Item

Abstract

We study global mean surface temperature records since 1850 and their potential forcings. We find long range correlations by the method of detrended fluctuation analysis in most data sets, in agreement with previous studies. As a predictive model, we employ a zero-dimensional energy balance model without memory that reproduces temperature data on the timescale of years. Even when driven with white noise, this model generates data that reproduce the observed long-range correlations. We are able to explain this with theoretical results for the AR(1) process, which demonstrates that even processes with exponentially decaying correlations yield nontrivial detrended fluctuation analysis results if the available data set is too short. This article gives new support to the scepticism about long memory in global mean temperatures and discusses further implications of our findings.