Detecting trends and biases in time series of ozonesonde data

In this paper, we discuss the complexities associated with the analysis and interpretation of trends in time series of ozonesonde data, focussing on approaches to deal with the special constraints imposed by the irregular nature of the ozonesonde data set. To improve upon earlier studies which have used monthly mean values and parametric techniques on ozonesonde data, a non-parametric statistical method is introduced to enable us to work with data from individual flights rather than with monthly mean values. To this end, ozone time series data are separated into their long-term, seasonal and short-term components to properly characterize the various scales of motion (climatic, annual and synoptic scale) embedded in the data set. We show that the statistical method used here meets the requirements for a reliable analysis of ozonesonde data. It is shown further that this approach enables us to estimate trends in the ozonesonde data with a very high degree of confidence. We then introduce a non-parametric technique for discerning sudden changes in time series data and discuss its usefulness in detecting potential biases in ozonesonde time series data, introduced by changes in instrumentation, flight time, preflight preparation and data reduction procedures. The results show that the method is able to detect discontinuities in the ozonesonde data which are supported by station histories. It is shown that the long-term trend estimates can be significantly affected by the presence of biases in the data. Although further research is necessary to adequately account for artificial breaks in the data at all heights and stations, there is an indication that the estimated upward trend in the raw tropospheric ozone data at Payerne, Hohenpeissenberg and Edmonton might be attributable, in part, to the presence of biases in the data.