Field testing a novel high residence positioning system for monitoring the fine‐scale movements of aquatic organisms

Abstract Acoustic telemetry is an important tool for studying the behaviour of aquatic organisms in the wild. VEMCO high residence (HR) tags and receivers are a recent introduction in the field of acoustic telemetry and can be paired with existing algorithms (e.g. VEMCO positioning system [VPS]) to obtain high‐resolution two‐dimensional positioning data. Here, we present results of the first documented field test of a VPS composed of HR receivers (hereafter, HR‐VPS). We performed a series of stationary and moving trials with HR tags (mean HR transmission period = 1.5 s) to evaluate the precision, accuracy and temporal capabilities of this positioning technology. In addition, we present a sample of data obtained for five European perch Perca fluviatilis implanted with HR tags (mean HR transmission period = 4 s) to illustrate how this technology can estimate the fine‐scale behaviour of aquatic animals. Accuracy and precision estimates (median [5th–95th percentile]) of HR‐VPS positions for all stationary trials were 5.6 m (4.2–10.8 m) and 0.1 m (0.02–0.07 m), respectively, and depended on the location of tags within the receiver array. In moving tests, tracks generated by HR‐VPS closely mimicked those produced by a handheld GPS held over the tag, but these differed in location by an average of ≈9 m. We found that estimates of animal speed and distance travelled for perch declined when positional data for acoustically tagged perch were thinned to mimic longer transmission periods. These data also revealed a trade‐off between capturing real nonlinear animal movements and the inclusion of positioning error. Our results suggested that HR‐VPS can provide more representative estimates of movement metrics and offer an advancement for studying fine‐scale movements of aquatic organisms, but high‐precision survey techniques may be needed to test these systems.


| INTRODUC TI ON
Recent advances in biotelemetry have revolutionized the scales at which aquatic organisms can be monitored in the wild, with data on the locations of individuals being collected more frequently and over larger geographic areas than previously possible (Baktoft et al., 2015;Binder, Holbrook, Hayden, & Krueger, 2016;Biesinger et al., 2013;Cooke et al., 2013). One area that has seen dramatic advancement has been the use of acoustic telemetry to gain accurate estimates (within a few meters) of the two-dimensional (2D) positions of aquatic organisms tagged with acoustic tags (Binder et al., 2016).
Acoustic telemetry positioning systems typically consist of several stationary receivers arranged in a regularly spaced array of near equilateral triangles or squares with overlapping detection ranges.
The positions of tagged individuals can then be calculated using the detection data collected from all receivers within the array based on time difference of arrival (TDOA) methodology (Biesinger et al., 2013;Espinoza, Farrugia, Webber, Smith, & Lowe, 2011;Smith, 2013). While the techniques and algorithms for transforming raw detection data to 2D positions are now well-established, the capability to obtain high-resolution temporal positional data over a large area (e.g. an entire lake) remains restricted due to inherent limitations of tag and receiver technologies and costs. For example, previous technologies with high temporal resolution capabilities required the use of cables to attach receivers to shore-based stations to fulfil power and clock synchronization requirements needed for a positioning system. The VEMCO positioning system (VPS) is a commonly used acoustic telemetry positioning algorithm that is based on a proprietary pulse-position modulation (PPM) coding scheme (69 and 180 kHz; VEMCO Ltd., Bedford, Nova Scotia, Canada) (Biesinger et al., 2013;Espinoza et al., 2011;Smith, 2013). Despite its regularity of use, the PPM coding scheme has some disadvantages that limit its ability to estimate high-resolution positional data to within seconds. PPM tags require a few seconds to transmit (Meckley, Holbrook, Wagner, & Binder, 2014), with a receiver having to detect all the transmission pulses, without interference from other tags, to properly decode the tag ID. For instance, if a single tag with a burst length (i.e. time to transmit the pulses) of 3 s and transmission delay (i.e. time between transmissions) of 5 s was in a VPS, the shortest possible positioning period is 8 s. However, when two or more tags are present, it is possible for transmissions to collide and not be detected, so transmission delay must be chosen to keep the collision rates at an acceptable level. Because of this, typical PPM VPS transmission periods exceed 1 min or longer. Moreover, this adverse effect of transmission interference on tag identification generally results in a positive relationship between tag transmission period and the number of tags being successfully detected by a receiver, thus limiting the temporal resolution of positioning using PPM VPS and/or the number of organisms that can be tracked in a small area.
The recent introduction of VEMCO high residence (HR) tags and receivers allows for the monitoring of aquatic animals at a temporal resolution <1 s, while maintaining traditional 180 kHz PPM technology. The HR tags emit a very short (<10 ms) transmission with its ID encoded that HR receivers can decipher, allowing for more tags to be detected with higher temporal resolution than PPM tags. As a result, when used with VPS algorithms, HR technology should allow researchers to monitor the spatial movements of many aquatic animals within a small area at high temporal resolution, with reduced collisions, thereby greatly expanding the ability to study the behaviour of many organisms simultaneously.
Here, we present results of the first documented field test of a VPS composed of HR receivers (hereafter, HR-VPS). We performed a series of stationary and moving trials with HR tags to evaluate the precision, accuracy and temporal capabilities of this technology. In addition, we present a sample of data obtained for five European perch Perca fluviatilis implanted with HR tags to illustrate how this technology can estimate the fine-scale behaviour of aquatic organisms.

