Internet & Technology

Variation in the performance of acoustic receivers and its implication for positioning algorithms in a riverine setting

Variation in the performance of acoustic receivers and its implication for positioning algorithms in a riverine setting
of 11
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
   Variation in the performance of acoustic receiversand its implication for positioning algorithms in ariverine setting Colin Ashley Simpfendorfer, Michelle RaNae Heupel, and Angela B. Collins Abstract: The performance of an array of data-logging single frequency acoustic receivers in the Caloosahatchee River(Florida, USA) was examined and the results incorporated into a positioning algorithm for animals tracked within thesystem. The mean code detection efficiency across all individual receivers and all download periods was 0.414 detec-tions per synchronization code. On average, the code rejection coefficient was approximately 4%, indicating that it wasonly a minor factor in reducing code detection efficiency. There were significant performance differences betweenstations and download periods, but no interaction between these two factors for all three metrics. Code detection effi-ciency, the rejection coefficient, and the noise quotient all showed significant variations with distance from the rivermouth and time since deployment. Comparison of position estimates with and without efficiency produced small differ-ences for bull sharks ( Carcharhinus leucas ) and cownose rays (  Rhinoptera bonasus ) monitored via this system. Rootmean square errors were higher for cownose rays (48 m) than for bull sharks (23 m). Mean differences for individualswere always slightly downstream because of the increasing code detection efficiency of upriver receivers. The results of this comparison indicated that the inclusion of code detection efficiency did not significantly improve the results of thepositioning algorithm. Résumé : Nous avons évalué la performance d’un réseau de récepteurs acoustiques à fréquence unique avec enregistre-ment des données dans la rivière Caloosahatchee (Floride, É.-U.) et incorporé les résultats dans un algorithme depositionnement pour les animaux suivis par le système. L’efficacité moyenne de détection des codes dans l’ensembledes récepteurs et toutes les périodes d’enregistrement est de 0,414 détections par code de synchronisation. En moyenne,le coefficient de rejet des codes est d’environ 4 %; c’est donc seulement un facteur mineur dans la réduction del’efficacité de la détection des codes. Il y a des différences significatives de performance entre les stations et les pério-des d’enregistrement, mais aucune interaction entre ces deux facteurs pour les trois métriques. L’efficacité de la détec-tion des codes, le coefficient de rejet et le quotient de bruit subissent tous des variations significatives en fonction dela distance de l’embouchure de la rivière et du temps écoulé depuis la mise en opération. Les comparaisons faites entenant compte ou non de l’efficacité des positions estimées de requins bouledogues ( Carcharhinus leucas ) et de raiesmourines (  Rhinoptera bonasus ) suivis par le système montrent de faibles différences. Les erreurs quadratiques moyen-nes sont plus élevées chez les raies mourines (48 m) que chez les requins bouledogues (23 m). Les différences moyen-nes au niveau des individus sont toujours légèrement vers l’aval, parce l’efficacité de détection des récepteurs augmentevers l’amont. Les résultats de cette comparaison indiquent que l’inclusion de l’efficacité de la détection des codesn’améliore pas les résultats de l’algorithme de positionnement.[Traduit par la Rédaction] Simpfendorfer et al.492 Introduction Data-logging single frequency acoustic receivers have be-come a popular tool for studying a variety of aspects of themovement, behavior, and life history of aquatic animals inrecent years (e.g., Klimley and Holloway 1999; Lowe et al.2003; Heupel et al. 2006). As popularity of this tool has in-creased, the variety and complexity of analysis techniquesapplied to receiver data has also increased. However, re-searchers are still discovering the intricacies of the equip- Can. J. Fish. Aquat. Sci. 65 : 482–492(2008) doi:10.1139/F07-180 © 2008 NRC Canada 482 Received 13 January 2007. Accepted 11 August 2007. Published on the NRC Research Press Web site at on 14 February2008.J19760 C.A. Simpfendorfer 1,2 and M.R. Heupel. 2 Center for Shark Research, Mote Marine Laboratory, 1600 Ken Thompson Parkway,Sarasota, FL 34236, USA. A.B. Collins. 3 Center for Shark Research, Mote Marine Laboratory, 1600 Ken Thompson Parkway, Sarasota, FL 34236, USA; andUniversity of South Florida, Department of Biology, 4202 East Fowler Avenue, Tampa, FL 33620, USA. 1 Corresponding author (e-mail: 2 Present address: Fishing and Fisheries Research Centre, School of Earth and Environmental Sciences, James Cook University,Queensland, 4811, Australia. 