Methods in Ecology and Evolution

Accepted Article
MS KIM WHORISKEY (Orcid ID : 0000-0001-7520-1016)
DR MARIE AUGER-MÉTHÉ (Orcid ID : 0000-0003-3550-4930)
DR LEE FG GUTOWSKY (Orcid ID : 0000-0003-1244-9465)
MR ROBERT J LENNOX (Orcid ID : 0000-0003-1010-0577)

Article type : Review

Editor : Luca Börger

Current and emerging statistical techniques for aquatic telemetry data: A guide

to analyzing spatially discrete animal detections

Kim Whoriskey1*, Eduardo G. Martins2, Marie Auger-Méthé3,4, Lee F.G. Gutowsky5,6,

Robert J. Lennox5,7, Steven J. Cooke5, Michael Power8, and Joanna Mills Flemming1

1
Department of Mathematics and Statistics, Dalhousie University, Halifax, B3H 4R2, Canada
2
Ecosystem Science and Management Program, University of Northern British Columbia,

Prince George, V2N 4Z9, Canada

3
Department of Statistics, University of British Columbia, Vancouver, V6T 1Z4, Canada
4
Institute for the Oceans and Fisheries, University of British Columbia, AERL, Vancouver,

V6T 1Z4, Canada

5
Fish Ecology and Conservation Physiology Laboratory, Department of Biology, Carleton

University, Ottawa, K1S 5B6, Canada

6
Aquatic Resource and Monitoring Section, Ontario Ministry of Natural Resources and

Forestry, Peterborough, K9L 1Z8, Canada

This article has been accepted for publication and undergone full peer review but has not
been through the copyediting, typesetting, pagination and proofreading process, which may
lead to differences between this version and the Version of Record. Please cite this article as
doi: 10.1111/2041-210X.13188
This article is protected by copyright. All rights reserved.
 7
NORCE Norwegian Research Centre Laboratory for Freshwater Ecology and Inland

Fisheries, Nygårdsporten 112, 5006 Bergen, Norway

Accepted Article
8
Department of Biology, University of Waterloo, Waterloo, N2L 3G1, Canada

*
Corresponding author. Email: [email protected]

Abstract

1. Telemetry, or the remote monitoring of animals with electronic transmitters and

receivers, has vastly enhanced our ability to study aquatic animals. Radio telemetry,

acoustic telemetry, and passive integrated transponders are three common

technologies that generate detection data – time-stamped tag-specific records that are

logged by receivers.

2. We review current statistical methods and comment on potential future directions for

analyzing detection data derived from fixed telemetry receiver arrays.

3. To illustrate how different methods may be used to achieve diverse study objectives,

we provide a case study dataset collected by an array of 42 acoustic telemetry

receivers on 187 bull trout in the Kinbasket Reservoir of British Columbia. To close,

we present a decision tree for guiding the selection of a method based on study

objectives and sampling design.

4. This paper provides both experienced and novice telemetry researchers with the

knowledge and tools to facilitate more comprehensive analysis of detection data and,

in so doing, ask a wide variety of ecological questions that will enhance our

understanding of aquatic organisms.

This article is protected by copyright. All rights reserved.

 Keywords

movement ecology; detection data; statistical methods; acoustic telemetry; radio telemetry;
Accepted Article
PIT tag; Ocean Tracking Network

Introduction

Aquatic animals live in habitats that create inherent challenges for those attempting to

study their ecology, behaviour, and physiology. Telemetry enables the remote monitoring of

free-living animals, whereby a signal emanating from a device (i.e., transmitter or tag) carried

by an animal transfers information to a receiver. The advent of telemetry tools has provided

researchers with effective means of studying aquatic animals in the streams, rivers, lakes,

estuaries, and oceans of the world (Lucas & Baras 2000; Hussey et al. 2015).

Three common telemetry technologies used with aquatic animals are radio and

acoustic telemetry, and passive integrated transponders (PIT). Radio telemetry uses radio

signals that are detected by an antenna affixed to a receiver, whereas acoustic telemetry uses

sound waves to transmit tag information to a hydrophone on a receiver. The transmitters of

both technologies are dependent on internal batteries that, along with the tag-animal size ratio

and tag settings, limit the duration of data collection. PIT tags rely on external energy derived

from an electromagnetic field emitted by receiver antennas, which prolongs the tag lifespan

but requires close proximity (Lucas & Baras 2000). Despite design differences (Lucas &

Baras 2000; Cooke et al. 2012), these three telemetry technologies all record one specific

kind of data: detection data that consist of time-stamped, tag-specific records registered and

stored by receivers when tagged animals are within range.

This article is protected by copyright. All rights reserved.

 Recently, a shift from mobile tracking towards using fixed receiving stations that

automatically log detections has led to a large number of tagged animals and extensive
Accepted Article
receiver coverage crossing geopolitical boundaries (Donaldson et al. 2014). The collection

and aggregation of large aquatic detection datasets has created both challenges and

opportunities for the study of wild aquatic animals (Lennox et al. 2017a). Although there

have been substantial developments in the statistical analysis of aquatic detection data, to our

knowledge there have been no attempts to synthesize the existing and emerging methods. Our

goal is to provide this synthesis. Although the methods we review are the most ubiquitous

(today), they are not exhaustive. In particular, because detection data are limited to collection

at discrete locations, we do not review methods for spatially continuous data (e.g., movement

paths collected by satellite telemetry devices). It is possible to obtain estimates of spatially

continuous data from detection data using positioning systems (e.g., Niezgoda et al. 2002;

Smith 2013) or by calculating centers of activity (Simpfendorfer, Heupel & Hueter 2002), in

which case other statistical methods not reviewed herein may be used, e.g., home range

analysis (Marshell et al. 2011), state-space models (Martins et al. 2014), or hidden Markov

models (Whoriskey et al. 2017). We also do not discuss software designed primarily for the

data management and visualization of aquatic detection data. These developments, e.g., the

Ocean Tracking Network Toolbox ([email protected]), ZoaTrack (Dwyer et al. 2015), the

Integrated Marine Observing System’s Animal Tracking Facility detection database (Hoenner

et al. 2018), and the R (R Core Team 2018) packages glatos (Holbrook, Hayden & Binder

2017), and VTrack (Campbell et al. 2012), provide high-quality standardized methods for

handling detection data; however, they typically do not incorporate a stochastic component.

This article is protected by copyright. All rights reserved.

 First, we review statistical methods for detection data derived from fixed telemetry

arrays in aquatic environments. To illustrate the differences between statistical methods,

Accepted Article
throughout the review we analyze a portion of a dataset collected on acoustically tagged bull

trout (Salvelinus confluentus). Then, we comment on potential future directions that could

help advance our understanding of how aquatic animals interact with each other, their

environment, and humans in a rapidly changing world. To close, we present a decision tree to

summarize the differences among the statistical methods and to help guide researchers on

how to analyze their detection data given the scientific questions of interest and sampling

design.

Illustrative Dataset

Between 2010-2012, 187 bull trout were acoustically tagged and monitored by an

array of 42 receivers deployed in the Kinbasket Reservoir of British Columbia, Canada

(Figure 1). The full dataset was previously analyzed in Martins et al. (2013) and Gutowsky et

al. (2016); for simplicity, we chose to analyze data collected only during January 2011. The

resulting dataset comprised three files: receiver metadata, that includes the identities and

locations of the deployed receivers, along with environmental information; tag metadata, that

consists of the unique tag id codes and other animal characteristics (e.g., length/weight/sex);

and detection data, i.e., the records of tags registered by receivers at a specific date and time.

Together, these data (hereafter “detection data”) provide a comprehensive view on individual

movements.

For any telemetry study, the question of interest and the spatiotemporal design of the

receiver deployments will influence the applicability of various statistical methods. Once a

method has been chosen, the detection data will need to be summarized into an appropriate

This article is protected by copyright. All rights reserved.

 response variable ( ). Examples include: counts of detections (Zhang et al. 2015), counts or

proportions of receivers visited within a specific time scale (Udyawer et al. 2015),
Accepted Article
presence/absence data (Dudgeon, Lanyon & Semmens 2013; Kessel et al. 2014a), time spent

in particular areas or residency indices (Ketchum et al. 2014; Kessel et al. 2014a), and

movement rates (Stich et al. 2015). We discuss the form of the response variable for each

method reviewed below, and use the Kinbasket dataset to illustrate the versatility of detection

data.

