Bioacoustic data fusion for western chimpanzee population monitoring
- Miranda Fittock
- Aug 27
- 4 min read
Updated: Aug 27
Author: Andrew Houldcroft, Doctoral Researcher, Cantanhez Chimpanzee Project / University of Exeter
Great apes, our closest living evolutionary relatives, are teetering on the edge of extinction, with all 14 non-human great ape taxa highly threatened at present, classified as either endangered or critically endangered by the IUCN Red List. The window of opportunity to conserve these charismatic species is closing – which begs the question of how?
Fundamental to effective conservation is a solid foundation of scientific evidence, such as robust estimates of density and distribution obtained via population monitoring programs. These state parameters allow us to assess population status (i.e., IUCN Red List), monitor changes through time, elucidate and mitigate key threats, identify key habitat areas and associations and critically, understand which conservation strategies actually work.
Population density and distribution are clearly vital to conservation, yet they are notoriously difficult to robustly estimate. This is particularly true for low density, group-living, wide-ranging cryptic taxa persisting in dense forest habitats – with great apes unfortunately being a prime example. Reflecting this, chimpanzee monitoring has largely depended on indirect nest transect surveys to date. Although more recently, conservation technologies such as camera traps, bioacoustics and even drones are quickly moving to the forefront.
With an ever-expanding variety of tools and techniques in the great ape surveying toolbox, researchers are often faced with the difficult task of choosing which single method is best. In reality, each survey approach comes with its own respective strengths and limitations. Camera traps for example, have a suite of robust protocols such as spatially explicit capture recapture (SECR) to estimate chimpanzee density. Yet, these surveys can be time-intensive and suffer from low detection rates leading to imprecise estimates and long deployment times. Bioacoustics on the other hand, can rapidly obtain confident estimates of presence-absence for chimpanzees, but there is a paucity of methods to directly estimate density. This is primarily owed to the inherent complexity of chimpanzee communication in terms of variation in call production rate.
When researchers are forced to pit methods against one another, the remarkable benefits of bioacoustics can be easily overlooked due to the challenges of directly estimating density. But what if researchers didn’t have to select a single method for population monitoring? What if we could leverage the strengths of bioacoustics to improve the accuracy, precision and spatial coverage of density estimates derived from other methods? These are the motivating questions behind my doctoral research into model-based data integration, also known as data fusion.
Data fusion models seek to provide a statistical description of how different data sources are realisations of the same underlying state process of interest (i.e., population density). In doing so, we can hope to leverage the collective strengths of a diverse range of datasets (i.e., large sample size, spatial coverage, high precision) to overcome their respective weaknesses. It is even possible to translate information across data currencies (i.e., detection-nondetection, counts, presence-only). It follows that the rich presence-absence and spatial distribution information offered by bioacoustics, could greatly improve density estimation with other methods on offer, such as camera trapping and indirect surveys of nests.
Investigating this, extensive fieldwork was conducted to monitor the critically endangered western chimpanzee (Pan troglodytes verus) in Cantanhez National Park (CNP), Guinea-Bissau in collaboration with the Institute of Biodiversity and Protected Areas (IBAP). Over the course of 4-months in early 2025, 20 BAR-LTs were rotated across 80 survey locations for an average of 15-days each. These locations comprised a diversity of soundscapes, including farmlands, orchards, secondary growth forest and dense tropical forest. In total, a staggering 10TB or approximately 30,000 hours of audio was collected, containing a rich diversity of acoustic signals including western chimpanzee pant hoots, screams and buttress drumming. In addition, a range of other non-human primates, such as the endangered king colobus (Colobus polykomos) and Temminck’s red colobus (Piliocolobus badius temminckii) were detected.
Automating the detection and classification of target species in this large quantity of raw audio data, is an important precursor to the proposed integrated modelling. For this purpose, a range of deep learning (DL) approaches are being explored and developed with collaborators at the University of Exeter. Once processed, the resulting bioacoustic detection-nondetection (occupancy) data will be integrated with camera trap occupancy and indirect nest survey data collected concurrently in CNP.
This doctoral research is currently in its nascent stage, but we hope that the methods under development, particularly those for bioacoustics, will play an integral role in not only landscape-scale, but also national and regional-level monitoring strategy moving forwards. We thank our primary research assistant Serifo Braima Dabo, IBAP collaborators, particularly Dr Aissa Regalla de Barros, Queba Quecuta, and Américo Nhanga Walanse Sanhá, as well as park guards, village chiefs and regulos for making this research possible. This research was funded by research grants from both the UKRI Centre for Doctoral Training in Environmental Intelligence and Re:wild Primate Action Fund.
Researcher Bio:
Andrew is an early-career conservation scientist, broadly interested in how spatial statistics and data integration can enhance our capacity to monitor wildlife populations and improve conservation outcomes for people and nature. Conservation technology is a central theme of his research, with bioacoustics being the latest addition to his monitoring toolkit. He previously worked in Spain deploying camera traps to monitor ungulate herbivory and movement, as well as analysing high-resolution drone-based imagery to perform habitat classification.
Words and images supplied by Andrew Houldcroft
With so many applications for bioacoustic technology, we're constantly inspired by our customers' projects and their findings.
Comments