Acoustic heterogeneity of tropical dry forest based on identification of landscape transformation

Abstract

ABSTRACT: The Colombian Tropical Dry Forest (TDF) is an important ecosystem due to its high levels of endemism. This ecosystem is currently under threat due to deforestation generated by cattle, mining, and urban development since more than 200 years. Therefore, it is urgent the need to carry out conservation activities which require to understand the ecological heterogeneity and the states of the sites. Traditionally, environmental conservation experts measure it through direct observation, but these methods are invasive to the study landscapes. A proficient alternative is the passive acoustic monitoring with the use of computational tools. However, there are no acoustic methods to determine the heterogeneity using successional states of sites. This document proposes a new method to automatically identify the transformation in separate sites within areas in the Colombian TDF. The methodology follows 5 steps: First, establish if the recording has a high noise level. If is a noisy recording, it is not analysed. Second, calculate the selected acoustics indices for the recording. Third, based on the recording hour the stage of daily acoustic patterns is selected. Four, Use GMM models to identify the transformation type. Five, calculate the proposed acoustic heterogeneity index. To achieve this, we did an analysis of acoustic variables to determine the most informative. It was proposed to include two new variables spectral centroid and spectral band-with, since these help to better identification of succesional states. Also, it was exploring the acoustic patterns found 3 stages with similar behavior: morning (5-8), day(8-17), and night (17-5). Our proposal was tested with a data-set provided by Alexander Von Humboldt Institute. This data-set consists of a group of acoustic recordings recorded in two local sites: La Guajira and Bolivar. The method to identify the transformation level achieved an F1 score of 92% and 90% for La Guajira and Bolivar regions. We use the Acoustic Heterogeneity index to create maps that allow to see similarities among the studied sites. Also, we found that the method can detect special sites that can be associated with anomalies in the landscapes. As far as the authors know, this proposal is the first method to find heterogeneity in ecosystems that perform high capabilities to create informative maps about the site states using acoustic indices analysis