Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Spatial and temporal variation in positioning probability of acoustic telemetry arrays: fine-scale variability and complex interactions

Spatial and temporal variation in positioning probability of acoustic telemetry arrays:... Background: As popularity of positional acoustic telemetry systems increases, so does the need to better under- stand how they perform in real-world applications, where variation in performance can bias study conclusions. Stud- ies assessing variability in positional telemetry system performance have focused primarily on position accuracy, or comparing performance inside and outside the array. Here, we explored spatial and temporal variation in positioning probability within a 140-receiver Vemco Positioning System ( VPS) array used to monitor lake trout, Salvelinus namay- cush, spawning behavior over 23 km in Lake Huron, North America. Methods: Variability in VPS positioning probability was assessed between August and November from 2012 to 2014 using 43 stationary transmitters distributed throughout the array. Various analyses were used to relate positioning probability to number of fish transmitters in the array, wave height, and thermal stratification. We also assessed the prevalence of ‘close proximity detection interference’ (CPDI) in our array by analyzing detection probability of 35 trans- mitters on collocated receivers. Results: Positioning probability of the VPS array varied greatly over time and space. Number of fish transmitters present in the array was a significant driver of reduced positioning probability, especially during lake trout spawning period when the fish were aggregated. Relationships between positioning probability and environmental variables were complex and varied over small spatial and temporal scales. One possible confounding variable was the large range of water depth over which receivers were deployed. Another confounding factor was the high prevalence of CPDI, which decreased exponentially with water depth and was less evident when wave heights were higher than normal. Conclusions: Some variables that negatively influenced positioning can be minimized through careful planning (e.g., number of tagged fish released, transmitter power level). However, results suggested that the acoustic environ- ment was highly variable over small spatial and temporal scales in response to complex interactions between many variables. Therefore, models that predict positioning or detection efficiencies as a function of environmental variables may not be attainable in most systems. The best defense against biased study conclusions is incorporation of in situ measures of system performance that allow for retrospective analysis of array performance after a study is completed. Keywords: Vemco Positioning System, Positional telemetry, Performance, Detection probability, Close proximity detection interference, Thermal stratification, Wave height, Signal code collision *Correspondence: tr.binder@gmail.com Center for Systems Integration and Sustainability, Department of Fisheries and Wildlife, Michigan State University, Hammond Bay Biological Station, 11188 Ray Rd., Millersburg, MI 49759, USA Full list of author information is available at the end of the article © 2016 Binder et al. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/ publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Binder et al. Anim Biotelemetry (2016) 4:4 Page 2 of 15 was estimated by the array for a given transmission, [16, Background 18]). Less is known about the effects of environmental Recent advances in aquatic animal telemetry technolo- variables (e.g., thermal stratification and waves) or more gies now provide researchers with an unprecedented complex processes, such as destructive code collisions or ability to track animal movements at fine spatial and the so-called close proximity detection interference (i.e., temporal scales, and answer behavioral and ecological detection interference as a result of transmission echoes questions that were previously beyond reach. One such being heard by nearby receivers; hereafter, CPDI; [22]), advancement that has become increasingly popular over on positioning probabilities. Nonetheless, because the the last decade is use of telemetry systems to estimate two-dimensional (2D) or even three-dimensional (3D) absence of evidence (i.e., positions) in telemetry studies is positions of transmitter-implanted animals using time not necessarily evidence of absence [13], understanding difference of arrival (TDOA) of acoustic transmissions at variation in positioning probability is critical to interpret- three or more acoustic receivers [1–4]. 2D and 3D tracks ing study results. from aquatic animals have been used to study behaviors Transmitter detections are the basis of position estima- ranging from broad spatial habitat use and home ranges tion; therefore, positioning probability should be influ - [4–7] to swimming speed [2, 8] and fine-scale responses enced by many of the same variables that drive variability to environmental stimuli [9, 10]. A variety of position in detection probability in presence/absence telemetry systems (e.g., environmental noise, aquatic vegetation, ing systems exist, each with its own set of strengths and biofouling; [13, 23, 24]). However, the issue of position- weaknesses, but they can be generally reduced to two ing probability in telemetry systems is complicated by categories: (1) cabled systems that use a single receiver the fact that the contribution of a given receiver to posi- with multiple hydrophones tethered with cables and tion estimates depends not only on the performance of (2) non-cabled systems that use multiple independent that receiver, but also on the performance of receivers receivers, each with a single independent hydrophone. Cabled systems tend to be limited in size and location of around it. Moreover, because questions addressed with deployments due to need for long cables between each positional telemetry arrays are often limited to finer hydrophone and receiver [3]. Non-cabled systems offer spatial and temporal scales than those addressed with more flexibility with respect to array size and are better the presence/absence systems, studies that use position- suited for remote locations [1, 11]. However, position ing systems may be more sensitive to biases resulting processing with non-cabled systems is more complicated from variability in performance [8, 11], particularly if the than with cabled systems because of the need to account measured response variable is based on the number of for differences between receiver clocks, which drift over positions returned by the system. time due to effects of temperature and subtle manufac In this study, we assessed spatial and temporal variabil- ity in positioning probability of a large acoustic telemetry turing differences. Nonetheless, non-cabled systems have positioning system (Vemco Positioning System; hereafter become increasingly popular due to ease of deployment VPS, Vemco Inc., Halifax, NS Canada) over three con- and flexibility to accommodate project designs and large secutive seasonal deployments. At the time of writing, study areas. this positional telemetry array was the largest ever con- Similar to the presence/absence telemetry systems that structed, consisting of 140 autonomous receivers and 43 provide coarse-scale behavioral data, positional telem- stationary transmitters with a spatial coverage of approx- etry systems are subject to performance variability [12], imately 23 km . Our specific objectives were: (1) to quan - which can complicate interpretation of animal tracks and tify the degree of spatial (<0.5  km ) and temporal (6  h) possibly bias study conclusions [8, 13]. Studies describ- variation in positioning probability that occurred over ing variation in detection probability of presence/absence three seasonal deployments between 2012 and 2014 and telemetry systems are abundant (reviewed in [12, 14]), (2) to determine whether variation in positioning proba- but perhaps due to the relative novelty of positional bility could be predicted by environmental variables (e.g., telemetry and also possibly a disconnect between the surface waves and water temperature) and other site-spe- end user (researcher) and the position estimation pro- cific variables such as signal code collisions and CPDI. cess, few papers have assessed spatial and temporal per- formance variability of positional telemetry systems [11, Methods 15–18]. The primary focus of most positional telemetry Study site and the Vemco Positioning System performance studies has been position accuracy. Position Spatial and temporal variation in performance of the accuracy has been well established as largely a function VPS (Vemco Inc.; Halifax, NS Canada) was assessed in a of the geometry of receivers relative to transmitters [1, large acoustic telemetry array designed to study spawning 11, 17, 19–21] and varies little in comparison with posi- behaviors of lake trout (Salvelinus namaycush) in northern tioning probability (i.e., the probability that a position Binder et al. Anim Biotelemetry (2016) 4:4 Page 3 of 15 Lake Huron, North America. The array, which consisted height and direction every 5  min for the duration of the of 140 VR2W-69 kHz autonomous receivers (Vemco Inc.; study. Halifax, NS Canada) and 43 stationary V16-6H transmit- At the end of each study season, receivers were ters (Vemco Inc.; Halifax, NS Canada), covered an area of retrieved and downloaded, and data files were sent to approximately 23  km and was deployed for between 87 Vemco for processing using their proprietary hyperbolic and 100 days each year from August to November, 2012– positioning algorithms [25]. Position estimates of trans- 2014. Stationary transmitters (hereafter, ‘sync tags’) with mitters were based on TDOA of each transmission at a known locations were deployed primarily to synchronize minimum of three and a maximum of six (limit set by clocks among receivers and secondarily to evaluate array manufacturer) receivers with synchronized clocks. When performance. Each sync tag transmitted a unique ID code a transmission was detected on more than six receiv- every 500–700 s (nominal delay = 600 s), with each time ers, positions were estimated using data from the first between transmissions (delay) being drawn from a uni- six receivers that detected a transmission based on lin- form distribution. The site encompassed several shoal and ear time-corrected detections. Hypothetically, the first reef areas with complex bathymetric features (Fig.  1) and six receivers that detected a transmission should rep- depths ranging from ~2 m to over 38 m. Vertical tempera- resent the six closest receivers to the transmitter, but in ture profiles in the study site were monitored using two practice that may not be true because of nonlinear drift lines of four temperature loggers (HOBO Water Temp of receiver clocks. The VPS returned a weight-averaged Pro v2; model U22-001; Onset Computer Corporation, position among all combinations of three receivers that Bourne MA) that measured temperature to a resolution detected each transmission, as well as position precision of 0.02 °C, with an accuracy of ±0.21 °C. A weather buoy estimates (horizontal position error; abbreviated ‘HPE’) (Tidas 900 buoy; S2 Yachts, Holland MI) was also moored that described the relative error sensitivity of each calcu- within the array. The buoy logged air temperature, surface lated position [25]. water temperature, wind speed and direction, and wave Spatial and temporal variability in positioning probability Spatial and temporal variation in VPS array performance in each study year was assessed using positioning prob- ability of 43 stationary sync tags distributed throughout the array (Fig.  1). Temporal variation in array perfor- mance was estimated by comparing performance metrics across 6-h time bins. Use of 6-h bins represented a com- promise between having enough transmissions (36 on average) to accurately estimate a positioning probability and being a short enough time interval to reflect envi - ronmental variability at an ecologically relevant scale. Probability of positioning each transmitter during each 6-h bin was calculated based on the ratio of observed −1 to expected positions (6 positions  h   ×  6  h  =  36 posi- tions expected). Subsequently, for each 6-h time bin, we used 2D cubic spline interpolation (R package ‘akima’; [26]) to estimate and visualize variation in positioning performance across the array. Mean array positioning probability during each 6-h bin was calculated from the 2D interpolations by taking the mean of all interpolated data points (spatial resolution of the interpolations was Fig. 1 Bathymetric map (in m) of the Drummond Island acoustic approximately 34  m ). Mean positioning probability of telemetry study site. A 140-receiver Vemco Positioning System ( VPS) the transmitters themselves was not used because trans- was deployed at the site during late summer and autumn each year mitters were not equally spaced in the array, which meant between 2012 and 2014. Symbols: cross = acoustic receiver ( VR2W ), blue circle = sync tag ( V16-6H), yellow square = temperature line some transmitters (particularly in the deep, less complex (containing 4 suspended HOBO temperature loggers), and orange areas of the array) represented a greater area of the array square = Tidas 900 weather buoy. Site S003, representing a specific than others. Due the random nature of sync tag transmis- sync tag referred to in the text, was labeled and located in the sions (i.e., uniform distribution between 500 and 700  s), southmost site in the array. Inset Location of the Drummond Island positioning probability estimates based on the mean of study site (red square) in Lake Huron, North America. Square indicates location only and is not to scale 36 transmissions per 6-h bin were subject to random Binder et al. Anim Biotelemetry (2016) 4:4 Page 4 of 15 error. Simulated transmission histories for 10,000 sync relative close proximity to a receiver are interrupted tags revealed a range of 34–37 transmissions during a by strong echoes off reflective surfaces [22], in