3 Present address: Florida Fish and Wildlife Conservation Commission, Fish and Wildlife Research Institute, 100 8th Avenue SE,St. Petersburg, FL 33701, USA.  ment and data it collects. Information recently released bythe manufacturer of one of these units has revealed that datarecorded by the units can be used to evaluate receiver per-formance (i.e., the ability to detect codes transmitted by atag) (Heupel et al. 2005).Several factors can affect the performance of receivers.First, large numbers of transmitters within the range of a re-ceiver can lead to high levels of signal collision (Voegeli etal. 1998; Lacroix and Voegeli 2000). The probability of these collisions is a function not only of the number of transmitters, but also the period between signal transmis-sions (which is specified at the time of manufacture) and thetime for signal transmission (which is a function of the cod-ing scheme used by the tag manufacturer). A second causeof reduced performance is a noisy environment at the fre-quency at which the receiver listens. Some of these units lis-ten over a relatively broad frequency range, meaning thatnoise does not necessarily need to be at the exact frequencyof the tags. Noise can be caused by biological componentsof the ecosystem (e.g., snapping shrimp; Heupel et al. 2006),physical factors (e.g., wind generating waves; Voegeli andPincock 1996; Klimley et al. 1998; Voegeli et al. 1998), orhuman influences (e.g., boat motors; Heupel et al. 2006).Third, performance can be reduced by the behavior of tagged animals that leads to rapid changes in the quality of the signal being transmitted. For example, animals that liveamong rocks or reefs may frequently have part of the pulsesequence that makes up the signal blocked, and so not all in-formation for a valid code is received. Fourth, the deploy-ment method of the receiver may affect performance. Forexample, the method and depth of deployment has beenshown to be important (Lacroix and Voegeli 2000), and theorientation of hardware used to moor receivers can interferewith signal reception (Clement et al. 2005). A final cause of reduced efficiency may be heterogeneity of the environmentrelative to the transmission of acoustic signals. For example,estuarine areas with strong stratification and clearly defineddensity gradients that can reflect or refract acoustic signals,areas of high suspended matter (Voegeli and Pincock 1996),or areas with high flow rates that can entrain small air bub-bles into the water (Thorstad et al. 2000; Lacroix et al. 2005)can reduce or interfere with signal transmission. Bottom to-pography may block signals, cause echoes or bend signals,and delay reception times (which for transmitters that usetimes between pulses may cause misidentification of codes).Researchers must therefore have a good understanding of theenvironment in which they are working, how animals willbehave, and how their equipment is set up to fully under-stand how the equipment will function and the type of datathey will collect.Information about the ability to determine performanceenables researchers to investigate if there are differences orchanges in acoustic receivers (e.g., between individual re-ceivers or over time) and so better understand the data col-lected and how it may affect the results of data analysis. Inparticular, analysis of data that requires the comparison of the number of codes received by different receivers mayneed to incorporate performance measures. For example, themean-position algorithm described by Simpfendorfer et al.(2002) that provides estimates of location (latitude and lon-gitude) based on data from arrays of acoustic receivers maybe affected if performance is consistently different betweenstations. This would occur because the locations are calcu-lated by weighting station positions by the number of signalsreceived in a given time period. If performance is differentbetween stations, then a station with better performance ismore likely to hear a transmitter (and so be over-representedin the data) than one with low efficiency. Thus the weight-ing factor in the calculations may need to take account of re-ceiver performance as well as the number of detections tomost accurately calculate location.In this study, the performance of a series of Vemco VR2data-logging acoustic receivers located in the estuarine por-tion of the Caloosahatchee River, Florida, was examined todetermine if there were differences between stations andover time and what factors may account for these differ-ences. A modified form of the mean-position algorithm(Simpfendorfer et al. 2002) that incorporated performancewas developed to describe the movement of tagged animalsalong the river. Finally, the results of a positioning algorithmthat incorporated code detection efficiency and one that didnot were compared for two species of elasmobranchs (bullsharks ( Carcharhinus leucas ) and cownose rays (  Rhinopterabonasus )) to determine if the inclusion of this metric hadany significant affect on the results. Materials and methods Study area The Caloosahatchee River connects Lake Okeechobee tothe southwest coast of Florida, USA (Fig. 1) and is a majorsource of fresh water to the Caloosahatchee Estuary andsouthern Charlotte Harbor. The river and estuary have beensubstantially