Detecting breakpoints in global temperature

Abstract. This paper aims to address the continuous debate of whether a hiatus occurred in the turn of this century. Several models have been employed to fit the global mean surface temperature data, and the results suggest that the allegation of an occurrence of hiatus lacks statistical evidence. However, these models had potential deficiencies in their capacity for detecting breakpoints, thereby weakening the arguments that deny the existence of a hiatus. To address this issue, we propose an improved sparse representation model, which can automatically segment and fit temperature records using piecewise polynomials. Simulations revealed improved detection performance; studies on five prominent global temperature records produced 2 to 6 breakpoints, none of which occurred after the year 1976, thus reinforcing arguments denying the existence of a 20th century hiatus.