Determination of new adjustment tables in order to bring radiosonde temperature and height measurements from different sonde types into relative agreement.

William G. Collins

NCEP/EMC

May 1999

1. Introduction

The changes in the sonde types in use and in the kind of corrections made at radiosonde launch sites has made it mandatory to re-examine the adjustments made by NCEP to the values of temperature and height from WMO coded messages. There are two main objectives in making adjustments to radiosonde temperatures and heights at the NCEP: 1) to make measurements within a region or across region boundaries commensurate, and 2) to make values as accurate as possible in an absolute sense.

There are different ways in which different sonde types may be compared. The most direct way is to fly them at the same time in close proximity. There are a number of studies that have used this technique, e.g. Nash and Schmidlin, 1987, Ivanov, et al, 1991, and Schmidlin, et al, 1995. If, among the types flown is an absolute standard, then both objectives 1) and 2) above may be fulfilled. However, there have been comparisons for only a limited number of the sonde types in use, and even for the available comparisons, the number of flights and conditions is limited. While this method must certainly be continued and expanded, other methods must also be pursued. Another method assumed that nighttime observations of temperature are accurate and derived daytime corrections from mean day-night differences. (See, e.g. McInturff, et al, 1979 and Julian, 1991.) Another possibility is to use satellite temperature retrievals as a common measurement to compare different sonde measurements. (E.g. McMillin, et al, 1988.)

All of the comparison methods mentioned above have limitations, not the least of which is that the sample size of all previously available comparisons is small (<100). There is another comparison method, to be used in this study that is universally applicable, and which can also rather easily be adapted to meet changes to the types of sondes in use. Namely, the reported radiosonde temperatures may be compared with a short-term forecast, interpolated to the sonde position. The evidence from model drift is that model errors are relatively small, providing a good approximation to an absolute error. But, more importantly, the mean model error varies smoothly in space, making comparison of different sonde types meaningful. Various methods of using comparison of temperature measurements with short-term forecasts is discussed by A. Nomura, P. Kallberg and S. Uppala, 1995.

2. Discussion of Proposed Method

The values of temperature (and height) reported in WMO coded messages are subject to many possible sources of error. First, the sensor may be enclosed by air that is not representative of surrounding air at the same elevation, thus making any "precise" measurement of the temperature unsuitable for anything by a micrometeorological use. Any sensor has a certain lag, making the measurement a mean, in some sense, over time (and thus space—vertical). The sensor is further influenced by conduction along its support, convective currents around it, and especially by its precise radiative environment. The values of temperature are reported at a discrete interval, and processed by algorithms whose workings are largely unknown and certainly often inconsistent, but yet supposed to follow the WMO guidelines. And finally, for some sonde types, there is an adjustment made to account for mean radiative effects. Fortunately, it appears that the processing methods do not differ at different stations for a given sonde type. It is true that there are several types of processing for the Vaisala RS80 temperature sensor, but these are distinguished as different sonde types, e.g. Vaisala RS80 Microarts and Vaisala RS80 Digicora.

Given the diversity of possible influences on the temperatures contained in a WMO message, a comparison with a short-term forecast (6-hr), whose errors in the mean are small and smoothly varying, seems most appropriate. A study about a year ago used data from 1997 to compute statistics of temperature observations-forecast. In keeping with the practice of Thiebaux, Pedder, and Gandin, the observation-forecast is called increment. The same computations, and others, were made for this study for all radiosonde temperature observations at mandatory levels from 1998. Many different statistics, to be discussed in the next section, were calculated. Confidence in these results is enhanced by the large agreement in the 1997 and 1998 results.

3. Temperature and Height Statistics

For each observation, the station, instrument type, pressure, and approximate solar elevation angle are known. The following basic statistics were calculated for each mandatory level observation:

1. temperature and height increments with respect to 6-hr forecast
2. day-night difference of increments
3. value of temperature and height
4. day-night difference of values
5. temperature and height increments with respect to analysis
6. day-night difference of analysis increments

For defining day-night differences, only observations taken 12 hours apart are used. The daytime observation is the one between 6 and 18 hours local time, while the nighttime observation is the other of the pair. It is noted that some of these statistics, while interesting, are not of much use for this study. Naturally, 1. is of nearly exclusive importance, but some information may be discussed relating to the others.

The statistics are first stratified by WMO block, instrument type, pressure, and solar elevation class. The solar elevation classes are for the following approximate elevations of the sun above the horizon: 1) < -10 degrees, 2) -10 to 15 degrees, 3) 15 to 30 degrees, 4) 30 to 60 degrees, and 5) 60 to 90 degrees. For each WMO block, instrument type, pressure, and solar elevation class, are collected the count, mean, and standard deviation. The values of the increments, for instance, may be compared for a given sonde type for various WMO blocks in which the sonde is used, and significant variation is found, even with counts in the hundreds and more. Therefore, it was decided that further averaging was necessary, and the values were further averaged across all WMO block. The values to be shown later are stratified only by instrument type, pressure, and solar elevation class. It is with this degree of averaging that the 1997 and 1998 results are very similar.

