Automated and repeated aerial observations of GPS-collared animals using UAVs and open-source electronics

Abstract

Telemetry has enabled ecologists to link animal movement trajectories and environmental features at a fine spatiotemporal resolution; however, the effects of social interactions on individual choice within large mobile groups remain largely unknown. Estimating the effect of social interaction in the wild remains challenging because existing long-term tracking tools such as GPS collars focus on the movements of a single individual and cannot observe the behavior of other individuals within the group. The progression of socially informed movement models requires measuring simultaneous trajectories of many individuals at once, as well as the instantaneous social cues to which individuals may be responding. The availability of low-flying unmanned aerial vehicles (UAVs) and low-cost open-source electronics presents a promising opportunity to collect fine-scale data on social interactions in order to advance our understanding of collective behavior. Here, we present a tracking system that enables the repeated localization and observation of a collared individual and its near neighbors using nadir video footage collected from a commercial UAV. We make use of open-source electronics combined with the UAV's in-built functionality that allows it to follow a stream of GPS locations to create an automated system that can follow a specific individual without user control. We demonstrate the tracking systems' performance by studying the group movements of a herd of Exmoor ponies (Equus ferus caballus), and as a proof of concept, we examine the position of the focal individual (collared animal) in relation to the center of the video frame. We also collect information about the focal individual's nearest neighbors. The automated animal observation tool is effective at consistently keeping the focal individual close to the center of the video frame, offering a new dimension to existing remote telemetry tools. For instance, the repeated observation of the same individual in different physiological states, seasons, and demographic groups potentially opens new avenues in collective movement ecology research. By making our design, software, and firmware freely available, we aim to encourage continuous improvements to collective behavior research and to facilitate replicable approaches across other species and ecosystems.

INTRODUCTION

In the last several decades, technological innovation has advanced rapidly in the field of movement ecology. This powerful synergy between science and technology has transformed the way we study ecology and has resulted in many exciting discoveries (Cagnacci et al., 2010). In the context of remote telemetry, technological developments have focused on GPS device miniaturization, increased temporal resolution, greater spatial accuracy, and longer battery life (Kays et al., 2015). As a result, vast amounts of high-accuracy animal location data for a variety of species can be collected (Bridge et al., 2011). Sensors, such as accelerometers, magnetometers, temperature sensors, and acoustic recorders, further enhance data collection by augmenting location data with information on physiological variables and energy expenditure (Boyers et al., 2019; Hooten et al., 2019; Martin Lopez et al., 2015; Qasem et al., 2012; Wilson et al., 2006). When combined with environmental data collected from remote sensing, aerial surveys, or transect sampling, animal movement patterns can be linked with physiological and environmental factors, providing an integrated view of the animal and its environment (Kays et al., 2015).

Although modern telemetry tools can provide high-resolution data on the movement and behavior of animals, they are limited in that they only allow for the tracking of a single individual and do not provide information on the instantaneous social cues and signals an individual animal may be observing. Many species move collectively, and social interactions play a significant role in decisions relating to when and where to move (Westley et al., 2018). Collective behaviors therefore drive many fundamental ecological processes (DeLellis et al., 2014; Sumpter, 2010), for instance, herds of ungulates find safety in numbers (Scheel, 1993), schools of fish display synchronized escape responses under predatory attack (Beauchamp, 2012; Herbert-Read et al., 2015), avian species such as penguins huddle for social thermoregulation (Ancel et al., 2015; Gilbert et al., 2010), and foraging routes are copied in species such as sparrows (Lima & Zollner, 1996).

