Comparisons between Vaisala RS90 and Snow White relative humidity measurements from the WMO GPS radiosonde Comparison in Brazil (2001) and Ascension Island (1999) Mr. R. Smout, Dr. J. Nash, Mr. D. Lyth and Mr. J. Elms Met Office, Beaufort Park, South Road, Easthampstead, Wokingham, Berkshire, RG40 3DN, UK Tel. +44 (0) 845 300 0300, Fax. +44 (0) 1344 85 5897, E-mail. radiosonde@metoffice.com Introduction The results presented in this paper are from two trials, the first in support of aircraft research at Ascension Island (1999) and then the WMO GPS trial held at Alcantara in Brazil 2001. A main aim of the tests for the Met Office was to evaluate and develop the humidity sensing of the Vaisala RS90 and Meteolabor Snow White chilled mirror hygrometer in a tropical climate. The Snow White sensing system is manufactured by Meteolabor, Switzerland and in these tests was flown with Sippican MKII radiosondes. Sippican, Meteolabor and Vaisala contributed to both tests. Operational Performance of Snow White At Ascension, 17 Snow Whites were flown with the Vaisala RS90 radiosonde and at Alcantara 21 SnowWhites. Snow White measures dewpoint or frost point depending on the state of the film on the mirror. In very dry conditions the peltier cooler on the Snow White may not be able to cool sufficiently to maintain the film. In this situation the system can be become insensitive to humidity increase until the dewpoint rise above the mirror temperature and the water film reforms. In these tests, the loss of the film in flight was not significant. Conversion of Snow White dewpoint to relative humidity was computed using the RS90 temperature measurements. As a 1ºC error in atmospheric produces around 6 per cent error in relative humidity when computing relative humidity from dewpoint, it is essential to have as accurate air temperatures as possible. The change in phase of the water surface on the Snow White mirror can be detected by comparison with the RS90 measurements on many flights. This usually occurs when the mirror temperature is between -20ºC and -30ºC. A correction was made after the basic Sippican in flight processing was completed to ensure that relative humidity relative to a water surface was computed for all heights. Figure. 1 illustrates the good agreement in measuring detailed vertical relative humidity structure as a function of time into flight, obtained from the Vaisala RS90 and Snow White systems in Alcantara. In Figure 2, the Snow White measurements showed a lot more structure in the vertical on smaller scales than the RS90. However, on this occasion the Vaisala ground station software had smoothed out the structures, which the RS90 sensor had sensed, see raw data. Unfortunately, the Snow White sensors deployed at Alcantara had a modified feedback control to speed up the sensor. This led to instability in measurements at lower temperatures, which had not occurred at Ascension Island. These fluctuations were not supported by raw RS90 measurements. The result was that Snow White measurements could not be safely be used as a reference below temperatures of -50ºC in the Brazil test. The oscillations in dewpoint temperature can be seen in the Snow White dewpoint trace between minutes 30 and 60 into flight in Fig. 3. The oscillations died away at about minute 60 and this leads to the suspicion that at the lowest internal radiosonde temperatures the power being delivered to the Peltier cooler for the Snow White mirror may have been inadequate to ensure cooling to the true frost point. In a similar set of dewpoint measurements from Ascension Island, see Fig. 6, the dewpoint trace was much less noisy than the Alcantara measurements. Large positive relative humidity anomalies (more than a hundred percent) were produced by Snow White at upper levels in early Alcantara flights. The Snow White system heats the sensor area when it is recognised that relative humidity is very high and rain drops or ice crystals could enter the sample volume, upsetting the film on the sensing mirror. The sensor volume heating was going on for much too long when it was initiated in thin cirrus cloud at Alcantara. This caused the dewpoint temperature to be much too high for up to 10 minutes after leaving cloud. Thus, additional covers were used to cover the duct entrance to the Snow White sensor area, during most of the Alcantara trial. This made it difficult for ice crystals to enter the sensor