Faculty of Science
USE OF UNMANNED AERIAL VEHICLES TO QUANTIFY SOCIAL AFFILIATIONS IN
BEEF CALVES
2017 | JUSTIN TERRANCE MUFFORD

B.Sc. Honours thesis - Biology

USE OF UNMANNED AERIAL VEHICLES TO QUANTIFY SOCIAL AFFILIATIONS IN
BEEF CALVES
by
JUSTIN TERRANCE MUFFORD

A THESIS SUBMITTED IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF

BACHELOR OF SCIENCE (CO-OP, HONS.)
in the
DEPARTMENT OF BIOLOGICAL SCIENCES
(Animal Biology)

This thesis has been accepted as conforming to the required standards by:
John S. Church (Ph.D.), Thesis Supervisor, Dept. Natural Resource Sciences
Garrett Whitworth (Ph.D.), Examining Committee member, Dept. Natural Resource Sciences
Matt Reudink (Ph.D.), Examining Committee member, Dept. Biological Sciences
Nancy Flood (Ph.D.), Examining Committee member, Dept. Biological Sciences

Dated this 16h day of May, 2017, in Kamloops, British Columbia, Canada

© Justin Terrance Mufford, 2017

ABSTRACT
We developed a novel method to quantify cow-calf social affiliations by using unmanned aerial
vehicles (UAVs) to quantify the distance between pairs of individuals in a herd. Using two UAVs
equipped with video-capable digital cameras, as well as photogrammetry and image analysis
software, we measured the distance between cow-calf pairs. We hypothesized that this distance
can be accurately measured by UAVs and has the potential to provide valuable information in
studies of animal behaviour and welfare. Twelve cow-calf pairs were uniquely identified with the
use of penning tags and coloured fabric squares glued to their backs. Aerial videos of the calves
held with their dams in a pasture were made over four days using UAVs. Recordings were made
during two periods on most days of study: Daytime (Day) and Evening (Eve). Calves were
acclimated to the UAVs by flying one UAV intermittently at distance of 10 m away from the calf
on the days before experimental recording began. We used two UAVs to video-capture the
following: (1) the location of all individuals (UAV flown at 100 m) and; (2) the identity of cowcalf pairs (UAV flown at 15 - 30 m). Still-images extracted from the UAV-acquired video
screenshots were used to produce orthomosaics using Agisoft Photoscan software to correct
geometric distortions from the original images. The orthomosaics captured all of the cows and
calves in a single image, from which we measured the distance between calves and their dams and
the distance between calves and non-related cows in the herd. We constructed a mixed model to
test the effect of time period and cow-calf relationship (related or non-related) on cow-calf
distance. Time period and cow-calf relationship each had a statistically significant effect (F1,849 =
101.79 and F1,849 = 26.75, respectively, P < 0.001) on the distance between cows and calves, but
there was no significant interaction between these variables (F1,849 = 1.35, P = 0.245). The mean
distance between calves and their dams (X̅ = 43.8 m, SD = 24 m) was smaller than that between
calves and non-related cows (X̅ = 55.6 m, SD = 28 m). The mean cow-calf distance was smaller
in the daytime (X̅ = 39.8 m, SD = 15 m) than in the evening (X̅ = 84.3 m, SD = 24 m). Distance
measurements were tested for accuracy using reference lengths that were measured in the field;
the mean difference between our software-measured length and the true length was 0.9 ± 0.7 m
(mean and standard deviation). Our novel and non-invasive method appears to be sufficiently
accurate and precise to study social behaviour in beef cattle in a field setting.
Thesis Supervisor: Dr. John Scott Church

ii

ACKNOWLEDGEMENTS
I would like to sincerely thank my supervisor, Dr. John Church, my honours committee, and other
collaborators involved in this project as well as my friends and family for their wonderful support
and help throughout the duration of this project. John, thank you for continuously encouraging and
empowering me to grow as an aspiring research student; I have always felt so appreciated and
trusted by you to develop creative ideas and work to the best of my ability. Nancy, thank you for
always providing me help and support and answering my never-ending supply of questions, even
in the most ridiculous situations. Thanks to my other collaborators for technical advice which truly
made a difference in the outcome of this project: Garrett Whitworth, Mark Paetkau and David Hill.
Thank you to my family for the love and support throughout periods of stress and exhaustion.
Thank you to my friends for their support, reminding me to laugh and to stop and smell the roses.
Thank you to our collaborators from both the AAFC Lethbridge Research Station and U of C for
use of the facilities and research herd. It was a pleasure to work with such a strong and unified
team in Lethbridge.

