Animal movement: the ideas

An animal wearing a GPS collar leaves a trail of points: where it was, and when. From those points ecologists learn how far animals travel, which habitats they use and avoid, where they cross roads, how they respond to people, and which corridors connect their populations. In this first part of the series you will learn what a track is made of, where tracking data come from and how they are stored, and you will build a first model in Terraflux that reads the tracks of eight fishers, small carnivores of the forests of North America, from a file of Movebank, and draws them over a map of land cover. At the end you will have the tracks on a map and a table of the animals, and you will know what to look at before any analysis.

The next parts build on this one: cleaning the tracks and measuring them (distances, speeds, directions, turns), projections, interpolation and simulated tracks, regular steps and the summaries of a track and the data of step selection functions, the method that relates movement to the landscape.

The idea#

A track is a sequence of locations in time#

A tracking device records fixes: at a given time, it tries to work out where it is. A GPS receiver does so from the signals of satellites; an Argos tag from the shift of its own signal received by satellites; a VHF collar is located by people with antennas on the ground. A successful fix gives a location (longitude and latitude, sometimes the height) and a timestamp. A failed fix gives only a time: under a dense canopy, in a burrow or in a steep valley the receiver may not see enough satellites.

The locations of one animal, in time order, form its track (also called a trajectory or a path). Two consecutive locations make a step, and a step has a length, a duration and a direction. Where one step meets the next, the animal turned by some angle: the turn angle.

A toy track: six locations, a step's length, a turn angle and a time lag — The parts of a track. Each orange dot is a location at a time t. A step joins two consecutive locations; its length is the distance between them and its time lag the time between them. At each location the animal turns from the direction it was going (dashed) to the direction of the next step: the turn angle.

Key idea: a track

A track is the sequence of locations \( (x_i, y_i) \) of one animal at the times \( t_0 < t_1 < \dots < t_n \). The step from location \( i \) to \( i + 1 \) has

a length \( d_i \), the distance between the two locations;
a time lag \( \Delta t_i = t_{i+1} - t_i \);
a speed \( v_i = d_i / \Delta t_i \), which is the average speed over the step (the animal did not move in a straight line at a constant speed between two fixes);
a direction (or heading, azimuth): the angle of the step, measured from north;
a turn angle: the direction of the step minus that of the step before, between -180° and 180° (0° straight on, positive to the right).

Everything about a track depends on the sampling interval \( \Delta t \): a fix every two minutes shows a wandering path, a fix every hour only its larger strokes.

The last point is the one to keep in mind through the whole series. The track is not the animal's path; it is a sample of it. The real path runs between the locations, and a longer interval cuts more of its corners. A track sampled every 10 minutes looks straighter and shorter than the same animal sampled every 2 minutes, so lengths, speeds and turn angles can only be compared between tracks sampled at the same rate. Part 4 shows how to make the rates equal.

Why movement matters#

Nathan et al. (2008) proposed a framework that most movement studies now follow, the movement ecology paradigm. An individual's movement results from four components: its internal state (why move: hunger, the search for a mate, fear), its motion capacity (how it moves: walking, flying), its navigation capacity (where to go) and the external factors (the landscape, the weather, other animals). A track records only the outcome, a sequence of locations, but by relating that outcome to the landscape and to time, ecologists infer the components behind it.

GPS collars made this possible for many species. Since the 2000s they have become light and cheap enough for animals of a few kilograms, and record a location every few minutes for weeks or months (Cagnacci et al., 2010). Kays et al. (2015) describe tracking as "an eye on life and planet": the movements of thousands of animals, shared in public archives, show how they respond to roads, cities, fences and a changing climate.

Where tracking data live: Movebank#

Movebank is the largest of these archives, run by the Max Planck Institute of Animal Behavior: researchers store their studies there, and many publish them in its data repository. Its data model (Kranstauber et al., 2011) gives every record the same columns, whatever the device, so that a file of one study reads like a file of another:

Column	What it holds
`timestamp`	the time of the record, in UTC (Coordinated Universal Time)
`location-long`, `location-lat`	the location, in degrees of longitude and latitude (WGS 84)
`individual-local-identifier`	the animal, by the name the study gave it
`tag-local-identifier`	the device it carried
`sensor-type`	the kind of sensor: gps, argos-doppler-shift, acceleration...
`visible`	false for the records Movebank hides (outliers flagged by people or by filters)

A file holds many more columns (the battery's voltage, the number of satellites, the estimated error...) that depend on the device. Movebank exports a study as a CSV file (text, one record per line), and Terraflux reads it as it is.

move2, amt and Terraflux#

Two R packages are the reference tools of the field. move2 (Kranstauber et al., 2024) reads Movebank's files and handles tracks as tables of points: measures, filters, interpolation, lines. amt, the animal movement tools (Signer et al., 2019), turns tracks into steps and builds the data of habitat selection analyses. Terraflux's operators named move.* do what their functions do, and give the same numbers: they are tested against both packages on the same data, to the last bit. Each operator's reference page names the function it reproduces (mt_read, track_resample, random_steps...), so you can follow a paper's methods written in R and repeat them here, step by step, in a graph.

