Smartphone Acoustic Monitoring for Urban Bird Biodiversity Mapping

Thesis Abstract

Urban biodiversity in cities faces rapid changes driven by anthropogenic noise, habitat fragmentation, and changing land use, yet systematic, scalable monitoring of bird communities remains fragmented due to resource-intensive field surveys. The study addresses the need for a cost-effective, scalable, ICT-enabled approach to map urban bird biodiversity using smartphone-acoustic monitoring, combining citizen-collected audio data with automated species identification to generate spatially explicit biodiversity indicators. The aim is to develop and validate a smartphone-based acoustic monitoring framework that reliably detects and localizes urban bird species, estimates species richness, and tracks temporal dynamics in relation to environmental drivers. Specific objectives are (1) to assemble a stratified sample of 40 urban neighborhoods representing varied land-use types (residential, commercial, green corridors) and to recruit 150 volunteer participants for standardized audio data collection over a six-month period; (2) to implement an end-to-end pipeline for audio data processing, including on-device pre-processing, cloud-based deployment of a Deep Convolutional Neural Network (CNN) classifier for 120 target species and an open-set detector for novel calls, and to validate classifier performance against expert-annotated reference datasets; (3) to combine acoustic detections with high-resolution environmental covariates (land cover, noise levels, meteorological conditions, and human activity indices) to estimate species richness and occupancy using hierarchical occupancy models and generalized additive models (GAMs); (4) to assess spatial and temporal patterns of biodiversity in relation to city structure and anthropogenic factors; and (5) to evaluate user engagement, data quality, and data governance implications for sustained citizen-science integration. The research adopts a pragmatic mixed-methods design under an ecological informatics framework, guided by the theory of ecosystem services and the acoustic niche concept. Population and sample comprise urban birds within the 40 neighborhoods, with 150 volunteers contributing roughly 18 hours of standardized audio per month, yielding an expected dataset exceeding 2,700 hours of) recordings, supplemented by 1,200 hours of expert-validated labeled audio from regional reference libraries. Data collection instruments include smartphone-enabled passive-recording apps configured for uniform sampling rates (44.1 kHz, 16-bit), calibrated noise assessment tools, a standardized protocol for recording duration and time-of-day sampling, a mobile survey for environmental metadata, and a centralized cloud repository with metadata standards ( Dublin Core-inspired). Validity and reliability are addressed through (i) cross-validation of the CNN classifier against held-out expert-labeled segments (precision, recall, F1-score), (ii) inter-annotator agreement for a subset of vocalizations (Cohen’s kappa), and (iii) test-retest analysis of environmental covariates. Data analysis proceeds in stages preprocessing and feature extraction (Mel spectrograms, MFCCs), species-level classification with transfer learning and data augmentation, occupancy and abundance modelling (multi-species occupancy with Varies of detection probability) using Bayesian hierarchical models, and spatial-temporal trend analysis via GAMs. A conceptual model integrates smartphone-derived acoustic detections, environmental covariates, and urban layout metrics to explain species richness and occupancy. Expected findings include high classifier accuracy for common urban species (precision >0.80, recall >0.75), robust occupancy estimates that reveal inverse relationships between noise intensity and passerine detections, and clear spatial gradients of biodiversity aligned with green-space connectivity. The study contributes to knowledge by demonstrating a scalable, participatory ICT-driven framework for urban biodiversity monitoring that merges citizen science with automated acoustic analytics, providing transferable methodologies for cities worldwide, and delivering actionable indicators for urban planning and conservation. The main conclusion anticipates that smartphone-based acoustic monitoring can produce reliable, cost-effective biodiversity maps at neighborhood scales when supported by standardized protocols, local reference libraries, and transparent data governance. Recommendations include the expansion to multi-city deployments, enhancement of real-time feedback to participants, integration with ecological corridor planning, and the development of open-access dashboards for policymakers and the public.