Drone-Based High-Resolution Tracking of Aquatic Vertebrates

Determining the small-scale movement patterns of marine vertebrates usually requires invasive active acoustic tagging or in-water monitoring, with the inherent behavioural impacts of those techniques. In addition, these techniques rarely allow direct continuous behavioural assessments or the recording of environmental interactions, especially for highly mobile species. Here, we trial a novel method of assessing small-scale movement patterns of marine vertebrates using an unmanned aerial vehicle that could complement longer-term tracking approaches. This approach is unlikely to have behavioural impacts and provides high accuracy and high frequency location data (10 Hz), while subsequently allowing quantitative trajectory analysis. Unmanned aerial vehicle tracking is also relatively low cost compared to single-use acoustic and GPS tags. We tracked 14 sharks for up to 10 min in a shallow lagoon of Heron Island, Australia. Trajectory analysis revealed that Epaulette sharks (Hemiscyllium ocellatum) displayed sinusoidal movement patterns, while Blacktip Reef Sharks (Carcharhinus melanopterus) had more linear trajectories that were similar to those of a Lemon shark (Negaprion acutidens). Individual shark trajectory patterns and movement speeds were highly variable. Results indicate that Epaulette sharks may be more mobile during diurnal low tides than previously thought. The approach presented here allows the movements and behaviours of marine vertebrates to be analysed at resolutions not previously possible without complex and expensive acoustic arrays. This method would be useful to assess the habitat use and behaviours of sharks and rays in shallow water environments, where they are most likely to interact with humans.


Introduction
Understanding animal movement requires information on foraging decisions, spatial use, and the distributions of populations [1]. While large-scale movements may be influenced by reproductive requirements or temperature [2], smaller-scale movement patterns of animals may be driven by food patches, habitat, or inter and intraspecific interactions between predator and prey, or commensal relationships [3,4]. Insight into how and why animals move within and across environments can be obtained by understanding the relationship between the environment and movement trajectories [5,6]. However, inshore aquatic areas, where the majority of marine vertebrates occur, are inherently difficult to sample and are the areas most threatened by human activities [7]. Gaining an understanding of the spatial dynamics of larger vertebrates in these coastal zones will facilitate better management of common anthropogenic threats and abatements.
interactions. In addition, the use of UAVs for tracking could allow supplemental small-scale tracking at much lower cost than other approaches, since the only cost is the initial purchase of the UAV.
This study trialled the use of UAVs in short-term tracking of marine vertebrates by tracking different species of sharks in a shallow reef lagoon environment. Each species has varied habitat use, perceived ability to move, and ecological requirements. Epaulette sharks (Hemiscyllium spp.) are small (<107 cm TL) tropical benthic cat sharks [48] that are primarily known for their extreme tolerance of hypoxic conditions [49][50][51]. While some studies suggest these sharks move very little during low tides, their movement and behaviour in reef lagoons are still poorly understood [52,53]. By contrast, Reef sharks (Carcharhinid spp.) are much larger (>1.5 m) and some of the most widely studied species of sharks [54][55][56], yet their habitat use in shallow lagoon environments is poorly understood due to the difficulties in accessing these environments. In addition, one Lemon shark was also tracked; a species that has been extensively studied in shallow environments [35,57]. We assessed whether trajectory analyses could be performed on these data, and whether patterns in trajectories could be detected within and between species of shark. These results assess the feasibility of this approach to tracking small-scale, high-accuracy movement and behaviour of marine vertebrates in shallow environments.