iii

TABLE OF CONTENTS
ABSTRACT .................................................................................................................................... ii
ACKNOWLEDGEMENTS ........................................................................................................... iii
List of figures ................................................................................................................................. iv
INTRODUCTION .......................................................................................................................... 1
Methods & Materials ...................................................................................................................... 6
Experimental Design................................................................................................................... 6
Habituation to UAVs .................................................................................................................. 7
Field Data Collection .................................................................................................................. 7
Data Acquisition ......................................................................................................................... 8
Measuring cow-calf distances ................................................................................................. 8
Testing data reliability .......................................................................................................... 11
Statistical Tests ......................................................................................................................... 11
Results ........................................................................................................................................... 12
Discussion ..................................................................................................................................... 14
Cow-Calf Affiliations ............................................................................................................... 14
Accuracy of Measurements....................................................................................................... 15
Future Use of UAVs in Studying Cattle Behaviour.................................................................. 17
Literature cited .............................................................................................................................. 20

LIST OF FIGURES
Figure 1. A diagram of a camera sensor receiving light rays at angle, taken from (Trac 2010a).
This diagram illustrates how an angled camera results in tilt displacement. The distance between
A’ and B’ is equal to the distance between C’ and D’ but this is distorted on the camera sensor as
shown by this diagram. ................................................................................................................... 4
Figure 2. A diagram of light rays entering a light sensor on a camera, taken from (Trac 2010b).
This schematic is an illustration of relief displacement. The distance between A’ and B’ is equal
iv

to the distance between C’ and D’ but this is distorted on the camera sensor as shown by this
diagram. .......................................................................................................................................... 5
Figure 3. A picture of calves and cows obtained by taking a screenshot of video recorded by a
Phantom 4 drone at an altitude of 20 m. Cows and calves were identified by coloured-felt and
numbered penning tags glued to their back with tagging cement. Each calf and its dam had the
same numbered tag and colour of fabric glued on their backs. Twelve different numbers and
colours were assigned to the herd to distinguish each pair. For example, in this photo the CalfDam pair that possesses the combination of the green felt fabric and penning tag #1 is distinguished
from the other Calf-Dam pair that possesses the magenta felt fabric and penning tag #3. The
combination of colour and number was necessary since sometimes the glare from the sunlight
made it difficult to identify the cattle by penning tag number alone. ............................................. 8
Figure 4a-d. Video screenshots of the cattle herd in the research pasture collected from a DJI
Phantom 4 drone at an altitude of 100 m (a-c). Multiple photos of the herd were orthorectified
together to make a single image (orthomosaic) of the entire herd (d). ........................................... 9
Figure 5. An example of an orthorectified image (orthomosaic) used to determine cow-calf
distances (a). The number of pixels between the penning tag of a cow and a calf were measured
using ImageJ. A reference length measured in the field was used to calibrate the image to obtain a
pixel to meter ratio. Two photos obtained from taking a screenshot from the UAV flying at a height
of 15 m are shown (b and c) around the orthomosaic (a). Photos screenshotted from the low-flying
drone were used to identify each individual in the orthomosaic. The magenta, blue, and green
arrows in photo “a” are matched with arrows in photos “b” and “c” to illustrate how the cattle in
the orthomosaic are matched with the photos obtained from the low-flying drone. .................... 10
Figure 6. An orthomosaic used to test the accuracy of the distance measurements made on ImageJ
software. In this example, reference 1 is a feeding trough used to calibrate the pixel to meter ratio
on ImageJ and reference 2 is the distance between two distinct fence posts used to compare our
software-measured length and the length measured in the field. .................................................. 11
Figure 7. A box plot showing the distribution of cow-calf proximity by relationship and time
period. Boxes indicate the interquartile range, lines across the boxes indicate the median, the
whiskers indicate the upper and lower limits, and the asterisks indicate outliers......................... 13
Figure 8. The distribution of cow-calf distances for individual pairs. .......................................... 14

