Putting the Significance of Spectral Peaks on the Level: Implications for the 1470-Yr Peak in Greenland δ18O

Abstract Spectral analysis of the Greenland Ice Sheet Project 2 (GISP2) δ 18 O record has been interpreted to show a 1/(1470 yr) spectral peak that is highly statistically significant ( p < 0.01). The presence such peak, if accurate, provides an important clue about mechanisms controlling glacial climate. As standard, however, statistical significance was judged relative null model, H 0 , consisting autoregressive order one process, AR(1). In this study, generalized using moving-average ARMA( q ). A rule thumb proposed for evaluating adequacy involving comparing expected and observed variances logarithm estimate, which are generally consistent insomuch as removal ARMA structure from time series results in approximately level estimate. An AR(1), or ARMA(1, 0), process shown be inadequate representation GISP2 structure, whereas higher-order processes result estimates. After suitably leveling accounting multiple hypothesis testing, multitaper estimation indicates insignificant. seeming prominence explained spectrum having selected on basis its anomalous. technique peaks also applicable other geophysical records. Significance Statement suitable necessary obtaining accurate test results, but means lacking. model presented form autoregressive, whose can gauged by variance log density. Application method found at