Upper-air temperature datasets are important for assessing climate change but are subject to many inhomogeneities due to changes in instruments and practices. A recently created radiosonde temperature dataset (Lanzante, Klein and Seidel 2003) (“LKS”) has been carefully scrutinized for temporal homogeneity and is therefore well-suited for trend analysis. This 81-station dataset covers the period from 1948 to 1997. The purpose of the RATPAC project is to increase its usefulness by extending it without introducing inhomogeneities that would interfere with trend assessment.

We have created large-scale mean upper-air temperature series through 2003 by using a semi-automated first difference method to append new radiosonde data to the LKS adjusted dataset, avoiding the time-consuming subjective examination used to create the original dataset. The new data comes from the Integrated Global Radiosonde Archive recently created at NCDC. The first difference method, used with station history information, tends to reduce inhomogeneities due to changes in instruments or procedures but introduces a random error related to outliers in the data. We estimate the uncertainty in trends in the resulting time series and compare the trends to those in other upper-air temperature products.