The data: eight fishers in New York#

The fisher (Pekania pennanti, formerly Martes pennanti) is a weasel of the northern forests of North America, about the size of a house cat: males weigh 3.5 to 6 kg, females about half as much. It was trapped out of much of its range in the 19^th and 20^th centuries and has since come back, including to the patchy, suburban forests around Albany, New York. LaPoint et al. (2013) put GPS collars on fishers there to see how they move through such a landscape, and whether the corridors predicted by models of landscape resistance are the ones the animals actually use. Their data are on Movebank (the study "Martes pennanti LaPoint New York") and are the example data of move2 and amt, which makes them a good set to learn with: you can compare every number of this series with the packages' documentation.

Download the file to follow along:

fishers.csv.gz (700 kB): Movebank's export of the study, compressed with gzip. Terraflux reads .csv, .csv.gz and Movebank's .zip files directly.

The file has 47,347 records of eight animals, three females (F1 to F3) and five males (M1 to M5), between February 2009 and May 2011:

animal	sex	records	locations	share located (%)	first	last	median interval (min)
F1	female	2,795	1,349	48.3	2011-02-11	2011-03-03	10
F2	female	4,689	3,004	64.1	2010-12-16	2011-01-04	2
F3	female	3,383	1,501	44.4	2011-01-21	2011-02-07	2.1
M1	male	1,990	919	46.2	2009-02-11	2009-03-04	15
M2	male	3,573	1,638	45.8	2011-02-10	2011-03-07	10
M3	male	5,013	2,436	48.6	2011-01-19	2011-02-11	2.2
M4	male	10,846	8,958	82.6	2010-02-09	2010-03-31	2
M5	male	15,058	13,099	87	2011-03-22	2011-05-28	2

The eight fishers. Records are the fix attempts; locations the ones that gave a position. The median interval is the time between two consecutive locations.

Two things stand out already. The animals were not followed at the same time, nor at the same rate: five of them were fixed about every 2 minutes, two every 10 and one every 15. And many attempts failed: for six animals less than two thirds of the records have a location, and for F3 fewer than half.

When to use it#

Reading and drawing the tracks is the first step of every movement analysis. The tracks then answer questions such as:

How far and how fast do animals move? Daily distances, speeds and the time spent moving or resting (part 2).
How much space do they use? The extent of their movements and how it grows over time, an indication of a home range or of dispersal (part 4).
When are they active? Movement by day and by night (part 4).
Which habitats do they select as they move? Step selection functions compare each step taken with the steps the animal could have taken (part 5).
Where do they cross a barrier, a road or a river? The steps that cross it, drawn as lines (part 2).

Terraflux does not yet estimate home ranges (minimum convex polygons, kernel densities, autocorrelated kernel densities) or behavioural states (hidden Markov models); for those, R's amt, ctmm and moveHMM read the same files. If your question is about where a species can live rather than how individuals move, see the species distribution models; if it is about how the landscape resists movement, see circuit connectivity and grid distances, whose resistance surfaces movement data can calibrate and test.

What you need#

Tracks: a Movebank file (CSV, CSV compressed with gzip, or a zip), or any table with a column of x (or longitude), one of y (or latitude), one of time and one naming the animal; move.as_tracks turns such a table, or a layer of points, into tracks (part 3).
Times as text in the ISO 8601 form (2010-03-05 17:02:00, with an optional fraction of a second and a time zone such as Z or -05:00), read as UTC; or numbers (seconds, days...) for simulated tracks.
A coordinate system: Movebank's locations are in longitude and latitude (EPSG:4326). The measures of part 2 work on the ellipsoid directly; part 3 shows when to project the tracks.
A map of the landscape for the figures and, in part 5, for habitat selection. Here, ESA's WorldCover map of land cover for 2021 (10 m), which Terraflux downloads with data.download.

Build the model#

The model has four nodes: one reads the file, one removes the records without a location, and two summarise the tracks, as lines and as a table of the animals.

Read the fishers' tracks (node 1), keep the records with a location (node 2), draw each track as a line (node 3) and list the animals (node 4).

