Detecting Changes through Time in the Variance of a Long-Term Hemispheric Temperature Record: An Application of Robust Locally Weighted Regression

Abstract Historic records of global or hemispheric temperature are an important source of information about climate change. In order to analyze such records statistically, it is necessary to have some knowledge about the behavior of their variances through time. A modified version of robust locally weighted regression, a nonparametric regression procedure, is used to study the behavior of the variance for a record of Southern Hemisphere temperature deviations. Sampling considerations suggest that the variance should decrease through time, as new recording stations are added to the sampling network. Surprisingly, the variance is found to remain virtually constant through time.