essence, single 6-h period, indicating a maximum of 6  % random causing the signal to collide with itself, and thus, not error in our positioning probability estimates. be properly decoded and logged on that receiver. We A generalized linear mixed-effect model with binomial searched for evidence of CPDI in our array using data error distribution (R Package ‘lme4’; [27]) was used to from 35 sync tags with collocated receivers (i.e., a model within- and between-sync tag variation in posi- receiver on the same mooring as the transmitter; Fig. 1). tioning probability against variables that could affect Prevalence of CPDI in our array was described by cal- detection of acoustic transmitters. Fixed effects in the culating the detection probability of each transmitter on model included the number of unique fish transmitters collocated receivers during each 6-h bin. As with posi- within detection range of the closest receiver (‘Unique- tioning probability, detection probability was calculated Fish’), wave height measured at the within-array weather as the ratio of observed detections to expected detec- buoy (‘WaveHt’), and degree of thermal stratification in tions (36 expected detections per 6-h time bin). Inter- the water column (i.e., difference between near-surface pretation of inter-site variability in detection probability and near-substrate temperature in the array, as measured was complicated by spatial variation in detection prob- by HOBO temperature logger lines; ‘DiffTemp’), as well as ability related to varying local environmental conditions their interactions. Sync tag site ID was included as a ran- (e.g., number of fish transmitters within detection range dom effect to account for inherent differences in position - of a receiver). To account for this variation, we stand- ing probability related to location of each transmitter in ardized detection probabilities on collocated receivers the array. to that on the non-collocated receiver with maximum The mixed-effect model revealed complex interactions detection probability for each transmitter. The new between variables, so rather than attempting to build a response variable, ‘relative detection probability,’ was the global model to describe variation in positioning prob- ratio of detection probability at the collocated receiver ability of our sync tags, we fit separate logistic regres - and detection probability for that transmitter at the sion models (R Package ‘stats’; [28]) for each transmitter non-collocated receiver (i.e., receiver not on the same site and examined spatial and temporal patterns in effect mooring as the transmitter) with maximum detection size (i.e., parameter estimates from logistic regressions) probability. The relationship between relative detection of three fixed-effect variables (i.e., UniqueFish, WaveHt, probability of transmitters and water depth was mod- and DiffTemp). Preliminary inspection of the data sug - eled using nonlinear, least squares regression (R Package gested that the relative influence of the variable Unique - ‘stats’). Fish on variation in array performance changed markedly between the lake trout pre-spawning and spawning peri- Results ods, and therefore, the two time periods were analyzed Temporal and spatial variability in positioning probability separately. Spawning period start dates were the same VPS array performance varied greatly within and among as were used previously at this site by Binder et  al. [29], years. Mean, array-wide 6-h positioning probability who used changepoint analysis to determine when male ranged from 8 to 90 % in 2012, from 38 to 92 % in 2013, lake trout implanted with pressure-sensing transmitters and from 59 to 95  % in 2014 (Fig.  2). The greater range moved from deep offshore water onto the shallow-water in VPS positioning performance in 2012 compared with spawning shoals. During each time period, parameter 2013 and 2014 was a result of extremely poor array per- estimates were inspected for obvious trends by map- formance during mid-to-late October and November ping them on bathymetry. We then tested for significant in 2012 (Fig.  2). Upon investigation, poor performance relationships between parameter estimates and trans- during this period was caused primarily by widespread mitter depth using linear regression (R Package ‘stats’). receiver memory saturation, which caused log files con - Where scatter plots revealed a nonlinear relationship, taining millisecond data required for positioning trans- segmented regression (i.e., broken-stick regression; R mitters to be overwritten by detection data, which at the Package ‘segmented’; [30]) was used in place of linear time was logged only to the nearest second (this issue regression. Comparisons between pre-spawning and was addressed in receiver firmware updates). The first spawning period parameter estimates were made using instance of this issue occurred on 06 October, and by paired t tests (R Package ‘stats’). the end of the season, 81 of 140 (59 %) receivers reached memory capacity and stopped logging millisecond data. Close proximity detection interference For this reason, 2012 VPS data collected on or after 06 CPDI is a phenomenon in acoustic telemetry whereby October were excluded from subsequent statistical transmission sequences of a transmitter located in analyses. Binder et al. Anim Biotelemetry (2016) 4:4 Page 5 of 15 Interestingly, positioning probability at some of the deeper receivers on the southwest side of the array (in particular S003; southernmost transmitter on the west side of array) consistently differed from other transmit - ter sites in the array. During the August dates, when mean whole-array positioning probability was relatively high, positioning probability at these sites was generally low (often less than 30  % of transmissions positioned; Fig.  3). Conversely, during the October and November dates when mean whole-array performance decreased, positioning probability at these sites tended to improve (see Additional file  1). Poor performance of these south- west receivers during summer months appeared to be related to thermal stratification in the water column, with positioning probability, particularly at site S003 (Fig.  1), improving during brief periods when little thermal strati- fication occurred in the array (Fig.  4). The negative rela - tionship between positioning performance and thermal stratification was also evident in among-year compari - sons, where performance at site S003 was greater in 2014, the year with the least thermal stratification, than in the previous 2  years (Fig.  4). Toward the end of each year when the thermal structure in the array became more homogenous, variation in positioning performance at this site was more closely related to the number of tagged fish within detection range of the site (Fig. 4). Attempts to develop a meaningful global model relating sync tag positioning probability to local environmental variables such as number of tagged fish present (Unique - Fish), wave height (WaveHt), and thermal stratification Fig. 2 Mean estimated Vemco Positioning System ( VPS) positioning (TempDiff ) revealed complex relationships and interac - probability (black line) in relation to wave height (red line) recorded tions that proved impossible to satisfactorily model using by the on-site weather buoy and the median number of unique fish transmitters (blue line) detected per receiver during each year of the the predictor variables available. As expected, positioning study. Mean positioning probability was calculated from 2D interpo- performance was negatively correlated with the number lations based on 43 sync tags distributed throughout the array (see of fish present within detection range across all receiv - Fig. 1). Each point on the graphs represents a 6-h time window ers during both pre-spawning and spawning periods (Fig.  5a, d). However, relationships between positioning probability and both wave height and thermal stratifica - Clear seasonal trends were evident in all three years, tion were site-specific. For example, at some sites posi - with better performance in August and September than tioning probability was negatively correlated with wave in October and November (Fig.  2). Within the array, a height, while at others a positive correlation occurred great deal of spatial variation occurred in positioning between positioning probability and wave height (Fig. 5b, probability. Within some 6-h time periods, position- e). Similar relationships were observed when comparing ing probability of sync tags in some parts of the array positioning probability against degree of thermal strati- was perfect (or near-perfect), while in other parts of the fication (Fig.  5c, f ). Moreover, at several sites, direction array, it was zero (Fig.  3). In general, positioning prob- of observed relationships was reversed during the pre- ability was poorer on the west side of the array (where spawning and spawning periods. the main lake trout spawning sites were located) than the A portion of the inter-receiver variability in relation- east side of the array, particularly in the autumn during ships between environmental conditions and positioning the lake trout spawning period. Nonetheless, all sync tags performance was related to the wide range of depths over displayed perfect positioning probability during at least which our array was deployed. However, relationships some periods, suggesting that spatial patterns were not between parameter estimates (i.e., estimates of effect a result of irregularities in array design (e.g., geometry, size) from transmitter-specific logistic regressions and specific receiver locations). Binder et al. Anim Biotelemetry (2016) 4:4 Page 6 of 15 Fig. 3 Estimated array-wide positioning probability (% positioning) of the Drummond Island Vemco Positioning System ( VPS) during three arbitrary 6-h time periods in 2012 and 2014. Estimates of positioning probability were interpolated based on 43 sync tags distributed throughout the array (see Fig. 1). A high degree of spatial and temporal variability occurred in positioning probability of the Drummond Island VPS array. See Additional file 1 for complete records for all 3 years of the study Binder et al. Anim Biotelemetry (2016) 4:4 Page 7 of 15 Fig. 4 Positioning probability (black line) of sync tag S003, the southernmost transmitter in the Drummond Island Vemco Positioning System ( VPS) array (see Fig. 1), relative to the median number of unique fish transmitters detected on receivers with line of sight and within 500 m of S003 and thermal stratification (bottom graph in each panel). Temperature profiles were interpolated from two separate temperature lines deployed in the array. Poor positioning probability at the end of the 2012 season was due to widespread receiver memory saturation, which began on 06 October. Each point on the graphs represents a 6-h time window. Temperatures are in °C depth of the transmitter were not always linear and dif- relationship between WaveHt and positioning probabil- fered between pre-spawning and spawning periods. Dur- ity) were generally negative for shallow transmitters and ing the pre-spawning period, when few fish were present positive for deep transmitters (Fig.  6a). A segmented in the array (Fig.  2), WaveHt parameter estimates (i.e., regression indicated a positive relationship between Binder et al. Anim Biotelemetry (2016) 4:4 Page 8 of 15 Fig. 5 Results of logistic regression analysis relating positioning probability of each sync tag (n = 43) to the number of fish transmitters detected on the closest receiver (‘UniqueFish’; a, d), wave height (‘WaveHt’; b, e), and difference between near-surface and near-substrate temperature in the array (‘DiffTemp’; c, f). Data from the lake trout spawning (right column) and pre-spawning (left column) periods are displayed separately. Red symbols indicate a significant negative relationship, blue symbols a significant positive relationship, and white symbols no significant relationship (α = 0.05) WaveHt parameter estimate and transmitter depth (seg- (segmented regression; t  =  −0.359, df  =  31, p  =  0.722). mented regression; t = 3.351, df = 31, p = 0.002) down to TempDiff parameter estimates were also positively cor - approximately 13.9 ± 2.4 m, after which WaveHt param- related with transmitter depth during the pre-spawning eter estimates remained relatively consistent across depth period (linear regression; t  =  3.723, df  =  33, p  <  0.001), Binder et al. Anim Biotelemetry (2016) 4:4 Page 9 of 15 Fig. 6 Relationship between logistic regression parameter estimates from positioning probability models (‘WaveHt’ = wave height, ‘Temp- Diff ’ = difference between near-surface and near-substrate water temperature) and water depth at 43 sync tag sites. Data from the lake trout spawning (right column) and pre-spawning (left column) periods were analyzed separately. Lines in b, c, and d depict results of linear regression analysis, while lines in a depict results of segmented regression. Significant relationships occurred between parameter estimates and water depth during the pre-spawning period, but not during the spawning period (α = 0.05) with negative parameter estimates predominant at shal- always negative) and did not vary with transmitter depth low depths and positive parameter estimates predomi- (linear regression; pre-spawning: t  =  1.471, df  =  33, nant at deeper depths (Fig.  6b). In contrast, during the p  =  0.151; spawning: t  =  −0.865, df  =  33, p  =  0.393). spawning period, neither WaveHt (linear regression; Nonetheless, UniqueFish parameter estimates were on t  =  0.673, df  =  33, p  =  0.506) nor TempDiff (linear average 2.47 (±1.33) times greater during the spawning regression; t = 0.283, df = 33, p = 0.779) parameter esti- period than during the pre-spawning period (paired t mates were significantly correlated with transmitter test; t  =  7.344, df  =  34, p  <  0.001), indicating a greater depth (Fig. 6c, d). negative relationship with positioning probability during Some of the discrepancy between the pre-spawning and the spawning period than non-spawning period. In fact, spawning period is likely due to the fact that the presence number of fish present in the array