Review of Current Statistical Methods

Generalized Modelling Framework

Researchers who use telemetry are often interested in determining whether there is a

relationship between animal movement patterns and a set of putative explanatory variables or

covariates. Because many of the possible response variables are non-Gaussian, traditional

statistical methods like analysis of variance and linear regression are not directly applicable.

Generalized linear models (GLMs) enable the modelling of non-Gaussian response variables

provided they follow a distribution belonging to the exponential family (Wood 2006). A

GLM links an observation to a set of covariates :

, (1)

where denotes the expectation of a random variable, is a monotonic link function,

and the vector contains entries that describe the relationship between and the

covariates (plus an intercept) contained in each row vector (Wood 2006).

This article is protected by copyright. All rights reserved.

 Because telemetry does not directly measure animal absence, researchers must decide

whether to interpret a lack of detections as absence and encode them as zeros within a
Accepted Article
dataset. The temporal resolution of the study directly affects the number of zeros in the

response, whereby many zeros will be included if animals are rarely detected over numerous

short time intervals. Furthermore, environmental features like topography, weather, and

biological noise, as well as collisions with other telemetry transmissions, can lead to false

absences (Cagua, Berumen & Tyler 2013). A dataset will be more difficult to accurately

model when the number of observed zeros is substantially greater than the number predicted;

such models may show evidence of overdispersion (when the response variance is larger than

expected) or lack of fit (Zuur et al. 2009). In these cases, zero-inflated models may provide

more accurate results (Zuur et al. 2009).

Both discrete and continuous covariates can be included in GLMs if they are linearly

related to the response. For non-linear relationships, generalized additive models (GAMs)

relate the response and covariates using a sum of smooth functions of the variables

(Wood 2006), e.g.,

(2)

Detection data have been related to covariates like lunar phase and tidal stage (Dudgeon,

Lanyon & Semmens 2013), water temperature (Kessel et al. 2014a; Udyawer et al. 2015),

discharge (Richard et al. 2014; Stich et al. 2015), and diel period (Ketchum et al. 2014;

Zhang et al. 2015). Temporal data can also be used, often by summarizing the response into

temporal blocks and including the blocks as a covariate. Blocks can be defined based on

species ecology (e.g., reproductive timing), or anthropogenically (e.g., by monthly intervals;

Matich & Heithaus 2014). When investigating a temporal trend in the response, temporal

autocorrelation should be checked and accounted for if the assumption of independence is

This article is protected by copyright. All rights reserved.

 violated, e.g., by incorporating lagged temporal variables (Kessel et al. 2014a) or including a

variance structure (Börger et al. 2006).

Accepted Article
Because most detection data are collected under the largely uncontrolled conditions of

the natural environment, some responses may only be independent when conditioned upon

other variables. These variables, also known as random effects, can be accounted for by

incorporating a second stochastic term into GLMs and GAMs to form generalized linear

mixed models (GLMMs) and generalized additive mixed models (GAMMs; Wood 2006). In

practice, random effects are often included to account for variation within sampling units. For

example, detection data are usually collected on a random subset of individuals from a

population. To conduct population-level inference, individual ID can be included as a random

effect with either, or both, an intercept and slope (Bolker et al. 2009). Random effects can

also be associated with space or time, e.g., receiver location (Ketchum et al. 2014) or age and

sampling year (Börger et al. 2006).

Generalized models were used to assess the factors affecting spatial distribution and

movement of bull trout in the full complement of the illustrative dataset (Gutowsky et al.

2016). Using a GLMM, Gutowsky et al. (2016) assessed the effects of year, season, sex, and

body size (covariates in ) on home range size (response ; 95% minimum convex polygon).

A GAMM was used to quantify the relationship between total displacement (response ; sum

of distances between receivers) and sex, body size, and smoothed month. Larger coefficient

values for spring and fall suggested that bull trout home ranges were larger in those seasons

than in winter and summer. Additionally, a positive sex-size interaction term suggested that

larger females moved farther than smaller females.

This article is protected by copyright. All rights reserved.

 Survival (time-to-event) analysis

Telemetry measures animal positions over time and changes in position can be related
Accepted Article
to important ecological events. For example, tagged animals may disperse or migrate

(Kawabata et al. 2010), interact with humans (Thorley, Youngson & Laughton 2007), pass an

obstacle (Castro-Santos & Haro 2003; Naughton et al. 2005; Martins et al. 2013), be

depredated (Danylchuk et al. 2007; Lennox et al. 2017b), or die (Curtis et al. 2015). These

events can be analyzed with GLMs using a binomial response, where study animals are

grouped into those that experience an event and those that do not. Survival analysis extends

the response by incorporating the time it takes for the event to occur (e.g., ) and

estimates the survival function, ,

, (4)

which describes the probability, , that an event will occur at some random time after

the set time (Pollock et al. 1989; Klein & Moeschberger 2003). The most common survival

function estimators are the non-parametric Kaplan-Meier and Nelson-Aalen estimators (Klein

& Moeschberger 2003). A log-rank test can be used to compare the estimated survival curves

of different groups (e.g., sex and reproductive state or moult stage; Pollock et al. 1989;

Huserbråten et al. 2013).

Further inference is possible with the hazard function , which describes the

conditional rate of an event occurring during a period of time { given that it has

not already been experienced (Klein & Moeschberger 2003):

. (5)

When the shape of the hazard or survival function is assumed, parametric survival analysis

can be performed with error distributions (e.g., Weibull) and this allows for predictive

extrapolation (Benoît et al. 2015). However, selecting a parametric hazard function requires

accurate knowledge of the true shape, which is not often known (White & Garrott 1990;

This article is protected by copyright. All rights reserved.

 Murray 2006). Consequently, using semi-parametric Cox proportional hazards regression can

be advantageous because there is no assumption about the hazard shape, yet the response can
Accepted Article
still be compared to a set of covariates (Murray 2006; Harrell 2015),

. (6)

The hazard is related to an arbitrary baseline hazard that is treated non-

parametrically, and a known parametric function of the covariates and their coefficients

(Klein & Moeschberger 2003). The Cox proportional hazards model is also less sensitive to

outlying observations than parametric models, but does require hazard proportionality which

can be verified graphically or by testing for independence between Schoenfeld residuals and

time (Harrell 2015). Violations of this assumption may be compensated for by fitting

stratified models (Harrell 2015).

In telemetry studies, animals often go undetected for extended periods, either because

they leave the detection range of the array or because they are inactive. The resulting

monitoring gaps can cause discontinuity in the hazard function (Murray 2006) and bias

survival estimates (Bunck, Chen & Pollock 1995). The Andersen-Gill estimator (Andersen &

Gill 1982) is a variation of the Cox proportional hazards model that uses a counting process

to account for discontinuous monitoring (Murray 2006; see e.g., Johnson et al. 2004). In

addition, individuals that fully drop out of the study before the event occurs can be censored

from survival analysis techniques without having to be removed entirely (Pollock,

Winterstein & Conroy 1989). For example, Topping and Szedlmayer (2011) used survival

analysis to study the residency time (event = emigration) of red snapper (Lutjanus

campechanus), and censored fish that either died before emigration or did not emigrate in

order to retain them in the analysis.

This article is protected by copyright. All rights reserved.

 An example of survival analysis using detection data can be found in Martins et al.

(2013), where the Kaplan-Meier estimator was used to compute the risk of bull trout
Accepted Article
unintentionally passing through hydro-electric dam turbines (the event of interest) from the

full Kinbasket dataset. The Kaplan-Meier estimator exhibits larger jumps in the survival

curve for the fall and winter, suggesting that the risk of passing through the dam was higher

during those seasons.

Mark-recapture models

Mark-recapture models are used for estimating movement or demographic attributes,

e.g., abundance or survival. These models are fit to data collected by capturing and marking a

sample of animals from a population, subsequent release, and re-sampling such that

additional samples can include both marked and unmarked animals (Amstrup, McDonald &

Manly 2005). When using telemetry, mark-recapture models are applicable if the tagging

procedure is considered the marking process, and detections are the recaptures. Few telemetry

studies record the presence of untagged animals (but see Dudgeon et al. 2015), therefore the

most applicable mark-recapture models incorporate data collected on tagged animals only,

which include known-fate, live-recapture, and recovery models (Lindberg 2012).