v

INTRODUCTION
Cattle raised by commercial producers form gregarious and complex social structures (Sowell et
al. 1999; Schein & Fohrman 1955). Understanding the social behaviour of cattle raised in these
commercial settings is important to optimize beef cattle production. For example, pregnant cows
prefer to isolate themselves from the herd in the hours prior to giving birth (Finger et al. 2014).
Commercial production systems that do not provide enough space for birthing cows risk interfering
with important bonding behaviours, which occur optimally in isolation from the rest of herd. This
could result in the breakdown of maternal behaviour, including rejection of the calf, which reduces
its ability to grow (von Keyserlingk & Weary 2007). Furthermore, social dominance plays a large
role in feeding behavior; overcrowding and improper grouping of cattle can negatively impact
feeding behaviour and therefor production (Grant & Albright 2001; Phillips & Rind 2001). While
there are a multitude of social behaviours that can be studied in cattle, the focus of this study was
social affiliations. Social affiliations, (aka social contacts, social interactions, and social
associations) are fine-scale behavioural interactions between two or more individuals within a herd
(e.g., cow-calf, cow-bull), which are often of interest to cattle producers (O’Neill et al. 2014).
Social affiliations have been investigated both quantitatively and qualitatively for decades, by
using field observations and with the use of technology such as video cameras, radio telemetry and
global positioning system (GPS) devices.
For behavioural research purposes, cattle can be fitted with collars equipped with global
positioning system (GPS) or global navigation satellite system (GNSS) devices, to obtain their
locations at consistent sampling frequencies (Handcock et al. 2009), which depend on the
ecological and/or behavioural questions being addressed (Anderson et al. 2012). The accuracy of
position measurement made using GPS varies between different models of collars and the data
analysis tools used to manage the raw GPS co-ordinates. The accuracy ranges from 1 or 2 m to 15
m (reviewed by Anderson et al. 2013). The accuracy has also been shown to be affected by the
topography and habitat of the area in which the GPS is used (Lewis et al. 2007).
In the past, cattle have also been fitted with collars equipped with ultra-high frequency (UHF)
radio devices to observe fine-scale social interactions (Swain & Hurley 2007). These devices,
known as contact loggers or proximity loggers, emit unique radio frequencies, and receive
emissions from other devices, to log the frequency and duration of close contacts between two
1

