Carlos VARGAS

The 2030 United Nations Agenda for Sustainable Development highlights the urgent need to address biodiversity loss and land degradation, which threaten ecosystems and livelihoods worldwide. This thesis contributes to these efforts by supporting the Payment for Environmental Services (PSA) program in Costa Rica as a case study for large-scale biodiversity Passive Acoustic Monitoring (PAM). We propose the FramewOrk for featuRe Extraction, viSualisation, and classificaTion of Soundscapes (FOREST), a modular Python-based framework that integrates preprocessing, dataset curation, feature extraction, visualisation, and predictive classification of ecological audio recordings. First, we establish a pipeline that transforms soundscapes into PyTorch tensors, consolidating a curated dataset of shape 249,660 × 6,016 and extracting five (5) statistical scalars and eleven (11) Ecological Acoustic Indices (EAIs). Second, we develop an evaluation framework comprising 3,577 experimental runs to systematically analyse the impact of individual features and their combinations on model performance. The results show that a subset of five (5) EAIs, namely NPP, BET, HTP, AEI, and HFQ, achieves robust and accurate classification. Complementary spidernet visualisations reveal distinct ecoacoustic profiles across the four (4) studied ecosystem regions (Reference Forest, Pasture, Natural Regeneration, and Plantation), supporting the interpretation of specific EAIs as proxy indicators of biodiversity. Third, we design and benchmark three (3) hybrid Deep Learning (DL)-based predictive models, ParaNet-CNN-LSTM, SeqNet-CNN-LSTM, and SeqNet-LSTM-CNN, against baseline models including Simple-SVM, Simple-CNN, Simple-LSTM, and ResNet1D. The comparative analysis shows that ParaNet-CNN-LSTM achieves the most consistent and reliable performance, with median accuracy values above 90% and maximum values exceeding 97.5% in the optimal range of 10–13 input features. The FOREST framework consolidates these contributions into an open-source, web-based application accessible at www.soundforest.app. While the approach is constrained by dataset imbalance, temporal assumptions, absence of metadata, and its limitation to the PSA program, the methodology provides a rigorous foundation. This thesis demonstrates that combining EAIs with DL-based hybrid architectures enables accurate classification of ecological soundscapes and offers a scalable path for future biodiversity monitoring across diverse ecosystems.