Causes of differing temperature trends in radiosonde upper air data sets

Differences between trends in different radiosonde temperature products resulting from the varying choices made by the developers of the data sets create obstacles for use of those products in climate change detection and attribution. To clarify the causes of these differences, one must examine results using a common subset of locations to minimize spatial sampling effects. When this is done for the Lanzante‐Klein‐Seidel (LKS) and Hadley Center (HadRT) radiosonde data sets, differences are reduced by at least one third. Differing homogeneity adjustment methods and differences in the source data are both important factors contributing to the remaining discrepancies. In contrast, subsampling the microwave sounding unit (MSU) satellite data sets according to the radiosonde coverage does not generally bring the trends in the satellite data closer to those in the radiosonde data so that adjustments and other processing differences appear to be the predominant sources of satellite‐radiosonde discrepancies. Experiments in which we subsample globally complete data sets provide additional insight into the role of sampling errors. In the troposphere, spatial sampling errors are frequently comparable to the trends for 1979–1997, while in the stratosphere the errors are generally small relative to the trends. Sampling effects estimated from National Centers for Environmental Prediction reanalysis and MSU satellite data for seven actual radiosonde networks show little consistent relation between sampling error and network size. These results may have significant implications for the design of future climate monitoring networks. However, estimates of sampling effects using the reanalysis and the satellite data sets differ noticeably from each other and from effects estimated from actual radiosonde data, suggesting that these globally complete data sets may not fully reproduce actual sampling effects.