individual animals. This method of quantifying social affiliations has been commonly employed
to study various aspects of social behaviour, such as interactions with other species (Drewe et al.
2012), and social preferences (Patison et al. 2010; Swain & Hurley 2007), as well as to identify
cows in estrus based on cow-bull contacts (O’Neill et al. 2014). Cow-calf social affiliations have
also been quantified manually, by visually estimating the distances between individuals
(Daleszczyk et al. 2004) and by estimating the coordinates of individual animals with the use of
field markers (Hirata et al. 2003) and references to field maps (Veissier et al. 1990). These
measurements have been used to study social preferences, maternal behaviour, and the dependence
of calves to their dams (Hirata et al. 2003; Veissier et al. 1990).
Studying social behaviour using GPS collars and radio telemetry technology has been shown to be
reliable; however, these devices can be costly and their use requires a high level of expertise.
Furthermore, problems associated with these devices include limited battery life, low accuracy
(Anderson et al. 2013), and a tendency to be damaged by weather and/or the wearer. Manual
observation has been proven to be cost-effective; however, observers in the field can impact the
behaviour of cattle thus affecting the validity of results. It has been demonstrated in standardized
behavioural tests that humans, whether they are stationary or moving, influence cattle behaviour
(reviewed by Waiblinger et al. 2006). Indeed, cattle, among other farm animals, develop
relationships with humans such that the animals can perceive human interaction as positive,
neutral, or negative, depending on their previous experiences and their underlying personality traits
(reviewed by Waiblinger et al. 2006). As an alternative to these techniques, we developed a novel
method to quantify cow-calf social affiliations using unmanned aerial vehicles (UAVs) to
determine the position of individuals in the field, using photogrammetry techniques and photo
editing software that is efficient and non-invasive.
Over the last decade, remarkable technological advances have been made in the development of
UAVs (Whitehead et al. 2014a; Whitehead et al. 2014b). UAVs are desirable for research because
of their low operational cost, ease of use, and flexibility in difficult terrain and adverse weather
conditions; they can be equipped with a variety of sensors (e.g., hyperspectral, multispectral,
infrared), produce little noise, reduce the risk of dangerous interactions with study subjects, and
eliminate the effect of human presence on animal behaviour (Pajares 2015; Moreland et al. 2015;
Matese et al. 2015; Klemas 2015; Koh & Wich 2012). These factors have led to a substantial
2

increase in the use of unmanned aerial vehicles in wildlife monitoring (reviewed by Linchant et
al. 2015), studies of wildlife ecology (reviewed by Anderson & Gaston 2013), and agriculture
(reviewed by Zhang et al. 2012). UAVs have been used to survey an incredibly diverse array of
taxa including songbirds (e.g., Spizella passerina) (Wilson et al. 2017), seabirds (e.g., Fregata
ariel) (Hodgson et al. 2016), seals (Halichoerus grypus) (Seymour et al. 2017), humpback whales
(Megaptera novaeangliae) (Hodgson et al. 2017), elephants (Loxodonta africana) (Vermeulen et
al. 2013), dugongs (Dugong dugon) (Hodgson et al. 2013) and black rhinos (Diceros bicornis)
(Mulero-Pazmany et al. 2014). Recently, they have also been used to study animal behaviour. In
2017, Hodgson et al. used quadcopters to follow humpback whales near the coast of Australia. In
this study, the drones were automated to follow pods of whales, allowing the researchers to observe
the travelling, diving and grouping of pods. Researchers at the USGS Alaska Science Centre flew
small unmanned aircraft over sea otter habitat to observe their foraging behaviour without
disturbing them (UAF 2017). UAVs have also been employed in the cattle industry to count cattle
in large feedlots with a reasonable level of accuracy, using both video cameras (Chamoso et al.
2014) and thermal cameras (Whitehead et al. 2014b). In addition, UAVs have been used to predict
feed intake in cattle kept in small feeding pens (Nyamuryekung’e et al. 2016). Currently, we are
the first to combine the use of UAVs and photogrammetry to study the fine-scale social behaviour
(i.e., social affiliations) of cattle in a field setting.
Photogrammetry is a set of techniques used to obtain quantitative and qualitative data on the
position of objects and the spaces between them from photographs (Ey-Chmielewska et al. 2015).
One common application of photogrammetry is the construction of 2-dimensional and 3dimensional topographic maps, which have been widely used in agriculture to obtain information
on crop health (reviewed by Candiago et al. 2015). Another common application of
photogrammetry is the construction of an orthomosaic, a 2-dimensional overhead-view image
constructed from multiple overlapping photos through a technique called orthorectification, which
corrects for geometric distortion caused by the nature of aerial photography (Laliberte et al. 2010;
Afek & Brand 1998). Geometric distortion in photos alters the true relative positions of points on
the image such that the length of an object, the length of a given area, or the distance between
different points on the image, can appear longer or shorter than it actually is. Distance and angle
of the camera (i.e., the perspective from which a photo is taken) can greatly distort the relative
positions between objects in an image, especially an aerial photograph captured at a tilted angle,
3

rather than perfectly vertical (tilt displacement) (Figure 1) (Verhoeven et al. 2012). Furthermore,
the magnitude of distortion worsens as the altitude at which the photo is being taken increases. A
tilt in the camera causes the distance between two points on the image to appear longer or shorter
than they actually are because when a camera sensor receives light at an angle (Figure 1), the light
rays hitting the camera sensor are displaced, which results in a distorted image. The higher the
altitude of an angled camera, the greater the displacement of light rays on the camera sensor.

