Modelling The Three Dimensional Space Use of Aquatic Animals Combining Topography and Eulerian Telemetry Data
Modelling The Three Dimensional Space Use of Aquatic Animals Combining Topography and Eulerian Telemetry Data
Modelling The Three Dimensional Space Use of Aquatic Animals Combining Topography and Eulerian Telemetry Data
DOI: 10.1111/2041-210X.13232
A P P L I C AT I O N
Correspondence
Eneko Aspillaga Abstract
Email: [email protected] 1. Passive acoustic telemetry provides the opportunity to monitor and contextual-
Funding information ize the movements of diverse aquatic animals. Despite depth being an essential
Ministerio de Educación, Cultura y Deporte, dimension along which many processes are organized, the Eulerian structure of
Grant/Award Number: AP2012-0141;
Ministerio de Ciencia e Innovación, Grant/ the acoustic telemetry data (movements perceived from fixed locations) and the
Award Number: CGL2016-78156-C2-1-R consequences of sound propagation in water hinders the incorporation of the ver-
Handling Editor: Marie Auger‐Méthé tical dimension into animal’s space use analyses.
2. Here, we propose a new data‐driven quantitative method to estimate 3D space use
from telemetry networks. The methodology is based on simulating large numbers of
stochastic synthetic paths, accommodating the detection probability around receiv-
ers and the depth information from transmitters and integrating the local topography.
3. The methodological protocol is explained in detail and tested with acoustic te-
lemetry data from the common dentex Dentex dentex in a Mediterranean marine
protected area. We present 3D space use estimations for the tagged individuals
and compare them with other 3D and 2D estimations derived with existing proba-
bilistic methods.
4. 3D space use estimations that incorporate topography provided a more compre-
hensive view of the movement ecology of tracked individuals, with relevant pieces
being missed by 2D representations. Our method generated realistic representa-
tions of the actual spatial co‐occurrence of individuals, including the spatio‐tem-
poral identification of relevant aggregation areas.
KEYWORDS
3D, aquatic ecology, movement ecology, space use, telemetry networks, topography, acoustic
telemetry, utilization distribution
1 | I NTRO D U C TI O N arboreal, or burrowing; Belant, Millspaugh, Martin, & Gitzen, 2012).
The relevance of the vertical dimension is especially conspicuous
Animal movement has been traditionally studied in a two‐dimen- in aquatic environments, where almost every environmental factor
sional (2D) domain, often omitting the vertical dimension, which varies with depth, generating sharp ecological gradients that affect
is essential for the activity of many organisms (e.g. aquatic, flying, species distribution and the potential space use of mobile organisms
Methods Ecol Evol. 2019;10:1551–1557. © 2019 The Authors. Methods in Ecology and | 1551
wileyonlinelibrary.com/journal/mee3
Evolution © 2019 British Ecological Society
|Methods in Ecology and Evolu
on
1552 ASPILLAGA et al.
(Aspillaga et al., 2017; Kitagawa et al., 2000). Moreover, interactions Our method assumes simple movement characteristics to gener-
among individuals (e.g. aggregations, avoidances) are often reflected ate synthetic paths, which follow the shortest pathway between
in a vertical stratification of the space use (Simpfendorfer, Olsen, two points and take into account the topography of the study
Heupel, & Moland, 2012; Vivancos, Closs, & Tentelier, 2016). In ma- site to avoid movements through hard boundaries (i.e. emerging
rine environments, movement analyses that are limited to the 2D landmasses and shallow areas). This contrasts with other 2D and
space provide a partial if not unrealistic view of the underpinning 3D space use approximations grounded on the Brownian motion
biological processes, and 3D approaches are required to obtain rele- model, which assume a diffusive movement model defined by a
vant ecological and behavioural conclusions (Belant et al., 2012; Lee, single parameter, the Brownian motion variance (Horne et al.,
Huveneers, Duong, & Harcourt, 2017). 2007; Kranstauber et al., 2012; Tracey et al., 2014). Our method,
Acoustic telemetry is the most widely used technique to mon- instead, incorporates the variability observed in the original data
itor the movement of aquatic animals (Hussey et al., 2015). In pas- and constructs consistent space use estimations by integrating
sive acoustic telemetry, animals tagged with acoustic transmitters large numbers of iterations of synthetic paths. This approximation
are monitored by a network of receivers placed at fixed locations accounts for movement barriers in a natural way, but implies the
in the study area (Heupel, Semmens, & Hobday, 2006), and thus, use of contextual data (physical characterizations of the study site)
movement is perceived in an Eulerian mode as presences, absences beyond the one provided by the acoustic array.
or transitions between discrete locations. The uncertainty regarding
the position of tagged animals is usually large (hundreds of meters),
as it depends on the distance at which acoustic signals are detect- 2 | OV E RV I E W O F TH E WO R K FLOW
able (i.e. the detection range).