altered over the past 100 years, including an ar-tificial link to Lake Okeechobee, intricate canal systems, twolocks to allow boat passage, and dams to regulate flow.These alterations have dramatically altered the freshwaterflow in this system, resulting in large fluctuations in salinityin the downstream portion of the river (SFWMD 2000). Thisstudy was completed in the estuarine portion of the river(Fig. 1) and encompassed approximately 28 km of river hab-itat. Upper reaches of the Caloosahatchee Estuary have natu-ral shoreline and native vegetation (primarily red mangroves,  Rhizophora mangle ). Closer to the mouth of the river, thehabitat has been largely altered by urbanization, as evi-denced by extensive canal developments (Fig. 1). Field methods A series of 20 VR2 acoustic receivers (Vemco/Amirix Ltd.)were deployed within the study site to passively track themovement of sharks and rays (Fig. 1). Methods for deployingreceivers have previously been described by Heupel andHueter (2001). Fourteen acoustic receivers were deployed inAugust 2003, and a further six were deployed in October2003. All receivers were continuously present within thestudy site for the duration of the project (through December2005). Receivers recorded the time, date, and identity of ani-mals fitted with acoustic transmitters that swam within rangeof the unit. Stations consisted of a single frequency receiverwith an omnidirectional hydrophone and had an approximatedetection range of 600 m for 16 mm diameter tags and 450 mfor 9 mm diameter tags (M.R. Heupel, unpublished data). © 2008 NRC Canada Simpfendorfer et al. 483  This detection range allowed animals to often be detected atmore than one station simultaneously. The receiver array thusallowed individuals to be continuously monitored for the en-tire period they were present within the study area. Data wereregularly downloaded from receivers in the field, with periodsof 7–42 days between servicing. Hydrophone stations werecleaned of all biofouling when serviced, and batteries werereplaced as required to ensure continuous operation. Waterquality parameters (temperature, salinity, conductivity, anddissolved oxygen) were recorded at the surface and bottom ateach station at the time of downloading using a YSI 85 hand-held meter.Sharks and rays were captured by longline or gill net,weighed, measured, and tagged with a dart tag (sharks) orcinch tag (rays). Sharks were surgically fitted with VemcoV16-4H RCODE 16 mm × 65 mm transmitters (outputpower 153 dB), and rays were externally fitted with VemcoV9-2H RCODE 8 mm × 23 mm transmitters (output power147 dB) (attached to the cinch tag placed through the spira-cle) for passive monitoring via the hydrophone array. Eachtransmitter was coded with a unique pulse series, operatedon 69.0 kHz at randomly spaced intervals between 45 and75 s, and had a battery life of at least 18 months (sharks) or6 months (rays). Random signal transmission times de-creased the probability that the signal from one transmitterwould continuously overlap and block the signal from an-other tag and so prevent detection by a receiver. A total of 36 bull sharks and 10 cownose rays were released into theCaloosahatchee River system by the authors during the studyperiod. In addition, a number of bull sharks and cownoserays released in an adjacent area (Pine Island Sound) movedinto the river, and common snook ( Centropomus unde-cimalis ) released upriver by another research group at timesmoved down river into the system (M.R. Heupel, unpub-lished data). Analysis of receiver performance Data files downloaded from each VR2 receiver containedtwo types of data: ( i ) header information that summarizedstudy details and the functioning of the receiver and( ii ) coded data that detailed the time and date of codes de-tected. The header information contains a number of param-eters that were used to investigate the performance of thereceiver during the period of deployment. These parametersreport metrics associated with the reception of code se-quences. All of the transmitters used in this study had a cod-ing scheme that had 4096 possible combinations using sevenpulses. The seven-pulse code had three sections. First, thetime between the first two pulses (called the synchronizationinterval) defines the coding scheme (in this case the 4096codes). For a code to be recognized, the tag must have avalid synchronization interval that falls within a very tightlydefined value and so is rarely replicated by underwaternoise. The second section of the code sequence, representedby the times between pulses two through five, encodes theunique identification number of the tag. Finally, the timesbetween pulses five and six and between six and seven areused as a checksum to validate the code. If the checksum isinvalid, the code detection is discarded. In addition, it shouldalso be noted that the receiver uses a blanking period(300 ms) after a valid pulse is detected. During the blankinginterval, no pulses will be detected. The blanking period is © 2008 NRC Canada 484 Can. J. Fish. Aquat. Sci. Vol. 65, 2008 Fig. 1. Map of the