appeared to be a pre - of lake trout transmitters within the array (‘UniqueFish’) dominant factor driving whole-array positioning prob- had a far greater influence on positioning probability dur - ability during the spawning period (Fig. 2). ing the spawning period (when far more fish were pre - sent in the array) than during the pre-spawning period. Close proximity detection interference UniqueFish parameter estimates were negative during One possible confounding factor in our analysis of posi- both pre-spawning and spawning period (i.e., relation- tioning probability was CPDI [22]. Thirty-five of 43 ship between UniqueFish and positioning probability was sync tags in our array were collocated with an acoustic Binder et al. Anim Biotelemetry (2016) 4:4 Page 10 of 15 Fig. 7 Boxplots displaying median (center line), 25th and 75th quartiles (bounding box), and range (dashed line) of detection probabilities on col- located receivers for 35 transmitters. Each panel shows results for a different year of the study (2012–2014). Individual detection probabilities were calculated based on 6-h time windows receiver. Examination of detection probability of sync collocated transmitters was strongly correlated across tags on collocated receivers revealed that this phenom- all three years (Pearson’s r  =  0.78, 0.75, and 0.85 for enon was widespread in our array (Fig.  7). Collocated 2012 vs. 2013, 2012 vs. 2014, and 2013 vs. 2014, respec- receivers never detected sync tags as well as other nearby tively, p  <  0.001 for all comparisons), which suggests surrounding receivers with line of sight to the transmitter that CPDI was location-specific and probably related to (i.e., maximum detection probability for a given transmit- physical characteristics of the immediate environment ter always occurred on a non-collocated receiver). Mean that changed little over time (i.e., lake bottom compo- maximum detection probability for each transmitter on sition and topography). Inter-site variation in mean nearby non-collocated receivers ranged from 0.67 to 0.91. relative detection probability was related negatively In contrast, mean detection probability on collocated to water depth at the site and was best approximated receivers averaged only 12  % of maximum detection with a decreasing exponential curve (relative detection −0.892 probability, ranging from 0.01 to 0.37 over the 3 years of probability  =  1.098  ×  depth , p  <  0.001 for both study. parameters; Fig.  8a), indicating that CPDI increased A high degree of intra-annual variability occurred exponentially with increased water depth. in detection probability of transmitters on collocated Within-site variability in relative detection probabil- receivers, both within and between sites (Fig.  7). How- ity of transmitters on collocated receivers was related ever, mean relative detection probability of individual to wave height (linear mixed-effect model, t  =  144.432, Binder et al. Anim Biotelemetry (2016) 4:4 Page 11 of 15 df = 36,010, p < 0.001; Fig. 8b). In general, mean relative detection probability of collocated transmitters was lower than average when wave height was less than approxi- mately 1.25 m and greater than average when wave height was greater than 1.25 m (Fig. 8b). No significant relation - ship existed between maximum detection probability and wave height (linear mixed-effect model, t  = −1.021, df = 36,010, p = 0.307), so the above relationship was not an artifact of decreased maximum detection probability at higher wave heights; mean maximum detection proba- bility of sync tags was greater than or equal to 0.77 (range 0.77–0.89) at all wave heights. Discussion Substantial spatial and temporal variability occurred in positioning probability of sync tags in our positional telemetry array. While spatial variation in positioning probability has been noted in other studies [15, 16], most studies have focused on receiver geometry and ignored temporal variability [16–18]. This trend may be due to the fact that most assessment studies were short in dura- tion [1, 15, 17], or used positional arrays with relatively small spatial coverage [1, 11, 15, 18]. The long duration, high environmental complexity, and large spatial extent of our VPS deployment provided a unique opportunity to explore and quantify within-array variability in posi- tioning probability. Some of the variability we observed could be explained by variables we measured in the field, but much of it could not, which highlights the complex nature of acoustics in natural systems. Some of the nega- tive relationships we observed could be controlled (e.g., number of tagged fish in the array; transmitter power level), or at the least minimized by careful planning and study design. Signal code collisions resulting from large numbers of fish transmitters in the array at the same time were a significant cause of decreased positioning probability in our system, particularly during the spawning period when they were the main driver of variation in array Fig. 8 a Relationship between mean relative detection probability performance. A signal code collision occurs when trans- of sync tags on collocated receivers and water depth at each loca- missions from two or more transmitters are detected tion. Data were normalized to the receiver with maximum detection simultaneously on the same receiver [31], preventing the probability for each transmitter. Individual points were color-coded receiver from properly decoding either signal. Probability by year. Data were best approximated with a declining exponential −0.892 of code collisions is a function of the number of trans- curve (relative detection probability = 1.098 × Depth ). b Dif- ference from mean relative detection probability of sync tags on col- mitters within range of a receiver, the duration of the located receivers across wave heights. Boxplots display median (center code signal, and the period between transmissions [23]. line), 25th and 75th quartiles (bounding box), and range (dashed line) The duration of a transmission, and thus the degree of of differences across 35 collocated sync tags during the three study susceptibility of a positioning system to signal code col- years. Detection probability on collocated receivers was generally lisions, is dependent on the coding scheme used, which lower than average when wave height was less than approximately 1.25 m and greater than average when wave height was greater than varies by manufacturer. With Vemco’s current coding 1.25 m scheme, the transmission duration is relatively long (up Binder et al. Anim Biotelemetry (2016) 4:4 Page 12 of 15 to ~5 s; [32]); therefore, depending on the nominal delay particularly if researchers are new to a study site, or the of the transmitters (i.e., period between successive trans- behavioral ecology of the animal is not well understood. missions), the probability of collisions can be quite high As has been observed in other positional telemetry at relatively low transmitter densities (see Fig. 9). studies [16, 18], positioning probability of sync tags at Signal code collisions have a high potential for creating our site was influenced by environmental variability. spatial and temporal bias in telemetry studies because the However, relationships between environmental vari- presence of the study animals themselves alters the per- ables and positioning probability in our array were loca- formance of the telemetry system. Therefore, investiga - tion- and time-specific. Steel et al. [18] reported variation tors should assess the potential for, and implications of, in the relative influence of environmental parameters signal code collisions (particularly if they are expected on positioning probability across three study systems to be heterogeneously distributed) both during the study (coastal, estuarine, and riverine), but our results indi- design stage and while interpreting results. In our case, cated that this variation can occur over relatively small the high prevalence of signal collisions in our system spatial scales. Our observations have important implica- during the spawning period was due to several factors, tions for acoustic telemetry studies in general. First, fine- including: (1) underestimation of lake trout annual mor- scale variation in response to environmental variables tality rate and spawning site fidelity [29], which caused indicated that the nature of these relationships was com- high numbers of transmitters to return to the system plex. Therefore, while general rules of thumb regarding during each year of the study (390 tagged trout released the effect of certain environmental variables on telemetry between 2010 and 2011), (2) high levels of aggregation system performance may be evident, development of a at relatively few spawning site locations [33], and (3) universally applicable predictive model [34] using easy- use of high-powered tags (V16-6H, 158  dB, 90  s nomi- to-measure variables is likely beyond reach prior to con- nal delay), which at times had detection ranges of several ducting a study. Second, because acoustic properties can kilometers. Based on our experience, we recommend vary over small distances, true optimization of an acous- adopting a conservative approach in study designs when tic telemetry system at some sites may require fine-scale determining how many transmitters to release in a study, range testing and development of spatially heterogeneous array designs (e.g., differential receiver spacing across an array). Water depth played an important role in the perfor- mance of our VPS array, particularly with respect to how the system responded to changes in wave height and thermal stratification. In general, shallow transmitters tended to be negatively affected by increased wave height and thermal stratification, while deeper transmitters often saw a boost in positioning probability under the same conditions. As far as we are aware, ours is the first study to report a positive relationship between position- ing probability and wave height or thermal stratification; however, we are doubtful that the increase in positioning probability is directly related to these variables. Rather, because both of these variables tend to reduce the dis- tance over which an acoustic transmitter is detected [23, 35], we hypothesize that the positive relationships are due to reduced transmission echoes and a consequent reduction in CPDI [22], which was most pronounced in deep water. This interpretation was supported by the Fig. 9 Estimated probability of a signal code collision (using Vemco’s observation that detection probability of transmitters on global coding scheme) based on the number of transmitters within collocated receivers increased as wave heights increased. detection range of a receiver. Collision probabilities were calculated Interestingly, depth effects were only observed during by simulating transmission histories (n = 10,000 transmissions) for the pre-spawning period. We attribute this to a masking between 1 and 50 transmitters assuming transmission duration of 5.12 s. Transmitters were programmed to transmit at random intervals effect by signal code collisions, which were more preva - between ±50 % of nominal delay. R code for the simulation is pro- lent during the spawning period than during the pre- vided as Additional file 2 spawning period. Binder et al. Anim Biotelemetry (2016) 4:4 Page 13 of 15 The high incidence of CPDI in our VPS array was likely or placing current receivers higher in the water column) due to the combined effect of our use of high-powered to improve performance in subsequent years. transmitters and a highly reflective acoustic environ - ment [22] in which substrate was dominated by the hard Conclusions surfaces of cobble, boulders, and bedrock. Although The acoustic telemetry community is constantly expand - evidence of CPDI occurred for all collocated receiver/ ing as new researchers adopt these technologies in their transmitter pairs, the prevalence of CPDI increased studies. Therefore, identification and discussion of issues exponentially with receiver depth, a characteristic we related to study design and data interpretation is valu- attributed to an exponential decline in ambient noise able to the community because it helps to improve the from surface sources (e.g., wind and waves). Preva- overall quality of data coming out of acoustic telemetry lence of CPDI is dependent on the coding scheme [21, studies, as well as their interpretation. Positional acoustic 22, 32] and, therefore, is likely to vary among telemetry telemetry, and more specifically VPS, is a relatively new systems. Thus, researchers should assess the potential technique for tracking movements of aquatic animals; for this phenomenon when designing their study and thus, a need exists to better understand how these sys- choose transmitter specifications and receiver place - tems perform in real-world applications. Only a few pub- ments that minimize the effect with the equipment they lished papers are available describing spatial or temporal plan to use. Without sufficient receiver overlap (i.e., variation in VPS performance, fewer still while the VPS redundancy), CPDI can create areas of low detection/ system was used to track animals, and none that have position probability in receiver arrays and curtains [22]. spanned the spatial scale and range of environmental Moreover, CPDI also has the potential to reduce accu- conditions that occurred in our study. Some of the vari- racy of positions returned from VPS arrays. The reason ables we identified as negatively influencing positioning for this inaccuracy is that position estimation will depend probability (e.g., signal code collisions) can be minimized on detections at receivers that are further away from the through careful planning, but our results also suggest transmitter, which could result in poor receiver geom- that the acoustic environment can be highly variable over etry, imprecise signal arrival time estimates, and conse- relatively small spatial and temporal scales, which if unac- quently higher horizontal position error [1, 18]. counted for, have the potential to bias study conclusions. Receiver memory saturation (which caused raw log files At