Figure 1. A diagram of a camera sensor receiving light rays at angle, taken from (Trac 2010a). This diagram illustrates how an
angled camera results in tilt displacement. The distance between A’ and B’ is equal to the distance between C’ and D’ but this is
distorted on the camera sensor as shown by this diagram.

4

The topography of an area can also distort the relative positions between two different points on a
photograph (Figure 2). This is known as relief displacement.

Figure 2. A diagram of light rays entering a light sensor on a camera, taken from (Trac 2010b). This schematic is an illustration
of relief displacement. The distance between A’ and B’ is equal to the distance between C’ and D’ but this is distorted on the
camera sensor as shown by this diagram.

The orthorectification process involves using software that builds a 3D model of a given area from
2D aerial photos. Mathematical computation is used to determine 3D coordinates from multiple
photos of overlapping images. The software then uses the 3D model to correct for geometric
distortions to build an accurate 2D orthomosaic. To our knowledge, few studies combine the use
of UAVs and photogrammetry techniques for animal conservation and management purposes.
Seymour et al. (2017) developed a method for automated detection and enumeration of grey seal
(Halichoerus grypus) colonies. This method consists first, of using a UAV to capture infrared and
RGB photos at designated intervals on an autonomous flight path; then these photos are
5

orthorectified and merged to create a single orthomosaic image; this image is further processed by
running algorithms programmed for automated detection and enumeration of individuals and
automated discrimination between pups and adults. Durban et al. (2015) piloted a technique using
UAVs for collecting anatomical information on killer whales (Orcinus orca). In this pilot project,
a UAV was deployed from the deck of a boat and used to capture overhead photos of the whales;
these photos were calibrated for measuring the length of the backs of individuals using computer
software.
The objective of this study was to develop and examine the use of a novel, non-invasive method
to quantify social affiliations of beef calves (Bos taurus) using UAVs and photogrammetry
techniques. We hypothesize that our method can be used to quantify the distance between cowcalf pairs in a field setting with a sufficient level of accuracy and precision to study behavioural
interactions between individual animals.

METHODS & MATERIALS
Experimental Design
Twelve Angus/cross beef cows (Bos taurus) and their single, 4-month-old calves were studied over
a 4-day period during the month of June, 2016 at the Agriculture & Agri-Food Canada (AAFC)
Lethbridge Research Centre, located near the city of Lethbridge, AB, Canada, while the cow-calf
pairs were on pasture. Each cow-calf pair was uniquely identified with the use of penning tags
(Wheel'n'Tree Enterprises, www.penningsupplies.com) and coloured fabric squares (22.9cm x
50.4cm) glued to their backs with tagging cement. We collected aerial data twice per day in two
different time periods: Daytime (Day) and Evening (Eve) for a total of six time periods throughout
the study (poor weather precluded data collection in two of the eight possible time periods). The
six time periods are designated Day 1 Daytime, Day 2 Daytime, Day 2 Evening, Day 3 Daytime,
Day 4 Daytime, and Day 4 Evening. The Daytime period was between 10:00 and 16:30 hours,
while Evening period was between 17:30 and 21:30 hours. We attempted to collect data every 15
minutes within each time period, but poor weather prevented the collection of data at certain times;
as a result, a different number of sampling points (i.e., times at which samples were taken) was
included in each time period: Day 1 Daytime, 7 points; Day 2 Daytime, 17 points; Day 2 Evening,
12 points; Day 3 Daytime, 13 points; Day 4 Daytime, 14 points; Day 4 Evening, 13 points (total =
76 sampling points). At each sampling point, we measured the distance between the calves and
6

