JASMINE sondes (Vaisala RS80) were launched using the Vaisala system on board the NOAA R. V. Ronald H. Brown. A total of 23 sondes were launched from April 10 - 17 for JASMINE leg 1, and 171 sondes were launched from May 6 - 30 for JASMINE leg 2, giving a total of 194 profiles. During the first leg nominally 3 soundings per day were collected, while nominally 6 soundings per day were collected during the second leg. The second cruise also consisted of two "star" ship track patterns in the Bay of Bengal 5 days long and about a week apart. Launches increased to 8 per day during this portion of the cruise, and CTD casts were increased to every 0.5^o. Sounding data include pressure, temperature, humidity and GPS-determined horizontal winds. Table 1 provides information on sounding locations, dates and general statistics. Table 2 details the launch statistics, including analysis of failed and shallow launches.

The sounding data were processed using basic assumptions about the tropical atmosphere, as well as ascent rate and adiabatic lapse rate, to detect questionable and bad values of T, P, RH, U and V. The processing is based on that used by NCAR/JOSS for the TOGA COARE soundings (Loehrer et al., 1996), where 5 basic flags are defined: 1=good, 2=questionable, 3=bad, 888=not assigned, and 999=missing. Full details of the criterion used to define these flags is provided in the readme file (readme.flags) which should be obtained when the data files and flag files are obtained from the JASMINE sounding website.

The basic results of the quality control are shown in Table 3. For both leg 1 and 2 the number of soundings with 100% and at least 50% of their wind data are shown in columns 2 and 3. These percentages are with respect to the number of successful launches from Table 2 (23 and 171 for legs 1 and 2, respectively). Also shown for both legs are the number of soundings with 100% of their temperature and humidity data in column 9. The log book from leg 2 allows us to look at wind losses and temperature and humidity biases as a function of sonde batch. To determine any relationship between sonde wind loss and batch, only the subset of soundings which attained pressures less than 100 mb were considered. This subset was chosen as it was not generally affected by weather or sonde related problems affecting the shallow soundings. From this subset of sondes (21 from leg 1 and 151 from leg 2), the number of sondes with less than 75% of their wind data were determined and are given in column 4. The further categorization of these sondes into their respective batches indicates that the later sondes had the most trouble with wind losses. There is currently no explanation for this result.

The temperature and humidity data did not have the loss problems experienced with the wind data, since these data rely only on sensors attached to the sonde, as opposed to the winds, which rely on the sonde to lock into at least 4 satellites. However, the sonde temperature and humidity sensors do appear to have a bias with respect to the IMET sensors located on the bow mast of the ship. Columns 10-17 indicate the difference between the IMET and surface sonde readings for the various batches. For the temperature sensor the difference does not seem to be affected by age and indicates the sondes are biased warm by about -0.5^oC. The relative humidity sensors, on the other hand, have a dry bias of about 4%, with the bias getting progressively worse with the age of the sonde. These numbers are roughly in agreement with those found by Chris Fairall in a separate analysis (personal communication). Section 3 discusses the temperature and humidity biases in detail and explains the optionally offered corrected humidity data made available at this website.

Initially sondes with no winds were terminated and a new sonde was launched. However, relaunching did not consistently result in getting winds so this procedure was not continued, mainly in order not to use up our supply of sondes. Similar, if not worse, loss of wind data was experienced by the KWAJEX program, which also used Vaisala sondes. While they found no pattern with wind loss and sonde batch, it has recently been discovered that most of their sondes were from 1999. Table 3 shows that JASMINE losses were greatest for this sonde batch. Vaisala is currently working on the GPS unit in their sondes which may have affected some batches ability to measure winds. However, a definite connection between JASMINE and KWAJEX wind data loss and the sonde unit has not been determined.

3. Temperature and Relative Humidity

A comparison of the surface sonde and IMET measurements is given in Table 3. A complete comparison of all the humidity sensors is provided in Table 4. The ETL and IMET temperatures agree very well, as do the sonde and staging bay temperatures. Comparing the two locations, it appears that the staging bay was somewhat warmer than the bow of the ship. For humidity it appears that the bow had slightly higher relative humidities compared to the staging bay, with the exception of the self-aspirating psychrometer measurement. As the sonde and digital sensors in the staging bay agree fairly well, it is thought that the psychrometer difference may be due to user error. The mixing ratios (q) indicate similar absolute moisture values at both the bow and staging bay areas, which is an encouraging result. However, as Table 3 shows, when the sondes are divided by batch there is a bias of 5% for the earlier batches. A temperature difference of -0.5 ^oC would give on the order of a +3% humidity difference between the staging bay and the bow of the ship. This is consistent with the later sonde batches but does not explain the 5% bias in the earlier batches. A correction for sonde bias was applied to the TOGA COARE sounding data (Lucas and Zipser, 1996), which experienced fairly severe biases for various reasons discussed in, for example, Cole (1993) and Loehrer et al. (1996). For JASMINE the bias is just out of range of the expected instrument accuracy, so we have decided to release the sonde data without a correction. Users who wish to have a corrected data set can obtain it in a separate file from the JASMINE sounding web site. The correction applied to the data is described below:

dq = 0.5*(q_Vaisala-q_sonde(0)) + 0.5*((q_sonde(1)+dqMO_65m)-(q_IMET+dqMO_10m))
dRH = dq/dq_sonde(1)

qVaisala	-	mean mixing ratio of SB#2&3
qsonde(0)	-	sonde mixing ratio at surface
qsonde(1)	-	sonde mixing ratio at first point above surface (~65m)
dqMO#m	-	expected difference in mixing ratio between # m and the surface based on Monin-Obukhov similarity theory.

Cole (1993) describes the Vaisala study which found that, based on the temperature behavior of the sonde bias, the correction dRH should be linearly decreased from 100% at the surface to 0% at 5 km. The sonde profile was therefore corrected in the following way: if height < 5 km, RH_corr = RH + dRH*(1-height/5), otherwise RH_corr = RH. The average correction at the surface is ~2.5%.

4. Summary

From our analysis of these data we emphasize the importance of the following procedures when collecting radiosonde data:

1. Record the sonde surface humidity, temperature and pressure readings. For the Vaisala system operating in normal mode these are replaced automatically with the ship's surface sensor data (the IMET sensors in the case of the Brown) by the program collecting the data. So,if not recorded by the launch person, these data are lost and cannot be used to determine any sonde sensor biases.
2. Record at least one independent measure of relative humidity and temperature in the launch area (if different from the surface sensor location), in order to determine any actual temperature/humidity differences that might appear as a bias in the sonde surface measurement.
3. Record the sonde serial number for every launch and the reason for termination (normal or otherwise). This information helps to pinpoint the reasons for bad launches and may help the manufacturer of the sonde to correct any problems should they be related to the sonde sensor itself.

5. References