| Study site
The Dubh Lochan is a natural, small (surface area = 10 ha), shallow (mean depth = 5 m), lowland freshwater lake located in Scotland, UK ( Figure 1a). The lake has a fine sediment substratum and its perimeter is surrounded by a 1-2 m boundary of macrophytes which rise from the bottom of the lake extending above the surface. During the study, the Secchi depth was on average 2 m and water temperatures, measured by sensors co-located on the acoustic receivers, which were set at various depths within the lake, averaged 10°C. The lake is closed to the public, with fishing and boating not permitted greatly reducing potential issues of surrounding noise on detection efficiency.

| Stationary and movement trials
Our test array consisted of seven underwater omnidirectional HR acoustic receivers (high residence 2 receiver 180 kHz; VEMCO Ltd.) positioned 77.1 ± 18.6 m (M ± SD) apart from each other with overlapping detection ranges (determined by a range test using our test tags, see Supporting Information) covering the northwest half of the lake (Figure 1b). Receivers were mounted pointing upwards to K E Y W O R D S acoustic telemetry, animal tracking, fine-scale positioning, fish behaviour, fish movement, high residence 2 receiver (HR2), VEMCO positioning system vertical aluminium rods, which were fixed in 20 kg buckets of concrete with two crossed steel rods at the base and deployed at the bottom the lake. Receivers and the GPS (Garmin GPS Map 60CSx; Garmin Ltd., Kansas City, Ohio, USA) used in the study were synchronized to comparable timestamps. Internal clocks of the receivers were synchronized using internally co-located transmitters ("sync tags"). The test tags used for range testing and stationary and movement trials were V5-HRs (180 kHz, VEMCO Ltd.) with a mean HR transmission period of 1.5 s (range 1-2 s).
To compare stationary position estimates of the HR-VPS with GPS-recorded locations of the test tags, we deployed three tags at haphazard locations throughout the positioning system at depths of 1-2 m (Table 1, Figure 1b). Tags were deployed separately at known locations each on a separate weighted anchor line, with a buoy extending to the surface. This procedure was repeated four times yielding 12 stationary positions for comparison. Position and deployment/recovery times were recorded for each stationary tag using the GPS (Table 1).
To compare moving tracks between the HR-VPS and the GPS, a V5-HR tag was placed on a weighted line 1-2 m below the water surface and towed using a boat outfitted with an electric motor.
Moving tests were conducted for 10 min, with five tests total and F I G U R E 1 (a) Bathymetry map of the Dubh Lochan, Scotland, UK (56.13N, −4.61W) and (b) maps indicating the locations of high residence 2 receivers making up the test array (n = 7 receivers) used in the stationary and movement trials and the full array (n = 13 receivers) used to track the movement of European perch as part of the larger study for which this equipment was tested were spread throughout the array (Table 2, Figure 2). Time and positions of the moving test were recorded every second by the GPS that was held above the tag during movement.
Following receiver recovery and downloads, raw data were processed by VEMCO into 2D positions of each test tag using hyperbolic positioning algorithms based on TDOA for each acoustic transmission detected by three or more receivers in the array (Espinoza et al., 2011;Smith, 2013). When a transmission is detected by three or more receivers, a position is calculated using every subset of three receivers, with a single position calculated on a weighted mean that favours the lowest error sensitivity (Smith, 2013). GPS records of tag position were not provided to VEMCO and were therefore compared to HR-VPS estimate positions independently. The accuracy of the GPS to record receiver and tag locations during the field tests was displayed as ±3 m, but the exact meaning of this error estimate (e.g. 50% confidence or 95% confidence) is not documented by Garmin.  (Table 1). For moving trials, accuracy was assessed as the distance between each estimated position and the time-matched recorded GPS position.
We then calculated the median, 5th and 95th percentiles and range of the accuracy and precision estimates for both stationary and moving trials (Tables 1, 2).