their dams as well as the distance between the calves and each of the other 11 cows in the herd.
Thus, at each sample point, we made 12 distance measurements for the calf-dam pairs and 122
distance measurements between calves and non-related cows. We then calculated the mean calfdam distance and the mean distance between calves and non-related cows at each time point.
This made two groups according to calf-cow relationship, which we called “Calf-Dam” and “CalfOther”. The four daytime periods (51 sample points) and the two evening periods (25 sample
points) made a Daytime (“Day”) group and Evening (“Eve”) group, respectively. We compared
the mean distance between cow-calf pairs by cow-calf relationship (i.e., Calf-Dam vs Calf-Other)
and by time period (i.e., Day vs Eve).

Habituation to UAVs
Two Phantom 4 DJI drones (Dà-Jiāng Innovations Science and Technology Co., Ltd) equipped
with 12-megapixel video cameras were used in the study. Three days prior to the data collection,
we allowed cattle to acclimatize to the UAVs hovering above and remaining stationary, in front
of, to the side of, and behind them, at a height of 10 m. In addition, we allowed cattle to acclimatize
to the UAVs flying past them up to speeds of 65km/hr at a height of 30 m. On the first day of the
habituation period, we observed that each individual showed reactions to the drone when the UAV
was within 10 m. The behaviours we observed included raising their head directly upwards to look
at the UAV when it was hovering in front and above the individual at a height of 10 m. We also
observed cattle raising their heads and turning their heads towards the side of their flank to look at
the UAV when it was hovering behind them or at a lateral distance of up to 10 m and a height of
10 m. After repeating this for three days, we observed that none of the cattle showed any
observable behavioural response to the UAVs when they were hovering within 10 m of them (in
any direction). We did not observe cattle responding to UAVs flying past them at a height of 30m
throughout the habituation period.

Field Data Collection
At each of the 76 sampling points throughout the data collection period, we flew both DJI Phantom
4 UAVs simultaneously, but at different altitudes. One pilot flew at an altitude of 100 m and the
other pilot flew at an altitude of between 15 and 30 m. Both pilots video recorded the herd at 4K
resolution and 24 frames per second (fps). The video recorded at 100 m was used to generate a
composite image of the entire herd and the video recorded at an altitude varying between 15 and
7

30 m was used to identify each individual cow and calf on the composite images (Figure 3) (See
data acquisition for more details).

Figure 3. A picture of calves and cows obtained by taking a screenshot of video recorded by a Phantom 4 drone at an altitude of
20 m. Cows and calves were identified by coloured-felt and numbered penning tags glued to their back with tagging cement. Each
calf and its dam had the same numbered tag and colour of fabric glued on their backs. Twelve different numbers and colours were
assigned to the herd to distinguish each pair. For example, in this photo the Calf-Dam pair that possesses the combination of the
green felt fabric and penning tag #1 is distinguished from the other Calf-Dam pair that possesses the magenta felt fabric and
penning tag #3. The combination of colour and number was necessary since sometimes the glare from the sunlight made it difficult
to identify the cattle by penning tag number alone.

Data Acquisition
Measuring cow-calf distances
At every sampling time point, we took screenshots of the collected video to obtain jpeg images of
the cattle herd. In most of the videos, the herd was too widely spaced to capture every individual
in a single frame. Thus, we used multiple, overlapping screenshots to capture the entire herd,
orthorectifying the photos together to generate a single, orthorectified image (an orthomosaic)
8

using photogrammetry software (Agisoft Photoscan Pro) (Figure 4a-d). We also did this when the
herd was tightly spaced together and didn’t require multiple photos since the orthorectification
process corrects for tilt displacement and relief displacement (geometric distortions of photos that
are caused by the camera aspect and the terrain of the area in which the photos are being taken).

Figure 4a-d. Video screenshots of the cattle herd in the research pasture collected from a DJI Phantom 4 drone at an altitude of
100 m (a-c). Multiple photos of the herd were orthorectified together to make a single image (orthomosaic) of the entire herd (d).