Despite the ecological importance of collective movement (Hughey et al., 2018), we currently lack effective tools for studying social interactions in animals in the wild. In small, highly cohesive groups, it may be possible to equip every individual with a GPS collar (Strandburg-Peshkin et al., 2015). However, in many species it is either not feasible to tag multiple individuals within a group, or the fission–fusion dynamics of group formation mean that individuals quickly separate (Couzin, 2006). Several studies have employed unmanned aerial vehicles (UAVs) (or drones) to study collective movement in situ (Hui et al., 2021; Iwamoto et al., 2022; Torney et al., 2018; Van Andel et al., 2015); however, there are many important ecological questions that cannot be addressed through short-term studies of this type. Tools that allow repeated observation of the same individual at multiple time points and in different environmental conditions while simultaneously observing other individuals within the group are required to investigate questions relating to leadership dynamics (Krause et al., 2000; Pettit et al., 2015), spatial aggregation of individuals within the group (Hansen et al., 2016), information sharing and cultural transfer within the group (Simons, 2004; Stewart & Harcourt, 1994; Ward et al., 2008), and individual personality (Sasaki et al., 2018). A promising approach to address these open questions in ecology is the use of customizable open-source electronics that are available from manufacturers such as Arduino, Raspberry Pi, and Adafruit. There is a growing use of these devices in ecology for monitoring and tracking wildlife (Alarcón-Nieto et al., 2018; Foley & Sillero-Zubiri, 2020; Greenville & Emery, 2016; Wild et al., 2022) as they offer a low-cost, highly flexible alternative to traditional commercial telemetry devices.

In this work, we propose an open-source low-cost system designed to enable repeated observations of GPS-collared individuals using UAVs. We combine a custom-built GPS collar and long-range (LoRa) radio transmitter with a commercial UAV and take advantage of the in-built capacity of the UAV to follow a stream of GPS locations. We first describe our system and then demonstrate the efficacy of the method by applying it to an example case study on the social dynamics of a small group of Exmoor ponies.

MATERIALS AND METHODS

An overview of our proposed system can be found in Figure 1 which illustrates the roles of the different components we employ and the manner in which they communicate with one another. In summary, we use an off-the-shelf commercial drone that automatically follows a stream of GPS locations that are transmitted from a collared animal via a base station and controller tablet. The GPS collar broadcasts high-frequency GPS fixes (1 Hz rate) to a base station using a LoRa radio transmitter. The base station receives GPS fixes and forwards them to the tablet via Bluetooth, which then controls the drone. A custom Android app installed on the tablet spoofs the tablet's GPS location by replacing its own location with the GPS coordinates from the collar. By implementing the drone's “follow-me” mode in this format, the drone is instructed to follow the collared animal rather than the tablet controlling the drone. As a result, the drone tracks and records nadir footage of a specific collared individual which is on average located at the center of the video frame. Detailed specifications and descriptions of components and the protocol of assembling both the base station and GPS collar are outlined in Appendix S1: Tables S1 and S2.

FIGURE 1. A schematic showing the key components of the automated tracking system setup including the GPS-collared animal, the operator, a base station, the tablet, and a drone as well as description of each component's role and how they are linked via various communication networks.

Hardware

Following the approach of Foley and Sillero-Zubiri (2020), we used a variety of microcontrollers and breakout modules developed by Adafruit Industries (www.adafruit.com), which manufactures a wide range of components for DIY electronics projects. All components employed were part of the Feather family, a range of microcontroller main boards (termed Feathers) and extension boards (termed FeatherWings) developed by Adafruit that provides a modular framework for creating bespoke devices by combining different microcontrollers, communication boards, and functional extensions, such as GPS modules. We created two custom devices for the system, a GPS tracker collar for transmitting locations, and a base station for receiving locations and forwarding to the controller tablet.

Custom GPS collar

To construct our custom GPS collar, we used an Adafruit Ultimate GPS FeatherWing (a low-power module for GPS location acquisition) and an Adafruit Feather M0 with RFM95 LoRa radio transceiver for wireless communication. Male and female stacking headers were used for connecting boards together, a uFL to RP-SMA antenna adapter cable connected the board and antenna which was then connected to a dipole swivel antenna (Appendix S1: Figure S2). We soldered the female stacking header onto the Adafruit Ultimate GPS FeatherWing while the male stacking header and the uFL connector were soldered onto the Adafruit Feather M0. Then, we connected the uFL to RP-SMA antenna adapter cable and the antenna. The GPS collar was powered by a 3.7-V, 2000-mAh rechargeable lithium-ion battery pack. To protect the assembled unit, we placed it in a 3D printed plastic casing custom designed in FreeCAD, an open-source 3D parametric modeler. The custom GPS collar, including the casing, electronics, and the halter, weighed 150 g.