complex study sites (e.g., sites with highly variable to be overwritten by detection data) late in 2012 caused depth, substrate types, or water chemistry), fine-scale a drastic decrease in whole-array positioning probability. range testing may allow researchers to optimize receiver At the time, no mechanism was in place to flag this issue array design; however, we acknowledge that such intense during either receiver download or position processing, pre-study testing is rarely practical. In most cases, the and because it occurred during the peak of lake trout best defense against making biased study conclusions due spawning, poor array performance was assumed to be to spatial and temporal changes in system performance due to signal code collisions. Indeed, it was not until we will be to incorporate methods for measuring that vari- started work on this manuscript that the memory issue ability into the study design. For most, that will involve was identified, which highlights the need for research - deploying stationary transmitters throughout the study ers to take responsibility for quality checking results site for the duration of the study, with the number and returned to them from positioning software. At a mini- location of these transmitters depending on the questions mum, we recommend researchers create visual displays being addressed and the complexity of the study site. (e.g., graphs or maps, e.g., Fig. 4) for assessing positioning At a minimum, this approach will allow researchers to performance over space and time. Unexplained changes interpret results in the context of array performance. In in performance should be investigated thoroughly to rule some cases, results may be standardized by incorporating out equipment failure or processing errors. Early assess- measures of system performance variability into analyses, ment of positioning performance, especially if assessed either through use of correction factors [36], or through from a pilot study, may also provide an opportunity to development of sophisticated statistical analyses that correct deficiencies in array design. For example, poor explicitly correct parameter estimates based on imperfect positioning probability on the southwest side of our array detection and positioning probabilities [37, 38]. during summer (top panels in Fig. 3) was likely due to the fact that transmissions on top of the reef had to transmit Availability of supporting data through the thermocline to reach surrounding receivers. The data supporting the results of this article are stored Had we identified this issue in 2012, changes could have in the Great Lakes Acoustic Telemetry Observation Sys- been made to the array (e.g., addition of more receivers tem (GLATOS) database (http://data.glos.us/glatos). Binder et al. Anim Biotelemetry (2016) 4:4 Page 14 of 15 4. Andrews KS, Tolimieri N, Williams GD, Samhouri JF, Harvey CJ, Levin PS. Data availability is subject to data sharing policies cur- Comparison of fine-scale acoustic monitoring systems using home range rently under development by GLATOS, the Great Lakes size of a demersal fish. Mar Biol. 2011;158:2377–87. Fishery Commission, and the United States Geological 5. Coates JH, Hovel KA, Butler JL, Klimley AP, Morgan SG. Movement and home range of pink abalone Haliotis corrugate: implications for restora- Survey. tion and population recovery. Mar Ecol Prog Ser. 2013;486:189–201. 6. Espinoza M, Farrugia TJ, Lowe CG. Habitat use, movements and site Additional files fidelity of the gray smooth-hound shark (Mustelus californicus Gill 1863) in a newly restored southern California estuary. J Exp Mar Biol Ecol. 2011;401:63–74. Additional file 1. Video showing spatial and temporal variation in posi- 7. Reubens JT, Pasotti F, Degraer S, Vincx M. Residency, site fidelity and habi- tioning performance of a 140-receiver Vemco Positioning System ( VPS) tat use of Atlantic cod (Gadus morhua) at an offshore wind farm using array over three lake trout spawning seasons. Estimates of positioning acoustic telemetry. Mar Environ Res. 2013;90:128–35. probability were interpolated based on performance of 43 stationary tags 8. Løkkeborg S, Fernö A, Jørgensen T. Eec ff t of position-fixing interval on (indicated by + symbol) scattered throughout the array. estimated swimming speed and movement pattern of fish tracked with a Additional file 2. R script for estimating, based on simulation, the proba- stationary positioning system. Hydrobiologia. 2002;483:259–64. biliity of a signal code collision based on the nominal delay, transmission 9. McMahan MD, Brady DC, Cowan DF, Grabowski JH, Sherwood GD. Using duration, and number of tags within detection range of a receiver. acoustic telemetry to observe the effects of a groundfish predator (Atlan- tic cod, Gadus morhua) on movement of the American lobster (Homarus americanus). Can J Fish Aquat Sci. 2013;70:1625–34. Authors’ contributions 10. Lynch BR, Rochette R. Circatidal rhythm of free-roaming sub-tidal green TRB led all aspects of study design, executed the field portion of the study, crabs, Carcinus maenas, revealed by radio-acoustic positional telemetry. and drafted the manuscript. TAH, CMH, and CCK participated in the design of Crustaceana. 2007;80:345–55. the study, provided consultation on statistical analyses, and helped to draft 11. Biesinger Z, Bolker BM, Marcinek D, Grothues TM, Dobarro JA, Lind- the manuscript. All authors read and approved the final manuscript. berg WJ. Testing an autonomous acoustic telemetry positioning system for fine-scale space use in marine animals. J Exp Mar Biol Ecol. Author details 2013;448:46–56. Center for Systems Integration and Sustainability, Department of Fisheries 12. Melnychuk MC: Detection efficiency in telemetry studies: definitions and and Wildlife, Michigan State University, Hammond Bay Biological Station, evaluation methods. In: Adams NS, Beeman JW, Eiler JH, editors. Telem- 11188 Ray Rd., Millersburg, MI 49759, USA. U.S. Geological Survey, Great Lakes etry techniques: a user guide for fisheries research. American Fisheries Science Center, Hammond Bay Biological Station, 11188 Ray Rd., Millersburg, Society, Bethesda, Maryland; 2012. p. 339–57. MI 49759, USA. Center for Systems Integration and Sustainability, Depart- 13. Payne N, Gillanders B, Webber D, Semmens J. Interpreting diel activity ment of Fisheries and Wildlife, Michigan State University, 1405 South Harrison patterns from acoustic telemetry: the need for controls. Mar Ecol Prog Road, 115 Manly Miles Building, East Lansing, MI 48823, USA. Ser. 2010;419:295–301. 14. Kessel ST, Cooke SJ, Heupel MR, Hussey NE, Simpfendorfer CA, Vagle S, Acknowledgements Fisk AT. A review of detection range testing in aquatic passive acoustic Thanks to F. Smith, S. Smedbol, A. Hillard, and J. MacAulay for processing the telemetry studies. Rev Fish Biol Fish. 2014;24:199–218. positions and for their helpful comments on the manuscript. Thanks also to 15. Baktoft H, Zajicek P, Klefoth T, Svendsen JC, Jacobsen L, Pedersen MW, R. Bergstedt, H. Thompson, E. Larson, C. Wright, J. Van Een, M. Lance ff wicz, L. March Morla D, Skov C, Nakayama S, Arlinghaus R. Performance assess- Lesmeister, D. Operhall, B. Lamoreux, S. Farha, Z. Wickert, J. Osga, Z. Holmes, ment of two whole-lake acoustic positional telemetry systems—Is reality S. Miehls, S. Seegert, J. Hinderer, R. Darnton, K. Smith, and P. Wigren for their mining of free-ranging aquatic animals technologically possible? PLoS assistance in the field and to P. Barbeaux, M. Ebener, R. Reining, A. Handziak, ONE. 2015;10:e0126534. and D. Pine for their assistance in procuring lake trout for tagging. This work 16. Bergé J, Capra H, Pella H, Steig T, Ovidio M, Bultel E, Lamouroux N. Prob- was funded by the Great Lakes Fishery Commission by way of Great Lakes ability of detection and positioning error of a hydro acoustic telemetry Restoration Initiative appropriations (GL-00E23010-3). This paper is contribu- system in a fast-flowing river: intrinsic and environmental determinants. tion number 20 of the Great Lakes Acoustic Telemetry Observation System Fish Res. 2012;125–126:1–13. (GLATOS) and contribution number 2012 of the USGS Great Lakes Science 17. Roy R, Beguin J, Argillier C, Tissot L, Smith F, Smedbol S, De-Oliveira E. Test- Center. Any use of trade, product, or firm names is for descriptive purposes ing the VEMCO Positioning System: spatial distribution of the probability only and does not imply endorsement by the U.S. Government. of location and the positioning error in a reservoir. Anim Biotelemetry. 2014;2:1. Competing interests 18. Steel AE, Coates JH, Hearn AR, Klimley AP. Performance of an ultrasonic The authors declare that they have no competing interests. telemetry positioning system under varied environmental conditions. Anim Biotelemetry. 2014;2:15. Received: 16 November 2015 Accepted: 11 January 2016 19. O’Dor RK, Andrade Y, Webber DM, Sauer WHH, Roberts MJ, Smale MJ, Voegeli FM: Applications and performance of radio-acoustic positioning and telemetry (RAPT ) systems. In Lagardère J-P, Anras M-LB, Claireaux G, editors. Advances in invertebrates and fish telemetry. Berlin: Springer; 1998. p. 1–8. (Developments in Hydrobiology, vol. 130). 20. Ehrenberg JE, Steig TW. A method for estimating the “position accuracy” References of acoustic fish tags. ICES J Mar Sci J Cons. 2002;59:140–9. 1. Espinoza M, Farrugia TJ, Webber DM, Smith F, Lowe CG. Testing a new 21. Yoshida N, MIitamura H, Aria N, Yamane H, Mitsunaga Y, Viputhanumas T, acoustic telemetry technique to quantify long-term, fine-scale move - Ratanachamnong D: Detection range and horizontal accuracy of a fine- ments of aquatic animals. Fish Res. 2011;108:364–71. scale positioning telemetry system at Kaeng Krachan Reservoir, Thailand. 2. Rillahan C, Chambers M, Howell WH, Watson WH III. A self-contained In: Proceedings of the design symposium on conservation of ecosystem system for observing and quantifying the behavior of Atlantic cod, Gadus ( The 13th SEASTAR2000 workshop), vol. 2; 2014. p. 9–14. morhua, in an offshore aquaculture cage. Aquaculture. 2009;293:49–56. 22. Kessel ST, Hussey NE, Webber DM, Gruber SH, Young JM, Smale MJ, Fisk 3. Cooke SJ, Niezgoda GH, Hanson KC, Suski CD, Phelan FJS, Tinline R, Philipp AT. Close proximity detection interference with acoustic telemetry: the DP. Use of CDMA acoustic telemetry to document 3-D positions of fish: importance of considering tag power output in low ambient noise relevance to the design and monitoring of aquatic protected areas. Mar environments. Anim Biotelemetry. 2015;3:5. Technol Soc J. 2005;39:17–27. Binder et al. Anim Biotelemetry (2016) 4:4 Page 15 of 15 23. Simpfendorfer CA, Heupel MR, Collins AB. Variation in the performance 33. Riley SC, Binder TR, Wattrus NJ, Faust MD, Janssen J, Menzies J, Marsden of acoustic receivers and its implication for positioning algorithms in a JE, Ebener MP, Bronte CR, He JX, Tucker TR, Hansen MJ, Thompson HT, riverine setting. Can J Fish Aquat Sci. 2008;65:482–92. Muir AM, Krueger CC. Lake trout in northern Lake Huron spawn on sub- 24. Heupel MR, Reiss KL, Yeiser BG, Simpfendorfer CA. Eec ff ts of biofouling on merged drumlins. J Great Lakes Res. 2014;40:415–20. performance of moored data logging acoustic receivers. Limnol Ocean- 34. Gjelland KØ, Hedger RD. Environmental influence on transmitter detec- ogr Methods. 2008;6:327–35. tion probability in biotelemetry: developing a general model of acoustic 25. Smith F. Understanding HPE in the Vemco Positioning System ( VPS). transmission. Methods Ecol Evol. 2013;4:665–74. Halifax, NS: Vemco Inc.; 2013. 35. Singh L, Downey NJ, Roberts MJ, Webber DM, Smale MJ, Van Den Berg 26. Akima H, Gebhardt A: Interpolation of irregularly and regularly spaced MA, Harding RT, Engelbrecht DC, Blows BM. Design and calibration of data. 2015. R package version 0.5-12. http://CRAN.R-project.org/ an acoustic telemetry system subject to upwelling events. Afr J Mar Sci. package=akima. 2009;31:355–64. 27. Bates D, Maechler M, Bolker BM, Walker S: lme4: Linear mixed-effects 36. Johnson BK, Ager AA, Findholt SL, Wisdom MJ, Marx DB, Kern JW, Bryant model using “Eigen” and S4. 2015. R package version 1.1-8. http://CRAN.R- LD: Mitigating spatial differences in observation rate of automated telem- project.org/package=lme4. etry systems. J Wildl Manag 1998;62:958–67. 28. R Core Team: R: A language and environment for statistical computing. 37. Martins EG, Gutowsky LF, Harrison PM, Patterson DA, Power M, Zhu DZ, 2015. https://www.R-project.org/. Leake A, Cooke SJ. Forebay use and entrainment rates of resident adult 29. Binder TR, Riley SC, Holbrook CM, Hansen MJ, Bergstedt RA, Bronte CR, fish in a large hydropower reservoir. Aquat Biol. 2013;19:253–63. He J, Krueger CC. Spawning site fidelity of wild and hatchery lake trout 38. Perry RW, Skalski JR, Brandes PL, Sandstrom PT, Klimley AP, Ammann A, (Salvelinus namaycush) in northern Lake Huron. Can J Fish Aquat Sci. MacFarlane B. Estimating survival and migration route probabilities of 2016;73:18–34. juvenile Chinook salmon in the Sacramento-San Joaquin River Delta. 30. Muggeo V: Segmented: an R package to fit regression models with North Am J Fish Manag. 2010;30:142–56. broken-line relationships. 2008. https://cran.r-project.org/web/packages/ segmented/index.html. 31. Voegeli FA, Lacroix GL, Anderson JM. Development of miniature pingers for tracking Atlantic salmon smolts at sea. Hydrobiologia. 1998;371:35–46. 32. Pincock DG. Understanding the performance of VEMCO 69 kHz single frequency acoustic telemetry. Halifax NS: Vemco Inc; 2008. Submit your next manuscript to BioMed Central and we will help you at every step: • We accept pre-submission inquiries • Our selector tool helps you to find the most relevant journal • We provide round the clock customer support • Convenient online submission • Thorough peer review • Inclusion in PubMed and all major indexing services • Maximum visibility for your research Submit your manuscript at www.biomedcentral.com/submit http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Animal Biotelemetry Springer Journals

Spatial and temporal variation in positioning probability of acoustic telemetry arrays: fine-scale variability and complex interactions

Loading next page...
