Estimating ecoacoustic activity in the Amazon rainforest through Information Theory quantifiers.

Area of Science:

• Ecology
• Bioacoustics
• Environmental Monitoring

Background:

• Automatic acoustic monitoring is crucial for early environmental stress detection.
• The Amazon's high acoustic diversity and data volume challenge manual analysis.
• Unsupervised methods are needed to overcome species-specific labeling difficulties.

Purpose of the Study:

• To develop an unsupervised ecoacoustic index to quantify soundscape complexity.
• To correlate this index with biodiversity and assess environmental stress.
• To create a method for analyzing large acoustic datasets from biodiverse regions.

Main Methods:

• Proposed the Ecoacoustic Global Complexity Index (EGCI) using Entropy, Divergence, and Statistical Complexity.
• Employed unsupervised machine learning to avoid individual species labeling.
• Mapped audio segments to a 2D-plane to visualize ecoacoustic dynamics.

Main Results:

• Demonstrated regularity in ecoacoustic richness across different temporal scales (hourly, daily).
• Successfully characterized the soundscape of the Mamirauá environmental protection area.
• Differentiated between species richness and environmental phenomena using the EGCI.

Conclusions:

• The EGCI provides an effective, unsupervised method for quantifying complex soundscapes.
• This index aids in understanding rainforest ecoacoustic dynamics and biodiversity.
• EGCI is a valuable tool for environmental monitoring and stress assessment in the Amazon.