Long-Range Memory in Millennium-Long ESM and AOGCM Experiments

Abstract

Consider the Earth’s global mean surface temperature time series (GMST) as a realization of a stochastic process. Based on a number of studies, a long-range memory (LRM) stochastic process seems to describe the GMST better than a shortrange memory model, such as the AR(1)-process. We want to study the persistence in climate model simulations, to find out if simulated temperature data exhibit the same LRM-properties as instrumental and paleo data. To infer whether the LRM originates from variations in external forcing or from internal variations in the climate system, both forced model runs and control runs are studied. LRM is characterized by an autocorrelation function decaying as a power law: •limt!1 C(t) / t −1, where is a scaling exponent describing the degree of persistence. •For a stationary LRM process: 0 < <1. •In this particular study, the persistence in Northern Hemisphere (NH) mean ST time series is determined by estimating by the DFA2 method. •We investigate the LRM in NH mean ST time series from millenium-long climate simulations and paleo data.