Base station

The components used to construct the base station include (1) FeatherWing Tripler Mini Kit that connects several featherwings boards together, (2) Adafruit FeatherWing monochrome organic light-emitting diode (OLED) for display purposes, (3) Adafruit LoRa radio FeatherWing transceiver for wireless communication, (4) Adafruit Feather nRF52840 Express for bluetooth communication, (5) RP-SMA tilt Swivel 1/2 wave whip omnidirectional antenna 2.4 GHz, (6) edge-launch SMA connector for connecting to whip antenna, and (7) male and female feather stacking headers for stacking boards together (Appendix S1: Figure S1). Soldered onto the FeatherWing Tripler Mini Kit were female feather stacking headers, while the male feather stacking headers were soldered onto the other components: the Adafruit FeatherWing monochrome OLED, the Adafruit (LoRa) radio FeatherWing, and the Adafruit Feather nRF52840 express. Then, the SMA connector was soldered onto the Adafruit LoRa radio FeatherWing and the whip antenna attached. Lastly, components were stacked together to make a single base station unit and powered by a power bank using a USB cable. To protect the base station, we placed it in a 3D-printed plastic housing that was again designed using FreeCAD as shown in Figure 2B. The design files for both the custom GPS collar and the base station are available here (Kavwele et al., 2024); however, for long-term field deployment, they will require modification to be made watertight and shockproof.

FIGURE 2. (A) A custom GPS collar attached to an adult Exmoor pony using a horse halter; (B) a fully assembled base station housed in a 3D-printed case; (C) the DJI Mavic 2 Pro drone used for video recording (photos by Cyrus M. Kavwele).

Unmanned aerial vehicle

We used the DJI Mavic 2 Pro drone (www.dji.com). The drone (shown in Figure 2C) is a quadcopter that can be controlled at a maximum distance of 6 km and has a maximum flight time of 27–31 min, and maximum speed of 20 m/s (no wind). The drone is equipped with a 20-megapixel Hasselblad camera which has 1″-type CMOS sensor and 4K resolution (3840 × 2160 pixel) recording at 30 fps. The camera is mounted on a stabilizing gimbal with control range pitch −90 to +30 (tilt). The drone is powered by a lithium-ion polymer intelligent flight battery with a capacity of 3850 mAh. Key specifications of the drone are as highlighted in Appendix S1: Table S3. To control the UAV, we used the third-party autonomous flight app Litchi (https://flylitchi.com/) as it has an in-built “follow-me” mode.

Software

All software used for the microcontrollers and the custom android app is open-source, released under the MIT license, and available from Kavwele et al. (2024).

The software for the GPS collar was designed so that it remains in a low-power state until the base station is in proximity, then when instructed to do so it enters a high-fix rate “GPS tracker mode” that sends a continuous stream of GPS locations to the base station via a peer-to-peer LoRa connection (see Algorithm 1).

1 AlgorithmPseudocode for custom GPS collar

From the perspective of the operator the system works as follows. Once the operator expects to be within range of a collared individual, they turn on the base station, and it begins broadcasting a wake-up message to any custom GPS collar attached to an animal that is within range. The base station broadcasts the message, waits 10 s for a response, and then, if no response is received, repeats the message (see Algorithm 2). If a GPS collar is within range and is in its listening mode, the collar will send an acknowledgement to the base station that includes a unique identifier for the custom GPS collar. The base station then prompts the operator to either ignore the GPS collar connection (which if selected, sends the base station back to broadcasting mode) or to connect to the GPS collar which involves sending a message that is addressed to the specific GPS collar instructing it to enter standby mode. During standby mode, both base station and GPS collar switch from a long-range communication mode to a short-range mode, and the GPS collar enters a high-fix rate GPS mode (1 Hz). Once the GPS collar has confirmed it has entered the short-range communication, high-fix rate mode, the base station prompts the operator to either disconnect and send the GPS collar back to sleep or to initiate its GPS tracker mode.