 
/lp/springer-journals/spatial-and-temporal-variation-in-positioning-probability-of-acoustic-fV9SQxf70P
Publisher
Springer Journals
Copyright
Copyright © 2016 by Binder et al.
Subject
Life Sciences; Animal Systematics/Taxonomy/Biogeography
eISSN
2050-3385
DOI
10.1186/s40317-016-0097-4
Publisher site
See Article on Publisher Site

Abstract

Background: As popularity of positional acoustic telemetry systems increases, so does the need to better under- stand how they perform in real-world applications, where variation in performance can bias study conclusions. Stud- ies assessing variability in positional telemetry system performance have focused primarily on position accuracy, or comparing performance inside and outside the array. Here, we explored spatial and temporal variation in positioning probability within a 140-receiver Vemco Positioning System ( VPS) array used to monitor lake trout, Salvelinus namay- cush, spawning behavior over 23 km in Lake Huron, North America. Methods: Variability in VPS positioning probability was assessed between August and November from 2012 to 2014 using 43 stationary transmitters distributed throughout the array. Various analyses were used to relate positioning probability to number of fish transmitters in the array, wave height, and thermal stratification. We also assessed the prevalence of ‘close proximity detection interference’ (CPDI) in our array by analyzing detection probability of 35 trans- mitters on collocated receivers. Results: Positioning probability of the VPS array varied greatly over time and space. Number of fish transmitters present in the array was a significant driver of reduced positioning probability, especially during lake trout spawning period when the fish were aggregated. Relationships between positioning probability and environmental variables were complex and varied over small spatial and temporal scales. One possible confounding variable was the large range of water depth over which receivers were deployed. Another confounding factor was the high prevalence of CPDI, which decreased exponentially with water depth and was less evident when wave heights were higher than normal. Conclusions: Some variables that negatively influenced positioning can be minimized through careful planning (e.g., number of tagged fish released, transmitter power level). However, results suggested that the acoustic environ- ment was highly variable over small spatial and temporal scales in response to complex interactions between many variables. Therefore, models that predict positioning or detection efficiencies as a function of environmental variables may not be attainable in most systems. The best defense against biased study conclusions is incorporation of in situ measures of system performance that allow for retrospective analysis of array performance after a study is completed. Keywords: Vemco Positioning System, Positional telemetry, Performance, Detection probability, Close proximity detection interference, Thermal stratification, Wave height, Signal code collision *Correspondence: tr.binder@gmail.com Center for Systems Integration and Sustainability, Department of Fisheries and Wildlife, Michigan State University, Hammond Bay Biological Station, 11188 Ray Rd., Millersburg, MI 49759, USA Full list of author information is available at the end of the article © 2016 Binder et al. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/ publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Binder et al. Anim Biotelemetry (2016) 4:4 Page 2 of 15 was estimated by the array for a given transmission, [16, Background 18]). Less is known about the effects of environmental Recent advances in aquatic animal telemetry technolo- variables (e.g., thermal stratification and waves) or more gies now provide researchers with an unprecedented complex processes, such as destructive code collisions or ability to track animal movements at fine spatial and the so-called close proximity detection interference (i.e., temporal scales, and answer behavioral and ecological detection interference as a result of transmission echoes questions that were previously beyond reach. One such being heard by nearby receivers; hereafter, CPDI; [22]), advancement that has become increasingly popular over on positioning probabilities. Nonetheless, because the the last decade is use of telemetry systems to estimate two-dimensional (2D) or even three-dimensional (3D) absence of evidence (i.e., positions) in telemetry studies is positions of transmitter-implanted animals using time not necessarily evidence of absence [13], understanding difference of arrival (TDOA) of acoustic transmissions at variation in positioning probability is critical to interpret- three or more acoustic receivers [1–4]. 2D and 3D tracks ing study results. from aquatic animals have been used to study behaviors Transmitter detections are the basis of position estima- ranging from broad spatial habitat use and home ranges tion; therefore, positioning probability should be influ - [4–7] to swimming speed [2, 8] and fine-scale responses enced by many of the same variables that drive variability to environmental stimuli [9, 10]. A variety of position in detection probability in presence/absence telemetry systems (e.g., environmental noise, aquatic vegetation, ing systems exist, each with its own set of strengths and biofouling; [13, 23, 24]). However, the issue of position- weaknesses, but they can be generally reduced to two ing probability in telemetry systems is complicated by categories: (1) cabled systems that use a single receiver the fact that the contribution of a given receiver to posi- with multiple hydrophones tethered with cables and tion estimates depends not only on the performance of (2) non-cabled systems that use multiple independent that receiver, but also on the performance of receivers receivers, each with a single independent hydrophone. Cabled systems tend to be limited in size and location of around it. Moreover, because questions addressed with deployments due to need for long cables between each positional telemetry arrays are often limited to finer hydrophone and receiver [3]. Non-cabled systems offer spatial and temporal scales than those addressed with more flexibility with respect to array size and are better the presence/absence systems, studies that use position- suited for remote locations [1, 11]. However, position ing systems may be more sensitive to biases resulting processing with non-cabled systems is more complicated from variability in performance [8, 11], particularly if the than with cabled systems because of the need to account measured response variable is based on the number of for differences between receiver clocks, which drift over positions returned by the system. time due to effects of temperature and subtle manufac In this study, we assessed spatial and temporal variabil- ity in positioning probability of a large acoustic telemetry turing differences. Nonetheless, non-cabled systems have positioning system (Vemco Positioning System; hereafter become increasingly popular due to ease of deployment VPS, Vemco Inc., Halifax, NS Canada) over three con- and flexibility to accommodate project designs and large secutive seasonal deployments. At the time of writing, study areas. this positional telemetry array was the largest ever con- Similar to the presence/absence telemetry systems that structed, consisting of 140 autonomous receivers and 43 provide coarse-scale behavioral data, positional telem- stationary transmitters with a spatial coverage of approx- etry systems are subject to performance variability [12], imately 23 km . Our specific objectives were: (1) to quan - which can complicate interpretation of animal tracks and tify the degree of spatial (<0.5  km ) and temporal (6  h) possibly bias study conclusions [8, 13]. Studies describ- variation in positioning probability that occurred over ing variation in detection probability of presence/absence three seasonal deployments between 2012 and 2014 and telemetry systems are abundant (reviewed in [12, 14]), (2) to determine whether variation in positioning proba- but perhaps due to the relative novelty of positional bility could be predicted by environmental variables (e.g., telemetry and also possibly a disconnect between the surface waves and water temperature) and other site-spe- end user (researcher) and the position estimation pro- cific variables such as signal code collisions and CPDI. cess, few papers have assessed spatial and temporal per- formance variability of positional telemetry systems [11, Methods 15–18]. The primary focus of most positional telemetry Study site and the Vemco Positioning System performance studies has been position accuracy. Position Spatial and temporal variation in performance of the accuracy has been well established as largely a function VPS (Vemco Inc.; Halifax, NS Canada) was assessed in a of the geometry of receivers relative to transmitters [1, large acoustic telemetry array designed to study spawning 11, 17, 19–21] and varies little in comparison with posi- behaviors of lake trout (Salvelinus namaycush) in northern tioning probability (i.e., the probability that a position Binder et al. Anim Biotelemetry (2016) 4:4 Page 3 of 15 Lake Huron, North America. The array, which consisted height and direction every 5  min for the duration of the of 140 VR2W-69 kHz autonomous receivers (Vemco Inc.; study. Halifax, NS Canada) and 43 stationary V16-6H transmit- At the end of each study season, receivers were ters (Vemco Inc.; Halifax, NS Canada), covered an area of retrieved and downloaded, and data files were sent to approximately 23  km and was deployed for between 87 Vemco for processing using their proprietary hyperbolic and 100 days each year from August to November, 2012– positioning algorithms [25]. Position estimates of trans- 2014. Stationary transmitters (hereafter, ‘sync tags’) with mitters were based on TDOA of each transmission at a known locations were deployed primarily to synchronize minimum of three and a maximum of six (limit set by clocks among receivers and secondarily to evaluate array manufacturer) receivers with synchronized clocks. When performance. Each sync tag transmitted a unique ID code a transmission was detected on more than six receiv- every 500–700 s (nominal delay = 600 s), with each time ers, positions were estimated using data from the first between transmissions (delay) being drawn from a uni- six receivers that detected a transmission based on lin- form distribution. The site encompassed several shoal and ear time-corrected detections. Hypothetically, the first reef areas with complex bathymetric features (Fig.  1) and six receivers that detected a transmission should rep- depths ranging from ~2 m to over 38 m. Vertical tempera- resent the six closest receivers to the transmitter, but in ture profiles in the study site were monitored using two practice that may not be true because of nonlinear drift lines of four temperature loggers (HOBO Water Temp of receiver clocks. The VPS returned a weight-averaged Pro v2; model U22-001; Onset Computer Corporation, position among all combinations of three receivers that Bourne MA) that measured temperature to a resolution detected each transmission, as well as position precision of 0.02 °C, with an accuracy of ±0.21 °C. A weather buoy estimates (horizontal position error; abbreviated ‘HPE’) (Tidas 900 buoy; S2 Yachts, Holland MI) was also moored that described the relative error sensitivity of each calcu- within the array. The buoy logged air temperature, surface lated position [25]. water temperature, wind speed and direction, and wave Spatial and temporal variability in positioning probability Spatial and temporal variation in VPS array performance in each study year was assessed using positioning prob- ability of 43 stationary sync tags distributed throughout the array (Fig.  1). Temporal variation in array perfor- mance was estimated by comparing performance metrics across 6-h time bins. Use of 6-h bins represented a com- promise between having enough transmissions (36 on average) to accurately estimate a positioning probability and being a short enough time interval to reflect envi - ronmental variability at an ecologically relevant scale. Probability of positioning each transmitter during each 6-h bin was calculated based on the ratio of observed −1 to expected positions (6 positions  h   ×  6  h  =  36 posi- tions expected). Subsequently, for each 6-h time bin, we used 2D cubic spline interpolation (R package ‘akima’; [26]) to estimate and visualize variation in positioning performance across the array. Mean array positioning probability during each 6-h bin was calculated from the 2D interpolations by taking the mean of all interpolated data points (spatial resolution of the interpolations was Fig. 1 Bathymetric map (in m) of the Drummond Island acoustic approximately 34  m ). Mean positioning probability of telemetry study site. A 140-receiver Vemco Positioning System ( VPS) the transmitters themselves was not used because trans- was deployed at the site during late summer and autumn each year mitters were not equally spaced in the array, which meant between 2012 and 2014. Symbols: cross = acoustic receiver ( VR2W ), blue circle = sync tag ( V16-6H), yellow square = temperature line some transmitters (particularly in the deep, less complex (containing 4 suspended HOBO temperature loggers), and orange areas of the array) represented a greater area of the array square = Tidas 900 weather buoy. Site S003, representing a specific than others. Due the random nature of sync tag transmis- sync tag referred to in the text, was labeled and located in the sions (i.e., uniform distribution between 500 and 700  s), southmost site in the array. Inset Location of the Drummond Island positioning probability estimates based on the mean of study site (red square) in Lake Huron, North America. Square indicates location only and is not to scale 36 transmissions per 6-h bin were subject to random Binder et al. Anim Biotelemetry (2016) 4:4 Page 4 of 15 error. Simulated transmission histories for 10,000 sync relative close proximity to a receiver are interrupted tags revealed a