Because of their high vertical resolution and long periods of record, radiosonde observations comprise an attractive dataset for detection of multi-decadal tropospheric and lower stratospheric temperature trends. However, the dataset is fundamentally an operational product, and was never intended for climate change detection. Consequently, data quality problems, particularly those associated with temporal homogeneity, introduce uncertainty into temperature trend estimates based on radiosonde data.
We will present quantitative estimates of the sensitivity of trends during 1959-95, and between 850 and 30 hPa, to the following: 1) choice of dataset; 2) method of estimating trends (parametric vs. non- parametric); 3) removal of abrupt discontinuities by two different objective methods for detecting change-points (Habermann, 1987, Bull. Seis. Soc. Amer.; Lanzante, 1996, Intl. J. Climatol.); and 4) the adjustment of time series using statistical methods combined with station history information. Our results are based on analysis of monthly CLIMAT reports compiled by Parker et al. (1997, Geophys. Res. Lett.) and monthly mean temperatures computed by the CARDS project (Eskridge et al. 1995, Bull. Amer. Meteor. Soc.) as well as station history information compiled by Gaffen (1996, NOAA Tech. Memo. ERL ARL-211).