Known fate models assume perfect detection probabilities, which rarely occur in

telemetry (Melnychuk 2012). Alternatively, live-recapture models are highly applicable for

analyzing detection data because they enable the joint estimation of detection probability and

demographic quantities. Among live-recapture models, the Cormack-Jolly-Seber model is

frequently used, often to estimate survival along migratory routes (e.g., Welch et al. 2009;

Moore et al. 2015). The Cormack-Jolly-Seber model is fitted using a multinomial likelihood

with two basic parameters: is the probability that an individual survives between

This article is protected by copyright. All rights reserved.

 detections, and is the probability that an individual is detected if alive and marked

(Amstrup, McDonald & Manly 2005). The response variable consists of a binary encounter
Accepted Article
history (absence = 0 and presence = 1) for every marked animal that is recorded on a discrete

temporal scale chosen by the researcher. If a single animal’s encounter history is encoded as

1101, where the first digit is the initial capture and tagging, then the associated encounter

probability would be

, (7)

if and are assumed constant through time, or

, (8)

if these probabilities are allowed to vary. Because the Cormack-Jolly-Seber model cannot

distinguish between mortality and emigration, survival estimates are more appropriately

termed apparent survival (Williams, Nichols & Conroy 2001).

A useful extension of the Cormack-Jolly-Seber model is the multi-state Arnason-

Schwarz model, which estimates survival and detection probabilities as a function of an

observed animal state (Amstrup, McDonald & Manly 2005). The states are assumed to follow

a first-order Markov process governed by transition probabilities (Amstrup, McDonald &

Manly 2005). It has been applied to detection data to estimate daily probabilities of horseshoe

crab spawning (Limulus polyphemus; reproductive state; Brousseau et al. 2004), survival of

downstream migrating Atlantic salmon (Salmo salar; location state; Holbrook, Kinnison &

Zydlewski 2011), and movement probabilities along walleye (Sander vitreus) migratory

routes (location state; Hayden et al. 2014).

Recovery mark-recapture models are useful when the recapture process is terminal

(Lindberg 2012). Information on deceased individuals can be jointly modeled with live

detection data using the Burnham model (Burnham 1993), which has been used to provide

more precise survival estimates (Sollmann et al. 2010) and to estimate the joint probability

This article is protected by copyright. All rights reserved.

 that tagged individuals were caught and reported (Martins et al. 2011). The Barker model is

an extension of the Burnham model useful for analyzing temporally continuous detection data
Accepted Article
(Barker 1997), and has been used to estimate the effects of gastric lavage on common snook

survival (Centropomus undecimalis; Barbour, Boucek & Adams 2012). Finally, Fouchet et al.

(2016) proposed an approach for temporally continuous data that combines survival analysis

with an inhomogenous Poisson process for modelling detection probability.

We fitted several Cormack-Jolly-Seber models using MARK (White & Burnham 1999)

and RMark (Laake 2013) to test whether sex or length were associated with weekly bull trout

survival, and whether the receiver array detection probability changed over time. We

compared candidate models using corrected Akaike’s information criterion (AICc), and

found that the best model estimated intercepts only for both survival and the detection

probability (Table 1). The apparent weekly survival probability was estimated at 0.91. In

addition, the detection probability was estimated at 0.69, which suggests that a combination

of receiver coverage/efficiency, environmental conditions, and fish behaviour limited the

array’s ability to detect bull trout.

Network analysis

Networks are mathematical objects consisting of nodes connected by edges (Dale &

Fortin 2010). They can be used to study animal movement by analyzing the relationships

between nodes, which can represent receivers or tagged animals separately (unipartite graphs;

e.g., Jacoby et al. 2012) or in the same graph (bipartite graphs; e.g., Finn et al. 2014). To

study movement, nodes are often specified as the stationary receivers and edges represent

either the directed or undirected movements of animals between receivers. Social

aggregations can also be studied when the animals are treated as nodes, e.g., by testing

whether there exist preferred associations among individuals (Stehfast et al. 2013).

This article is protected by copyright. All rights reserved.

 A network’s response variable is an adjacency matrix, which describes the

connections between pairs of nodes (Farine & Whitehead 2015). For example, the adjacency
Accepted Article
matrix

(9)

R1 R2 R3

describes a system of three receivers ( and ), where three movements were

recorded from to and four records indicate animals staying at . Adjacency matrices

for null networks can also be of interest, e.g., to document potential direct routes among

receivers. For the above example, the following null network

(10)

R1 R2 R3

indicates that movements are possible between all pairs of receivers except from to in

either direction. Once the adjacency matrix is defined, visuals and metrics can be calculated

that describe the network connectivity (Dale & Fortin 2010). For example, the node degree is

the number of incoming and outgoing edges of a node, which describes the amount of traffic

through a receiver and therefore may indicate areas of importance (Jacoby et al. 2012; Farine

& Whitehead 2015), whereas edge density is the fraction of observed edges to all

theoretically possible edges, and can help indicate the amount of random/non-random

movement (Jacoby et al. 2012).

Network metrics can be compared amongst groups, e.g., to determine sex-specific

differences in movement (Jacoby et al. 2012) or preferred areas (Stehfast et al. 2015). When

the groups are not necessarily known a priori, community detection algorithms can be used to

identify groups of receivers or animals that are closely related, for example to identify home

ranges (Finn et al. 2014). If an observed network can be classified as a theoretical network

This article is protected by copyright. All rights reserved.

 pattern, then known properties can be interpreted (e.g., Fox and Bellwood 2014). Finally,

disrupting the networks by removing nodes and studying the subsequent network
Accepted Article
fragmentation can help to assess the effects of habitat disruption (Jacoby et al. 2012) and the

protective capabilities of potential marine reserves (Espinoza et al. 2015).

Direct hypothesis testing on network measures is possible using GLMMs, but the

assumption of independence may be violated (Farine & Whitehead 2015). Permutation and

randomization techniques provide non-parametric methods for hypothesis testing by

comparing an observed statistic to those calculated from randomly generated networks (Dale

& Fortin 2010), and have been used to assess whether animals are moving randomly

(Espinoza et al. 2015). In addition, networks can be compared to each other or other dyadic

variables using a Mantel test which assesses the correlation between two matrices (Urban et

al. 2009; Farine & Whitehead 2015), e.g., to test whether yellowfin tuna (Thunnus albacares)

social associations are related to pre-defined cohorts (Stehfast et al. 2013). Relationships

between networks and more than one dependent variable can be evaluated using the multiple

regression quadratic assignment procedure (Farine & Whitehead 2015), which has been used

to assess the effect of environmental variables on small-spotted catshark (Scyliorhinus

canicula) movements (Jacoby et al. 2012).

We applied network analysis to the bull trout dataset (Figure 2) after summarizing

detection data into directed movements between pairs of receivers. Using the R package

igraph (Csárdi & Nepusz 2006) and treating the receivers as nodes, we plotted a network

for each sex making the size of each node proportional to its degree and using weighted edges

to represent the number of directed movements between nodes. These networks suggest that

the main pool of the reservoir experiences more fish traffic compared to either of the reaches,

and therefore likely contains important bull trout habitat. In addition, a Mantel test between

This article is protected by copyright. All rights reserved.

 the two networks suggested a weak but statistically significant (r=0.17, P<0.05) correlation in

movement patterns between the males and females.

Accepted Article
Future Directions

Gaussian random fields

Gaussian random fields (GRFs) are a promising approach for analyzing detection data

within a spatial context. Specifically, GRFs estimate the residual spatial correlation

remaining after accounting for measured explanatory variables (Thorson & Minto 2015). In

fisheries, they have been used to model the spatial dependence of population processes and to

understand the relationship between fish distribution and habitat (Thorson & Minto 2015;

Thorson et al. 2015; Carson, Shackell & Mills Flemming 2017). With telemetry data, GRFs

have been used to show how the number of at-sea seal encounters co-varied with bathymetry

and distance to the seal haul-out site (Carson & Mills Flemming 2014). In that study, the

receiver locations changed through time; here, we demonstrate the potential of GRFs for

detection data collected at fixed locations by investigating whether the presence of the dam

affects bull trout distribution using the illustrative dataset. Because the GRF is a flexible

hierarchical model, these data could have been modeled in several different ways. For

example, we could have: modeled the duration or number of detections (e.g., Carson & Mills

Flemming 2014); accounted for false absences by using a zero-inflated distribution (e.g.,

Cosandey-Godin et al. 2015); or used a state-space model to account for technological error

(e.g., Thorson et al. 2015).

We assumed that the number of individuals detected at a given receiver location

(of which there are ) was Poisson distributed:

. (11)

This article is protected by copyright. All rights reserved.

