Since the TAR, considerable effort has been devoted to assessing and improving the quality of the radiosonde temperature record (see Appendix 3.B.5.1). A particular aim has been to reduce artificial changes arising from instrumental and procedural developments during the seven decades (1940s–2000s) of the radiosonde record (Free and Seidel, 2005; Thorne et al., 2005a; Karl et al., 2006). Comparisons of several adjustment methods showed that they gave disparate results when applied to a common set of radiosonde station data (Free et al., 2002). One approach, based on the physics of heat transfer within the radiosonde, performed poorly when evaluated against satellite temperature records (Durre et al., 2002). Another method, comparison with satellite data (HadRT (Hadley Centre Radiosonde Temperature Data Set); Parker et al., 1997), is limited to the satellite era and to events with available metadata, and causes a reduction in spatial consistency of the data. A comprehensive intercomparison (Seidel et al., 2004) showed that five radiosonde data sets yielded consistent signals for higher-frequency events such as ENSO, the Quasi-Biennial Oscillation (QBO) and volcanic eruptions, but inconsistent signals for long-term trends.
Several approaches have been used to create new adjusted data sets since the TAR. The Lanzante-Klein-Seidel (LKS; Lanzante et al., 2003a,b) data set, using 87 carefully selected stations, has subjectively derived bias adjustments throughout the length of its record but terminates in 1997. It has been updated using the Integrated Global Radiosonde Archive (IGRA; Durre et al., 2006) by applying a different bias adjustment technique (Free et al., 2004b) after 1997, creating a new archive (Radiosonde Atmospheric Temperature Products for Assessing Climate; RATPAC). Another new radiosonde record, HadAT2 (Hadley Centre Atmospheric Temperature Data Set Version 2, successor to HadRT), uses a neighbour comparison approach to build spatial as well as temporal consistency. A third approach (Haimberger, 2005) uses the bias adjustments estimated during data assimilation into model-based reanalyses to identify and reduce inhomogeneities in radiosonde data. Despite the risk of contamination by other biased data or by model bias, the resulting adjustments agree with those estimated by other methods. Rather than adjusting the data, Angell (2003) tried to reduce data quality problems by removing several tropical stations from his radiosonde network.
Despite these efforts to produce homogeneous data sets, recent analyses of radiosonde data indicate that significant problems remain. Sherwood et al. (2005) found substantial changes in the diurnal cycle in unadjusted radiosonde data. These changes are probably a consequence of improved sensors and radiation error adjustments. Relative to nighttime values, they found a daytime warming of sonde temperatures prior to 1971 that is likely to be spurious and then a spurious daytime cooling from 1979 to 1997. They estimated that there was probably a spurious overall downward trend in sonde temperature records during the satellite era (since 1978) throughout the atmosphere of order 0.1°C per decade globally. The assessed spurious cooling is greatest in the tropics (0.16°C per decade for the 850 to 300 hPa layer) and least in the NH extratropics (0.04°C per decade). Randel and Wu (2006) used collocated MSU data to show that cooling biases remain in some of the LKS and RATPAC radiosonde data for the tropical stratosphere and upper troposphere due to changes in instruments and radiation correction adjustments. They also identified problems in night data as well as day, indicating that negative biases are not limited to daytime observations. However, a few stations may have positive biases (Christy and Spencer, 2005).
The radiosonde data set is limited to land areas, and coverage is poor over the tropics and SH. Accordingly, when global estimates based solely on radiosondes are presented, there are considerable uncertainties (Hurrell et al., 2000; Agudelo and Curry, 2004) and denser networks – which perforce still omit oceanic areas – may not yield more reliable ‘global’ trends (Free and Seidel, 2005). Radiosonde records have an advantage of starting in the 1940s regionally, and near-globally from about 1958. They monitor the troposphere and lower stratosphere; the layers analysed are described below and in Figure 3.16. Radiosonde-based global mean temperature estimates are given in Figure 3.17, presented later.
Figure 3.16. Vertical weighting functions (grey) depicting the layers sampled by satellite MSU measurements and their derivatives, and used also for radiosonde and reanalysis records. The right panel schematically depicts the variation in the tropopause (that separates the stratosphere and troposphere) from the tropics (left) to the high latitudes (right). The fourth panel depicts T4 in the lower stratosphere, the third panel shows T2, the second panel shows the troposphere as a combination of T2 and T4 (Fu et al., 2004a) and the first panel shows T2LT from the UAH for the low troposphere. Adapted from Karl et al. (2006).