2 AlgorithmPseudocode for base station

If the operator selects the GPS tracker mode, the base station starts receiving the high-fix rate locations from the GPS collar on the animal and forwards them to the tablet via a Bluetooth link. At this stage, the custom app on the tablet begins to spoof the tablet's GPS location, making it appear as if the tablet is located with the GPS-collared animal when it is actually held by the operator. The operator then launches the drone, and once it is launched, they activate the in-built “follow-me” mode in the Litchi drone controller with a prespecified altitude. This causes the drone to start tracking the focal individual (collared animal) and record nadir footage of the focal individual and its near neighbors within the herd. The initiation and termination of the recording are controlled by the operator. While the base station is receiving GPS locations, the operator has the choice to either keep the base station receiving and listening for location messages or send a sleep command to the GPS collar which disconnects the communication channel and sends the GPS collar back to its low-power state. Typically, the operator will take control of the drone at this point; however, in the absence of user control or GPS locations being received, the “follow-me” mode will automatically deactivate, and the drone will remain airborne at the last GPS location received.

CASE STUDY Study species and location

We conducted a field test of our framework at the University of Glasgow's Cochno Farm and Research Center (4°24.467′ E, 55°56.237′ N) which covers approximately 66 ha and is about a 25-min drive to the north of the university. The facility is home to an Exmoor pony (Equus ferus caballus) herd that was introduced in November 2020 and consisted of four mares, one with a foal, and a stallion at the time of introduction. The herd is part of a breed conservation initiative and currently supports research into Exmoor pony conservation and rewilding. Ponies are true grazers that form small herds with a preference for open habitats (Vermeulen, 2015), and they influence ecological succession by preventing shrub and tree overgrowth, which in turn hinders forest succession (Hagstrup et al., 2020). At the time of data collection, more ponies had been introduced, bringing the total to 11 individuals, which included seven mares, one stallion, one subadult, and two foals.

Data collection

We collected data between 9 April 2022 and 2 October 2022, on five separate days. The GPS collar was deployed and collected at the start and end of each field test, and no chemical immobilization was required due to the ponies' familiarity with their handler and fondness for carrots. The GPS collar was attached to the side of a horse halter, strapped on to an adult mare, and adjusted to avoid discomfort.

Once the GPS collar was deployed, we relocated to a location approximately 100 m away from the herd and prepared a takeoff and landing zone. This distance was selected to minimize any potential disturbance to the herd caused during the setup and launch of the drone. We activated the base station by connecting a power source and once connected to the tablet via bluetooth, it began to periodically broadcast the wake-up message to any GPS collar within range. The base station continued to broadcast the wake-up message until the deployed GPS collar entered its listening mode (see Algorithm 1) at which point we were able to initiate the connection between the base station and GPS collar, and we then placed the GPS collar in standby mode awaiting the instruction to enter the full GPS tracker mode.

After successfully placing the GPS collar in standby mode, we launched the drone and climbed away from the herd at a 45° angle up to 80 m above ground level to ensure that the ponies' natural behavior was not disrupted. Climbing at an angle enhances stability and control, counters the airflow around the drone that affects stability, and is energy-efficient since the drone uses forward thrust to gain altitude. We considered 80 m above ground level to be the optimal height for the drone as it resulted in high-resolution imagery, and the ponies did not respond to the noise of the drone at this altitude. Once the drone was airborne at the target altitude, we initiated the GPS tracker mode of the GPS collar and activated the “follow-me” mode on the drone controller application. The final stage of tracking was to begin GPS spoofing on the tablet, after which the drone automatically flew toward the collared individual, positioned itself directly above and began to track the collared animal and record nadir footage. Recording bouts lasted for about 15–20 min depending on weather conditions, and repeated flights were undertaken with breaks in between to change battery.