 We linked the mean of the distribution, , to the linear predictor, , through a log-link

function:
Accepted Article
. (12)

In turn, was related to the distance between the receiver and the dam through the linear

equation:

, (13)

where represents the log of the number of detections expected when the distance to the

dam ( ) and the spatial random effect ( both have no effect, and is the regression

coefficient for . The random effect accounts for the effect of unknown spatial factors

influencing the response and we model it as a GRF, meaning that for any ,

where D is the domain, we let be distributed as a multivariate normal:

. (14)

Here, is a n x n covariance matrix where the (i,j)th element of is defined by the Matérn

covariance structure, which for is defined as:

, (15)

where is the Euclidean distance between receiver locations and , the smoothness

parameter is set equal to 1, and the spatial scale  and marginal variance are both

estimated. As in Thorson et al. (2015), we used R-INLA (Illian, Sørbye & Rue 2012) to

simplify model implementation with stochastic partial differential equations (SPDEs) and

TMB (Kristensen et al. 2016) to estimate the model parameters.

Model fitting resulted in parameter estimates of (95%CI: 1.73-3.12) and

(95% CI: -0.063 – -0.022), indicating that as distance to the dam increased the

number of individuals detected decreased. This may result in part because the distance from

the dam increases proportional to the distance from the main lacustrine habitat for most

This article is protected by copyright. All rights reserved.

 locations within the reservoir. In addition, the spatial correlation of the GRF accounts for

some of the bull trout distribution not explained by distance from the dam (Figure 3).
Accepted Article
Accounting for spatial correlation and measurement error

We expect that methods for estimating spatial correlation associated with animal

movement will grow in popularity as receiver coverage and method documentation continue

to expand. While we proposed the GRF as a flexible method for modelling animal movement

data with a spatial component, other spatial methods exist that could be applied to detection

data. For example, one terrestrial study used a spatial (and temporal) correlation structure

within a GLMM to assess the factors affecting home range size of radio-tracked roe deer

(Capreolus capreolus; although home range in this case was calculated from spatially

continuous detection data, similar principles would apply to responses calculated from

discrete detection data; Börger et al. 2006). Network autocorrelation models (Leenders 2002)

could be used to estimate the correlation between network attributes caused by receiver

location. In addition, spatial capture-recapture models are a spatial extension of mark-

recapture models, and are well established in terrestrial studies with encounter data like those

generated by camera trapping (Royle et al. 2014). These models involve hierarchical

modelling of a spatial point process of unobserved animal activity centres and a detection

probability function depending on distance from the activity centres (Efford & Fewster

2013). Despite similarities with terrestrial encounter data, we have seen few studies that

apply spatial capture-recapture methods to aquatic detection data, but see Raabe, Gardner &

Hightower (2014), who studied the survival and movement of PIT-tagged American shad

(Alosa sapidissima) using these methods.

This article is protected by copyright. All rights reserved.

 Many studies have investigated the measurement error of aquatic tracking technology by

estimating the detection efficiency (the frequency with which a receiver will detect a fish
Accepted Article
within its given range; Simpfendorfer, Heupel & Collins 2008), detection range (the

probability of detection given distance from a receiver; Kessel et al. 2014b), or the frequency

of false detections (when a receiver logs a false ID or detects an absent animal; Heupel,

Semmens & Hobday 2006). However, few studies incorporate this information into their

biological inferences. Those that do may use it to pre-process their data (e.g., Kessel et al.

2014a, Hoenner et al. 2018), or directly incorporate measurement error into the statistical

method (Simpfendorfer, Heupel & Collins 2008; Pedersen & Weng 2013; Winton et al.

2018). Measurement error can additionally be used to help numerically optimize the

spatiotemporal design of a receiver array before deployment, resulting in a study design with

enhanced ability to acquire high-quality data (Pedersen, Burgess & Weng 2014).

State-space models are hierarchical models that can pair a measurement equation with a

model for animal movement, and simultaneously estimate both processes (Auger-Méthé et al.

2017). Two notable examples with detection data include: 1) a nonparametric function for

detection probability paired with an Ornstein-Uhlenbeck movement process to estimate the

home range of a humphead wrasse (Cheilinus undulatus; Pedersen & Weng 2013); and 2) a

Gaussian decay measurement equation coupled with a binomial spatial point process to

estimate centres of activity of a black sea bass (Centropristis striata; Winton et al. 2018).

State-space models have gained popularity for analyzing spatially continuous animal

movement data, likely because of their flexibility – multiple measurement error distributions

can be included and matched specifically to the tracking technology (e.g., Winship et al.

2012), and the movement process can range from individual models of movement (e.g.,

Auger-Méthé et al. 2017) to GRFs (e.g., Thorson et al. 2015). We believe that state-space

This article is protected by copyright. All rights reserved.

 models could provide a framework that improves the reliability of statistical analyses of

detection data.
Accepted Article
Broadening the scope of animal movement analyses

Telemetry technology will continue to improve technically in ways that will increase

study longevity, target more species or life stages, and expand the scope of data collection

(Lennox et al. 2017a). Study designs will also evolve, as auxiliary biological and

environmental variables are collected during sampling or independently, and as telemetry

networks facilitate the sharing of resources and multi-species data (Lennox et al. 2017a). As a

result, telemetry studies will have the potential to generate massive, interdisciplinary datasets,

and statistical methods for analyzing such complex data will have to adapt appropriately. In

the future, movement ecologists may look to the burgeoning research field of human

mobility, which has exploded since the advent of the smartphone with GPS tracking and

geolocated social media postings (Thums et al. 2018). Because humans and non-human

animals appear to conform to similar ecological principles, e.g., site fidelity, aggregation, and

sociality (Meekan et al. 2017), movement ecologists will have the opportunity to readily

appropriate big data approaches from human mobility studies (Thums et al. 2018).

Discussion

To aid researchers in matching a statistical method to their data and study objective,

we devised a decision tree which we present in Figure 4. We recognize that accounting for

every possible study design would be unrealistic, therefore we suggest that researchers utilize

our decision tree as a first general guide through some of the possible statistical methods, not

an exhaustive instruction catalog. We hope that a tree will help researchers narrow their

selection, but then we strongly suggest that this is followed by comprehensive study of the

This article is protected by copyright. All rights reserved.

 chosen method(s), and its(their) accompanying assumptions. To that end, we summarize our

guide below.
Accepted Article
We suggest using mark-recapture methods when studying population dynamics,

especially when the detection ability of the array is suspect. However, note that the most

commonly used mark-recapture method, the Cormack-Jolly-Seber model, is dependent upon

the following assumptions: 1) tags are not lost and do not fail; 2) survival is not influenced by

the tag or tagging procedure; and 3) survival and detection probability do not vary among

tagged animals. Preliminary laboratory studies assessing tag attachment and retention can

help to determine the risk of tag loss or failure (Holbrook et al. 2013). Pilot studies can be

used to assess whether tagging influences survival (Furey et al. 2016); however, tagging may

negatively affect multiple traits in a cumulative way such that the full influence is not

understood by assessing the effect on survival alone (Bodey et al. 2018). In fact, tagging can

affect traits like growth, swimming performance, and social interactions (Jepsen et al. 2015),

and these potential effects should be carefully considered in any analysis. Finally, some of the

factors affecting individual variation in survival and detection (e.g., sex, age) can be

incorporated into the Cormack-Jolly-Seber model through stratification or regression analysis

(Williams et al. 2001).

To understand the occurrence of an event when temporal records for the event exist,

consider using survival analysis. Survival analysis can be used to understand the survival of

tagged animals; however, it is distinguishable from mark-recapture via their response

variables. Survival analysis requires a temporal value for the response (e.g., ),

whereas mark-recapture uses a discrete time series (e.g., 1001101), and time is often

incorporated by allowing probabilities to be dynamic. Although a temporal response can also

This article is protected by copyright. All rights reserved.

 be modelled using GLMs/GLMMs/GAMs/GAMMs, survival analysis accounts for the fact

that for some study animals it may not be possible to determine whether they experience the
Accepted Article
event. Censoring is appropriate as long as the probability of being censored is independent of

the probability of the event (Harrell 2015).

To describe the connectivity among receiver locations or tagged animals, we

encourage the use of network analysis which provides easily interpretable visualizations of

this connectivity. However, network analysis does assume that all the nodes of a system are

represented in the graph (Dale & Fortin 2010) and is therefore more useful for datasets

collected by many receivers/individuals. Nodes and edges must be carefully defined in order

to accurately represent the study system; any deviations from the true network, for example

through data transformation, inclusion of false absences, or exclusion of individuals, can

significantly impact network measures and the overall network structure (Farine and

Whitehead 2015).