Drone Piloting and Tracking
Monitoring the movement patterns of sharks was conducted in April 2017 and 2018 at mid to low tides in a shallow reef habitat at Heron Island, Great Barrier Reef, Australia (23.4423 • S, 151.9148 • E). These shallow reefs are dominated by sand separated by patch reef, and are isolated from the ocean during low tides [58]. Lower tides were monitored in this study to allow easier visibility through wave breaks. Wind patterns on the tracking days ranged from 5 to 17 knots. The drone used in this study was an off-the-shelf DJI Phantom 3 Professional ® , flown in GPS mode with the DJI GO app on an iPad Mini. The stock camera carried by this drone has an f/2.8 lens with a 94 • field of view which is capable of recording 1080p footage at over 30 frames per second and to which we fitted a polarised lens to reduce reflections on the surface of the water. Flight caching (which records over 50 flight parameters, including altitude, horizontal and vertical speeds, and GPS coordinates in real time) was enabled, and the drone was manually piloted over the scientific research zone at an altitude of 15 m when searching for Epaulette sharks, and 40 m when searching for larger sharks. The UAV was flown in a "S" search pattern until a shark was spotted, then the drone was immediately lowered from 15 to 2-3 m altitude and manually hovered directly above the shark. Flown in this fashion, the GPS location of the drone was directly above the shark, allowing the recorded flight path of the drone to act as a reliable proxy for the position of the shark. A high-resolution 1080p video at 30 fps was recorded of the entire 'track' session of each shark (Figure 1), which also allowed us to determine when a track was started during analyses (record on and off are parameters stored by the UAV) and record behaviours of interest. Once a shark was spotted, the UAV was used to track it until the shark was lost (e.g., due to surface glint or uncertainty that it was hiding in a coral patch) or until the battery of the drone was at 30% (after which long-term battery life can be affected). See Table 1 for a summary of this approach. Only one shark was tracked at a time. This research was conducted under Macquarie University's animal ethics research permit ARA 2016/019. Table 1. Methodological process for drone-based animal tracking flights.
Step Process 1 Check for appropriate weather conditions (low wind <10 kph = optimal, maximum of 30 kph).
2 Take off with drone and fully-charged battery, fly with camera pointed directly down.
3 Fly S-shaped search pattern above area of interest at 15-50 m altitude, depending on size of target species. Adjust height accordingly.

4
Once target species is spotted, immediately centre animal in the dead centre of viewing screen and descend to 2-5 m altitude.

5
Start recording video to record behaviours and aid in data processing, 6 Use gentle and consistent pilot inputs to remain directly above animal in this fashion. Avoid jagged and large inputs (this is true of any remote piloting, but especially in this case).

7
Continue tracking until animal is lost or battery gets to the 30% mark (this will prevent long-term battery degradation).

8
Return drone and land.

1
• New trends toward environmentally friendly materials • New materials through nanotechnology routes MATBIM2019 will provide a platform to bring together researchers from academia and industry, as well as students, practitioners, and experts in the field to share new ideas and contribute the latest findings related to the packaging materials science and technology. The conference will feature the following: • Invited lectures from leading experts (academic and industrial) • Sessions for additional contributed presentations • Poster session • Poster awards • Exhibition area for sponsors • Gala dinner at the fascinating "Sala del Cenacolo" at the Museum of Science and Technology (MUST) • Tour at the Museum of Science and Technology (MUST) • Discounted rates for students

Measurement Accuracy
Measurement accuracy, in this case the measured coordinate error relative to the true position of the animal, should be measured when testing tracking solutions, as it can affect how results are interpreted [59]. Previous studies have found that when topographical data are not of interest (as is the case here), the horizontal accuracy of UAV inbuilt GPS data can be in the sub-10 cm range without requiring ground control points (GCPs) [60], and thus we estimated that the measurement accuracy from our UAV is in a similar range. In environments with more than 15 GPS satellites in range, this accuracy is likely to be greater still, and conversely accuracy would be lower in environments with fewer than 10 satellites in range.
While we did not quantify measurement accuracy during this study, the greatest source of measurement error associated with this method is likely to be pilot error. This would cause the UAV to deviate from its position above the tracked animal. In windy conditions, piloting skill becomes more important to produce accurate measurements. Wind-induced error could be corrected by attempting to fly over a stationary point for 30 s before a track was commenced. The movement over this time could then be used to correct movement rates for any subsequent tracking data. Quantifying manual flight accuracy is more difficult, but pilots could undergo a standardised flight path (or 'track') and Drones 2018, 2, 37 5 of 14 use error against the true path to correct for errors during subsequent flights. We would expect that measurement error would decrease over time as pilots develop their skills with this approach.