Data processing and analysis

To locate animals within video frames, we used established computer vision techniques, specifically employing the you only look once (YOLO) (Redmon et al., 2016) single-shot object detector. YOLO object detectors are a family of deep convolutional neural networks that predict bounding boxes and class probabilities in a single pass through the network and have been applied in several ecological computer vision tasks (Kavwele et al., 2022; Roy et al., 2023; Torney et al., 2019).

For our application, we employed an implementation of YOLOv3 (Torney et al., 2019) that had been pretrained on drone images collected from a study of wild horses in Portugal (Inoue et al., 2020) in combination with a detection linking algorithm to create tracks (Wojke et al., 2017). Due to the similarity in the study animals, the YOLOv3 object detector differentiated objects from the background satisfactorily; hence, no further fine-tuning was required. We used this method to detect every individual within a frame of the video and create tracks for individuals and then employed a manual process of track inspection to link broken tracks and to determine the collared individual. The procedure involved inspection of video footage at points where a track ended and manually linking any subsequent new track that was associated with the same individual. The track ID of the individual wearing the GPS collar was noted based on its position to the center of the frame and its coloration. We took the xy-pixel location of the center of the bounding box associated with an individual as its location and used this value to calculate nearest-neighbor distances and the Euclidean distance to the center of the video frame.

Performance of the aerial observation system

We firstly evaluated the performance of our system by examining the average position of the focal individual (collared animal) relative to the center of the video frame. The results revealed that the distance of the focal individual to the center of the video frame is on average approximately twice the focal individual's body length, as shown in Figure 3A. Furthermore, as indicated by the dotted line in Figure 3A, the average distance of all other individuals within the frame to the center of the video frame are nine times the body length. This is illustrated in Figure 3C, which is a cropped version of the original, where the blue dot indicates the center of the frame while the focal individual is outlined with a red bounding box. The fact that the focal individual is typically located at the center of the frame provides a straightforward way to identify this individual when analyzing the herd's trajectories. Once all individual's are tracked, the collared individual will be associated with the trajectory that has the shortest average distance to the center of the frame.

FIGURE 3. (A) The distribution of the collared individual's distance to the center of the frame. For comparison, the dotted line shows the average distance from the center of the video frame to all other individuals within the frame. (B) The distribution of the collared individual's nearest-neighbor distances, as an example of behavior analyzed from data collected using a custom-built tracker. Body length was calculated based on the maximum length of the bounding box surrounding the focal individual in each video. (C) A cropped still frame showing the collared individual bounded by a red box with the center of the frame indicated by a blue dot (photo by Cyrus M. Kavwele).

We also provide an example of a behavioral metric that can be extracted from data collected with the automated tracking system. In Figure 3B, we show density plots of the near neighbor distances for the first, second, and third nearest neighbors. This analysis shows that the nearest neighbor is typically found within a distance equivalent to one body length of the focal individual, while the second and third nearest neighbors are within a distance of approximately 2.5 times the body length. While investigating near neighbor distances represents a very simple example of the type of analysis that may be performed using repeated observations of a single collared individual, it provides insight into the spatial structure of the herd and if repeated for multiple individuals may reveal different levels of sociality between individuals or within a single individual at different points in time.

DISCUSSION

We introduce an innovative approach for developing a tracking system utilizing low-flying drones and affordable programmable open-source electronics (Cressey, 2017). By utilizing data obtained from Exmoor ponies, we assessed the effectiveness of our tracking system by analyzing the position of the collared individual relative to the center of the video frame. Moreover, we demonstrated the system's capability to gather individual-level data within the social context of a group, expanding the range of potential questions that such data can address. This automated tracking system is novel in its ability to allow users to observe collared individuals across multiple time points, in different seasons and habitats while collecting data on the individual's near neighbors.

