Theoretical and empirical investigations of echolocation in bat groups

Abstract

Animals in groups gain a variety of advantages from their membership. Group membership also simultaneously comes with a variety of costs. From a sensory perspective being part of a group challenges the individual sensory system with the multitude of signals to be dealt with. Much work has gone into understanding how passive sensing animals that act as ‘receivers’ of signals (eg. vision and audition), manage the sensory challenge of groups. Multiple passive sensing animals may perceive their environment and signals in it without majorly affecting the perception of their neighbour. Active sensing animals in contrast emit probes of energy to detect their surroundings. In active sensing groups, it is expected that group-members will mutually interfere, or jam each other’s sensory systems. Echolocating bats emit intense calls, and use the returning echoes to detect their surroundings. In groups, due to mutual jamming, individual bats are expected to suffer a severe drop in echo-detection. Despite this expectation, echolocating bats are very gregarious, and show impressive collective behaviours. In this thesis I investigate how active sensing echolocating bats manage to echolocate in groups using a combination of computational simulations and field studies, while also contributing to the methods that ease acoustic tracking and the analysis of echolocation calls.In Chapter 2, I quantify the sensory challenge of group echolocation. In groups, the returning echoes of each bat will be overlapped by the intense calls and echoes of their neighbours. I estimate the detriment in echo detection that bats may experience with increasing group size using computational simulations. I build an experimentally parametrised model implementing details of bat audition, sound propagation and group geometry. I find that bats may still be detecting echoes in group sizes of up to a hundred. Bats in such large groups however may be detecting only one neighbour occasionally once every three calls. The model assumed a simplified auditory system, and thus represents a lower-bound for echo detection. My model represents the first attempt at a biologically parametrised model of group echolocation. The results raise the question about the severity of group echolocation and estimates the sensory input available for collective motion in bat aggregations.Chapter 3 is an observational study in the field looking into the echolocation of high duty-cycle bats in groups. High duty-cycle bats emit long calls with short pauses in between. Their long calls and frequent call emission increases the likelihood of call-echo overlaps even in small groups. Due to the challenge of analysing overlapping calls, not much work has been done studying high duty-cycle bat groups, and have primarily been in flightroom conditions. Using audio and video recordings of free-flying bats in a cave, I analyse the difference in echolocation when high duty cycle bats fly alone and when in groups of upto four bats. I develop a package to automate the segmentation and measurment of individual calls into their component parts (described in Chapter 7). I also develop a method to analyse audio with overlapping calls and use it in conjunction with simulations to understand if bats alter their echolocation in groups. The results suggest no major changes in call parameters between solitary and group-flying bats. The study contributes know-how in the analysis of overlapping calls and automation of individual call analysis. The study highlights the robustness of bat echolocation, and stresses the importance of field studies to characterise the capabilities of active sensing animals.From high duty-cycle bats, Chapter 4 reports the details of another observational study looking at group echolocation in free-flying low duty-cycle bats. Low duty-cycle bats emit short calls with long silences in between. Despite the punctuated calling behaviour of low duty-cycle bats, results from modelling in Chapter 2 show that echo detection can already be affected from group sizes of 30 bats onwards. I present the methods and investigative potential behind what I call the Ushichka dataset. Ushichka is a multi-channel, multi-sensor dataset of Myotis myotis and Myotis blythii bats echolocating over a range of group sizes between 1-~30 in a cave chamber. The dataset consists of synchronised microphone and thermal-camera arrays, along with a LiDAR scan of the cave chamber. The microphone arrays capture the call emissions, while camera arrays capture flight trajectories. The LiDAR scan provides a contextual 3D record of the volume. Given the position, call emission and LiDAR data, we can for the first time reconstruct the sensory inputs of individual bats in groups to great accuracy. Analysing multi-bat audio brings its own challenges such as call overlaps and multi-channel correspondence. However, it is my opinion that the observed group sizes of upto ~30 bats corresponds to a ‘Goldilocks’ zone, where current methods may perform satisfactory acoustic tracking. Unlike comparable studies, Ushichka, is to my knowledge, the first such dataset to record the collective behaviour of bats in the wild with multiple sensors simultaneously.Chapter 5 marks the beginning of a series of methodological reports contributing to the study of group echolocation. Multi-microphone arrays are central to studies of echolocation. Arrays provide access to the 3D position of the calling bat, but also add to the logistical effort during field work. Most arrays consist of microphones placed on bulky frames that are difficult to carry. Their typically rectilinear forms stand out in natural settings and may result in artifactual inspection behaviours by the animals themselves. In place of frames, placing microphones freely in the field also brings the burden of having to measure microphone positions each time. In Chapter 5 I present the results of a collaboration towards a frame-less, measurement-free approach to acoustic tracking. The workflow involves freely placing microphones and recording a series of common sounds on all channels. The time-differences-of-arrival between channels are then used to estimate microphone positions automatically. In this report we show the sucessful estimation of freely-placed microphones in a cave setting to within ± 4cm of ground-truthed measurements. This is the first time such a methodology has been applied in the field of echolocation, and it promises to expand the freedom and scale of multi-microphone arrays under field and laboratory settings.The accuracy of acoustic tracking is affected by a host of factors such as array geometry, source sound type and location of sound emission. When designing a microphone array from scratch, or when characterising an array post-hoc, it is important to understand the baseline accuracy the system will show. Chapter 6 presents the tacost software package that generates simulated multi-channel audio according to user-specified scenarios. While tacost does not perform acoustic tracking itself, it generates the simulated data to allow the user to compare the consequences of various design decisions. tacost is a tool to assist the optimisation of acoustic tracking systems during the conception phase, and post-hoc analysis after recordings have been performed.The echolocation call is a common sensory ‘unit’ for investigations. The acoustic parameters of a call and its spectro-temporal structure are tightly linked with the behaviour at hand. Common approaches to measure echolocation calls include using automated inhouse-scripts or manual measurements. Inhouse-scripts suffer from a lack of public scrutiny, while manual measurements are biased and do not scale well with sample size. In Chapter 7, I present the itsfm software package to automate the segmentation and measurement of echolocation calls. I implement a commonly described method to segment CF-FM calls, along with introducing a new algorithm. The new algorithm is consistently more accurate at segmentation than the commonly described method. Even though originally developed for the analysis of CF-FM calls, the routines in itsfm are also of interest for bioacousticians at large.In conclusion, I briefly describe key findings and outline my vision of future research based on the three-pronged approach to study active sensing in groups. The three-pronged approach consists of 1) advancing techniques to aid field studies and analysis of audio with overlapping calls, 2) conducting controlled experiments to better estimate the sensory abilities of individuals and 3) using the collected data from field studies and controlled experiments to generate and parametrise computational models. This thesis provides a glimpse of the advancements that the three-pronged approach can provide to active sensing with its contributions to computational modelling, field observations and new techniques.