volume and trigger the heating. Previous test work in the UK had shown that the design of the Snow White sensor duct ensured adequate flow of air past the sensor even when the duct entrance was almost entirely covered. Excessive heating of the sensor volume was reduced although not totally eliminated in the remaining flights Interpretation of in flight comparisons Figure 3 shows dewpoint temperature measurement from flight 11 as a function of time into flight. Here minute 60 corresponds to a pressure of 88 hPa and is close to the tropopause. Vaisala values of dewpoint were computed from temperature and relative humidity measurements and Snow White values are actual measurements. Fig 4. Shows the corresponding temperature and relative humidity measurements. On this flight the RS90 and Snow White relative humidity measurements were very close up till minute 50 (pressure 144 hPa, temperature -70ºC. After this the RS90 appeared to be slower in response than the Snow White, but as noted earlier the frostpoint reported by the Snow White was probably too low because of problems with the battery supply at the lowest temperatures. Figure 5 shows a similar temperature and relative humidity plot from the trial on Ascension Island. At minute 49 the Snow White shows a much larger positive increment in relative humidity than the Vaisala RS90. This was at a temperature of –65 ºC and a pressure of about 170 hPa. At minute 72 the Snow White's relative humidity increased and this is taken as a symptom of the gradual failure of the peltier cooler to deliver sufficient cooling to give the correct frostpoint. When the remainder of the relatively reliable comparisons are reviewed, the main discrepancies in response between the two humidity systems seem to start at about 150 hPa, and temperatures around –70 ºC. This requires further investigation, once Snow White has been improved to ensure that adequate cooling can be supplied to pressures down to 50 hPa. would be that the humidity sensors are sensing air at a higher temperature than ambient after the air was heated by passing over some of the surfaces surrounding the sensors. Certainly, the RS90 sensors at Brazil agreed with a Thygan reference before launch to within 1 per cent in daytime and nighttime conditions. On the ground , the surface winds were strong and the air reaching the RS90 sensors was not usually passing over any heated surface before encountering the sensor. For the temperature band -15 to -30ºC, the conditions differed between the tests, with high humidity not encountered at Ascension. Systematic differences between Vaisala RS90 and Snow White remained lower than 4 per cent at night, whilst the Vaisala RS90 measurements were clearly lower by around 6 per cent in the daytime comparisons. Relative humidity differences between the Sippican Hygristor used with Snow White Day and Snow White Night are shown for the temperature band 0 to 15 ºC (heights 3 to 5 km) in Fig. 7, and for the temperature band –15 to –30 ºC (approximate heights 8 to 10 km) in Fig. 9. The relative humidity differences between the Vaisala Rs90 and Snow white for the same temperature bands are shown in Figs. 8 and 10. Note the humidity difference scales are quite different between the Sippican and Vaisala RS90 plots. The thick black line indicates nighttime comparisons whilst the light grey lines are for daytime comparisons. The results from Ascension Island are indicated by diamonds and the result from Brazil are indicated by circles. The reproducibility of the Sippican relative humidity measurements is estimated at between 3 and 6 per cent (better at high than low humidity) from the results in Fig. 11, compared to better than 2 per cent for the Vaisala RS90 measurements. Thus, the sample sizes in these tests were not totally adequate to define longterm systematic bias for Sippican to the same degree of certainty as with the Vaisala system. At low humidity (less than 20 per cent) the Sippican sensors indicated values 5 to 10 per cent higher than Snow White in Brazil and about 10 per cent lower than Snow White in Ascension. This is a clear demonstration of the main weakness of the carbon hygristor sensor, that sensor measurements are not reproducible at low humidity. This problem was larger at the lower temperatures ,as seen in Fig.9. At high humidity, the hygristors used with the Snow White at night in Brazil had a strong positive bias, whilst