Smartphone-based acoustic monitoring for Southeast Asian forest biodiversity decline

Thesis Abstract

Rapid declines in forest biodiversity across Southeast Asia necessitate scalable, cost-effective monitoring approaches that can operate in remote, canopy-dense environments. This study addresses the problem of sparse, sporadic biodiversity data by evaluating smartphone-based acoustic monitoring as a scalable tool to detect and track changes in forest vertebrate and avian communities, with implications for conservation prioritization and land-use policy. The overarching aim is to determine whether citizen-accessible smartphone recorders, coupled with automated sound-event analysis, can reliably reflect biodiversity trends over time and space in Southeast Asian forests. Specific objectives are (1) to characterize the acoustic signature diversity of target taxa (birds, small mammals, amphibians) across four protected and four degraded forest sites in Peninsular Malaysia, Sumatra, and Borneo over 24 months; (2) to develop and validate an end-to-end pipeline that combines field-recording protocols, on-device preprocessing, cloud-based species classification using a convolutional neural network (CNN) model trained on region-specific call libraries, and statistical aggregation of acoustic indices; (3) to compare smartphone-derived biodiversity indicators with traditional point-count surveys and camera-trap data using regression analysis and Bland–Altman agreement; (4) to assess temporal and spatial trends in acoustic diversity metrics (Shannon index, acoustic evenness, and Delta Tonal Index) in relation to habitat integrity, edge effects, and rainfall seasonality; (5) to evaluate user engagement, data quality, and operational feasibility to inform scalable implementation frameworks for local communities and conservation agencies. Methodologically, the study adopts a mixed-methods design, integrating quantitative acoustic analyses with qualitative assessments of operational feasibility. The population comprises forested habitats within Southeast Asia, with sampling across eight sites selected to capture gradients of disturbance. A stratified purposive sampling approach will recruit 40 local citizen-science volunteers who will record 10-minute audio samples weekly using standardized smartphone applications and external mics, yielding an expected dataset of approximately 1,920 hours of passive acoustic recordings per site over two years. In addition, trained field ecologists will conduct parallel 12-hour monthly point-count surveys and deploy camera traps to provide independent biodiversity benchmarks. The primary data collection instrument is a smartphone-based audio recorder app augmented by a standardized calibration protocol, supplemented by portable weather meters to capture ambient conditions. The analytical framework integrates (i) deep-learning-based species classification using a CNN trained on region-specific acoustic libraries, (ii) computation of acoustic indices (Acoustic Complexity Index, Entropy, and Normalized Performance Time Series) and biodiversity indicators (H’ and Evenness), (iii) generalized additive models (GAMs) to relate acoustic metrics to habitat variables (Fragmentation Index, Normalized Difference Vegetation Index, canopy cover) and climatic covariates (precipitation, humidity), and (iv) Bland–Altman and linear regression analyses to compare smartphone-derived indices with conventional survey data. The study will apply theoretical lenses from the Species-Aecology framework and the Information Theory of Biodiversity, alongside the Theory of Technological Acceptance to interpret citizen-science engagement and data quality. Expected findings include (a) a robust, region-specific acoustic classifier achieving precision and recall above 0.75 for target taxa across sites, (b) significant associations between acoustic diversity indices and habitat integrity measures, with GAMs explaining up to 60% of observed variance in biodiversity indicators, (c) measurable concordance between smartphone-derived metrics and traditional survey counts, though smartphone data will exhibit higher sensitivity to sampling effort and environmental noise, and (d) clear seasonal and spatial patterns in biodiversity signals corresponding to monsoonal patterns and forest edge effects. The study is anticipated to demonstrate that smartphone-based acoustic monitoring can detect early signals of biodiversity decline, complementing conventional methods and enabling rapid, scalable assessments across large landscapes. The contribution to knowledge includes (i) a validated, scalable acoustic monitoring pipeline tailored for Southeast Asian forests combining citizen science with machine learning and ecological modeling, (ii) empirical evidence on the reliability and limitations of smartphone-derived biodiversity indicators relative to traditional methods, (iii) a comparative multi-site framework linking acoustic signals to habitat and climate drivers, and (iv) practical guidelines for implementation, including data governance, volunteer training, and quality assurance. The main conclusion is that smartphone-based acoustic monitoring, when underpinned by region-specific call libraries, standardized protocols, and robust analytical models, provides a credible, cost-effective means to monitor forest biodiversity at scale in Southeast Asia. Recommendations include expanding citizen-science training programs, investing in cloud-based processing infrastructure, integrating acoustic monitoring with habitat restoration planning, and establishing collaborative data-sharing agreements to inform regional conservation policy.