Data Preparation
Logged flight data were downloaded from the UAV using iTunes, and a csv file for each flight log was extracted using the TXTlogtoCSVtool [61], though many other freely available online tools exist that allow similar data to be extracted from the flight logs. The DJI series of UAVs record flight logs at a frequency of 10 Hz, but only GPS coordinates, altitude, time, and the status of video recording were used in this study. The 'recording on' readings in the flight logs, which indicate that video was being recorded, were used to identify tracks of interest within the flight logs, which included data from the moment the UAV was switched on. Flight altitude was then used to assess when the 'track' commenced, designated as when the UAV lowered itself to recording height (2-3 m altitude). The primary variables of interest for this research were GPS coordinates and time of data point, since those allow the extrapolation of other variables of interest. Data for all flight logs were then collated before being loaded into R software V 3.4.4 through RStudio [62].
For a detailed procedure for data processing, we have included an attachment with the R script produced by this study (Supplementary Information). On some days, flight conditions were near the upper limits of the capabilities of the DJI Phantom 3 in GPS mode (~30 km/h): This made many corrections necessary to compensate for gusts of wind, and it was difficult to maintain the drone directly above the shark. This could result in a total track distance that was greater than that of the shark or inaccuracies in the recorded track. To reduce these effects and produce a more homogenous track, 5-s localised smoothing was used on the GPS tracks. Since the DJI Phantom records data at 10 Hz, this meant tracks were smoothed over 50 data points. Smoothed tracks were calculated using the 'ave' function for creating figures of tracks with the ggmap package [63] using satellite imagery provided by Google. For trajectory analysis, GPS coordinates were transformed into metric X and Y coordinates using the 'rgdal' package and transformed into individual tracks using a processing function (Supplemental Data).

Data Analysis
The aim of these analyses was to extract quantitative trajectory metrics that could be used to characterise individual and species-specific movement patterns on small scales. The method and variables extracted here from the data are just some possibilities, and other approaches, such as state-space models or hidden Markov models could be applied here, depending on the desired metrics [64,65]. The trajr package [66] was used throughout the data analyses. Trajr provides a toolkit for trajectory analysis by including numerous metrics to quantitively describe the movement behaviour of animals. All tracks were initially smoothed using the TrajsmoothSG function to correct for inaccuracies in flight tracks. Variables available from this package extracted from each track were: Mean speed, the standard deviation of the speed, total track distance, total track duration, sinuosity, Emax (a measure of straightness), and 'pause interval'. The 'pause interval' was used to estimate the frequency per minute of pauses in movement where speed was below 0.2 m·s −1 for at least 5 s. This was used as a way of assessing rest or foraging events when the sharks were relatively immobile. The 0.2 m·s −1 cut-off was chosen as roughly a quarter of the mean movement speed for the species. Trajr can calculate the periodicity of movement patterns; however, the number of data points in the tracks presented here (>4000) caused processing bottlenecks that resulted in calculation errors of unknown magnitude, so this measure was not included. The relationships between species of shark and the variables listed above were then examined visually using a Principle Correspondence Analysis.

Results
In total, twenty-six 15-to 20-min flights were conducted searching for and tracking sharks. Fourteen sharks were successfully tracked for three to eleven minutes, including nine Epaulette sharks  Table 2). The trajectories of individual sharks were highly variable, both within and between species, with some sharks having large numbers of pauses, and others swimming continuously during the tracks (Figure 4). Principle Correspondence Analysis suggested that sinuosity, interval numbers, and Emax (straightness) were species-specific characteristics, while the standard deviation of movement speed was independent of species ( Figure 5). Three natural feeding or attempted feeding events were recorded: Two for Blacktip Reef sharks and one for Epaulette sharks. In Blacktip Reef sharks, these feeding events were associated with more sinuous and less straight trajectories (sharks 12 and 13, Table 2).

