Minimum spanning tree as a new, robust repertoire size comparison method: simulation and test on birdsong

摘要

The comparison of acoustic complexity across individuals is often essential for understanding the evolution of acoustic signals. In many animal taxa, as a proxy of acoustic complexity, repertoire size is intensively studied; however, its estimation is challenging in species with large repertoires, as this process is time-consuming and may involve considerable subjectivity for the classification of signal elements. Here, we propose a novel application of the minimum spanning tree (MST) method for comparing individuals' signal complexity, an approach that does not require classification process. We suggest that the differences in the MST length predict the differences in the repertoire sizes between individuals. To evaluate these proposals, first, we performed a simulation study investigating the effect of the practically important variables (repertoire size, number of acoustic parameters, sample size, distribution of element types and within-group variance) on the MST length. Second, we compared repertoire size estimates from the same song data from male collared flycatchers obtained using the fully manual, computer-aided manual and MST methods. In our simulation study, we found that the repertoire size strongly correlated with MST length. We also found significant effects of sample size, number of parameters and within-group variance, as well as how uniformly the samples were distributed between the groups, on the MST length. Our empirical data also revealed a strong correlation between the computer-aided manual estimation of repertoire sizes and MST length, which was comparable to the correlation between the estimations of repertoire size obtained using the two different manual methods. Therefore, we suggest using the MST method to compare the acoustic complexity among individuals in birds and other animals, with the practical restrictions suggested by our simulation results.