If spatial correlation is of interest then researchers should use spatially explicit

methods like GRFs. It is possible to incorporate spatial information into some of the other

statistical methods we have described. For example, with network analysis a receiver node

can include a location, and with regression-type analyses (e.g., Cormack-Jolly-Seber models

or GLMs) spatial references can be incorporated as covariates. However, spatial models are

distinguishable from these methods because they estimate spatial correlation, which when

ignored can invalidate analyses by violating the assumption of independence (Thorson &

Minto 2015). In addition, estimating spatial correlation can show how

unobserved/unmeasured variables correlated in space affect the response variable (Carson &

Mills Flemming 2014; Thorson et al. 2015). Network autocorrelation models, spatial capture-

This article is protected by copyright. All rights reserved.

 recapture, or spatial generalized models (not reviewed here; see Zuur et al. 2009) can also be

used to estimate spatial correlation.

Accepted Article
Finally, for most other scenarios, we recommend using GLMs/GLMMs or

GAMs/GAMMs. Generalized modelling is arguably the most accessible statistical method

presented in this paper in terms of documentation and application. It is also flexible, as

several different response variables can be used, both linear and non-linear covariate

relationships are possible, and random effects and correlative structures can be included

(Zuur, Ieno & Saveliev 2017). However, these methods come with their own assumptions

(e.g., distribution assumptions of the residuals) and complexities, therefore we would

encourage readers to consult more specific guides (e.g., Bolker et al. 2009) before

implementation.

Conclusion

Telemetry is increasingly used to track aquatic animals. This has led to a massive

expansion in the volume and detail of ensuing movement data, and significant growth in the

availability of suitable statistical methods. It is often no longer sufficient to rely on relatively

simple descriptive analytical techniques, yet choosing from among available methods can be

daunting. We reviewed advanced statistical methods useful for detection data in order to

introduce them to aquatic telemetry users and provide researchers with the tools necessary for

more comprehensive detection data analysis. We focused specifically on detection data

recorded in aquatic environments, which can differ in small but substantial ways from those

collected in terrestrial studies. For example, the camera traps often used in terrestrial studies

can detect previously unknown/unmarked individuals, whereas acoustic, radio, and PIT

receivers can only identify tagged individuals, thus hindering our ability to estimate

population size from these data using mark-recapture methods. However, some of the

This article is protected by copyright. All rights reserved.

 methods mentioned here (e.g., spatially explicit capture-recapture) have been established in

terrestrial studies for 10+ years, and minor modifications could significantly enhance the
Accepted Article
analysis of aquatic detection data. Going forward, we recommend that aquatic ecologists look

towards terrestrial studies and other fields like human mobility to help motivate the statistical

advances that will be needed to analyze detection datasets that are rapidly growing in both

size and complexity.

Acknowledgements

The authors are indebted to several funding agencies. This research was supported by the

Ocean Tracking Network (OTN), a Canadian Statistical Sciences Institute (CANSSI)

Collaborative Team Project, and grants from the Natural Sciences and Engineering Research

Council of Canada (NSERC) to MAM, JMF, SJC, and MP. KW has also been supported by

an NSERC postgraduate scholarship and a Killam Predoctoral Scholarship, while MAM

received an NSERC postdoctoral fellowship. SJC acknowledges the Canada Research Chairs

Program. The data used for the illustrative case study was collected as part of the NSERC

HydroNet Strategic Network Grant Program with additional support from BC Hydro. This

paper is a contribution of the ideasOTN committee of OTN. The authors sincerely thank

Jonathon Pye and Fred Whoriskey for comments and support for the manuscript, as well as

Luca Börger and two anonymous reviewers for their crucial input that significantly improved

the paper.

This article is protected by copyright. All rights reserved.

 Data Accessibility

Supplementary code and data for the illustrative analyses are available via GitHub at
Accepted Article
https://doi.org/10.5281/zenodo.2628108. The data are available publicly through the OTN

data portal at either https://doi.org/10.14286/2019-kinbasket-bulltrout or

https://members.oceantrack.org/.

Author Contributions

EGM conceived the project idea. SJC and MP provided the data. KW and MAM analyzed the

data, with input from EGM and LFGG. KW led the writing of the manuscript, but all authors

contributed significantly to all drafts and approved the manuscript for publication.

References

Amstrup, S.C., McDonald, T.L., & Manly, B.F.J. (2005) Handbook of Capture-Recapture

Analysis. Princeton University Press, Princeton, N.J.

Andersen, P. & Gill, R. (1982) Cox’s regression model for counting processes, a large sample

study. The Annals of Statistics, 10, 1100-1120.

Auger-Méthé, M., Albertsen, C.M., Jonsen, I.D., Derocher, A.E., Lidgard, D.C., Studholme,

K.R., Bowen, W.B., Crossin, G.T., & Mills Flemming, J. (2017) Spatiotemporal

modelling of marine movement data using Template Model Builder (TMB). Marine

Ecology Progress Series, 565, 237-249.

Barbour, A.B., Boucek, R.E., & Adams, A.J. (2012) Effect of pulsed gastric lavage on

apparent survival of a juvenile fish in a natural system. Journal of Experimental Marine

Biology and Ecology, 422-423, 107–113.

This article is protected by copyright. All rights reserved.

 Barker, R.J. (1997) Joint modeling of live-recapture, tag-resight, and tag-recovery data.

Biometrics, 53, 666–677.

Accepted Article
Benoît, H.P., Capizzano, C.W., Knotek, R.J., Rudders, D.B., Sulikowski, J.A., Dean, M.J.,

Hoffman, W., Zemeckis, D.R., & Mandelman, J.W. (2015) A generalized model for

longitudinal short-and long-term mortality data for commercial fishery discards and

recreational fishery catch-and-releases. ICES Journal of Marine Science, 72, 1834-1847.

Bodey, T.W., Cleasby, I.R., Bell, F., Parr, N., Schultz, A., Votier, S.C., & Bearhop, S. (2018)

A phylogenetically controlled meta-analysis of biologging device effects on birds :

Deleterious effects and a call for more standardized reporting of study data. Methods in

Ecology and Evolution, 9, 946-955.

Bolker, B.M., Brooks, M.E., Clark, C.J., Geange, S.W., Poulsen, J.R., Stevens, M.H.H., &

White, J.-S.S. (2009) Generalized linear mixed models: A practical guide for ecology and

evolution. Trends in Ecology and Evolution, 24, 127–135.

Börger, L., Franconi, N., Ferretti, F., Meschi, F., De Michele, G., Gantz, A., & Coulson, T.

(2006) An integrated approach to identify spatiotemporal and individual-level

determinants of animal home range size. The American Naturalist, 168, 471-485.

Brousseau, L.J., Sclafani, M., Smith, D.R., & Carter, D.B. (2004) Acoustic-tracking and

radio-tracking of horseshoe crabs to assess spawning behavior and subtidal habitat use in

Delaware Bay. North American Journal of Fisheries Management, 24(4), 1376–1384.

Bunck, C.M., Chen, C.L., & Pollock, K.H. (1995) Robustness of survival estimates from

radio-telemetry studies with uncertain relocation of individuals. The Journal of Wildlife

Management, 59, 790-794.

Burnham, K.P. (1993) A theory for combined analysis of ring recovery and recapture data.

Marked individuals in the study of bird populations (eds J.D. Lebreton & P.M. North), pp.

199–213. Birkhäuser Verlag, Basel, Switzerland.

This article is protected by copyright. All rights reserved.

 Cagua, E.F., Berumen, M.L., & Tyler, E.H.M. (2013) Topography and biological noise

determine acoustic detectability on coral reefs. Coral Reefs, 32, 1123-1134.

Accepted Article
Campbell, H.A., Watts, M.E., Dwyer, R.G., & Franklin, C.E. (2012) V-Track: software for

analysing and visualising animal movement from acoustic telemetry detections. Marine

and Freshwater Research, 63, 815-820.

Carson, S., & Mills Flemming, J. (2014) Seal encounters at sea: A contemporary spatial

approach using R-INLA. Ecological Modelling, 291, 175-181.

Carson, S., Shackell, N., & Mills Flemming, J. (2017) Local overfishing may be avoided by

examining parameters of a spatio-temporal model. PLoS ONE 12, e0184427.

Castro-Santos, T., & Haro, A. (2003) Quantifying migratory delay: a new application of

survival analysis methods. Canadian Journal of Fisheries and Aquatic Sciences, 60, 986–

996.