Discussion
This study demonstrates that UAV-based tracking is a prospective tool in obtaining information on the spatial and temporal movement of marine animals at high resolutions. UAVs have previously been trialled in shallow inshore areas to obtain information about distribution patterns and inter and intraspecific behaviours of large marine vertebrates [43,45]. However, the potential of UAVs to effectively track movement and record the behaviour of marine vertebrates has only been assessed using hover flight over specific structures [4]. Understanding the small-scale movement, habitat use, and fine-scale foraging behaviour of larger vertebrates can help to guide the management of these habitats, yet these data are inherently difficult to obtain. To date, information on small-scale movement of marine animals is usually acquired with acoustic tags via an array of fixed receivers [67] or active tracking of the animal with a manually driven vessel [32]. More recently, VEMCO high residence tags and receivers (HR-VPS) present positional data on aquatic animals at high temporal resolutions and fine-scales [68]. However, these methods are labour-intensive either in terms of Drones 2018, 2, 37 9 of 14 deploying and maintaining receivers or tracking animals, require invasive surgery, and are expensive due to tag (often single-use) and receiver costs. Utilising a UAV is noninvasive technique that does not require the deployment of fixed structures and is relatively cost-effective, given that the predominant expense is the one-off purchase of a drone. Furthermore, while acoustic telemetry and active tracking provide fine scale positioning and habitat utilisation, the noncontinuous timepoints at which data are obtained (generally every 1-3 min at best, Espinoza et al. [67]) fail to provide information about when animals are feeding and why an animal is altering its trajectory and speed. Being able to assess this in real time continuously, coupled with a previously georeferenced assessment of the surrounding habitat, should enable researchers to infer immediate relationships between habitat and behaviour [69]. While combining tagging with technologies such as Crittercam [70] can effectively bridge this gap, the cost and requirement of tagging the animals remains. Our study demonstrates that UAV technology is a cost-effective tool for obtaining high-resolution information on the movement and behaviour of marine vertebrates, providing utilisation data of habitats such as the reef flat. We suggest that movement studies relying on satellite and/or acoustic tags on animals visible from the surface consider the addition of UAV tracking to provide additional, continuous higher-resolution movement and behaviour data in areas of interest, or when tagging is unfeasible or undesirable.
The drone-based tracking approach demonstrated here proved to be an effective tool to track the fine-scale movements and behaviours of marine vertebrates in shallow reef environments. While the number of animals tracked here is too small to infer ecological patterns, the data provided by these analyses demonstrate the potential for this approach. Analysis of shark trajectories in our research revealed a high variation in movement patterns within and between species. Mesopelagic shark movement patterns were more linear than the more sinuous movement patterns of benthic Epaulette sharks. One broad objective was to determine whether the movement patterns of Epaulette sharks were restricted to higher tides, and if these movements were predominantly nocturnal and crepuscular, as studies have suggested [52,53]. Epaulette sharks observed during low tides were highly active, with mean movement rates of 0.38 m·s −1 and a maximum tracked distance of 222 m over roughly 10 min. Previous research had indicated that the habitat ranges of these sharks were relatively restricted (30-49 m in Heupel and Bennett [53] over 1-725 days). While our results do not provide information on residency, they suggest that Epaulette sharks are highly active within their habitat ranges. Since all our observations were conducted during daylight, our results also imply that Epaulette movement is not exclusively crepuscular and nocturnal [52]. Three natural feeding events across all tracked sharks were recorded. The approach provided in this study was a suitable means to extract high-resolution movement data of sharks in a shallow reef environment and would likely be useful in shallow beach environments where human-shark interactions frequently occur. Clearly, longer tracking trajectories and larger sample sizes are needed here to verify these behaviours, and UAV technology provides an exciting proposition for further exploring the behaviour of these marine vertebrates.

Considerations for Use of This Methodology
Drone-based animal tracking provides many benefits over conventional tracking approaches, but there are limitations in its use. Chief among those is the limited battery life of current drones, which limits the upper extent of tracking time. With the previous-generation DJI Phantom 3 Pro used here, tracking time was effectively limited to~15 min at the most in ideal conditions. The current-generation DJI Phantom 4 Pros would extend that track time to~25 min, and consistent improvements in battery and drone technology should continually increase this maximum track time. Pilot fatigue can then become a limitation, as tracking in this fashion requires constant flight inputs and concentration. Even if battery technology allows longer flights, expecting tracks of over an hour using this approach may be unfeasible with the use of one UAV and pilot. An alternate to this would be to have several drones and pilots that take turns flying over the animal. While we saw no clear patterns of change in behaviour once we positioned the UAV above the animals, direct testing examining UAV impacts on behaviour should be conducted and may vary between species, though recent studies have found that the noise effects of UAVs underwater are minor [71] and a recent review suggests small electric UAVs like the one here are unlikely to have an impact [72]. Weather conditions can also limit the use of this methodology, since wind clutter on the water's surface reduces visibility. We found wind speeds of less than 10 knots were ideal, though tracking could occur in up to 17 knot winds, and the approaches described in detail in Joyce et al.
[73] will help users to limit the environmental effects during drone flights. Finally, since drones can only record horizontal movement of animals in this manner, the vertical movements of the animals are not recorded.

Conclusions
Understanding the spatial dynamics of animal populations is a core aspect of conservation and management. While acoustic and pop-up archival tagging can provide information about movement speeds and habitat preferences at small scales, the high costs of these approaches, stress to animals from the tagging process, and lower resolution preclude their use in some scenarios. Obtaining information on the frequency of feeding events as well as the microhabitats that are being utilised can provide further information on their ecology and drive the management of these animals. This is particularly important in areas that have high human/marine vertebrate interactions, such as inshore reef flats or shallow coastal areas. The use of drone-based animal tracking should allow the expansion of new research areas and, in conjunction with acoustic and satellite tagging approaches, help to build cohesive life-histories of marine vertebrates.