The continuous advancement and refinement of telemetry, increased computational power, and robust mathematical models have enabled researchers to delve into the intricacies of movement ecology and behaviors (Hooten et al., 2017; Joo et al., 2020; Mennill et al., 2012; Nathan et al., 2022; Northrup et al., 2022), unlocking new avenues for exploration. Despite significant progress in movement ecology, the study of localized social interactions within collectively moving species in the wild has not progressed at the same pace. This discrepancy arises from the limited capabilities of existing tracking tools, which fail to provide simultaneous fine-scale trajectories for each individual in a group over extended periods. However, with the development of our automated tracking system, it becomes feasible to acquire fine-scale data on the social behavior of a collared animal in the wild. The ability to repeatedly observe an individual within its social context presents a unique opportunity to investigate complex behavioral questions, such as decision-making, personality, leadership roles, and the role of social cues in movement decisions under varying physiological and environmental conditions.

Obtaining simultaneous trajectories of non-focal individuals within a group will enable researchers to explore social interactions. For example, it allows for investigating how individuals respond to social cues such as the influence of nearest neighbors or group average heading (Dalziel et al., 2016), potentially unveiling stereotyped individual responses to collective attributes of the group. This tracking system could also be employed to study social behavior in the context of post-reintroduction monitoring. As an example, Mertes et al. (2019) utilized GPS location data to monitor the post-release movement of reintroduced scimitar-horned oryx (Oryx dammah). Therefore, adopting our automated tracking system could enhance data collection by providing insights into the cultural transmission of information (Berdahl et al., 2018) and the leadership roles of naive individuals introduced into experienced populations. In the realm of social interactions and dynamics, investigations can be conducted into aspects such as the formation of “leader–follower” relationships, consensus building as clusters explore new habitats, and the influence of nearest neighbors, among other factors. However, acquiring trajectories of non-focal individuals within a herd depends on factors such as the spatial extent of the herd, the flight altitude, and the camera field of view. Thus, a comprehensive understanding of the study system and the behavior of the target species is crucial when designing behavioral studies involving non-focal animals.

Along with the new ecological questions that can be addressed through repeated observations of the same individual, the use of GPS collars to control and position a drone will potentially also facilitate existing approaches to studying collective behavior in the field. The communication system has an expected range of 15 km in the long-range mode and 2 km in the short- range mode depending on the terrain and the height of the base station. Further, the GPS coordinates of the collar are transmitted to the base station and can be displayed on the tablet. Beyond the tracking capacity the system will therefore make finding animals more straightforward in the field and once located will facilitate staying with the herd. Strategies for constant observation of a herd, such as the relay system of drones described in Koger et al. (2023), could in principle also be automated with multiple controller tablets receiving locations from the base station.

Our automated tracking system presents a valuable tool for facilitating the acquisition of collective behavior data in the wild if its full potential is realized. However, when considering drone usage, there are several limitations and challenges that need to be overcome. The legal framework governing the use of UAVs may acts as a barrier to fully harnessing their potential in ecological studies (Witczuk et al., 2018). For instance, long-distance flights become impossible due to the requirement to maintain visual line of sight with the drone. This necessitates frequent relocation of UAV operators, which can be time-consuming. Furthermore, legal regulations vary across countries, resulting in inconsistencies in drone operations. The flight time of a UAV depends on its model, payload, and battery size and is greatly reduced by high wind speeds and/or low temperatures (Beaver et al., 2020; Torney et al., 2018). In addition to regulations and battery life, UAVs are also affected by prevailing weather conditions and poor visibility during precipitation, fog, or haze restricts operations.