Cooke, S.J., Hinch, S.G., Lucas, M.C., & Lutcavage, M. (2012) Chapter 18 – Biotelemetry

and biologging. Fisheries Techniques, 3rd edn (eds A.V. Zale, D.L. Parrish, & T.M.

Sutton), pp. 819-860. Bethesda, MD: American Fisheries Society.

Cosandey-Godin, A., Krainski, E.T., Worm, B., & Mills Flemming, J. (2015) Applying

Bayesian spatiotemporal models to fisheries bycatch in the Canadian Arctic. Canadian

Journal of Fisheries and Aquatic Sciences, 72, 186-197.

Csárdi G., & Nepusz, T. (2006) The igraph software package for complex network research,

InterJournal, Complex Systems, 1695.

Curtis, J.M., Johnson, M.W., Diamond, S.L., & Stunz, G.W. (2015) Quantifying delayed

mortality from barotrauma impairment in discarded red snapper using acoustic telemetry.

Marine and Coastal Fisheries, 7, 434-449.

Dale, M.R.T., & Fortin, M.-J. (2010) From graphs to spatial graphs. Annual Review of

Ecology, Evolution, and Systematics, 41, 21-38.

This article is protected by copyright. All rights reserved.

 Danylchuk, S.E., Danylchuk, A.J., Cooke, S.J., Goldberg, T.L., Koppelman, J., & Philipp,

D.P. (2007) Effects of recreational angling on the post-release behavior and predation of
Accepted Article
bonefish (Albula vulpes): The role of equilibrium status at the time of release. Journal of

Experimental Marine Biology and Ecology, 346, 127-133.

Donaldson, M.R., Hinch, S.G., Suski, C.D., Fisk, A.T., Heupel, M.R., & Cooke, S.J. (2014)

Making connections in aquatic ecosystems with acoustic telemetry monitoring. Frontiers

in Ecology and the Environment, 12, 565-573.

Dudgeon, C.L., Lanyon, J.M., & Semmens, J.M. (2013) Seasonality and site fidelity of the

zebra shark, Stegostoma fasciatum, in southeast Queensland, Australia. Animal Behaviour,

85, 471–481.

Dudgeon, C.L., Pollock, K.H., Braccini, J.M., Semmens, J.M., & Barnett, A. (2015)

Integrating acoustic telemetry into mark–recapture models to improve the precision of

apparent survival and abundance estimates. Oecologia, 178, 761-772.

Dwyer, R.G., Brooking, C., Brimblecombe, W., Campbell, H.A., Hunter, J., Watts, M.E., &

Franklin, C.E. (2015) An open Web-based system for the analysis and sharing of animal

tracking data. Animal Biotelemetry, 3, 1.

Efford, M.G., & Fewster, R.M. (2013) Estimating population size by spatially explicit

capture-recapture. Oikos, 122, 918–928.

Espinoza, M., Lédée, E.J.I., Simpfendorfer, C.A., Tobin, A.J., & Heupel, M.R. (2015)

Contrasting movements and connectivity of reef-associated sharks using acoustic

telemetry: implications for management. Ecological Applications, 25, 2101-2118.

Farine, D.R., & Whitehead, H. (2015) Constructing, conducting, and interpreting animal

social network analysis. Journal of Animal Ecology, 84, 1144-1163.

This article is protected by copyright. All rights reserved.

 Finn, J.T., Brownscombe, J.W., Haak, C.R., Cooke, S.J., Cormier, R., Gagne, T., &

Danylchuk, A.J. (2014) Applying network methods to acoustic telemetry data: Modeling
Accepted Article
the movements of tropical marine fishes. Ecological Modelling, 293, 139-149.

Fouchet, D., Santin Janin, H., Sauvage, F., Yoccoz, N.G., & Pontier, D. (2016) An R package

for analysing survival using continuous-time open capture–recapture models. Methods in

Ecology and Evolution, 7, 518–528.

Fox, R.J., & Bellwood, D.R. (2014) Herbivores in a small world: network theory highlights

vulnerability in the function of herbivory on coral reefs. Functional Ecology, 28, 642-651.

Furey, N.B., Hinch, S.G., Bass, A.L., Middleton, C.T., Minke Martin, V., & Lotto, A.G.

(2016) Predator swamping reduces predation risk during nocturnal migration of juvenile

salmon in a high-mortality landscape. Journal of Animal Ecology, 85, 948–959.

Gutowsky, L.F.G., Harrison, P.M., Martins, E.G., Leake, A., Patterson, D.A., Power, M., &

Cooke, S.J. (2016) Interactive effects of sex and body size on the movement ecology of

adfluvial bull trout (Salvelinus confluentus). Canadian Journal of Zoology, 94, 31-40.

Harrell, F. (2015) Regression modeling strategies: with applications to linear models, logistic

and ordinal regression, and survival analysis. Springer, Berlin.

Hayden, T.A., Holbrook, C.M., Fielder, D.G., Vandergoot, C.S., Bergstedt, R.A., Dettmers,

J.M., Krueger, C.C., & Cooke, S.J. (2014) Acoustic telemetry reveals large-scale

migration patterns of walleye in Lake Huron. PLoS ONE, 9, e114833.

Heupel, M.R., Semmens, J.M., & Hobday, A.J. (2006) Automated acoustic tracking of

aquatic animals: scales, design and deployment of listening station arrays. Marine and

Freshwater Research, 57, 1-13.

Hoenner, X., Huveneers, C., Steckenreuter, A., Simpfendorfer, C., Tattersall, K., Jaine, F.,

Atkins, N., Babcock, R., Brodie, S., Burgess, J., Campbell, H., Heupel, M., Pasquer, B.,

Proctor, R., Taylor, M.D., Udyawer, V., & Harcourt, R. (2018) Australia’s continental-

This article is protected by copyright. All rights reserved.

 scale acoustic tracking database and its automated quality control process. Scientific Data,

5, 180206.
Accepted Article
Holbrook, C.M., Kinnison, M.T., & Zydlewski, J. (2011) Survival of migrating Atlantic

salmon smolts through the Penobscot River, Maine: a prerestoration assessment.

Transactions of the American Fisheries Society, 140, 1255–1268.

Holbrook, C.M., Perry, R.W., Brandes, P.L., and Adams, N.S. (2013) Adjusting survival

estimates for premature transmitter failure: a case study from the Sacramento-San Joaquin

Delta. Environmental Biology of Fishes, 96, 165–173.

Holbrook, C.M., Hayden, T., & Binder, T. (2017) glatos: A package for the Great Lakes

Acoustic Telemetry Observation System. R package version 0.1.3. URL

https://gitlab.oceantrack.org/GreatLakes/glatos.

Huserbråten, M.B.O., Moland, E., Knutsen, H., Olsen, E.M., André, C., & Stenseth, N.C.

(2013) Conservation, spillover and gene flow within a network of Northern European

marine protected areas. PLoS ONE, 8, e73388.

Hussey, N.E., Kessel, S.T., Aarestrup, K., Cooke, S.J., Cowley, P.D., Fisk, A.T., Harcourt,

R.G., Holland, K.N., Iverson, S.J., Kocik, J.F., Mills Flemming, J.E., & Whoriskey, F.G.

(2015) Aquatic animal telemetry: A panoramic window into the underwater world.

Science, 348, 1255642.

Illian, J.B., Sørbye, S.H., & Rue, H. (2012) A toolbox for fitting complex spatial point

process models using integrated nested Laplace approximation (INLA). The Annals of

Applied Statistics, 6, 1499–1530.

Jacoby, D.M.P., Brooks, E.J., Croft, D.P., & Sims, D.W. (2012) Developing a deeper

understanding of animal movements and spatial dynamics through novel application of

network analyses. Methods in Ecology and Evolution, 3, 574-583.

This article is protected by copyright. All rights reserved.

 Jepsen, N., Thorstad, E.B., Havn, T., & Lucas, M.C. (2015) The use of external electronic

tags on fish: an evaluation of tag retention and tagging effects. Animal Biotelemetry, 3:49.
Accepted Article
Johnson, C.J., Boyce, M.S., Schwartz, C.C., & Haroldson, M.A. (2004) Modeling survival:

application of the Andersen-Gill model to Yellowstone grizzly bears. Journal of Wildlife

Management, 68, 966-978.

Kawabata, Y., Okuyama, J., Asami, K., Okuzawa, K., Yoseda, K., & Arai, N. (2010) Effects

of a tropical cyclone on the distribution of hatchery-reared black-spot tuskfish Choerodon

schoenleinii determined by acoustic telemetry. Journal of Fish Biology, 77, 627-642.

