Statistical Significance of Trends and Trend Differences in Layer-Average Atmospheric Temperature Time Series

B. D. Santer, J. S. Boyle, J. J. Hnilo, K. E. Taylor

Program for Climate Model Diagnosis and Intercomparison, Lawrence Livermore National Laboratory, Livermore, CA 94550, U.S.A.

T. M. L. Wigley, D. Nychka

National Center for Atmospheric Research, Boulder, Colorado

D. J. Gaffen

NOAA Environmental Research Laboratories, Air Resources Laboratory, Silver Spring, MD 20910

D. E. Parker

Hadley Centre for Climate Prediction and Research, United Kingdom Meteorological Office, Bracknell, U.K.


Abstract

This paper examines trend uncertainties in layer-average free atmosphere temperatures arising from the use of different trend estimation methods. It also considers the significance of individual trends and of trend differences between datasets. We focus on data from satellite and radiosonde measurements and from two recent reanalysis projects. To facilitate intercomparison, we compute from reanalyses and radiosonde data deep-layer temperatures equivalent to those estimated from the satellite-based Microwave Sounding Unit (MSU). Trend analyses and comparisons are given over a range of time intervals.

We compare linear trends based on minimization of absolute deviations and minimization of squared deviations. Differences are generally less than 0.05°C/decade over 1959-96. Over 1979-93, they exceed 0.10°C/decade for lower tropospheric time series and 0.15°C/decade for the lower stratosphere. Trend fitting by minimization of absolute deviations degrades the previously reported global lower tropospheric trend agreement of 0.03°C/decade between the MSU and radiosonde data.

To assess trend significance, we employ two methods to account for temporal autocorrelation effects. When our preferred method is applied, virtually none of the individual 1979-93 trends in deep-layer temperatures are significantly different from zero. We also test the significance of a trend in d(t), the time series of differences between a pair of datasets. Use of d(t) removes variability common to the time series being compared and facilitates identification of small but meaningful trend differences. Roughly 30% of the dataset comparisons show significant differences in lower tropospheric trends, primarily related to differences in measurement system. Significant trends in d(t) are also found between different versions of the MSU channel 4 data, and between channel 4 time series bases on globally-complete MSU data and on MSU data subsampled with the incomplete rediosonde coverage.

We also compute the 95% confidence intervals for individual trends and show that these overlap for almost all datasets considered. For lower tropospheric trends over 1979-93, the adjusted 95% confidence intervals for all datasets encompass both zero and the model-projected trends due to anthropogenic effects.


Back to Publications List

To the Climate Analysis Section Home Page


Hongjun Zhang: zhangho@ucar.edu