Understanding the fundamental ecology of endangered species is essential for effective conservation, yet this remains challenging for elusive species inhabiting tropical forests. For the endangered Bornean white-bearded gibbon (Hylobates albibarbis), much of the available ecological information derives from peat swamp forests, while comparatively little is known from other forest types that make up a large part of its range. Passive acoustic monitoring provides an opportunity to address this gap, enabling the study of species’ vocal behaviour over larger spatial and temporal scales than previously possible.
We deployed eight autonomous recording units across three forest types in Central Kalimantan, Indonesia, collecting 23,244 hours of acoustic data over 18 months. A pretrained deep learning detector was applied to identify great calls, performed by female gibbons as part of morning duets and used as a key indicator for comparing population density. We identified 83,956 great calls and examined how daily calling activity varied across habitats and in response to seasonal rainfall.
Daily calling activity differed significantly among forest types, consistent with expected differences in gibbon population density. Significant temporal variation in calling behaviour was observed consistently across habitats. We documented negative short-term and positive long-term effects of rainfall on calling activity. Daily calling activity peaked 51-52 days following rainfall events, with effect sizes increasing with rainfall dose, suggesting that calling activity reflects a lagged phenological fruiting response to seasonal rainfall.
Our findings highlight the importance of accounting for variable vocalisation rates in acoustic monitoring, particularly when evaluating the additive effects of anthropogenic disturbance and climate change on species behaviour and ecology. We emphasise the value of incorporating spatial data to strengthen ecological inferences from acoustic datasets, and demonstrate the power of deep learning for long-term monitoring of species’ vocal behaviour, providing deeper ecological understanding across increasingly broad spatiotemporal scales.

Eight autonomous recording units (ARUs) were deployed across three forest types (lowland heath, low pole, and mixed swamp) in the Mungku Baru Education and Research Forest, Central Kalimantan, Indonesia, from July 2018 to December 2019. The ARUs recorded each day continuously, saving audio as 1-hour waveform audio files (WAV). To ensure full coverage of morning gibbon duets, recordings between 04:00 and 10:00 were selected, and only days with no missing data were included, leaving 23,244 hours of audio. 

We then applied a pretrained deep learning automated detector, described in Owens et al. (2024), which identified 83,596 H. albibarbis great calls. These are listed in "detector_output.csv", along with associated spatiotemporal data. These detections were aggregated to estimate the number of great calls per day at each ARU (daily call rate), and whether or not a great call was detected on a given day at each ARU (daily call presence). Daily call rate and daily call presence are presented in "daily_data.csv", along with associated spatiotemporal data. To estimate daily rainfall accumulations for our study site we used the PERSIANN-CDR V3 dataset (Ashouri et al., 2015), downloaded via the CHRS data portal (Center for Hydrometeorology and Remote Sensing, University of California, Irvine). Daily rainfall data is presented in "daily_rainfall.csv". To assess detector performance, subsets of the detections were manually validated to test if precision and recall varied across conditions. Information regarding these subsets is presented in "precision_validation.csv" and "recall_validation.csv".

All statistical analyses were conducted in R (R Core Team, 2023). The analysis pipeline for evaluating detector precision and recall can be reproduced using the provided script, "detector_validation.R". The analysis pipeline for assessing the timing of detected calls, and the effect of habitat and rainfall on great call rate and presence can be reproduced using the provided script, "call_analysis.R". 

• Ashouri, H., Hsu, K.-L., Sorooshian, S., Braithwaite, D. K., Knapp, K. R., Cecil, L. D., Nelson, B. R., & Prat, O. P. (2015). PERSIANN-CDR: Daily precipitation climate data record from multisatellite observations for hydrological and climate studies. Bulletin of the American Meteorological Society, 96(1), 69–83. https://doi.org/10.1175/BAMS-D-13- 00068.1
• Center for Hydrometeorology and Remote Sensing (CHRS). (n.d.). PERSIANN-CDR V3 Daily Precipitation Data [Data set]. University of California, Irvine. Retrieved April 1, 2026 from https://chrsdata.eng.uci.edu/
• Owens, A. F., Hockings, K. J., Imron, M. A., Madhusudhana, S., Mariaty, Setia, T. M., Sharma, M., Maimunah, S., Van Veen, F. J. F., & Erb, W. M. (2024). Automated detection of Bornean white-bearded gibbon (Hylobates albibarbis) vocalizations using an open-source framework for deep learning. Journal of the Acoustical Society of America, 156(3), 1623–1632. https://doi.org/10.1121/10.0028268