range of 34–37 transmissions during a by strong echoes off reflective surfaces [22], in essence, single 6-h period, indicating a maximum of 6  % random causing the signal to collide with itself, and thus, not error in our positioning probability estimates. be properly decoded and logged on that receiver. We A generalized linear mixed-effect model with binomial searched for evidence of CPDI in our array using data error distribution (R Package ‘lme4’; [27]) was used to from 35 sync tags with collocated receivers (i.e., a model within- and between-sync tag variation in posi- receiver on the same mooring as the transmitter; Fig. 1). tioning probability against variables that could affect Prevalence of CPDI in our array was described by cal- detection of acoustic transmitters. Fixed effects in the culating the detection probability of each transmitter on model included the number of unique fish transmitters collocated receivers during each 6-h bin. As with posi- within detection range of the closest receiver (‘Unique- tioning probability, detection probability was calculated Fish’), wave height measured at the within-array weather as the ratio of observed detections to expected detec- buoy (‘WaveHt’), and degree of thermal stratification in tions (36 expected detections per 6-h time bin). Inter- the water column (i.e., difference between near-surface pretation of inter-site variability in detection probability and near-substrate temperature in the array, as measured was complicated by spatial variation in detection prob- by HOBO temperature logger lines; ‘DiffTemp’), as well as ability related to varying local environmental conditions their interactions. Sync tag site ID was included as a ran- (e.g., number of fish transmitters within detection range dom effect to account for inherent differences in position - of a receiver). To account for this variation, we stand- ing probability related to location of each transmitter in ardized detection probabilities on collocated receivers the array. to that on the non-collocated receiver with maximum The mixed-effect model revealed complex interactions detection probability for each transmitter. The new between variables, so rather than attempting to build a response variable, ‘relative detection probability,’ was the global model to describe variation in positioning prob- ratio of detection probability at the collocated receiver ability of our sync tags, we fit separate logistic regres - and detection probability for that transmitter at the sion models (R Package ‘stats’; [28]) for each transmitter non-collocated receiver (i.e., receiver not on the same site and examined spatial and temporal patterns in effect mooring as the transmitter) with maximum detection size (i.e., parameter estimates from logistic regressions) probability. The relationship between relative detection of three fixed-effect variables (i.e., UniqueFish, WaveHt, probability of transmitters and water depth was mod- and DiffTemp). Preliminary inspection of the data sug - eled using nonlinear, least squares regression (R Package gested that the relative influence of the variable Unique - ‘stats’). Fish on variation in array performance changed markedly between the lake trout pre-spawning and spawning peri- Results ods, and therefore, the two time periods were analyzed Temporal and spatial variability in positioning probability separately. Spawning period start dates were the same VPS array performance varied greatly within and among as were used previously at this site by Binder et  al. [29], years. Mean, array-wide 6-h positioning probability who used changepoint analysis to determine when male ranged from 8 to 90 % in 2012, from 38 to 92 % in 2013, lake trout implanted with pressure-sensing transmitters and from 59 to 95  % in 2014 (Fig.  2). The greater range moved from deep offshore water onto the shallow-water in VPS positioning performance in 2012 compared with spawning shoals. During each time period, parameter 2013 and 2014 was a result of extremely poor array per- estimates were inspected for obvious trends by map- formance during mid-to-late October and November ping them on bathymetry. We then tested for significant in 2012 (Fig.  2). Upon investigation, poor performance relationships between parameter estimates and trans- during this period was caused primarily by widespread mitter depth using linear regression (R Package ‘stats’). receiver memory saturation, which caused log files con - Where scatter plots revealed a nonlinear relationship, taining millisecond data required for positioning trans- segmented regression (i.e., broken-stick regression; R mitters to be overwritten by detection data, which at the Package ‘segmented’; [30]) was used in place of linear time was logged only to the nearest second (this issue regression. Comparisons between pre-spawning and was addressed in receiver firmware updates). The first spawning period parameter estimates were made using instance of this issue occurred on 06 October, and by paired t tests (R Package ‘stats’). the end of the season, 81 of 140 (59 %) receivers reached memory capacity and stopped logging millisecond data. Close proximity detection interference For this reason, 2012 VPS data collected on or after 06 CPDI is a phenomenon in acoustic telemetry whereby October were excluded from subsequent statistical transmission sequences of a transmitter located in analyses. Binder et al. Anim Biotelemetry (2016) 4:4 Page 5 of 15 Interestingly, positioning probability at some of the deeper receivers on the southwest side of the array (in particular S003; southernmost transmitter on the west side of array) consistently differed from other transmit - ter sites in the array. During the August dates, when mean whole-array positioning probability was relatively high, positioning probability at these sites was generally low (often less than 30  % of transmissions positioned; Fig.  3). Conversely, during the October and November dates when mean whole-array performance decreased, positioning probability at these sites tended to improve (see Additional file  1). Poor performance of these south- west receivers during summer months appeared to be related to thermal stratification in the water column, with positioning probability, particularly at site S003 (Fig.  1), improving during brief periods when little thermal strati- fication occurred in the array (Fig.  4). The negative rela - tionship between positioning performance and thermal stratification was also evident in among-year compari - sons, where performance at site S003 was greater in 2014, the year with the least thermal stratification, than in the previous 2  years (Fig.  4). Toward the end of each year when the thermal structure in the array became more homogenous, variation in positioning performance at this site was more closely related to the number of tagged fish within detection range of the site (Fig. 4). Attempts to develop a meaningful global model relating sync tag positioning probability to local environmental variables such as number of tagged fish present (Unique - Fish), wave height (WaveHt), and thermal stratification Fig. 2 Mean estimated Vemco Positioning System ( VPS) positioning (TempDiff ) revealed complex relationships and interac - probability (black line) in relation to wave height (red line) recorded tions that proved impossible to satisfactorily model using by the on-site weather buoy and the median number of unique fish transmitters (blue line) detected per receiver during each year of the the predictor variables available. As expected, positioning study. Mean positioning probability was calculated from 2D interpo- performance was negatively correlated with the number lations based on 43 sync tags distributed throughout the array (see of fish present within detection range across all receiv - Fig. 1). Each point on the graphs represents a 6-h time window ers during both pre-spawning and spawning periods (Fig.  5a, d). However, relationships between positioning probability and both wave height and thermal stratifica - Clear seasonal trends were evident in all three years, tion were site-specific. For example, at some sites posi - with better performance in August and September than tioning probability was negatively correlated with wave in October and November (Fig.  2). Within the array, a height, while at others a positive correlation occurred great deal of spatial variation occurred in positioning between positioning probability and wave height (Fig. 5b, probability. Within some 6-h time periods, position- e). Similar relationships were observed when comparing ing probability of sync tags in some parts of the array positioning probability against degree of thermal strati- was perfect (or near-perfect), while in other parts of the fication (Fig.  5c, f ). Moreover, at several sites, direction array, it was zero (Fig.  3). In general, positioning prob- of observed relationships was reversed during the pre- ability was poorer on the west side of the array (where spawning and spawning periods. the main lake trout spawning sites were located) than the A portion of the inter-receiver variability in relation- east side of the array, particularly in the autumn during ships between environmental conditions and positioning the lake trout spawning period. Nonetheless, all sync tags performance was related to the wide range of depths over displayed perfect positioning probability during at least which our array was deployed. However, relationships some periods, suggesting that spatial patterns were not between parameter estimates (i.e., estimates of effect a result of irregularities in array design (e.g., geometry, size) from transmitter-specific logistic regressions and specific receiver locations). Binder et al. Anim Biotelemetry (2016) 4:4 Page 6 of 15 Fig. 3 Estimated array-wide positioning probability (% positioning) of the Drummond Island Vemco Positioning System ( VPS) during three arbitrary 6-h time periods in 2012 and 2014. Estimates of positioning probability were interpolated based on 43 sync tags distributed throughout the array (see Fig. 1). A high degree of spatial and temporal variability occurred in positioning probability of the Drummond Island VPS array. See Additional file 1 for complete records for all 3 years of the study Binder et al. Anim Biotelemetry (2016) 4:4 Page 7 of 15 Fig. 4 Positioning probability (black line) of sync tag S003, the southernmost transmitter in the Drummond Island Vemco Positioning System ( VPS) array (see Fig. 1), relative to the median number of unique fish transmitters detected on receivers with line of sight and within 500 m of S003 and thermal stratification (bottom graph in each panel). Temperature profiles were interpolated from two separate temperature lines deployed in the array. Poor positioning probability at the end of the 2012 season was due to widespread receiver memory saturation, which began on 06 October. Each point on the graphs represents a 6-h time window. Temperatures are in °C depth of the transmitter were not always linear and dif- relationship between WaveHt and positioning probabil- fered between pre-spawning and spawning periods. Dur- ity) were generally negative for shallow transmitters and ing the pre-spawning period, when few fish were present positive for deep transmitters (Fig.  6a). A segmented in the array (Fig.  2), WaveHt parameter estimates (i.e., regression indicated a positive relationship between Binder et al. Anim Biotelemetry (2016) 4:4 Page 8 of 15 Fig. 5 Results of logistic regression analysis relating positioning probability of each sync tag (n = 43) to the number of fish transmitters detected on the closest receiver (‘UniqueFish’; a, d), wave height (‘WaveHt’; b, e), and difference between near-surface and near-substrate temperature in the array (‘DiffTemp’; c, f). Data from the lake trout spawning (right column) and pre-spawning (left column) periods are displayed separately. Red symbols indicate a significant negative relationship, blue symbols a significant positive relationship, and white symbols no significant relationship (α = 0.05) WaveHt parameter estimate and transmitter depth (seg- (segmented regression; t  =  −0.359, df  =  31, p  =  0.722). mented regression; t = 3.351, df = 31, p = 0.002) down to TempDiff parameter estimates were also positively cor - approximately 13.9 ± 2.4 m, after which WaveHt param- related with transmitter depth during the pre-spawning eter estimates remained relatively consistent across depth period (linear regression; t  =  3.723, df  =  33, p  <  0.001), Binder et al. Anim Biotelemetry (2016) 4:4 Page 9 of 15 Fig. 6 Relationship between logistic regression parameter estimates from positioning probability models (‘WaveHt’ = wave height, ‘Temp- Diff ’ = difference between near-surface and near-substrate water temperature) and water depth at 43 sync tag sites. Data from the lake trout spawning (right column) and pre-spawning (left column) periods were analyzed separately. Lines in b, c, and d depict results of linear regression analysis, while lines in a depict results of segmented regression. Significant relationships occurred between parameter estimates and water depth during the pre-spawning period, but not during the spawning period (α = 0.05) with negative parameter estimates predominant at shal- always negative) and did not vary with transmitter depth low depths and positive parameter estimates predomi- (linear regression; pre-spawning: t  =  1.471, df  =  33, nant at deeper depths (Fig.  6b). In contrast, during the p  =  0.151; spawning: t  =  −0.865, df  =  33, p  =  0.393). spawning period, neither WaveHt (linear regression; Nonetheless, UniqueFish parameter estimates were on t  =  0.673, df  =  33, p  =  0.506) nor TempDiff (linear average 2.47 (±1.33) times greater during the spawning regression; t = 0.283, df = 33, p = 0.779) parameter esti- period than during the pre-spawning period (paired t mates were significantly correlated with transmitter test; t  =  7.344, df  =  34, p  <  0.001), indicating a greater depth (Fig. 6c, d). negative relationship with positioning probability during Some of the discrepancy between the pre-spawning and the spawning period than non-spawning