Using image processing software (ImageJ), we measured the distance between calves and their
mother cows (Calf-Dam group) and the distance between calves and non-related cows (Cow-Other
group). The cow-calf distances can be quantified on the orthomosaic by measuring the number of
pixels between two points. Using the ImageJ software, we marked the penning tag on the back of
every individual in the orthomosaic image and obtained the x- and y-coordinates of each mark.
These coordinates were used to geometrically determine the cow-calf distances. A pixel to meter
ratio was obtained by calibrating the image with a known reference length to measure the cowcalf distances in meters. Thus, we measured the distance between every individual in the herd. The
9

cattle in the research pasture aggregated in different areas of the pasture throughout the study, and
we generated an orthomosaic image for each of the areas. Thus, we used different reference
objects, one for each of the areas: a feeding trough that was 3.51 m in length, and the distance
between two distinct fence posts, which were 42.6 m apart for one set of posts and 30 m for the
other set of posts (Figure 5).

Figure 5. An example of an orthorectified image (orthomosaic) used to determine cow-calf distances (a). The number of pixels
between the penning tag of a cow and a calf were measured using ImageJ. A reference length measured in the field was used to
calibrate the image to obtain a pixel to meter ratio. Two photos obtained from taking a screenshot from the UAV flying at a height
of 15 m are shown (b and c) around the orthomosaic (a). Photos screenshotted from the low-flying drone were used to identify
each individual in the orthomosaic. The magenta, blue, and green arrows in photo “a” are matched with arrows in photos “b”
and “c” to illustrate how the cattle in the orthomosaic are matched with the photos obtained from the low-flying drone.

10

Testing data reliability
We sought to determine if our measurements on ImageJ software were both accurate (i.e., were
distances measured using the software close to the true distance measured in the field) and precise
(i.e., were they consistent). To do this, we tested two orthomosaics in each of our six time periods,
except for Day 2 Daytime, for which we had only one orthomosaic in this time period that
contained the reference objects needed to make comparisons. This gave us a total sample of 11
orthomosaics to represent the error of our total data set of 76 orthomosaics. For each of the 11
orthomosaics, we calibrated the pixel to meter ratio using a reference object in the image that we
measured in the field, then measured the length of a second reference object that was also measured
in the field (Figure 6) to make measurement comparisons. These references were described earlier.
This process was repeated three times for each orthomosaic (to test our consistency), giving a total
of 33 comparisons to determine the difference between the length of a reference object measured
using ImageJ software and the length of that object as measured in the field.

Figure 6. An orthomosaic used to test the accuracy of the distance measurements made on ImageJ software. In this example,
reference 1 is a feeding trough used to calibrate the pixel to meter ratio on ImageJ and reference 2 is the distance between two
distinct fence posts used to compare our software-measured length and the length measured in the field.

Statistical Tests
The cow-calf proximity could have been affected by the following variables: the day of study (days
1 - 4), the time period of the study (daytime or evening), and the relationship between the cow and
the calf (whether the cow is the mother of the calf or the cow is unrelated to the calf). The cow11

calf proximity significantly differed by day (P < 0.001), as an ANOVA. To remove the effect of
day in further statistical analysis, the data were standardized by day such that the mean was equal
to zero and the standard deviation was equal to one. The standardized data was used in a mixed
model to examine the effect of time period, relationship, and the interaction between period and
the relationship. We constructed the mixed model with the cow-calf distance as the response
variable, time period and relationship as fixed effects and individual calf as a random effect. The
mixed model was used as it accounts for pseudo replication (i.e., measurements were made on the
same cows and calves at each sampling point) in our study. All tests were done using JMP 13
statistical software.

