Identifying disturbed habitats: A new method from acoustic indices

Monitoring and preventing changes in ecosystems is one of the most difficult challenges for biologists. Currently, soundscape analysis (the analysis of acoustic sounds emitted by the ecosystems) have been accepted as a technique that allows assessment of the biodiversity in natural landscapes. To measure the quality of habitats using recordings, acoustic indices have been developed. These indices are divided into alpha indices (within-group indices) and beta indices (between-group indices). Alpha acoustic indices attempt to represent different attributes of a habitat (e.g., evenness, richness, and heterogeneity), unlike beta acoustic indices that focus on estimating dissimilarity between communities. However, there is not a single index that can abstract all the components from a complex biological system in order to characterize habitats. Furthermore, acoustic indices exhibit patterns along the day that hinder the direct analyzing of habitats. In this study, we go beyond the strategy of choosing a particular index, providing a methodology that includes an integration of several alpha acoustic indices. This integration is performed using classification methods (multilayer neural network and one-class support vector machine) that allow characterizing and identifying predefined habitat prototypes (mature forest, secondary growth, and pasture). An accuracy of 0,89 ± 0,01 was obtained using this methodology to classify these three habitats with different degree of disturbance. An additional experiment were performed to validate the methodology and prove that works a more finite resolution (identification of three forests). The results of the methodology represent the contribution of this study: the integration of the acoustic indices to identify types of habitats and a new monitoring complementary tool, which alerts if new samples taken in these habitats start to be distant to the prototype habitat behaviors.