4. Results

a. average for all sonde types

The average difference between the temperature and height observations and the 6-hr forecast (from the NCEP’s Global Data Assimilation System) are of interest to judge the magnitude of the overall difficulty with the observed measurements. Fig. 1 shows the temperature increment average across all sonde types and solar elevation classes for the 1998 data. Also shown is the 12-hour difference in the increment, as well as the standard deviations in these statistics. Between 850 and 50 hPa, the increment magnitude is less than about 0.5 degrees (generally negative). The magnitude of the temperature increment is larger both near the surface and in the upper stratosphere. The reason for the increase near the surface is not known, but likely due to boundary layer processes inadequately accounted for in the model. The increase in the stratosphere is likely due to radiative effects on the temperature sensors. Fig. 2 shows the same statistics for height. In both, note that the mean 12-hour increment differences are quite small, indicating that the predominant sonde type processing procedures perform adjustments to remove this difference. (Since Vaisala RS80 is the predominant sonde type and it does make such adjustments, this is a reasonable conclusion.)

b. mean Vaisala RS80 sonde

The ECMWF does not make adjustments to the Vaisala RS80 sonde temperature measurements (as far as the author is aware). However, even though the Vaisala RS80/Digicora is the most widely used type, it was decided to average all the various incarnations of the RS80 to form a reference sonde for this study. The following table lists all the sonde types in use in 1998, including types 37, 52, 60, 61, 62 and 63 which are Vaisala RS80 sonde types.

Table 1. Radiosonde types in use in 1998

62

The average Vaisala RS80 sonde temperature increments are shown in Fig. 3. The broad agreement with the global average results is noted, and the variation with solar elevation class may be seen. The height increment statistics are likewise shown in Fig. 4.

Since it is unknown what part of the mean increment is contributed by model error, and what part is contributed by all that can happen with the sonde measurements, it seems reasonable to adjust all sonde measurements to the mean Vaisala RS80 reference sonde. When representative absolute errors become known for the Vaisala RS80 sonde, then these can be used as a further adjustment to all sonde measurements. The Figs. 5 and 6 show the day-night differences of temperature and height increments for the reference sonde for the different solar elevation classes. The variations are quite small between 925 and 50 hPa.

c. adjustments for some representative sonde types

Adjustments, relative to the reference sonde, have been calculated for all sonde types. Only a few representative types will be shown. As a check, Figs. 7 and 8 show the temperature and height adjustments for the Vaisala RS80/Digicor or Marwin sonde type. The temperature adjustments are generally less than +/- 0.2 degrees and –2 to 13 meters.

The adjustments for the NOAA/VIZ Type B2 sonde are shown in Figs. 9 and 10. The temperature adjustments stratify nicely with the solar elevation angle and generally increase in magnitude with height, attaining values as large as 2.4 degrees. The temperature adjustment at lower elevations is small. In practice, they might be set to zero in the lower troposphere. The height adjustments likewise stratify well with solar elevation angle. For the sun above +15 degrees, the adjustments grow with height, attaining 130 m for the sun at high elevation at 10 hPa.

Further examples a given in Figs. 11 and 12 for the Shanghai Radio sonde and in Figs. 13 and 14 for the Indian Meteorological Center Type MK3 sonde. They show rather different characteristics.

5. Proposal

In looking at the results, one might easily question what confidence may be had in the details of the derived adjustments. The counts for different sonde types, and for a given sonde type, are rather variable, with especially few counts at the highest elevations and for the highest solar elevation class. In most cases, however, the counts are in the hundreds to thousands (even tens of thousands). Further, as already stated, the mean increments for 1997 and 1998 compare closely. Therefore, when the count is above, say 100, it seems reasonable to propose using the adjustments as calculated, filling in other values with smooth variation with more certain values. When absolute calibration is available, then a further adjustment can be made for all sonde types.

The following tables are derived from the procedure outlined above. They are put closely in the format used by the presently operational scheme, RADCOR. However, the number of solar elevation classes has been reduced as the data do not seem to justify further refinement. Table 2 gives information on the reference (mean RS80) sonde, first the counts and then the mean values. Following is Table 3, giving the adjustment values of temperature and height for the mandatory pressures from 1000 to 10 hPa, at each solar elevation class, for each radiosonde type.

Table 2. Counts and increments for the reference sonde. The solar classes follow in order: <-10 to 15, 15 to 30, 30 to 60 and 60 to 90 degrees. The columns are labeled with the pressure.

Table 3. Adjustment values for each sonde type. The solar classes follow in order: <-10 to 15, 15 to 30, 30 to 60 and 60 to 90 degrees. The columns are labeled with the pressure.