Kessel, S.T., Chapman, D.D., Franks, B.R., Gdamke, T., Gruber, S.H., Newman, J.M., White,

E.R., & Perkins, R.G. (2014a) Predictable temperature-regulated residency, movement and

migration in a large, highly mobile marine predator (Negaprion brevirostris). Marine

Ecology Progress Series, 514, 175-190.

Kessel, S.T., Cooke, S.J., Heupel, M.R., Hussey, N.E., Simpfendorfer, C.A., Vagle, S., &

Fisk, A.T. (2014b) A review of detection range testing in aquatic passive acoustic

telemetry studies. Reviews in Fish Biology and Fisheries, 24, 199-218.

Ketchum, J.T., Hearn, A., Klimley, A.P., Peñaherrera, C., Espinoza, E., Bessudo, S., Soler,

G., & Arauz, R. (2014) Inter-island movements of scalloped hammerhead sharks (Sphyrna

lewini) and seasonal connectivity in a marine protected area of the eastern tropical Pacific.

Marine Biology, 161, 939–951.

Klein, J. P., & Moeschberger, M. L. (2003) Survival analysis: techniques for censored and

truncated data. Springer Science & Business Media, Inc., New York, N.Y.

Kristensen, K., Nielsen, A., Berg, C.W., Skaug, H., & Bell, B. (2016) TMB: automatic

differentiation and Laplace approximation. Journal of Statistical Software, 70, 1–21.

This article is protected by copyright. All rights reserved.

 Laake, J.L. (2013) RMark: An R interface for analysis of capture-recapture data with MARK.

AFSC Processed Report 2013-01. Alaska Fisheries Science Centre, NOAA, National
Accepted Article
Marine Fisheries Service, Seattle WA. [accessed 3 June 2018].

Leenders, R.Th.A.J. (2002) Modeling social influence through network autocorrelation:

constructing the weight matrix. Social Networks, 24, 21-47.

Lennox, R.J., Aarestrup, K., Cookie, S.J., Cowley, P.D., Deng, Z.D., Fisk, A.T., Harcourt,

R.G., Heupel, M., Hinch, S.G., Holland, K.N., Hussey, N.E., Iverson, S.J., Kessel, S.T.,

Kocik, J.F., Lucas, M.C., Mills Flemming, J., Nguyen, V.M., Stokesbury, M.J.W., Vagle,

S., Vanderzwaag, D.L., Whoriskey, F.G., & Young, N. (2017a) Envisioning the Future of

Aquatic Animal Tracking: Technology, Science, and Application. Bioscience, 67, 884-

896.

Lennox, R.J., Filous, A., Danylchuk, S.C., Cooke, S.J., Brownscombe, J.W., Friedlander,

A.M., & Danylchuk, A.J. (2017b) Factors influencing postrelease predation for a catch-

and-release tropical flats fishery with a high predator burden. North American Journal of

Fisheries Management, 37, 1045-1053.

Lindberg, M.S. (2012) A review of designs for capture–mark–recapture studies in discrete

time. Journal of Ornithology, 152, 355–370.

Lucas, M.C., & Baras, E. (2000) Methods for studying spatial behaviour of freshwater fishes

in the natural environment. Fish and Fisheries, 1, 283-316.

Marshell, A., Mills, J.S., Rhodes, K.L., & McIlwain, J. (2011) Passive acoustic telemetry

reveals highly variable home range and movement patterns among unicornfish within a

marine reserve. Coral Reefs, 30, 631-642.

Martins, E.G., Hinch, S.G., Patterson, D.A., Hague, M.J., Cooke, S.J., Miller, K.M.,

Lapointe, M.F., English, K.K., & Farrell, A.P. (2011) Effects of river temperature and

This article is protected by copyright. All rights reserved.

 climate warming on stock-specific survival of adult migrating Fraser River sockeye

salmon (Oncorhynchus nerka). Global Change Biology, 17, 99–114.

Accepted Article
Martins, E.G., Gutowsky, L.F.G., Harrison, P.M., Patterson, D.A., Power, M., Zhu, D.Z.,

Leake, A., & Cooke, S.J. (2013) Forebay use and entrainment rates of resident adult fish in

a large hydropower reservoir. Aquatic Biology, 19, 253-263.

Martins, E.G., Gutowsky, L.F.G., Harrison, P.M., Mills Flemming, J.E., Jonsen, I.D., Zhu,

D.Z., Leake, A., Patterson, D.A., Power, M. & Cooke, S.J. (2014) Behavioral attributes of

turbine entrainment risk for adult resident fish revealed by acoustic telemetry and state-

space modeling. Animal Biotelemetry, 2, 13.

Matich, P., & Heithaus, M.R. (2014) Multi-tissue stable isotope analysis and acoustic

telemetry reveal seasonal variability in the trophic interactions of juvenile bull sharks in a

coastal estuary. Journal of Animal Ecology, 83, 199–213.

Meekan, M.G., Duarte, C.M., Fernández-Gracia, J., Thums, M., Sequeira, A.M.M., Harcourt,

R., & Eguíluz, V.M. (2017) The ecology of human mobility. Trends in Ecology and

Evolution, 32, 198-210.

Melnychuk, M.C. (2012) Detection efficiency in telemetry studies: definitions and evaluation

methods. Telemetry techniques: a user guide for fisheries research (eds N.S. Adams, J.W.

Beeman, & J.H. Eiler), pp. 339-357. American Fisheries Society, Bethesda, MD.

Moore, M.E., Berejikian, B.A., Goetz, F.A., Berger, A.G., Hodgson, S.S., Connor, E.J., &

Quinn, T.P. (2015) Multi-population analysis of Puget Sound steelhead survival and

migration behavior. Marine Ecology Progress Series, 537, 217–232.

Murray, D.L. (2006) On improving telemetry-based survival estimation. Journal of Wildlife

Management, 70, 1530-1543.

Naughton, G.P., Caudill, C.C., Keefer, M.L., Bjornn, T.C., Stuehrenberg, L.C., & Peery, C.A.

(2005) Late-season mortality during migration of radio-tagged adult sockeye salmon

This article is protected by copyright. All rights reserved.

 (Oncorhynchus nerka) in the Columbia River. Canadian Journal of Fisheries and Aquatic

Sciences, 62, 30-47.

Accepted Article
Niezgoda, G., Benfield, M., Sisak, M., & Anson, P. (2002) Tracking acoustic transmitters by

code division multiple access (CDMA)-based telemetry. Hydrobiologia, 483, 275-286.

Pedersen, M.W., & K.C. Weng. (2013) Estimating individual animal movement from

observation networks. Methods in Ecology and Evolution, 4, 920-929.

Pedersen, M.W., Burgess, G., & Weng, K.C. (2014) A quantitative approach to static sensor

network design. Methods in Ecology and Evolution, 5, 1043-1051.

Pollock, K.H., Winterstein, S.R., & Conroy, M.J. (1989) Estimation and analysis of survival

distributions for radio-tagged animals. Biometrics, 45, 99-109.

Pollock, K.H., Winterstein, S.R., Bunck, C.M., & Curtis, P.D. (1989) Survival analysis in

telemetry studies: the staggered entry design. Journal of Wildlife Management, 53, 7-15.

R Core Team. (2018) R: A Language and Environment for Statistical Computing. R

Foundation for Statistical Computing. Vienna, Austria. URL https://www.R-project.org/

[accessed 23 April 2018]

Raabe, J.K., Gardner, B., & Hightower, J.E. (2014) A spatial capture–recapture model to

estimate fish survival and location from linear continuous monitoring arrays. Canadian

Journal of Fisheries and Aquatic Sciences, 71, 120–130.

Richard, A., Bernatchez, L., Valiquette, E., & Dionne, M. (2014) Telemetry reveals how

catch and release affects prespawning migration in Atlantic salmon (Salmo salar).

Canadian Journal of Fisheries and Aquatic Sciences, 71, 1730–1739.

Royle, J.A., Chandler, R.B., Sollmann, R., & Gardner, B. (2014) Spatial capture-recapture.

Academic Press, Waltham, Massachusetts.

Schwarz, C.J., Schweigert, J.F., & Arnason, A.N. (1993) Estimating migration rates using

tag-recovery data. Biometrics, 49, 177–193.

This article is protected by copyright. All rights reserved.