those hygristors used on the other night flight without Snow white were within 5 per cent of the Vaisala measurements. Thus overall the hygristor sensors in Brazil appeared similar to those used at Ascension. Figures 11 and 12 show the standard deviations associated with the comparisons between Sippican and Vaisala RS90 and the Snow White. Here, the results from the two tests have been combined together. Figs. 7 to 12 indicate that the systematic bias between the sensors tends to be larger than the random errors for both sets of sensors. Conclusion The results from the tests between the Vaisala RS90 and Snow White in Ascension and Brazil indicate that it should be possible to produce a working reference for future assessment of radiosonde relative humidity measurements by combining the two systems together, given current technical problems are resolved. In the temperature band 0 to 15 ºC, the Vaisala RS90 comparisons with Snow White were extremely consistent between the two test sites. At night the sensors agreed very closely at high humidity and the Snow Whites were about 4 per cent lower than the Vaisala measurements at relative humidity of about 20 per cent. Some of the difference at low relative humidity would have originated from hysteresis in the Vaisala sensing system. In the day the Vaisala measurements were about 5 per cent lower than the Snow White measurements at high relative humidity. The measurements of the two systems in daytime cloud suggest that most of the difference arises from a low bias in the Vaisala measurements. A reason for this The results demonstrate that there are significant daynight differences between operational relative humidity sensor measurements.This should not be considered surprising because of the methods of sensor exposure. In the future it will be necessary to implement a suitable radiation correction scheme for humidity sensors, similar to that used for temperature sensors. Comparison statistics from the middle troposphere Improved accuracy of relative humidity measurements in the tropics is a significant issue for progress in numerical weather prediction. Further tropical radiosonde tests will be required to ensure optimum accuracy measurements for future operations. Snow White RS90 RS80 RS90 RAW RS90 Snow White Fig 1. Comparison of daytime RS90 and Snow–white temperature and relative humidity measurements, showing typical output from both RS90 edited (used for the reported values) and raw data files before smoothing algorithms were applied. Fig 4. Comparison of temperatures and relative humidity from the same flights as Fig. 3. Snow White RS90 Sippican Snow White RS90 edited RS90 raw Fig 2. Comparison of daytime RS90 and Snow–white temperature and relative humidity measurements. Snow White Vaisala RS90 Vaisala RS80 Fig 3. Comparison of dew point temperatures measured by the Snow-white and values computed from relative humidty and temperature of the Vaisala radiosondes Fig 5. Temperature and relative humidity comparison between Snow White, Vaisala RS90 and Sippican MKII from Ascension 1999. Snow White Vaisala RS90 Fig 6. Comparison of dew point temperatures measured by the Snow White and values computed from the Vaisala RS90. Humidity Difference from Snow White (%) Humidity Difference From Snow White (%) 15 10 5 0 -5 -10 Sippican MKII -15 0 20 40 60 80 4 3 2 1 0 -1 -2 -3 -4 Vaisala RS90 -5 -6 100 0 20 Relative Humidity (%) Humidity Difference from SnowWhite (%) Humidity Difference From Snow White (%) Sippican MKII 5 0 -5 -10 -15 -20 40 60 80 100 Fig 8. Vaisala RS90 - Snow White Day Night Differences from Ascension Island & Brazil 0°C to 15°C 10 20 60 Relative Humidity (%) Fig 7. Sippican MKII - Snow White Day Night Differences from Ascension Island & Brazil 0°C to 15°C 0 40 80 5 Vaisala RS90 3 1 -1 -3 -5 -7 100 0 20 40 60 80 100 Relative Humidity (%) Relative Humidity (%) Fig 10. Vaisala RS90 - Snow White Day Night Differences from Ascension Island & Brazil -30°C to -15°C 10 10 9 9 8 8 Standard Deviation (%) Standard Deviation (%) Fig 9. Sippican MKII - Snow White Day Night Differences from Ascension Island & Brazil -30°C to -15°C 7 6 5 4 3 2 7 6 5 4 3 2 1 1 0 0 0 20 40 60 80 100 Relative Humidity (%) Fig 11. Standard Deviations for figures 7 & 8 Vaisala RS90 circles, Sippican MkII squares 0 20 40 60 80 100 Relative Humidity (%) Fig 12. Standard Deviations for figures 9 & 10 Vaisala RS90 circles, Sippican MkII squares