The tracking system is highly customizable, allowing for the incorporation of user-defined requirements and the integration of additional components to enhance performance, thus expanding its applicability across various biomes and species. The flexibility and potential for future modifications are discussed further below. To begin with, the use of programmable microcontrollers enables researchers to modify the software to suit their specific research needs. For example, they can adjust the frequency of GPS fix rates, determine the type of data to store, and set the frequency of low-power states, among other user-specific settings. Changing the drone camera sensors is also a possibility. UAV-based platforms can leverage a wide range of sensors available on the market, such as RGB cameras, hyperspectral, thermal, LIDAR, and multispectral sensors (Sun et al., 2021). Employing different sensors can enhance the data collected, such as evaluating plant health and assessing the influence of stress on ungulate collective behavior using multispectral sensors (Wang et al., 2019). Another example is the utilization of thermal infrared sensors, which can discriminate endotherms from their surroundings (Beaver et al., 2020; Burke et al., 2019) and enable the detection of animals in low-light conditions, such as herds in dense woodlands.

Drones have become a versatile tool for ecological studies and have been employed in studies of species distribution and abundance (Corcoran et al., 2021), human–wildlife conflict mitigation (Gorkin III et al., 2020), and habitat structure and its influence on collective decision-making (Strandburg-Peshkin et al., 2017). Our proposed methodology has the potential to broaden the scope of ecological questions that can be addressed in this context since the system has the ability to collect data on the surrounding physical and biotic environment (as discussed in Koger et al., 2023) alongside capturing footage of both focal and non-focal individuals and their interactions. For instance, the drone footage can be analyzed to extract information on landscape features, human disturbance, habitat structure, and resource availability, and as such, the extracted data would allow ecologists to infer group navigation and decision-making processes and investigate dynamic, individual-level responses to environmental features. Such empirical data on fine-scale movement can yield valuable insights into an animal's memory of the landscape, particularly when encountering linear structures or features like rivers, roads, or resource patches, among others.

The 3D housing cases can be customized to fit various ungulate species, ranging in size from medium to mega-herbivores found in different biomes. The objective is to ensure a perfect fit, watertightness, and robustness to withstand any shock or disturbance caused by the collared animal. We emphasize, however, that our 3D cases were not tested for robustness since we attached and detached them at the end of every filming event; therefore, modifications would be necessary before field deployment. Additionally, future modifications could explore ways to harness green energy by using a mini solar panel that charges the battery or the use of vibration energy harvesters, which convert the kinetic energy of the collared animal into electrical energy (Zhang et al., 2021). Such changes would eliminate the need to recover the GPS tracker for charging or replacement, significantly increasing its lifespan from several months to years and resulting in more extensive data, spanning multiple seasons. Without additional charging, we estimate a 2000-mAh lithium-ion battery will provide approximately 20 h of high-frequency observation and would last approximately 6 months if entering listening mode once per day.

In conclusion, our low-cost, automated tracking system has been tested and demonstrated functionality for repeatedly observing collared animals, enabling the collection of fine-scale behavioral data on species that live in groups. This system represents a significant advance, allowing for the transition from individual tracking to the simultaneous tracking of all individuals in cohesive groups with the focal individual at the center of the frame. Consequently, this bespoke tool empowers researchers to explore ecological questions with unprecedented levels of detail and complexity, yielding fresh insights into the movement ecology and collective behavior of group-navigating species in the wild.

ACKNOWLEDGMENTS

This work was supported by the University of Glasgow's Lord Kelvin/Adam Smith (LKAS) PhD scholarship for Cyrus M. Kavwele. Colin J. Torney is supported by a James S. McDonnell Foundation Studying Complex Systems Award. We are grateful to the Bio-electronics lab under the leadership of Jakub Czyzewski, University of Glasgow. We are also grateful to Majaliwa M. Masolele and Dennis Minja for their assistance in the field.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT

Data, code, and utilities (Kavwele et al., 2024) are available from Zenodo: https://doi.org/10.5281/zenodo.10963780.

ETHICS STATEMENT

Handling of the Exmoor pony (Equus ferus caballus) was approved by the Research Ethics Committee in the School of Biodiversity, One Health and Veterinary Medicine, University of Glasgow, under the approval number EA45/19.