Accurate depth data are often provided by the acoustic transmit- The method to estimate 3D space use probabilities has been devel-
ters attached to animals (Veilleux et al., 2016), but this information oped in the r programming environment (R Core Team, 2017) and
is rarely taken into account when estimating the space use of fish. In implemented in the fishtrack3d package (https://github.com/aspil
recent years, several probabilistic approaches have been proposed laga/fishtrack3d). The workflow is divided into three steps: (a) em-
to quantify volumetric space use by incorporating the vertical dimen- pirical characterization of detection probabilities around receivers,
sion into location‐based methods, such as kernel density estimations (b) generation of synthetic paths and (c) assembly of 3D volumes.
(Simpfendorfer et al., 2012; Worton, 1989), or into movement‐based The main features of each of the steps are sketched below, but we
methods, such as the Brownian bridge movement model (Horne, provide a detailed explanation in a vignette included as supporting
Garton, Krone, & Lewis, 2007; Kranstauber, Kays, Lapoint, Wikelski, information (Appendix S1) and within the fishtrack3d package.
& Safi, 2012; Tracey et al., 2014). However, these methods require
a higher precision in the locations of animals, which is only possible
2.1 | Empirical characterization of detection
when acoustic receivers are placed on a relatively close grid, allow-
probabilities
ing the calculation of centres of activity (Simpfendorfer, Heupel, &
Hueter, 2002) or the application of more sophisticated triangulation Two main factors modulate the probability of being detected by
methods (e.g. VEMCO Positioning System; Steel, Coates, Hearn, & an acoustic receiver: the distance between the transmitter and
Klimley, 2014). Typical coast‐lined configurations of acoustic arrays the receiver (in relation to the acoustic range) and the presence of
are problematic given that non‐mesh architectures prevent highly physical impediments for the transmission of acoustic signals. The
resolved spatial locations and complicate the aforementioned volu- acoustic range is tested by placing several receivers and one or more
metric space use estimations. transmitters at increasing distances in between to determine the
Another challenge of movement analyses in two and three di- ratio of emitted signals that are detected at each distance (Kessel
mensions is the incorporation of topographical characteristics af- et al., 2014). We model the average decay in detection probability
fecting space use estimations. Prominent types of landforms, such with distance by fitting a logistic regression model to an empirical
as mountains or rivers in terrestrial ecosystems, or emerged land- range test carried out in the study area. In addition, the presence
masses in aquatic ones can pose insurmountable barriers to the of emerged landmasses or heterogeneous rocky bottoms generate
movement of animals, a fact that should be incorporated in the es- areas of acoustic shadows within the potential range of the receiver
timation of the space use, yet rarely is included in the analysis (e.g. array, from which tagged animals cannot be detected. We use a
Benhamou & Cornélis, 2010; Wilson, Hanks, & Johnson, 2017). viewshed analysis to identify the acoustic shadows generated by
Here, we propose a new computational method to estimate 3D large emerged landmasses, to then incorporate their effect during
space use probability distributions from passive acoustic teleme- the simulation of synthetic paths. Viewshed is a commonly used ter-
try data. It is an empirically driven method, which integrates the rain analysis technique that recognizes the geographical area that is
location uncertainty of coarse acoustic telemetry data by simu- visible from a specific location (Haverkort, Toma, & Zhuang, 2009),
lating large numbers of stochastic 3D trajectories (i.e. synthetic which is also applicable to sound propagation to detect, over a large
paths) according to the observed detection probabilities around scale, the areas from which acoustic transmitters could be detected
receivers and the movements of individuals within the array. by each receiver.