Model file (TOML)

format = "terraflux/1"

[[nodes]]
id = "read"
op = "move.read_tracks"
params = { path = "fishers.csv.gz" }
note = "The fishers of LaPoint et al. (2013), Movebank's export"

[[nodes]]
id = "located"
op = "vector.filter"
inputs = { vector = "read.tracks" }
params = { expression = "not isnull(geometry.x)" }
note = "The records with a location"

[[nodes]]
id = "lines"
op = "move.track_lines"
inputs = { tracks = "located.vector" }

[[nodes]]
id = "animals"
op = "move.track_data"
inputs = { tracks = "read.tracks" }

Read the tracks. Add move.read_tracks (library subject Animal movement) and set its path to fishers.csv.gz in the Node pane (the button beside the setting opens a file dialog that lists Movebank's files). The node reads every record as a point in longitude and latitude, with all the file's columns as attributes. It reads the columns as R and pandas would: numbers as numbers, the flags true and false as 1 and 0, the rest as text. It then orders the records by animal and by time, which every other move node needs; Movebank's files are usually in that order already, but not always (the setting sort = false keeps the file's order, as move2 does). The animal is the column individual-local-identifier; when an animal carried more than one tag during a study, the node adds a column individual-tag-local-identifier that names the animal and the tag, and takes the tracks from it, as move2 does.
Keep the records with a location. The 14,443 records whose fix failed have a time but no coordinates: they come into Terraflux as points with an empty geometry. Most measures are undefined for them, so drop them with vector.filter (subject Vectors) and the expression not isnull(geometry.x): geometry.x is the longitude of each point, missing where there is no point, and isnull tests for a missing value. Connect read.tracks to the filter's input vector. The filter keeps the order of the records, so its output is still sorted.
Draw each track as a line. move.track_lines joins the locations of each animal in time order into one line, with the number of locations (n_locations) and the columns that do not change within the track (the animal's name, its tag, the study). Connect located.vector to its input tracks.
List the animals. move.track_data gives one row per animal with those same constant columns and no geometry: a table of the individuals, like move2's mt_track_data. It can read the records before or after the filter; here it reads them all.

Run the model with Ctrl+Enter on the last node, or from the command line:

terraflux run fishers.toml --out results

The command writes the model's outputs: add them to the file, one section per output, such as

[outputs.lines]
from = "lines.lines"

Vectors are written as GeoPackage files (results/lines.gpkg), which QGIS, R's sf and Python's geopandas read; vector.write writes one where you choose, in any format GDAL writes.

In the application

Beside any map, the viewer's Layers keep outputs on the map: Add a layer lists every vector output of the run; add the tracks or their lines to draw them over the land cover, reprojected to it, and choose their style. Hover over a location to read its attributes (its timestamp, its animal, the collar's columns): the quickest way to check that the time and the animal are the columns you expect.

Read the results#

The tracks on the map#

The tracks of the eight fishers over the land cover — The eight fishers' tracks (move.track_lines) over ESA WorldCover 2021 near Albany, New York: tree cover in green, cropland in yellow, built-up land in pink, water in blue. The river crossing from the west is the Mohawk, which joins the Hudson (running north to south near 73.7° W). The pink crosses are runways: Albany's airport near M3 and F3, a smaller airfield north of M4.

The first look at a set of tracks answers simple questions that are easy to forget later. Are the locations where they should be? (A location in the sea, or a thousand kilometres away, is a fix gone wrong or a collar being shipped.) How large an area does each animal cover? Here the answer is a few kilometres: the fishers stay in the forest patches between suburbs, roads and fields, and no track crosses the Hudson. Do the tracks overlap? F3's track lies under M3's, which hides it at this scale, and M1's and F2's share the same woods two years apart.

When each animal was followed#

The period each fisher was tracked — When each fisher was tracked, and how many locations it gave. Each animal was followed for two to ten weeks, in late winter or spring, in one of three years.

The animals were followed for 2 to 10 weeks each, in winter and spring, in 2009, 2010 and 2011. An analysis that pools them pools different seasons and years: a difference between two animals may be a difference between two winters. Studies report this: the number of animals, the dates and the number of locations of each.

A day of a fisher#

One day of M4: its locations, and the attempts that failed — One day of M4, from noon on 5 March 2010 to noon the next day (local time). Left: its 248 locations, coloured by the hours since noon. Right: the times of the fixes (green) and of the attempts that gave no location (grey).

Zooming in on one day shows the sampling. M4 was located 248 times in 24 hours, but not regularly: the fixes come in dense runs in the evening and at night and are sparse by day, and 33 attempts gave no location. The fisher rested at the east end of its range in the afternoon (the tight cluster of dark points), set off in the evening, looped through the woods all night, and was resting again in the west the next morning (the yellow points). The steps between its fixes add up to about 12 km, though some of that is the error of the fixes (part 2). This is typical of GPS data: the interval is a setting of the collar, but the fixes actually obtained depend on the animal (resting in a hollow tree blocks the signal) and on the collar's battery. Part 4 turns such irregular tracks into steps of a regular interval.