| Fish movement data
Following field testing of the HR-VPS, the acoustic array was ex-  GPS accuracy was ±3 m (50% CEP) during the trials. HR-VPS, high residence-VEMCO positioning system; CEP, circular error probable. a Estimated as the distance between each position estimated by HR-VPS from GPS recorded position for that trial. b Estimated as the distance between each position estimated by HR-VPS from the median position estimated by HR-VPS for that trial.
the preceding position that was less than a set time. Next, it would again recalculate all transmission periods and delete the first one that was less than that set time. This process was repeated until the new dataset contained only positions that had transmission periods that were less than the set time. For each fish/dataset, we calculated the following common movement metrics: total distance travelled, mean turning angle and mean speed. We then used linear regression to determine how estimates of the movement metrics changed as a function of transmission period.

| Stationary and movement trials
The median precision (  TA B L E 2 Summary of movement (M) trials including the duration (nearest minute), number of positions estimated by HR-VPS and the median [5th-95th percentiles] and range of accuracy estimates F I G U R E 2 Results of stationary tag trials (S1-S12), including the GPS recorded position of the tag for each trial (blue triangle), all positions estimated by the high residence-VEMCO positioning system algorithm (red dots), and the location of the receivers in the array for reference (black squares). Estimated positions are semitransparent to allow denser regions of position accumulation to opaquer. Note, S10 was placed inside of the thick macrophyte lining the edge of the lake, to which the outline of the lake corresponds   probable]), meaning that 50% of all measurements would be within a radius of 3 m and 95% CEP would be within 6 m (Misra & Enge, 2010). Moreover, because the HR-VPS analysis uses sync tag data to measure distances between neighbouring receivers, it is likely that the HR-VPS positions are more accurate in a relative sense with respect to one another, but not necessarily with respect to their actual locations on the earth measured by the handheld GPS through the triangulation of satellites. A source of error in the movement trials could be the movement of the tag under and to the side of the boat during the movement tests; thus, causing a small discrepancy in the spatial locations of the GPS and the transmitting tag, although the magnitude of error stemming from this factor is difficult to determine. Whatever the case, these results suggest that estimated position accuracy will only be as good as the accuracy of the GPS coordinates for receivers and as advancements in telemetry continue to progress more accurate surveying techniques for determining receiver deployment locations will become necessary.

