Examining the temporal structure and information entropy of leopard seal calling bouts

Abstract

Leopard seals (Hydrurga leptonyx) produce sequences of stereotyped sounds, or bouts, during their breeding season. The seals share common sounds but combine them in individually distinctive sequences. This study examines the underlying structure of the calling bouts by estimating the information entropy of the sound sequences with three entropy estimators. The independent identically distributed (IID) model estimates entropy from the simple frequencies of each sound. The Markov model estimates entropy from the frequency of pairs of sounds. Finally, the nonparametric sliding window match length (SWML) estimator exploits a relationship between the information entropy and the average subsequence match length. A better model for a given sequence achieves a lower entropy estimate. This study analyzed the calling bouts of 35 leopard seals recorded during the 1992-1994 and 1997-1998 Antarctic field seasons. The decrease of entropy estimates between the IID and Markov models for all seals analyzed confirmed the presence of temporal structures in the bouts. For twenty-one of thirty-five seals, the SWML entropy estimate was not significantly less than the Markov estimate, suggesting that a first-order Markov model accurately represents the structure of their bouts. [Work supported by the Australian-American Fulbright Commission and the Australian Research Council.]