Abstract
This paper describes long‐term changes of global atmospheric temperature, using a strict Bayesian approach which considers three different models to describe the time series: the constant model, the linear model and a change point model. The change point model allows the description of nonlinear annual rates of change with associated confidence intervals. We calculate the probabilities of each of the three models and average finally over these models to obtain the expected functional behaviour and rate of change in temperature with annual resolution. We apply this procedure to a new homogenized Radiosonde Atmospheric Temperature Products for Assessing Climate (RATPAC‐A) data set. Annual mean temperature for 13 pressure levels from the surface to 30 hPa is examined. Residual sums of squares reveal that Bayesian‐model‐averaged function descriptions and rates of changes are especially useful and informative for the surface, troposphere and tropopause and less appropriate for the stratosphere. From the surface up to the tropopause (200–100 hPa), the results reveal that the change point model provides the best data fit. Despite the occurrence of two volcanic eruptions El Chicón (1982) and Mt. Pinatubo (1991), the stratosphere (70–30 hPa) shows a preference for the linear model (60%). The near surface changes exhibit comparatively high change point probability around 1985 and 1995, whereas those at the tropopause level are highest between 1995 and 2000. For the surface and troposphere the model‐averaged functional behaviour increases quite constantly, whereas the model‐averaged functional behaviour for the tropopause decreases until the end of the 1990s and increases from 2000 onwards. The limitations of the currently used radiosonde data render interpretation of the observed changes difficult. Additionally undetected change points may result from our limited model space. In future it should be tested whether a multiple change point model provides a better data description for the stratosphere. Copyright © 2008 Royal Meteorological Society