What to report#

From this first model a methods section takes: the number of animals and of locations; the period of tracking of each; the scheduled interval of the fixes and the share of attempts that gave a location (the fix success rate); and the source of the data, with its DOI. For these fishers: 8 animals, 32,904 locations from 47,347 attempts (69.5 %), intervals of 2, 10 or 15 minutes, February 2009 to May 2011 (LaPoint et al., 2013, with the DOI of the data).

Settings that change the answer#

move.read_tracks has one setting besides the path: sort. With sort = true (the default) the records are put in order of animal (in the order the animals first appear) and time; ties keep the file's order. Every other move node expects this order and stops with an error when the times of a track go backwards. Keep the default unless you know the file is ordered and want its order exactly.

The other move nodes have two settings in common, time and track: the columns of the time and of the animal. They are found by their usual names (timestamp, time, datetime... for the time; individual-tag-local-identifier, individual-local-identifier, track, id, animal... for the animal), so with Movebank's files you never set them. With a table of your own, name the columns there, or set the settings. Changing the track column changes the answer: tracks by tag instead of by animal split an animal whose collar was replaced into two tracks, and the step between the old collar's last fix and the new one's first disappears.

Common pitfalls#

Time zones. Movebank stores times in UTC, and Terraflux reads every time without a zone as UTC. The fishers' winter night is between about 22:00 and 12:00 UTC. A table of your own in local time must say so (a zone such as -05:00 at the end of each time), or the day and night of part 4 will be shifted.
Records without a location. They are kept by move.read_tracks, because some analyses need them (the fix success rate, or the other sensors of a collar). Drop them before measuring; the measures stop with an error if some remain, and tell you to.
Tracks out of order. A table joined or appended from several files may not be sorted. Keep sort = true, or sort it with move.as_tracks.
Different sampling rates. Do not compare distances, speeds or turn angles between animals fixed at different intervals; resample them first (part 4).
Names of species. The study, and the file's individual-taxon-canonical-name, use the former name Martes pennanti; the species has been Pekania pennanti since a genetic study moved it out of Martes. Search for both in the literature.

Going further#

Clean the tracks and measure them: Movebank's filters, duplicates, distances, speeds, directions and turns.
Projections, interpolation and simulated tracks.
Regular steps and track summaries.
The data of step selection functions.
The operators' reference: move.read_tracks, move.track_lines, move.track_data, vector.filter.

References#

Cagnacci, F., Boitani, L., Powell, R. A., & Boyce, M. S. (2010). Animal ecology meets GPS-based radiotelemetry: a perfect storm of opportunities and challenges. Philosophical Transactions of the Royal Society B, 365(1550), 2157-2162. https://doi.org/10.1098/rstb.2010.0107
Kays, R., Crofoot, M. C., Jetz, W., & Wikelski, M. (2015). Terrestrial animal tracking as an eye on life and planet. Science, 348(6240), aaa2478. https://doi.org/10.1126/science.aaa2478
Kranstauber, B., Cameron, A., Weinzerl, R., Fountain, T., Tilak, S., Wikelski, M., & Kays, R. (2011). The Movebank data model for animal tracking. Environmental Modelling & Software, 26(6), 834-835. https://doi.org/10.1016/j.envsoft.2010.12.005
Kranstauber, B., Safi, K., & Scharf, A. K. (2024). move2: R package for processing movement data. Methods in Ecology and Evolution, 15(9), 1561-1567. https://doi.org/10.1111/2041-210X.14383
LaPoint, S., Gallery, P., Wikelski, M., & Kays, R. (2013). Animal behavior, cost-based corridor models, and real corridors. Landscape Ecology, 28(8), 1615-1630. https://doi.org/10.1007/s10980-013-9910-0
LaPoint, S., Gallery, P., Wikelski, M., & Kays, R. (2013). Data from: Animal behavior, cost-based corridor models, and real corridors. Movebank Data Repository. https://doi.org/10.5441/001/1.2tp2j43g
Nathan, R., Getz, W. M., Revilla, E., Holyoak, M., Kadmon, R., Saltz, D., & Smouse, P. E. (2008). A movement ecology paradigm for unifying organismal movement research. Proceedings of the National Academy of Sciences, 105(49), 19052-19059. https://doi.org/10.1073/pnas.0800375105
Signer, J., Fieberg, J., & Avgar, T. (2019). Animal movement tools (amt): R package for managing tracking data and conducting habitat selection analyses. Ecology and Evolution, 9(2), 880-890. https://doi.org/10.1002/ece3.4823
Zanaga, D., Van De Kerchove, R., Daems, D., De Keersmaecker, W., Brockmann, C., Kirches, G., ... Arino, O. (2022). ESA WorldCover 10 m 2021 v200. Zenodo. https://doi.org/10.5281/zenodo.7254221