study site showing the locations (circles) of data-logging acoustic receivers. Inset shows location of study site onthe west coast of Florida, USA.  used to prevent echoes causing false pulses, but has impor-tant implications for the number of pulses recorded whenmultiple tags are transmitting within the range of a receiver.Four code reception parameters in the header informationwere used to calculate performance metrics for this study:( i ) number of valid synchronization intervals (an estimate of how many codes were transmitted, hereafter referred to assynchs); ( ii ) the number of valid pulses detected (hereafterreferred to as pulses); ( iii ) the number of codes rejected be-cause of invalid checksums; and ( iv ) the number of validcode detections. The first two parameters are likely to be un-derestimates of true values because of the use of blankingperiods after pulses are heard, which means that when pulsesequences from two or more tags overlap, there is a proba-bility that some of the pulses will not be heard. In addition,the probability of detecting a code pulse declines with in-creasing distance between the point of transmission and re-ceiver. Thus, tags located further from receivers are likely tohave a reduced chance of being fully detected.Three metrics of receiver performance were calculatedfrom the four code reception parameters (Table 1). Thesemetrics — code detection efficiency, rejection coefficient,and noise quotient — were calculated for each receiver eachtime it was downloaded, and the parameters represented per-formance during the period between the current downloadand the previous download. The code detection efficiencyand rejection coefficient were simple ratios and havestraightforward interpretations. However, the noise quotientis more complex and can be used to identify two situations.First, if there was a large amount of environmental noise thatgenerated pulses that the receiver records, then there wouldbe many more pulses than could possibly be generated bythe tags present (represented by the synchs multiplied by thenumber of pulses in a code), and the value will be a large,positive number. Second, if many tags were present withinthe detection range of a receiver, there would be a high prob-ability of signal collisions and so the number of synchs mul-tiplied by the code length (in this study comprising sevenpulses) would be high relative to the number of pulses, andthe value would be negative. When the only pulses recordedare those from valid code transmissions, the value of thenoise quotient will be zero.Performance metrics were analyzed using a general linearmodel (GLM), with receiver number and download period asfactors, as well as the receiver × period interaction term.When required, log(  x  + 1) transforms were used to normal-ize the performance metrics. If the interaction term was notsignificant ( P > 0.05), then the GLM was rerun without it toimprove the predictive power of the model because of thelarge number of receivers (20) and periods (46) in the analy-sis. Since the noise quotient was not a ratio, its value was atleast in part a function of the time between downloads. Toaccount for this in the GLM, the number of days since thelast download was included as a covariate.To investigate factors that may have influenced the perfor-mance of receivers, regression analysis using the mean val-ues estimated by the GLM was used to determine if therewere significant relationships with a range of factors. Fac-tors included distance from the mouth of the river, numberof different codes received (as an index of the probability of signal collisions), total number of detections, time betweendownloads and cleaning (except for the noise quotient), sa-linity regime (as indicated by the daily mean bottom salinityat the Cape Coral Bridge, 10 km from the river mouth), tem-perature regime (as indicated by the mean temperatureacross all stations on the days of downloading), and riverflow rates (as indicated by the flow rate at the FranklinLocks, 42 km from the river mouth). Where the data sug-gested more complex relationships, multiple regressionswere used to investigate the combined effects of environ-mental factors as well as deployment parameters. Multipleregressions were carried out using stepwise forward inclu-sion of factors. Factors for potential inclusion were timesince deployment, total number of detections for the period,time since last download, temperature, salinity, and freshwa-ter inflow rate. Data on daily salinity and flow rates wereobtained from the South Florida Water Management District,which maintains continuous data-loggers at several locationson the Caloosahatchee River. Estimation of tag position To estimate the position of individuals within the Caloo-sahatchee River, it was assumed that the array was best rep-resented as a linear system, a common approach in riverinesettings (Burrell et al. 2005). The mean-position algorithmdescribed by Simpfendorfer et al. (2002) was adapted byassigning the distance from the mouth of the river of eachstation rather than the latitude or longitude. The distance of