 Simpfendorfer, C.A., Heupel, M.R., & Hueter, R.E. (2002) Estimation of short-term centers

of activity from an array of omnidirectional hydrophones and its use in studying animal
Accepted Article
movements. Canadian Journal of Fisheries and Aquatic Sciences, 59, 23-32.

Simpfendorfer, C.A., Heupel, M.R., & Collins, A.B. (2008) Variation in the performance of

acoustic receivers and its implication for positioning algorithms in a riverine setting.

Canadian Journal of Fisheries and Aquatic Sciences, 65, 482-492.

Smith, F. (2013) Understanding HPE in the VEMCO positioning system (VPS). VEMCO

Document #: DOC-005457-01. URL http://vemco.com/ [accessed 27 September 2018]

Sollmann, R., Furtado, M.M., Jácomo, A.T.A., Tôrres, N.M., & Silveira, L. (2010) Maned

wolf survival rate in central Brazil. Journal of Zoology, 282, 207–213.

Stehfast, K.M., Patterson, T.A., Dagorn, L., Holland, K.N., Itano, D., & Semmens, J.M.

(2013) Network analysis of acoustic tracking data reveals the structure and stability of fish

aggregations in the ocean. Animal Behaviour, 85, 839-848.

Stehfast, K.M., Patterson, T.A., Barnett, A., & Semmens, J.M. (2015) Markov models and

network analysis reveal sex-specific differences in the space-use of a coastal apex

predator. Oikos, 124, 307-318.

Stich, D.S., Kinnison, M.T., Kocik, J.F., & Zydlewski, J.D. (2015) Initiation of migration and

movement rates of Atlantic salmon smolts in fresh water. Canadian Journal of Fisheries

and Aquatic Sciences, 72, 1339–1351.

Thorley, J.L., Youngson, A.F., & Laughton, R. (2007) Seasonal variation in rod recapture

rates indicates differential exploitation of Atlantic salmon, Salmo salar, stock components.

Fisheries Management and Ecology, 14, 191-198.

Thorson, J.T., Skaug, H.J., Kristensen, K., Shelton, A.O., Ward, E.J., Harms, J.H., &

Benante, J.A. (2015) The importance of spatial models for estimating the strength of

density dependence. Ecology, 96, 1202-1212.

This article is protected by copyright. All rights reserved.

 Thorson, J.T. & Minto, C. (2015) Mixed effects: a unifying framework for statistical

modelling in fisheries biology. ICES Journal of Marine Science, 72, 1245-1256.

Accepted Article
Thums, M., Fernández-Gracia, J., Sequeira, A.M.M., Eguíluz, V.M., Duarte, C.M., &

Meekan, M.G. (2018) How big data fast tracked human mobility research and the lessons

for animal movement ecology. Frontiers in Marine Science, 5, 21.

Topping, D.T., & Szedlmayer, S.T. (2011) Site fidelity, residence time and movements of red

snapper Lutjanus campechanus estimated with long-term acoustic monitoring. Marine

Ecology Progess Series, 437, 183-200.

Udyawer, V., Read, M., Hamann, M., Simpfendorfer, C.A., & Heupel, M.R. (2015) Effects

of environmental variables on the movement and space use of coastal sea snakes over

multiple temporal scales. Journal of Experimental Marine Biology and Ecology, 473, 26–

34.

Urban, D.L., Minor, E.S., Treml, E.A., & Schick, R.S. (2009) Graph models of habitat

mosaics. Ecology Letters, 12, 260-273.

Welch, D.W., Melnychuk, M.C., Rechisky, E.R., Porter, A.D., Jacobs, M.C., Ladouceur, A.,

McKinley, R.S., & Jackson, G.D. (2009) Freshwater and marine migration and survival of

endangered Cultus Lake sockeye salmon (Oncorhynchus nerka) smolts using POST, a

large-scale acoustic telemetry array. Canadian Journal of Fisheries and Aquatic Sciences,

66, 736–750.

White, G.C., & Garrott, R.A. (1990) Analysis of wildlife radio-tracking data. Academic

Press, San Diego, California.

White, G.C., & Burnham, K.P. (1999) Program MARK: survival estimation from populations

of marked animals. Bird Study, 46(supp1), S120-S139.

This article is protected by copyright. All rights reserved.

 Whoriskey, K., Auger-Méthé, M., Albertsen, C.M., Whoriskey, F.G., Binder, T.R., Krueger,

C.C., & Mills Flemming, J. (2017) A hidden Markov movement model for rapidly
Accepted Article
identifying behavioral states from animal tracks. Ecology and Evolution, 7, 2112-2121.

Williams, B.K., Nichols, J.D., & Conroy, M.J. (2001) Analysis and management of animal

populations: Modeling, estimation, and decision making. Academic Press, San Diego,

California.

Winship, A.J., Jorgensen, S.J., Shaffer, S.A., Jonsen, I.D., Robinson, P.W., Costa, D.P., &

Block, B.A. (2012) State-space framework for estimating measurement error from double-

tagging telemetry experiments. Methods in Ecology and Evolution, 3, 291-302.

Winton, M.V., Kneebone, J., Zemeckis, D.R., & Fay, G. (2018) A spatial point process

model to estimate individual centres of activity from passive acoustic telemetry data.

Methods in Ecology and Evolution, 9, 2262-2272.

Wood, S.N. (2006) Generalized additive models: An introduction with R. Taylor & Francis

Group, LLC, Boca Raton, FL.

Zhang, Y., Xu, Q.Q., Alós, J., Liu, H., Xu, Q.Q., & Yang, H. (2015) Short-term fidelity,

habitat use and vertical movement behavior of the black rockfish Sebastes schlegelii as

determined by acoustic telemetry. PLoS ONE, 10, e0134381.

Zuur, A.F., Ieno, E.N., Walker, N.J., Saveliev, A.A., & Smith, G.A. (2009) Mixed effects

models and extensions in ecology with R. Springer Science + Business Media, New York,

NY.

Zuur, A.F., Ieno, E.N., & Saveliev, A.A. (2017) Beginner’s guide to spatial, temporal, and

spatial-temporal ecological data analysis with R-INLA. Volume 1: Using GLM and

GLMM. Highland Statistics Ltd., Newburgh, UK.

This article is protected by copyright. All rights reserved.

 Tables

Table 1 Model fits with corrected AIC (AICc) values from the mark recapture analysis of the
Accepted Article
illustrative bull trout dataset. Construction of the model is given by the Model Formula,

where Φ(.) denotes the effects related to the apparent survival probability and p(.)

denotes those related to the probability of detection. ~1 denotes an intercept only model.

DeltaAICc is the difference in AICc from the best model.

Model Formula # Parameters AICc DeltaAICc

Φ(~1)p(~1) 2 258.0 0.00
Φ(~length)p(~1) 3 259.4 1.41
Φ(~sex)p(~1) 3 260.0 2.02
Φ(~sex * length)p(~1) 5 260.6 2.59
Φ(~sex + length)p(~1) 4 261.5 3.42
Φ(~1)p(~time) 4 261.5 3.50
Φ(~length)p(~time) 5 263.3 5.29
Φ(~sex)p(~time) 5 263.6 5.58
Φ(~sex * length)p(~time) 7 264.6 6.57
Φ(~sex + length)p(~time) 6 265.4 7.39

Figures

Figure 1 Study location of the illustrative dataset, i.e., the Kinbasket reservoir in British

Columbia, Canada, with the location of the dam in yellow, and the receiver locations in dark

blue. Detection range was assumed to be 500m, a distance shorter than the width of either

Reach in most places.

Figure 2 Sex-specific results from applying network analysis to the illustrative dataset.

Yellow represents the position of the dam. Circles denote the receiver positions, and are

weighted based on their node degree, i.e., the number of incoming and outgoing edges of a

This article is protected by copyright. All rights reserved.

 node. Edges are weighted based on the number of directed bull trout movements between

receivers. Males are on the left (green), and females are on the right (orange).
Accepted Article
Figure 3 Results from the GRF analysis on the illustrative dataset. Panel A) shows the mesh

calculated by the INLA SPDE; panel B) represents the expected number of bull trout across

the reservoir returned by the full model; panel C) represents the expected number of fish as

influenced by distance from the dam; and panel D) represents the expected number of fish

based on the effect of the GRF only. Yellow represents the position of the dam.

Figure 4 Decision tree for identifying appropriate statistical methodologies for analyzing

detection data collected by acoustic, radio, or PIT telemetry.

This article is protected by copyright. All rights reserved.

Accepted Article

This article is protected by copyright. All rights reserved.

Accepted Article

This article is protected by copyright. All rights reserved.