Comparative analysis of bioacoustic indices: Performance and limitations

Passive acoustic monitoring (PAM) increasingly uses bioacoustic indices to assess biodiversity, although their application remains controversial. Therefore, this study evaluates the performance and limitations of five commonly used indices: Acoustic Entropy ( H s ), von Neumann Entropy ( H v ), Acoustic Diversity Index (ADI), Bioacoustic Index (BIO), and Acoustic Evenness Index (AEI), across different levels of species richness and two taxonomic groups. For this purpose, we used two datasets: (i) 26,028 synthetic assemblages with controlled species richness, including birds, anurans, and background-noise segments; and (ii) 10,163 manually annotated raw soundscapes containing bird vocalizations, including calls and songs, and anuran calls, with and without noise filtering. The results indicate that index values may be misleading when distinguishing background-audio segments from soundscapes with varying levels of species richness, as background-audio segments exhibited values similar to, or even higher than, those observed in soundscapes with high species richness. The use of noise filtering to minimize the problem of noisy segments showed inconclusive results, modifying the correlations between indices and species richness in an inconsistent way. In some cases, it converted positive correlations into negative ones for bird calls and songs, or negative correlations into positive ones for anuran calls. Therefore, it was not possible to establish a consistent rule ensuring that the filter should always be applied. Furthermore, two case studies showed that the distribution of spectral energy and the syllable repetition rate over time significantly influence index values, indicating that these indices are not reliable proxies for species richness.