The radiosonde record gives important information about upper air temperature trends and variability, but suffers from numerous instrument and procedural changes that may distort those trends. Because radiosonde stations are typically more widely spaced than surface stations, and because instrument changes are often simultaneous throughout an entire country, adjustments based on comparisons of individual records with neighboring station data are not usually useful. Statistical methods for the detection of discontinuities are insufficient by themselves because they do not distinguish natural from artificial changes. The three main approaches that have been used to adjust radiosonde temperature data are: (1) physical modeling of the radiation-induced errors of individual instrument types (R. Eskridge and J. Luers, National Climatic Data Center); (2) comparison with satellite data (D. Parker and M. Gordon, U.K. Met Office, and J. Christy, University of Alabama); and (3) “expert committee” decisions based on multiple criteria (J. Lanzante and S. Klein, NOAA Geophysical Fluid Dynamics Laboratory, and D. Seidel, NOAA Air Resources Lab).

At a workshop in October 2000, methods to identify and adjust for inhomogeneities in radiosonde records were discussed and the results of applying those methods to a common set of 12 radiosonde stations were compared. Fewer than 50% of the discontinuities found by different groups at the same stations were within one year of each other. Even when the dates coincided, the amounts and even the signs of the adjustments at individual pressure levels varied widely. For some stations and levels, the effect of adjustments on trends is comparable in size to the trends themselves. These comparisons suggest that individual station time series adjusted by different methods can differ substantially, with potentially significant effects on local or regional trends.

A new project is underway to create an improved subset of radiosonde temperature data from the CARDS database. Starting with an 87-station dataset for 1958-1997 adjusted using the expert committee approach, the project will include additional stations and extend the dataset to the present, with provision for future updating at NCDC.