ASPILLAGA et al. Methods in Ecology and Evolu
on |
1553
The acoustic range and the distribution of the acoustic shadows paths. We present the study area as a three‐dimensional grid of vox-
present spatial and temporal variations due to physical and biological els with fixed horizontal and vertical resolution (i.e. 10·10·1 m3 in
factors, such as sound diffraction, hydrographic conditions (e.g. cur- our example), to then calculate the average percentage of time that
rents, swell), habitat characteristics and environmental noise (Heupel synthetic paths spend in each voxel. These average values are finally
et al., 2006), which are omitted in this first version of our method due processed by applying a three‐dimensional kernel estimation, using
to the difficulty of quantifying them. However, one of the advantages the “ks” package for r (Duong, 2007), to generate smooth 3D‐UDs
of our method is its high flexibility to accommodate additional data and that are easier to manipulate and plot. 3D‐UDs can be used to cal-
further modifications in future, such as the inclusion of habitat‐ or re- culate the volume sizes and spatial overlaps for different probability
ceiver‐specific acoustic range models that account for spatial heteroge- contours, or be plotted using 3D visualization engines such as the
neity (e.g. using mixed effects models, Selby et al., 2016), or information “rgl” (Adler et al., 2018) and “plotly” (Sievert, 2018) packages for r or
from fixed control tags to account for temporal variation. Moreover, external software such as paraview. (Ahrens, Geveci, & Law, 2005).
high‐resolution cartographies and voxel‐based 3D viewshed algorithms
might be used to improve the characterization of acoustic detection
landscapes, at the expense of increasing the computational cost. 3 | TE S TI N G TH E M E TH O D
20 20
10 10
0 0
0 20 40 60 80 100 0 20 40 60 80 100
No. of synthetic paths No. of synthetic paths
collapsing the 3D‐UDs (i.e. by summing the UDs of all the depth lay-
4.2 | 3D versus 2D space use estimations
ers), to be compared with the output of the dBBMM approach. We
used each of the 3D and 2D‐UD estimates to calculate the overlap in- The 3D space use probability estimations for common dentex in-
dices between individuals, measured as the proportion of cells/voxels dividuals provided a spatially explicit view of the location of their
in common within the 50% and 95% UD probability contours. We also home ranges (Figure 2, App #1 in Supporting Information Appendix
characterized the size variations in the estimated space use volumes S3). A clear vertical and horizontal segregation was observed among
conditioned to the length of the tracks and the number of synthetic individuals, which was impossible to discern separately by means of
paths used. We integrated different periods (i.e. 1, 7, 30, and 180 days) 2D space use estimations or the distribution of depth data (Figure 2).
as a surrogate of path lengths and increased the number of synthetic When considering the horizontal plane only, the results obtained
paths from 1 to 100. Finally, we quantified the spatial co‐occurrence from both the collapsed 3D‐UDs and 2D‐UDs calculated with the
of the common dentex during the spawning period (taking place be- dBBMM were highly consistent (Figures 3 and 4b). This robustness is
tween May and June), when individuals punctually migrate to deep remarkable given that the collapsed 3D‐UDs make minimal assump-
rocky bottoms to breed (Aspillaga, 2017). In particular, we computed tions on the underlying nature of the movement process, while the
the overlap between 3D‐UD week‐long estimations for some tracked 2D‐UDs were calculated assuming a conditional Brownian movement
individuals outside and within the spawning period, separately for day between locations (Horne et al., 2007; Kranstauber et al., 2012).
and night periods as defined by local sunset and sunrise times pro- The most relevant differences between the resulting 3D space
vided by the NOAA Solar Calculator (www.esrl.noaa.gov/gmd/grad/ use estimates were related to the topography, which allowed the
solcalc/). effective exclusion of unfrequented shallow areas and unsuitable
emerged landmasses and the redistribution of the UD in those areas
that were in fact suitable for individuals (Figures 3 and 4c, and App
4 | R E S U LT S A N D D I S CU S S I O N
#2 in Supporting Information Appendix S3). By handling and incor-
porating geographical barriers into space use characterizations, our
4.1 | Optimizing the number of synthetic paths
data‐driven approach clearly led to more accurate and realistic 3D‐
Generating each synthetic path proved a computationally intense UD and 2D‐UD estimates (Supporting Information Appendix S2).