| D ISCUSS I ON
Although our results demonstrate that the HR-VPS system can provide positional data at unprecedented temporal scales, there are some areas that require further study to determine the limitations of the system and the degree of advance beyond previous technologies. For example, it would be useful to test the system over a range of environmental conditions including environments with differing bathymetry, macrophyte abundance, current speeds, ice conditions and many other abiotic factors, as similar factors have also been found to impact detection efficiency for PPM acoustic technology (e.g. Huveneers et al., 2016;Steel, Coates, Hearn, & Klimley, 2014).
For example, stationary test S10 was purposefully placed within a surrounding mass of macrophytes and resulted in very few useable detections by our positioning system, indicating that the HR-VPS positioning system may not be suitable for monitoring aquatic organisms that frequently use or spend long periods of time under dense cover. Furthermore, and irrespectively of absolute position accuracy, our results show that such a high-resolution tracking system is capable of accurately reflecting complex path tortuosity (Figure 3f).
Tortuosity is an important feature of movement (Benhamou, 2004) that remains difficult to encapsulate using classical low-resolution tracking technologies, despite the immense improvements made in animal movement modelling in the last years (Hooten, King, & Langrock, 2017). Encapsulating the natural tortuosity of an animal's movements using high-resolution tracking technology can in fact contribute, to provide appropriate estimates of trajectory characteristics (i.e. bearings and speed for continuous-time models or turning angles and step length for discrete-time models) to be used, in combination with environmental covariates, to depict existing behavioural modes (also called behavioural states or processes) and switches between behavioural modes (Parton & Blackwell, 2017). The downside of increasing spatio-temporal resolution of animal tracks becomes, however, apparent when step length (distance between successive positions) becomes smaller or equivalent to positioning error, generating jagged movement tracks (see Figure 4)    Easting (m) behavioural modes and switches occurring at different spatiotemporal scales and overcoming problems related to sampling rates (Baktoft, Gjelland, Økland, & Thygesen, 2017;Baktoft et al. 2015;Calabrese, Fleming, & Gurarie, 2016;Fleming et al., 2014;Pedersen, Righton, Thygesen, Andersen, & Madsen, 2008).
The study of individual variation in behavioural and physiological traits has experienced a surge of interest over the last decade, with much of this work being done on aquatic organisms and especially fish (Burton, Killen, Armstrong, & Metcalfe, 2011;Killen, Marras, Metcalfe, McKenzie, & Domenici, 2013;Metcalfe, Van Leeuwen, & Killen, 2016). However, we still have very little understanding of the ecological consequences of inter-individual phenotypic variation due to our limited ability to measure the movements of organisms in the natural aquatic environment. Our results suggest that the HR-VPS technology should close this critical gap, finally providing accurate measures of spontaneous activity, foraging ability and habitat preferences in the wild. This will allow individual variation in traits such as metabolic rate, stress responsiveness and personality of multiple individuals in a confined area to be directly related to movement patterns (migration, foraging habits and spawning aggregations) in free-ranging animals (Baktoft et al., 2016;Laskowski et al., 2016;Treberg, Killen, MacCormack, Lamarre, & Enders, 2016).
From a conservation perspective, this technology will facilitate the study of how animal movements change in response to natural and anthropogenic change in variables such as temperature, oxygen and food availability. Furthermore, the direct behavioural responses of aquatic animals to human activities such as boat noise (Simpson et al., 2016), ecotourism (Heyman, Carr, & Lobel, 2010) and fishing pressure (Tsuboi, Morita, Klefoth, Endou, & Arlinghaus, 2015) can be more accurately assessed.

ACK N OWLED G EM ENTS
The authors thank the staff at the Scottish Center for Ecology and the F I G U R E 5 Comparison of estimates of common movement metrics calculated at various transmission periods for the movement tracks of each fish presented in Figure 4, including (a) total distance travelled, (b) mean speed and (c) mean turning angle. The transmission period of 4 s is the mean transmission period of the high residence tag implanted in the fish. Transmission period of 7, 10, 15, 30, 45 and 60 s were obtained by degrading the data from the fish data. Note that points are slightly offset on the x-axis at each transmission period to prevent overlap of data points