Effect of Interannual Variations on Detection of Climate Change

The global mean temperature increase since 1880 as observed to date and forecast through 2070 is approximately exponential in nature, with interannual variations, some of which are decadal in length. Most analyses of detection of temperature increase assume a change which is linear in time or a linear variation to a change point after which the slope changes. The plots of global mean temperature do not have a break point. In this paper an exponential model is assumed for which the coefficient a of the exponent and the characteristic (e-folding) time tau are to be determined from the temperature record. The data include interannual variations and errors, so that a and tau are random variables. For the first phase of this study the biases and standard deviations of these two parameters are computed by Monte-Carlo simulation and by an approximate analytic expression. If one has only data for the early part of the time period the standard deviations are large for a and tau. For even sparse data later in the time period, a and tau can be determined with greater accuracy.

The global mean temperature record shows a cooling during the 1940s and a slowing of warming in the 2000-2010 decade. For the second part of the study, climate oscillations of various periods are assumed. Temperature oscillations of a few years are added to the data. These simply add to the noise level. However, temperature changes with periods comparable to the length of the data set can overpower the trend prediction.