and time‐consuming step (45 ± 8 s for a 100‐location path on a
computer with 4GB RAM); hence, the method might be optimized
4.3 | Spatial overlap during the spawning season
by adjusting the number of synthetic paths that are required to
reach a stable 3D‐UD solution. Short tracking periods consisting of A 3D approach provides a unique view of the space use patterns of
fewer locations require large numbers of synthetic paths in order fishes, in particular, when studying the spatial overlap between individ-
to reduce the variability of the resulting space use estimations uals. The overlap indices of any given pair of 3D‐UDs were lower on av-
(Figure 1). Long tracking periods, by contrast, contain many more erage than the overlap values of the 2D‐UDs, but without any apparent
data points that already capture most of the space use variability; correlation between them (Figure 4a). Correspondingly, the aggregating
hence, robust space use volumes are obtained with fewer synthetic behaviour of common dentex during the spawning season (Aspillaga,
paths. In our case, at a temporal resolution of 30 min, one‐day‐long 2017) is strongly evidenced when analysing 3D overlaps. At night, the
tracks (41 ± 8 fixes per track) required almost 70 synthetic paths to 3D overlap between pairs of individuals was significantly higher in the
stabilize the size of the space use volume estimations. In contrast, spawning season compared to the non‐spawning season (Kruskal–
30‐ or more‐day long tracks (more than 1,382 ± 60 fixes per track) Wallis rank‐sum test, H = 34.6, df = 1, p‐value < 0.05, overlaps for the
required only half of the sampling effort, that is about 35 synthetic 95% contour level; Figure 5). These differences were not significant if
paths (Figure 1). 2D‐UD overlap data were used (Kruskal–Wallis rank‐sum test, H = 2.8,
ASPILLAGA et al. Methods in Ecology and Evolu
on
1555 |
F I G U R E 3 Bi‐dimensional representations of utilization distributions of two common dentex individuals (rows) calculated with different
methods: the dynamic Brownian bridge movement model (dBBMM, first column), 3D kernel density estimation on centres of activity
(CoA, second column) and the new proposed method without and with incorporating topographic information (third and fourth columns,
respectively). 3D‐UDs were collapsed to the horizontal plane by vertically summing the UDs of all the depth layers. Polygons indicate, from
the inside out, the 50% and 95% contour levels
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
2D overlap index (dBBMM) 2D overlap index (dBBMM) 3D overlap index (CoA)
F I G U R E 4 Comparison of the overlap indices calculated based on different UD estimations: 2D‐UDs estimated with the Brownian bridge
movement model (dBBMM) versus 3D‐UDs (a) and dimensionally reduced 3D‐UDs (i.e. collapsed to the horizontal plane) (b), and 3D‐UDs
calculated by applying a 3D kernel density estimation on centres of activity (CoA) versus 3D‐UDs. Each point corresponds to the overlap
values between a pair of individuals, for the 50% and 95% probability contours (different colours)
1556 |
Methods in Ecology and Evolu
on ASPILLAGA et al.
(a) 3D−UD − 50% contour (b) 3D−UD − 95% contour F I G U R E 5 Overlap indices between
0.6 No−spawning Spawning No−spawning Spawning volumetric (3D‐UD, a–b) and 2‐
1.0
0.5 dimensional (2D‐dBBMM, c–d) space use
estimations of common dentex individuals
Overlap index
0.8
Overlap index
0.4 Day
Night
0.3 0.6 for consecutive weeks before and during
the spawning period. Panels (a) and (c)
0.2 0.4
correspond to 50% UD contour levels,
0.1 0.2
and panels (b) and (d) to 95% UD contour
0.0 0.0 levels. The weekly and day/night 3D‐UDs
Mar Apr Apr Apr Apr May May May Mar Apr Apr Apr Apr May May May
31−06 07−13 14−20 21−27 28−04 05−11 12−18 19−25 31−06 07−13 14−20 21−27 28−04 05−11 12−18 19−25 of all the individuals can be interactively
explored in the shiny app #3 in Appendix
(c) 2D−UD (dBBMM) − 50% contour (d) 2D−UD (dBBMM) − 95% contour
S3
No−spawning Spawning No−spawning Spawning
1.0 1.0
0.8
Overlap index
0.8
Overlap index
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
Mar Apr Apr Apr Apr May May May Mar Apr Apr Apr Apr May May May
31−06 07−13 14−20 21−27 28−04 05−11 12−18 19−25 31−06 07−13 14−20 21−27 28−04 05−11 12−18 19−25
df = 1, p‐value = 0.1, for the 95% contour level). In accordance with pre- CGL2016‐78156‐C2‐1‐R). EA was supported by the Spanish Ministry
vious works based on 3D approaches (e.g. Vivancos et al., 2016; Lee et of Education (FPU grant, ref. AP2012‐0141). We would also like to
al., 2017), our results corroborate that 2D estimations overestimate the thank five anonymous reviewers and the associate editor for their
spatial overlap because they do not account for the segregation of in- valuable comments that substantially improved the manuscript.
dividuals across depth, and therefore, they provide a partial and biased
view of individual behavioural processes. In comparison with other 3D‐
AU T H O R S ’ C O N T R I B U T I O N S
UD estimation methods, our method redistributes the UD according to
the topography and provides spatially explicit space use estimations. E.A. conceived the project and led the writing of the manuscript;
In the case of the common dentex, our method detected aggregations E.A., K.S. and F.B. developed the method; F.B. and B.H. tested the
of individuals between 40 and 50 meters in the south‐eastern part of method; E.A. and B.H. provided the acoustic telemetry data; all au-
the Medes Islands (App #3 in Supporting Information Appendix S3). thors made significant contributions to the manuscript and gave final
This example illustrates why topography‐wise 3D‐UD estimations are approval for publication.
a valuable tool to detect and visualize susceptible areas, providing pre-
cise information to focus on conservation efforts.