period. In fact, spawning period is likely due to the fact that the presence number of fish present in the array appeared to be a pre - of lake trout transmitters within the array (‘UniqueFish’) dominant factor driving whole-array positioning prob- had a far greater influence on positioning probability dur - ability during the spawning period (Fig. 2). ing the spawning period (when far more fish were pre - sent in the array) than during the pre-spawning period. Close proximity detection interference UniqueFish parameter estimates were negative during One possible confounding factor in our analysis of posi- both pre-spawning and spawning period (i.e., relation- tioning probability was CPDI [22]. Thirty-five of 43 ship between UniqueFish and positioning probability was sync tags in our array were collocated with an acoustic Binder et al. Anim Biotelemetry (2016) 4:4 Page 10 of 15 Fig. 7 Boxplots displaying median (center line), 25th and 75th quartiles (bounding box), and range (dashed line) of detection probabilities on col- located receivers for 35 transmitters. Each panel shows results for a different year of the study (2012–2014). Individual detection probabilities were calculated based on 6-h time windows receiver. Examination of detection probability of sync collocated transmitters was strongly correlated across tags on collocated receivers revealed that this phenom- all three years (Pearson’s r  =  0.78, 0.75, and 0.85 for enon was widespread in our array (Fig.  7). Collocated 2012 vs. 2013, 2012 vs. 2014, and 2013 vs. 2014, respec- receivers never detected sync tags as well as other nearby tively, p  <  0.001 for all comparisons), which suggests surrounding receivers with line of sight to the transmitter that CPDI was location-specific and probably related to (i.e., maximum detection probability for a given transmit- physical characteristics of the immediate environment ter always occurred on a non-collocated receiver). Mean that changed little over time (i.e., lake bottom compo- maximum detection probability for each transmitter on sition and topography). Inter-site variation in mean nearby non-collocated receivers ranged from 0.67 to 0.91. relative detection probability was related negatively In contrast, mean detection probability on collocated to water depth at the site and was best approximated receivers averaged only 12  % of maximum detection with a decreasing exponential curve (relative detection −0.892 probability, ranging from 0.01 to 0.37 over the 3 years of probability  =  1.098  ×  depth , p  <  0.001 for both study. parameters; Fig.  8a), indicating that CPDI increased A high degree of intra-annual variability occurred exponentially with increased water depth. in detection probability of transmitters on collocated Within-site variability in relative detection probabil- receivers, both within and between sites (Fig.  7). How- ity of transmitters on collocated receivers was related ever, mean relative detection probability of individual to wave height (linear mixed-effect model, t  =  144.432, Binder et al. Anim Biotelemetry (2016) 4:4 Page 11 of 15 df = 36,010, p < 0.001; Fig. 8b). In general, mean relative detection probability of collocated transmitters was lower than average when wave height was less than approxi- mately 1.25 m and greater than average when wave height was greater than 1.25 m (Fig. 8b). No significant relation - ship existed between maximum detection probability and wave height (linear mixed-effect model, t  = −1.021, df = 36,010, p = 0.307), so the above relationship was not an artifact of decreased maximum detection probability at higher wave heights; mean maximum detection proba- bility of sync tags was greater than or equal to 0.77 (range 0.77–0.89) at all wave heights. Discussion Substantial spatial and temporal variability occurred in positioning probability of sync tags in our positional telemetry array. While spatial variation in positioning probability has been noted in other studies [15, 16], most studies have focused on receiver geometry and ignored temporal variability [16–18]. This trend may be due to the fact that most assessment studies were short in dura- tion [1, 15, 17], or used positional arrays with relatively small spatial coverage [1, 11, 15, 18]. The long duration, high environmental complexity, and large spatial extent of our VPS deployment provided a unique opportunity to explore and quantify within-array variability in posi- tioning probability. Some of the variability we observed could be explained by variables we measured in the field, but much of it could not, which highlights the complex nature of acoustics in natural systems. Some of the nega- tive relationships we observed could be controlled (e.g., number of tagged fish in the array; transmitter power level), or at the least minimized by careful planning and study design. Signal code collisions resulting from large numbers of fish transmitters in the array at the same time were a significant cause of decreased positioning probability in our system, particularly during the spawning period when they were the main driver of variation in array Fig. 8 a Relationship between mean relative detection probability performance. A signal code collision occurs when trans- of sync tags on collocated receivers and water depth at each loca- missions from two or more transmitters are detected tion. Data were normalized to the receiver with maximum detection simultaneously on the same receiver [31], preventing the probability for each transmitter. Individual points were color-coded receiver from properly decoding either signal. Probability by year. Data were best approximated with a declining exponential −0.892 of code collisions is a function of the number of trans- curve (relative detection probability = 1.098 × Depth ). b Dif- ference from mean relative detection probability of sync tags on col- mitters within range of a receiver, the duration of the located receivers across wave heights. Boxplots display median (center code signal, and the period between transmissions [23]. line), 25th and 75th quartiles (bounding box), and range (dashed line) The duration of a transmission, and thus the degree of of differences across 35 collocated sync tags during the three study susceptibility of a positioning system to signal code col- years. Detection probability on collocated receivers was generally lisions, is dependent on the coding scheme used, which lower than average when wave height was less than approximately 1.25 m and greater than average when wave height was greater than varies by manufacturer. With Vemco’s current coding 1.25 m scheme, the transmission duration is relatively long (up Binder et al. Anim Biotelemetry (2016) 4:4 Page 12 of 15 to ~5 s; [32]); therefore, depending on the nominal delay particularly if researchers are new to a study site, or the of the transmitters (i.e., period between successive trans- behavioral ecology of the animal is not well understood. missions), the probability of collisions can be quite high As has been observed in other positional telemetry at relatively low transmitter densities (see Fig. 9). studies [16, 18], positioning probability of sync tags at Signal code collisions have a high potential for creating our site was influenced by environmental variability. spatial and temporal bias in telemetry studies because the However, relationships between environmental vari- presence of the study animals themselves alters the per- ables and positioning probability in our array were loca- formance of the telemetry system. Therefore, investiga - tion- and time-specific. Steel et al. [18] reported variation tors should assess the potential for, and implications of, in the relative influence of environmental parameters signal code collisions (particularly if they are expected on positioning probability across three study systems to be heterogeneously distributed) both during the study (coastal, estuarine, and riverine), but our results indi- design stage and while interpreting results. In our case, cated that this variation can occur over relatively small the high prevalence of signal collisions in our system spatial scales. Our observations have important implica- during the spawning period was due to several factors, tions for acoustic telemetry studies in general. First, fine- including: (1) underestimation of lake trout annual mor- scale variation in response to environmental variables tality rate and spawning site fidelity [29], which caused indicated that the nature of these relationships was com- high numbers of transmitters to return to the system plex. Therefore, while general rules of thumb regarding during each year of the study (390 tagged trout released the effect of certain environmental variables on telemetry between 2010 and 2011), (2) high levels of aggregation system performance may be evident, development of a at relatively few spawning site locations [33], and (3) universally applicable predictive model [34] using easy- use of high-powered tags (V16-6H, 158  dB, 90  s nomi- to-measure variables is likely beyond reach prior to con- nal delay), which at times had detection ranges of several ducting a study. Second, because acoustic properties can kilometers. Based on our experience, we recommend vary over small distances, true optimization of an acous- adopting a conservative approach in study designs when tic telemetry system at some sites may require fine-scale determining how many transmitters to release in a study, range testing and development of spatially heterogeneous array designs (e.g., differential receiver spacing across an array). Water depth played an important role in the perfor- mance of our VPS array, particularly with respect to how the system responded to changes in wave height and thermal stratification. In general, shallow transmitters tended to be negatively affected by increased wave height and thermal stratification, while deeper transmitters often saw a boost in positioning probability under the same conditions. As far as we are aware, ours is the first study to report a positive relationship between position- ing probability and wave height or thermal stratification; however, we are doubtful that the increase in positioning probability is directly related to these variables. Rather, because both of these variables tend to reduce the dis- tance over which an acoustic transmitter is detected [23, 35], we hypothesize that the positive relationships are due to reduced transmission echoes and a consequent reduction in CPDI [22], which was most pronounced in deep water. This interpretation was supported by the Fig. 9 Estimated probability of a signal code collision (using Vemco’s observation that detection probability of transmitters on global coding scheme) based on the number of transmitters within collocated receivers increased as wave heights increased. detection range of a receiver. Collision probabilities were calculated Interestingly, depth effects were only observed during by simulating transmission histories (n = 10,000 transmissions) for the pre-spawning period. We attribute this to a masking between 1 and 50 transmitters assuming transmission duration of 5.12 s. Transmitters were programmed to transmit at random intervals effect by signal code collisions, which were more preva - between ±50 % of nominal delay. R code for the simulation is pro- lent during the spawning period than during the pre- vided as Additional file 2 spawning period. Binder et al. Anim Biotelemetry (2016) 4:4 Page 13 of 15 The high incidence of CPDI in our VPS array was likely or placing current receivers higher in the water column) due to the combined effect of our use of high-powered to improve performance in subsequent years. transmitters and a highly reflective acoustic environ - ment [22] in which substrate was dominated by the hard Conclusions surfaces of cobble, boulders, and bedrock. Although The acoustic telemetry community is constantly expand - evidence of CPDI occurred for all collocated receiver/ ing as new researchers adopt these technologies in their transmitter pairs, the prevalence of CPDI increased studies. Therefore, identification and discussion of issues exponentially with receiver depth, a characteristic we related to study design and data interpretation is valu- attributed to an exponential decline in ambient noise able to the community because it helps to improve the from surface sources (e.g., wind and waves). Preva- overall quality of data coming out of acoustic telemetry lence of CPDI is dependent on the coding scheme [21, studies, as well as their interpretation. Positional acoustic 22, 32] and, therefore, is likely to vary among telemetry telemetry, and more specifically VPS, is a relatively new systems. Thus, researchers should assess the potential technique for tracking movements of aquatic animals; for this phenomenon when designing their study and thus, a need exists to better understand how these sys- choose transmitter specifications and receiver place - tems perform in real-world applications. Only a few pub- ments that minimize the effect with the equipment they lished papers are available describing spatial or temporal plan to use. Without sufficient receiver overlap (i.e., variation in VPS performance, fewer still while the VPS redundancy), CPDI can create areas of low detection/ system was used to track animals, and none that have position probability in receiver arrays and curtains [22]. spanned the spatial scale and range of environmental Moreover, CPDI also has the potential to reduce accu- conditions that occurred in our study. Some of the vari- racy of positions returned from VPS arrays. The reason ables we identified as negatively influencing positioning for this inaccuracy is that position estimation will depend probability (e.g., signal code collisions) can be minimized on detections at receivers that are further away from the through careful planning, but our results also suggest transmitter, which could result in poor receiver geom- that the acoustic environment can be highly variable over etry, imprecise signal arrival time estimates, and conse- relatively small spatial and temporal scales, which if unac- quently higher horizontal