Earth science: Statistical structure of geomagnetic reversals

Abstract

from D. V. Kent REVERSALS of the Earth's magnetic field have not occurred at a constant average rate over the past 170 Myr 1,2. Shifts in the mean reversal rate at various times appear to be separated by periods of tens of mil­ lions of years in which the average rate of reversals remains nearly constant. Con­ clusions drawn from these and other statis­ tical properties of reversal processes have been based on analyses of early versions of the geomagnetic polarity timescale (OPTS). More recent versions of the OPTS for the past 170 Myr incorporate a variety of new data from marine magnetic anom­ alies, magneto-biostrategraphic correl­ ations and improved age calibrations, re­ sulting in a new picture of the reversal be­ haviour of geomagnetic fields3.4 . McFadden and MerrilP have analysed the mean rate of reversals using a sliding window with a fixed number of reversals. This reveals an approximately linear trend, increasing from less than one reversal per million years at about 86 Myr BP Gust after the 30 Myr interval of constant normal polarity in the Cretaeceous) to over four per million years for the past few million years. Lowrie first arrived at a similar trend using a sliding window of fixed dur­ ation 4 • Then he and I documented such reversal behaviour in a different revised OPTS5. McFadden and Merrill suggest a change in the structure of the rate vari­ ations at about 10 Myr BP, although it has been suggested4,6 that several maxima in reversal frequency associated with an ap­ parent 15 to 20 Myr fluctuation are super­ posed on the long term linear trend. lems remain to be resolved before much confidence can be placed in our knowledge of the statistical structure of reversals. One long-standing but central issue concerns the presence or absence of short polarity intervals. It is clear that the distribution of observed polarity intervals is generally too deficient in short intervals for it to be fully compatible with either an exponential dis­ tribution or a Poisson reversal process in which the probability for the next reversal to occur is constant with time. Instead, it has been shown 2 ,7 that a gamma distribution provides a good fit to the observed polarity intervals, at least for the past 86 Myr. In a companion paper to reference 3, McFadden argues that the observed dis­ tribution embodies a Poisson process combined with a filtering process that preferentially removes the record of short polarity intervals, giving rise to the ob­ served gamma distribution 7 • Accordingly, McFadden and MerrilP calculate that the actual reversal frequency was 46 per cent higher than is observed, so that about half of the observed polarity intervals represent a concatenation of more than one int­ erval of the actual reversal sequence; magnetostratigraphers, take note. The alternative is that short intervals seldomly occurred. This would imply a renewal process in which the probability per unit time of a reversal is not constant and that the occurrence of each reversal is not independent of earlier ones. Difficult though it is to obtain firm evid­ ence of short polarity intervals, because they are at or below the threshold of resolution of most palaeomagnetic studies, it will be important to do so because oftheir dramatic effect on the statistics of geo­ magnetic reversals according to our calcul- Molecular biology