DATA AC C E S S I B I L I T Y
Adler, D., Murdoch, D., Nenadic, O., Urbanek, S., Chen, M., Gebhardt, A.,
Eddelbuettel, D. (2018). rgl: 3D Visualization Using OpenGL. Retrieved
AC K N OW L E D G E M E N T S
from https://cran.r-projec t.org/package=rgl
Ahrens, J., Geveci, B., & Law, C. (2005). ParaView: An End-User Tool for
We are grateful to X. Roijals for his technical support at the
Large-Data Visualization. In C. D. Hansen and C. R. Johnson (Eds.),
Computational Biology Laboratory (CEAB‐CSIC). The work has The Visualization Handbook (pp. 717–731). Oxford, UK: Elsevier Inc.
been supported by the Spanish Ministry of Research (Grant doi:10.1016/B978-012387582-2/50038-1.
ASPILLAGA et al. Methods in Ecology and Evolu
on
1557 |
Aspillaga, E. (2017). Movement ecology of coastal fishes in a marine pro- R Core Team. (2017). R: A Language and Environment for Statistical
tected area: Implications for management and conservation. Barcelona, Computing. Vienna, Austria: R Foundation for Statistical Computing.
Spain: Universitat de Barcelona. /10.13140/RG.2.2.17120.79360 Retrieved from https://www.r-projec t.org/
Aspillaga, E., Bartumeus, F., Linares, C., Starr, R. M., López‐Sanz, À., Díaz, Selby, T. H., Hart, K. M., Fujisaki, I., Smith, B. J., Pollock, C. J., Hillis‐Starr, Z.,
D., … Hereu, B. (2016). Ordinary and extraordinary movement be- … Oli, M. K. (2016). Can you hear me now? Range‐testing a submerged
haviour of small resident fish within a mediterranean marine pro- passive acoustic receiver array in a Caribbean coral reef habitat. Ecology
tected area. PLoS ONE, 11(7), 1–19. https://doi.org/10.1371/journ and Evolution, 6(14), 4823–4835. https://doi.org/10.1002/ece3.2228
al.pone.0159813 Sievert, C. (2018). plotly for R. Retrieved from https://plotly-book.cpsie
Aspillaga, E., Bartumeus, F., Starr, R. M., López‐Sanz, À., Linares, C., vert.me.
Díaz, D., … Hereu, B. (2017). Thermal stratification drives movement Simpfendorfer, C. A., Heupel, M. R., & Hueter, R. E. (2002). Estimation
of a coastal apex predator. Scientific Reports, 7(1), 526. https://doi. of short‐term centers of activity from an array of omnidirectional
org/10.1038/s41598-017-00576-z hydrophones and its use in studying animal movements. Canadian
Belant, J. L., Millspaugh, J. J., Martin, J. A., & Gitzen, R. A. (2012). Multi‐ Journal of Fisheries and Aquatic Sciences, 59(1), 23–32. https://doi.
dimensional space use: The final frontier. Frontiers in Ecology and org/10.1139/f01-191
the Environment, 10(1), 11–12. https://doi.org/10.1890/12.WB. Simpfendorfer, C. A., Olsen, E. M., Heupel, M. R., & Moland, E. (2012).
003 Three‐dimensional kernel utilization distributions improve estimates
Benhamou, S., & Cornélis, D. (2010). Incorporating movement behav- of space use in aquatic animals. Canadian Journal of Fisheries and
ior and barriers to improve kernel home range space use estimates. Aquatic Sciences, 69(3), 565–572. https://doi.org/10.1139/f2011-179
Journal of Wildlife Management, 74(6), 1353–1360. https://doi. Steel, A., Coates, J., Hearn, A., & Klimley, A. (2014). Performance of
org/10.2193/2009-441 an ultrasonic telemetry positioning system under varied envi-
Duong, T. (2007). ks: Kernel density estimation and kernel discriminant ronmental conditions. Animal Biotelemetry, 2(1), 1–17. https://doi.
analysis for multivariate data in R. Journal of Statistical Software, org/10.1186/2050-3385-2-15
21(7), 1–16. https://doi.org/10.1139/F2011-179 Tracey, J. A., Sheppard, J., Zhu, J., Wei, F., Swaisgood, R. R., & Fisher, R. N.
