Automated Acoustic Monitoring for Urban Bat Diversity Assessment

Thesis Abstract

Urban environments increasingly fragment bat habitats, yet rapid urbanization and noise pollution obscure accurate assessment of bat diversity. This study develops and tests an automated acoustic monitoring framework to quantify urban bat diversity, distribution, and activity patterns, addressing the problem of inconsistent, labor-intensive survey methods and limited temporal resolution in conventional monitoring. The aim is to produce a scalable, ethics-compliant methodology that integrates autonomous recording devices, machine learning classification, and ecological modelling to inform urban biodiversity management. The objectives are to (1) deploy a network of autonomous ultrasonic recorders across three urban districts with varying green space indices, (2) develop and validate a supervised deep-learning classifier for open-access bat echolocation calls to identify species and functional guilds with ?85% precision, (3) quantify bat activity and species richness across temporal scales (seasonal, diurnal, and weather-driven variations), (4) examine associations between bat diversity and urban features (green corridors, building density, noise levels, and artificial lighting) using generalized linear mixed models, and (5) propose a decision-support framework for city planners to mitigate bat-habitat fragmentation. The study employs a mixed-methods design guided by niche theory and urban ecology paradigms, with a population comprising bat communities in a metropolitan region characterized by mosaic habitats. A stratified random sampling strategy is used to select 60 sampling points across three land-use zones (green-core parks, semi-urban fringes, and built-up corridors), with recording campaigns conducted continuously for six months using 20 low-cost autonomous detectors capable of capturing high-frequency echolocation calls. Data collection instruments include calibrated ultrasonic recorders (0.1–150 kHz), ground-truthing via nocturnal netting for a subset of sites (n=12) to validate species identifications, and environmental sensors measuring temperature, humidity, wind, ambient noise, and light intensity. The analytic framework comprises three integrated streams. First, signal processing and species identification employ a convolutional neural network (CNN) trained on a curated library of local echolocation calls, complemented by a random forest classifier for uncertain detections, with model performance validated against ground-truth identifications and cross-validated using k-fold techniques. Second, ecological modelling uses generalized linear mixed models (GLMMs) to relate bat activity and species richness to habitat variables, with site as a random effect and temporal covariates (month, urban temperature, moon phase) as fixed effects. Third, spatial analysis applied to detect habitat corridors and fragmentation patterns utilizes network analysis and Spatial Autoregressive (SAR) models to account for spatial autocorrelation. Data will be processed with open-source software including R and Python-based pipelines, and model selection will rely on information criteria (AIC/BIC) and cross-validation. Expected findings include robust species identifications for at least six urban bat species, evidence of higher activity and richness in green-core parks and along linear ecological networks, and significant associations between bat diversity and green space metrics, reduced lighting intensity, and lower nocturnal noise in targeted districts. The study anticipates diurnal and seasonal patterns aligned with literature on insect prey availability and commuting foraging strategies, with weather variables (temperature and humidity) explaining a substantial portion of variance in activity. The contribution to knowledge lies in delivering a validated, scalable automated monitoring framework tailored to urban contexts, advancing methodological approaches in bioacoustics through an end-to-end pipeline that combines acoustic signal processing, machine learning, and spatial-temporal ecological modelling. The findings will offer actionable guidance for urban planning, such as prioritizing connectivity corridors, implementing bat-friendly lighting schemes, and enhancing green infrastructure to sustain diverse bat assemblages. Conclusions emphasize that automated acoustic monitoring can provide high-resolution, repeatable, and cost-effective assessments of urban bat diversity, enabling proactive management of urban ecosystems. Recommendations include expanding the monitoring network to additional cities, integrating citizen-science noise data to augment spatiotemporal coverage, and developing open-access dashboards for stakeholders to monitor urban bat diversity in near-real time.