position error [1, 18]. counted for, have the potential to bias study conclusions. Receiver memory saturation (which caused raw log files At complex study sites (e.g., sites with highly variable to be overwritten by detection data) late in 2012 caused depth, substrate types, or water chemistry), fine-scale a drastic decrease in whole-array positioning probability. range testing may allow researchers to optimize receiver At the time, no mechanism was in place to flag this issue array design; however, we acknowledge that such intense during either receiver download or position processing, pre-study testing is rarely practical. In most cases, the and because it occurred during the peak of lake trout best defense against making biased study conclusions due spawning, poor array performance was assumed to be to spatial and temporal changes in system performance due to signal code collisions. Indeed, it was not until we will be to incorporate methods for measuring that vari- started work on this manuscript that the memory issue ability into the study design. For most, that will involve was identified, which highlights the need for research - deploying stationary transmitters throughout the study ers to take responsibility for quality checking results site for the duration of the study, with the number and returned to them from positioning software. At a mini- location of these transmitters depending on the questions mum, we recommend researchers create visual displays being addressed and the complexity of the study site. (e.g., graphs or maps, e.g., Fig. 4) for assessing positioning At a minimum, this approach will allow researchers to performance over space and time. Unexplained changes interpret results in the context of array performance. In in performance should be investigated thoroughly to rule some cases, results may be standardized by incorporating out equipment failure or processing errors. Early assess- measures of system performance variability into analyses, ment of positioning performance, especially if assessed either through use of correction factors [36], or through from a pilot study, may also provide an opportunity to development of sophisticated statistical analyses that correct deficiencies in array design. For example, poor explicitly correct parameter estimates based on imperfect positioning probability on the southwest side of our array detection and positioning probabilities [37, 38]. during summer (top panels in Fig. 3) was likely due to the fact that transmissions on top of the reef had to transmit Availability of supporting data through the thermocline to reach surrounding receivers. The data supporting the results of this article are stored Had we identified this issue in 2012, changes could have in the Great Lakes Acoustic Telemetry Observation Sys- been made to the array (e.g., addition of more receivers tem (GLATOS) database (http://data.glos.us/glatos). Binder et al. Anim Biotelemetry (2016) 4:4 Page 14 of 15 4. Andrews KS, Tolimieri N, Williams GD, Samhouri JF, Harvey CJ, Levin PS. Data availability is subject to data sharing policies cur- Comparison of fine-scale acoustic monitoring systems using home range rently under development by GLATOS, the Great Lakes size of a demersal fish. Mar Biol. 2011;158:2377–87. Fishery Commission, and the United States Geological 5. Coates JH, Hovel KA, Butler JL, Klimley AP, Morgan SG. Movement and home range of pink abalone Haliotis corrugate: implications for restora- Survey. tion and population recovery. Mar Ecol Prog Ser. 2013;486:189–201. 6. Espinoza M, Farrugia TJ, Lowe CG. Habitat use, movements and site Additional files fidelity of the gray smooth-hound shark (Mustelus californicus Gill 1863) in a newly restored southern California estuary. J Exp Mar Biol Ecol. 2011;401:63–74. Additional file 1. Video showing spatial and temporal variation in posi- 7. Reubens JT, Pasotti F, Degraer S, Vincx M. Residency, site fidelity and habi- tioning performance of a 140-receiver Vemco Positioning System ( VPS) tat use of Atlantic cod (Gadus morhua) at an offshore wind farm using array over three lake trout spawning seasons. Estimates of positioning acoustic telemetry. Mar Environ Res. 2013;90:128–35. probability were interpolated based on performance of 43 stationary tags 8. Løkkeborg S, Fernö A, Jørgensen T. Eec ff t of position-fixing interval on (indicated by + symbol) scattered throughout the array. estimated swimming speed and movement pattern of fish tracked with a Additional file 2. R script for estimating, based on simulation, the proba- stationary positioning system. Hydrobiologia. 2002;483:259–64. biliity of a signal code collision based on the nominal delay, transmission 9. McMahan MD, Brady DC, Cowan DF, Grabowski JH, Sherwood GD. Using duration, and number of tags within detection range of a receiver. acoustic telemetry to observe the effects of a groundfish predator (Atlan- tic cod, Gadus morhua) on movement of the American lobster (Homarus americanus). Can J Fish Aquat Sci. 2013;70:1625–34. Authors’ contributions 10. Lynch BR, Rochette R. Circatidal rhythm of free-roaming sub-tidal green TRB led all aspects of study design, executed the field portion of the study, crabs, Carcinus maenas, revealed by radio-acoustic positional telemetry. and drafted the manuscript. TAH, CMH, and CCK participated in the design of Crustaceana. 2007;80:345–55. the study, provided consultation on statistical analyses, and helped to draft 11. Biesinger Z, Bolker BM, Marcinek D, Grothues TM, Dobarro JA, Lind- the manuscript. All authors read and approved the final manuscript. berg WJ. Testing an autonomous acoustic telemetry positioning system for fine-scale space use in marine animals. J Exp Mar Biol Ecol. Author details 2013;448:46–56. Center for Systems Integration and Sustainability, Department of Fisheries 12. Melnychuk MC: Detection efficiency in telemetry studies: definitions and and Wildlife, Michigan State University, Hammond Bay Biological Station, evaluation methods. In: Adams NS, Beeman JW, Eiler JH, editors. Telem- 11188 Ray Rd., Millersburg, MI 49759, USA. U.S. Geological Survey, Great Lakes etry techniques: a user guide for fisheries research. American Fisheries Science Center, Hammond Bay Biological Station, 11188 Ray Rd., Millersburg, Society, Bethesda, Maryland; 2012. p. 339–57. MI 49759, USA. Center for Systems Integration and Sustainability, Depart- 13. Payne N, Gillanders B, Webber D, Semmens J. Interpreting diel activity ment of Fisheries and Wildlife, Michigan State University, 1405 South Harrison patterns from acoustic telemetry: the need for controls. Mar Ecol Prog Road, 115 Manly Miles Building, East Lansing, MI 48823, USA. Ser. 2010;419:295–301. 14. Kessel ST, Cooke SJ, Heupel MR, Hussey NE, Simpfendorfer CA, Vagle S, Acknowledgements Fisk AT. A review of detection range testing in aquatic passive acoustic Thanks to F. Smith, S. Smedbol, A. Hillard, and J. MacAulay for processing the telemetry studies. Rev Fish Biol Fish. 2014;24:199–218. positions and for their helpful comments on the manuscript. Thanks also to 15. Baktoft H, Zajicek P, Klefoth T, Svendsen JC, Jacobsen L, Pedersen MW, R. Bergstedt, H. Thompson, E. Larson, C. Wright, J. Van Een, M. Lance ff wicz, L. March Morla D, Skov C, Nakayama S, Arlinghaus R. Performance assess- Lesmeister, D. Operhall, B. Lamoreux, S. Farha, Z. Wickert, J. Osga, Z. Holmes, ment of two whole-lake acoustic positional telemetry systems—Is reality S. Miehls, S. Seegert, J. Hinderer, R. Darnton, K. Smith, and P. Wigren for their mining of free-ranging aquatic animals technologically possible? PLoS assistance in the field and to P. Barbeaux, M. Ebener, R. Reining, A. Handziak, ONE. 2015;10:e0126534. and D. Pine for their assistance in procuring lake trout for tagging. This work 16. Bergé J, Capra H, Pella H, Steig T, Ovidio M, Bultel E, Lamouroux N. Prob- was funded by the Great Lakes Fishery Commission by way of Great Lakes ability of detection and positioning error of a hydro acoustic telemetry Restoration Initiative appropriations (GL-00E23010-3). This paper is contribu- system in a fast-flowing river: intrinsic and environmental determinants. tion number 20 of the Great Lakes Acoustic Telemetry Observation System Fish Res. 2012;125–126:1–13. (GLATOS) and contribution number 2012 of the USGS Great Lakes Science 17. Roy R, Beguin J, Argillier C, Tissot L, Smith F, Smedbol S, De-Oliveira E. Test- Center. Any use of trade, product, or firm names is for descriptive purposes ing the VEMCO Positioning System: spatial distribution of the probability only and does not imply endorsement by the U.S. Government. of location and the positioning error in a reservoir. Anim Biotelemetry. 2014;2:1. Competing interests 18. Steel AE, Coates JH, Hearn AR, Klimley AP. Performance of an ultrasonic The authors declare that they have no competing interests. telemetry positioning system under varied environmental conditions. Anim Biotelemetry. 2014;2:15. Received: 16 November 2015 Accepted: 11 January 2016 19. O’Dor RK, Andrade Y, Webber DM, Sauer WHH, Roberts MJ, Smale MJ, Voegeli FM: Applications and performance of radio-acoustic positioning and telemetry (RAPT ) systems. In Lagardère J-P, Anras M-LB, Claireaux G, editors. Advances in invertebrates and fish telemetry. Berlin: Springer; 1998. p. 1–8. (Developments in Hydrobiology, vol. 130). 20. Ehrenberg JE, Steig TW. A method for estimating the “position accuracy” References of acoustic fish tags. ICES J Mar Sci J Cons. 2002;59:140–9. 1. Espinoza M, Farrugia TJ, Webber DM, Smith F, Lowe CG. Testing a new 21. Yoshida N, MIitamura H, Aria N, Yamane H, Mitsunaga Y, Viputhanumas T, acoustic telemetry technique to quantify long-term, fine-scale move - Ratanachamnong D: Detection range and horizontal accuracy of a fine- ments of aquatic animals. Fish Res. 2011;108:364–71. scale positioning telemetry system at Kaeng Krachan Reservoir, Thailand. 2. Rillahan C, Chambers M, Howell WH, Watson WH III. A self-contained In: Proceedings of the design symposium on conservation of ecosystem system for observing and quantifying the behavior of Atlantic cod, Gadus ( The 13th SEASTAR2000 workshop), vol. 2; 2014. p. 9–14. morhua, in an offshore aquaculture cage. Aquaculture. 2009;293:49–56. 22. Kessel ST, Hussey NE, Webber DM, Gruber SH, Young JM, Smale MJ, Fisk 3. Cooke SJ, Niezgoda GH, Hanson KC, Suski CD, Phelan FJS, Tinline R, Philipp AT. Close proximity detection interference with acoustic telemetry: the DP. Use of CDMA acoustic telemetry to document 3-D positions of fish: importance of considering tag power output in low ambient noise relevance to the design and monitoring of aquatic protected areas. Mar environments. Anim Biotelemetry. 2015;3:5. Technol Soc J. 2005;39:17–27. Binder et al. Anim Biotelemetry (2016) 4:4 Page 15 of 15 23. Simpfendorfer CA, Heupel MR, Collins AB. Variation in the performance 33. Riley SC, Binder TR, Wattrus NJ, Faust MD, Janssen J, Menzies J, Marsden of acoustic receivers and its implication for positioning algorithms in a JE, Ebener MP, Bronte CR, He JX, Tucker TR, Hansen MJ, Thompson HT, riverine setting. Can J Fish Aquat Sci. 2008;65:482–92. Muir AM, Krueger CC. Lake trout in northern Lake Huron spawn on sub- 24. Heupel MR, Reiss KL, Yeiser BG, Simpfendorfer CA. Eec ff ts of biofouling on merged drumlins. J Great Lakes Res. 2014;40:415–20. performance of moored data logging acoustic receivers. Limnol Ocean- 34. Gjelland KØ, Hedger RD. Environmental influence on transmitter detec- ogr Methods. 2008;6:327–35. tion probability in biotelemetry: developing a general model of acoustic 25. Smith F. Understanding HPE in the Vemco Positioning System ( VPS). transmission. Methods Ecol Evol. 2013;4:665–74. Halifax, NS: Vemco Inc.; 2013. 35. Singh L, Downey NJ, Roberts MJ, Webber DM, Smale MJ, Van Den Berg 26. Akima H, Gebhardt A: Interpolation of irregularly and regularly spaced MA, Harding RT, Engelbrecht DC, Blows BM. Design and calibration of data. 2015. R package version 0.5-12. http://CRAN.R-project.org/ an acoustic telemetry system subject to upwelling events. Afr J Mar Sci. package=akima. 2009;31:355–64. 27. Bates D, Maechler M, Bolker BM, Walker S: lme4: Linear mixed-effects 36. Johnson BK, Ager AA, Findholt SL, Wisdom MJ, Marx DB, Kern JW, Bryant model using “Eigen” and S4. 2015. R package version 1.1-8. http://CRAN.R- LD: Mitigating spatial differences in observation rate of automated telem- project.org/package=lme4. etry systems. J Wildl Manag 1998;62:958–67. 28. R Core Team: R: A language and environment for statistical computing. 37. Martins EG, Gutowsky LF, Harrison PM, Patterson DA, Power M, Zhu DZ, 2015. https://www.R-project.org/. Leake A, Cooke SJ. Forebay use and entrainment rates of resident adult 29. Binder TR, Riley SC, Holbrook CM, Hansen MJ, Bergstedt RA, Bronte CR, fish in a large hydropower reservoir. Aquat Biol. 2013;19:253–63. He J, Krueger CC. Spawning site fidelity of wild and hatchery lake trout 38. Perry RW, Skalski JR, Brandes PL, Sandstrom PT, Klimley AP, Ammann A, (Salvelinus namaycush) in northern Lake Huron. Can J Fish Aquat Sci. MacFarlane B. Estimating survival and migration route probabilities of 2016;73:18–34. juvenile Chinook salmon in the Sacramento-San Joaquin River Delta. 30. Muggeo V: Segmented: an R package to fit regression models with North Am J Fish Manag. 2010;30:142–56. broken-line relationships. 2008. https://cran.r-project.org/web/packages/ segmented/index.html. 31. Voegeli FA, Lacroix GL, Anderson JM. Development of miniature pingers for tracking Atlantic salmon smolts at sea. Hydrobiologia. 1998;371:35–46. 32. Pincock DG. Understanding the performance of VEMCO 69 kHz single frequency acoustic telemetry. Halifax NS: Vemco Inc; 2008. Submit your next manuscript to BioMed Central and we will help you at every step: • We accept pre-submission inquiries • Our selector tool helps you to find the most relevant journal • We provide round the clock customer support • Convenient online submission • Thorough peer review • Inclusion in PubMed and all major indexing services • Maximum visibility for your research Submit your manuscript at www.biomedcentral.com/submit

Journal

Animal BiotelemetrySpringer Journals

Published: Jan 28, 2016

References