Haverkort, H., Toma, L., & Zhuang, Y. (2009). Computing visibility on ter- (2014). Movement‐based estimation and visualization of space use in
rains in external memory. Journal of Experimental Algorithmics, 13(1), 3D for wildlife ecology and conservation. PLoS ONE, 9(7), e109065.
1.5. https://doi.org/10.1145/1412228.1412233 https://doi.org/10.1371/journal.pone.0101205
Heupel, M. R., Semmens, J. M., & Hobday, A. J. (2006). Automated acous- Veilleux, M. A. N., Lapointe, N. W. R., Webber, D. M., Binder, T. R.,
tic tracking of aquatic animals: Scales, design and deployment of Blanchfield, P. J., Cruz‐Font, L., … Cooke, S. J. (2016). Pressure sensor
listening station arrays. Marine and Freshwater Research, 57, 1–13. calibrations of acoustic telemetry transmitters. Animal Biotelemetry,
https://doi.org/10.1071/MF05091 4(1), 3. https://doi.org/10.1186/s40317-015-0093-0
Horne, J. S., Garton, E. O., Krone, S. M., & Lewis, J. S. (2007). Analyzing Vivancos, A., Closs, G., & Tentelier, C. (2016). Are 2D space‐use analyses
animal movements using Brownian bridges. Ecology, 88(9), 2354– adapted to animals living in 3D environments? A case study on a fish shoal.
2363. https://doi.org/10.1890/06-0957.1 Behavioral Ecology, arw175. https://doi.org/10.1093/beheco/arw175
Hussey, N. E., Kessel, S. T., Aarestrup, K., Cooke, S. J., Cowley, P. D., Wilson, K., Hanks, E. M., & Johnson, D. S. (2017). Estimating animal
Fisk, A. T., … Whoriskey, F. G. (2015). Aquatic animal telemetry: A utilization densities using continuous‐time Markov chain models.
panoramic window into the underwater world. Science, 348(6240), Methods in Ecology and Evolution, 12(10), 3218–3221. https://doi.
1255642. https://doi.org/10.1126/science.1255642 org/10.1111/ijlh.12426
Kessel, S. T., Cooke, S. J., Heupel, M. R., Hussey, N. E., Simpfendorfer, Worton, B. J. (1989). Kernel methods for estimating the utilization distri-
C. A., Vagle, S., & Fisk, A. T. (2014). A review of detection range bution in home‐range studies. Ecology, 70(1), 164–168. https://doi.
testing in aquatic passive acoustic telemetry studies. Reviews in org/10.2307/1938423
Fish Biology and Fisheries, 24(1), 199–218. https://doi.org/10.1007/
s11160-013-9328-4
Kitagawa, T., Nakata, H., Kimura, S., Itoh, T., Tsuji, S., & Nitta, A. S U P P O R T I N G I N FO R M AT I O N
(2000). Effect of ambient temperature on the vertical distribu-
tion and movement of Pacific bluefin tuna Thunnus thynnus ori- Additional supporting information may be found online in the
entalis. Marine Ecology Progress Series, 206, 251–260. https://doi. Supporting Information section at the end of the article.
org/10.3354/meps20 6251
Kranstauber, B., Kays, R., Lapoint, S. D., Wikelski, M., & Safi, K. (2012).
A dynamic Brownian bridge movement model to estimate utilization
How to cite this article: Aspillaga E, Safi K, Hereu B,
distributions for heterogeneous animal movement. The Journal of
Animal Ecology, 81(4), 738–746. https://doi.org/10.1111/j.1365-2656. Bartumeus F. Modelling the three‐dimensional space use of
2012.01955.x aquatic animals combining topography and Eulerian telemetry
Lee, K. A., Huveneers, C., Duong, T., & Harcourt, R. G. (2017). The ocean data. Methods Ecol Evol. 2019;10:1551–1557. https://doi.
has depth: Two‐ versus threedimensional space use estimators in
org/10.1111/2041-210X.13232
a demersal reef fish. Marine Ecology Progress Series, 572, 223–241.
https://doi.org/10.3354/meps12097