 8 ascents when all four GPS radiosondes were launched simultaneously.
Fig 1RMS Differences from reference (average of Sippican, Graw and Vaisala), Camborne Dec2000
These results demonstrated that the errors in the GPS wind measurements of all 4 systems were usually less than
0.5 ms-1 in each component. The 4 systems could be flown together successfully, as was later achieved in Brazil.
- GPS height comparisons
Fig 2 shows differences of simultaneous height measurements compared with the Sippican GPS height as reference. The Sippican and Vaisala RS80 measurements were geopotential height, with the RS80 measurement derived from PTU measurements and the Sippican values form adjusted GPS geometric height. The MODEM and Graw measurements were purely geometric heights. The positive bias of the RS80 in the upper stratosphere results from negative pressure sensor errors in the RS80.
Fig 2 Comparison of simultaneous height measurements against Sippican measurements, Camborne December 2000. Standard deviations of GPS heights with respect to Sippican increase with height from about 10m near the ground up to between 40 and 60m at 30 km.
In the 2002 test, Vaisala RS90 geopotential heights were smaller than simultaneous Sippican GPS geopotential heights by about 50m at 30 km. Graw heights were larger than Sippican by about 80m. These results imply that a combination of GPS radiosondes should provide a suitable height reference for testing other radiosondes in future, although some work is required to clarify the processing procedures used by the various manufacturers.
4.Rate of response of temperature sensors
All four new generation radiosondes tested have small fast-response temperature sensors, and the resultant differences in temperature measurements are most readily seen at low pressures. The temperature response of the Modem and Graw sensors is seen to be much faster than that of the RS80 in Fig 3(a) at a pressure close to 38 hPa in 2000. In Fig.3 (b) the RS90 and Sippican Microsonde II (chip thermistor) nighttime measurements were very similar and the DFM-97 was only marginally slower in response in 2002. This increased speed of response should lead to smaller radiation induced temperature errors, but the benefit will only be obtained if the sensor coating has low emissivity in the infrared and high reflectivity at visible and UV wavelengths. At least one of the sensors could probably benefit from an improved sensor coating to minimise variations in errors caused by changes in the infrared environment during ascent.
Fig 3(a) Comparison of Modem GL98, Graw DFM97 Fig 3(b) Comparison of Graw DFM97, Sippican and Vaisala RS80 temperatures at 38 hPa chip and Vaisala RS90 temperatures at 10 hPa
13/12/00 1425 UTC Camborne. 17/1/02 2042 UTC Camborne
5.Relative Humidity Comparisons
The data sets obtained in December 2000 and January 2002were not large enough to merit a detailed examination of the performance of the relative humidity sensors, which requires separation of the measurements into daytime and night time conditions. However, if day and night measurements are combined, then within the temperature band 0 to –20 ºC, all four systems agreed on average to better than 5 per cent for the range of relative humidity from 95 per cent to 45 per cent. Discrepancies between sensors were larger than this at very high and low humidity. Both Graw and Modem sensor measurements showed occasional anomalies; for instance the Modem measurements in 2 flights were low by more than 10 per cent relative humidity suggesting that all production issues with respect to the sensors may not be fully developed. Fig.4 shows a comparison between the four systems where all were functioning typically. The good agreement in the detailed vertical relative humidity structure can be seen.
Fig. 4 Simultaneous comparison of relative humidity measurements from December 2000
In low level cloud, the average RS80 raw humidity was 105 per cent, the RS90 raw was 103 per cent, the Sippican measurement was 97.5 per cent, the Graw was 97 per cent and the Modem was 99.5 per cent [small sample]. These Vaisala radiosondes were not suffering from chemical contamination, but were probably reading too high and higher than most UK operational radiosondes, see below. A possible explanation may be that the RS80 sensors had been chemically contaminated in the factory during calibration. Subsequent storage for about 3 months with the desiccant in the radiosonde pack might remove the contamination and produce sensors reading too high a value at high humidity.
5.1.Evaluation of operational RS80 relative humidity measurements in cloud in the UK
Calibration of the radiosonde at high humidity can be checked when there is large amounts of low cloud at the time of launch. If the observer reported 8/8 stratus or fog at the time of launch there is little doubt that the radiosonde encountered saturated conditions within the boundary layer. Operational statistics from Sippican soundings in the UK confirmed that maximum relative humidity recorded in cloud during 2001 seldom-reached 100 per cent, whereas in earlier years the sensors often recorded 100 per cent.
Fig. 5 shows histograms of daytime and night-time maximum humidity sensed in low stratus cloud by the current operational Vaisala RS80 radiosonde used on Met Office sites in the UK for the last 3 years. The statistics have been derived from raw data archives since this takes out the rounding down to 100 per cent applied by the Vaisala processing software when the sensor indicates values higher than 100 per cent. At night, for conditions when it is sure that the radiosonde passed through cloud, the average relative humidity measured by the sensor was close to 100 per cent, see Table 3. The average humidity measured in similar conditions by day is about 1.5 per cent less than the nighttime average. This day-night difference is probably due to the effect of solar heating of the RS80 humidity sensor and is similar for full stratus cover, and for slightly lower cloud amounts. The histograms show that in the last 3 years about 8 per cent of all RS80 observations made in low stratus in the UK have measured relative humidity <96 per cent. In future it is hoped that these outliers can be eliminated from operational observations.
Fig. 5 Histograms of relative humidity sensed by radiosondes in the UK where there has been full stratus cloud cover at low levels between June 99 & May 02. Values above 100 per cent are reported as 100 per cent to the user
LAUNCH / DAYTIME / NIGHTTIMEPERIOD / conditions / CLOUD / NUM / AVE % / S.D. / NUM / AVE% / S.D.
01/06/99-31/05/02 / DRY / 8/8 Stratus only / 180 / 98.9 / 3.3 / 285 / 100.3 / 3.2
01/06/99-31/05/02 / DRY / 6 or 7/8 all low / 1994 / 96.3 / 3.7 / 1464 / 97.9 / 3.4
01/06/95-31/05/02 / ALL / 7 or 8/8 all low / 5013 / 99.0 / 4.2 / 5240 / 100.5 / 4.0
Table 3 Daytime and night time maximum relative humidity sensed in passing through low cloud at UK operational stations with manual cloud observations
6.Summary
The comparison of GPS wind measurements confirmed that a combination of GPS systems would produce a suitable working reference for wind measurements. This has already been applied in testing a military radiotheodolite system in the UK. Similarly, GPS radiosondes which decode the GPS signals can clearly act as height references once the differences between geometric and geopotential heights are understood and the correct links implemented. The preparations in the UK led to a successful WMO GPS Radiosonde Comparison in Brazil. The very good quality radiosonde transmitters used by Graw and Modem radiosondes facilitated this. The improved resilience of the Vaisala RS90G to radar interference compared to the RS80G was apparent.
Whilst the speed of response of temperature sensors has improved, more attention should be paid to the coating of the sensors and the positioning of the sensors. Too many modern radiosonde have temperature sensors that can be deployed in a wide variety of positions, rather than in a fixed position relative to the radiosonde body and relative to the supporting wires for the sensor.
Relative humidity sensors are gradually improving, but enough thought has not been given to exposure problems. The Met Office is continuing with a test plan agreed with Vaisala. This will involve testing of a radiosonde with more flexible control electronics that will allow optimised heating cycles to be developed for the current RS90 humidity sensor suite.
Reference: [1] Recent Worldwide GPS Radiosonde Performance - Incorporating the review of WMO GPS Questionnaire, 2001" John Elms WMO web page under IMOP programme