each station from the river mouth was calculated using ageographic information system (ArcView 3.3, ESRI Corpo-ration, Redlands, California). In addition, the ability to ac-count for the code detection efficiency of receivers wasincorporated into the calculations. as consistent differencesbetween receivers may bias the probability of detection.The mean position (river distance, X  km ) for a particulartime period (30 min in this study) was  X w X w iiiii km = ∑∑ © 2008 NRC Canada Simpfendorfer et al. 485 Metric Formula CommentsCode detection efficiency (cde) cde = D  /  S  Ranges from 0 to 1Rejection coefficient (rc) rc = C   /  S  Ranges from 0 to 1Noise quotient (nq) nq = P – ( S  ·cl) Can be positive or negative Note: D , number of valid detections; S  , number of synchs; C  , number of codes rejected because of invalid checksums; P , number of pulses detected; and cl, number of pulses used to make a valid code(for the tags used in this study, cl = 7). Table 1. Metrics used to evaluate the performance of Vemco VR2 acoustic receivers in theCaloosahatchee River estuary.  where X  i was the distance from station i to the river mouth,and w i was the weighting factor that incorporated the num-ber of valid signals received at each station during the timeperiod covered by the calculations and the code detection ef-ficiency of the receivers. There were two potential ways tocalculate w i depending on how efficiency changed over time.If there was no change in a receiver’s code detection effi-ciency (cde) over time, or if the change in efficiency of eachstation was the same over time (i.e., the interaction term inthe GLM was not significant), then w n i i i = /( )cdewhere n i is the number of valid signals received at station i during the time period, and cde i is the efficiency of receiver i across all periods. The mean station values estimated by theGLM were used to represent cde i . However, in the situationwhere a significant interaction in code detection efficiencyby station and time occurs (i.e., the efficiency of stationsdoes not change in a consistent fashion over time), then w n it i it  = /( )cdewhere cde it  is the code detection efficiency of station i dur-ing period t  . Mean values of interaction terms would be usedto represent cde it  .To determine the level of influence of inclusion of codedetection efficiency in mean position estimates of river dis-tance, the algorithm was also run assuming that the effi-ciency of all stations had a constant value of 1.0. Thisanalysis was carried out for four bull sharks that were resi-dent for periods longer than 4 months and for three cownoserays that were present for periods longer than 1 month. Dif-ferences between estimates were calculated from each indi-vidual and for each species combined. Root mean squareerrors (RMSE) were also calculated in the same way. Results Performance metrics Code detection efficiency at each download across all re-ceivers varied considerably, from 0.011 to 0.757 detections·synch –1 . The mean number of detections per synch was0.414, suggesting that on average, less than half of the codestransmitted were detected. There was no significant interac-tion in detection efficiency between receiver number and pe-riod (GLM, F  = 0.368, df = 701, P = 0.368), indicating thatchanges in this metric at receivers were consistent over time.The GLM without the interaction term indicated that therewere significant differences in code detection efficiency byreceiver (Fig. 2 a ; F  = 15.56, df = 19, P < 0.0001) and period(Fig. 3 a ; F  = 8.95, df = 45, P < 0.0001). Code detection effi-ciency increased significantly with increased distance fromthe mouth of the river (Table 2). The only significantchanges in code detection efficiency over time were de-creases with time since initial deployment and with timesince the previous download. There were no significant lin-ear relationships with river flow, salinity regime, or tempera-ture.The rejection coefficient was consistently low, rangingfrom 0.000 to 0.110 rejections·synch –1 . The mean rate of code rejection due to invalid checksum values was 0.041.The distribution of the rejection coefficient was skewed, anda log(  x  + 1) transform was applied to normalize the data.There was no significant interaction in the rejection coeffi-cient between receiver number and period (GLM, F  = 4.09,df = 759, P = 0.217). The GLM of the rejection coefficientwithout the interaction term indicated that there were signifi-cant differences in the rejection coefficient between receivernumber (Fig. 2 b ; F  = 3.82, df = 19, P < 0.0001) and period(Fig. 3 b ; F  = 5.37, df = 45, P < 0.0001). There was a signifi- © 2008 NRC Canada 486 Can. J. Fish. Aquat. Sci. Vol. 65, 2008 Fig. 2. Variations in receiver performance metrics with distancefrom the mouth of the Caloosahatchee River: ( a ) code detectionefficiency, ( b ) rejection coefficient, and ( c ) noise quotient. Pointsindicate mean values estimated by a general linear model withreceiver number and period as factors; error bars indicate onestandard error.
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!