RESULTS
Time period had a significant effect on cow-calf proximity (F1,849 = 101.79, P < .001) as did cowcalf relationship (F1,849 = 26.75, P < .001). There was no statistically significant interaction between
time period and cow-calf relationship (F1,849 = 1.35, P = 0.245). The mean cow-calf distance (n =
576) in the daytime was 39.8 ± 15 m (mean ± standard deviation) and in the evening (n = 288) was
84.3 ± 24 m (Figure 7). The mean proximity (n = 72) of the Calf-Dam group was 43.8 ± 24 m
(mean ± standard deviation) and the mean proximity (n = 792) of the Calf-Other group was 55.6
± 28 m (mean ± standard deviation).

12

Figure 7. A box plot showing the distribution of cow-calf proximity by relationship and time period. Boxes indicate the interquartile
range, lines across the boxes indicate the median, the whiskers indicate the upper and lower limits, and the asterisks indicate
outliers.

At an individual level, the cow-calf distance ranged between 0.5 m to 304 m (Figure 8). The long
tail of the distribution is skewed to the right, showing that the cow-calf distances substantially
greater than the mean are less common.

13

900
800
700

Frequency

600
500
400
300
200
100
0

0

40

80

120

160

200

240

280

Proximity (m)

Figure 8. The distribution of cow-calf distances for individual pairs.

In the 11 orthomosaics that we used to test the accuracy and precision of our measurements, the
mean difference between the measured length of a reference object and its true length measured in
the field was 0.9 ± 0.7 m (mean and standard deviation). The maximum difference was 2.2 m.
Within each orthomosaic, we measured the reference length three times and calculated the mean
(n = 3); to test our precision, we calculated the standard deviation of each of the 11 means. The
mean standard deviation within each orthomosaic (n = 11) was 0.2 ± 0.1 m (mean and standard
deviation) and the max was 0.5 m.

DISCUSSION
Cow-Calf Affiliations
We examined the effects of time period and cow-calf relationship on cow-calf proximity. In both
the evening and daytime periods, calves were closer to their dams than they were to other cows in
the herd. This was expected since the calves require nursing from their dams. This is also consistent
with observations made in previous cow-calf affiliation studies (Swain & Hurley 2007; Hirata et
al. 2003).
14

All cow-calf pairs, regardless of their relationship (i.e., Calf-Dam or Calf-Other), were closer
together in the daytime period than in the evening period, showing that the entire herd is more
closely spaced in the daytime. The average cow-calf distance in the evening was twice as much as
in the daytime. This strong diurnal variation in cow-calf distances may largely be attributed to the
behaviour of the dams. Beef cattle exhibit a “hiding” predator-avoidance strategy in which dams
leave their calves behind during maintenance activities (grazing, ruminating, and resting) and then
return to the calf for nursing activities (grooming and milking) (von Keyserlingk & Weary 2007).
Young animals generally reduce their activity to minimize predator risk in species that employ the
“hiding” strategy (in contrast, in species employing a “following” strategy, the young stay closer
to their mothers at all times ) (Fisher et al. 2002; Carl & Robbins 1988). Indeed, cow-calf distances
in beef cattle have been shown to be largest when the dams are grazing, resting and ruminating
and lowest when the calves are nursing (Hirata et al. 2003). Video observation showed that the
cows in this study spent a greater proportion of time grazing and walking in the evening periods
than in the daytime periods; in the latter, they seemed to spend a greater proportion of time lying
down. Since this study was carried out near the hottest time of the year in Lethbridge, AB, it may
have been more comfortable for cattle to forage in the evening; especially since there was no shade
in the research pasture. This may be a reasonable explanation for the strong diurnal variation in
the cow-calf distance differences observed in this study.

Accuracy of Measurements
There are several sources of error that may have potentially affected the accuracy and precision of
our distance measurements. Orthomosaics are generated by orthorectifying multiple photos
together through a computational process using photogrammetry software. We used photos that
were screenshots of video captured by camera-equipped drones flown at an altitude of 100 m above
ground level. Because we manually flew the drones when recording video, the height at which the
drone was flying during recording may have varied among our 76 sample points; however, this
variation was likely no greater than two meters. Even though the camera is gimbal-stabilized, we
experienced strong winds in the field location in Lethbridge, AB, which may have tilted the camera
angle, and moved the drone to a slightly different altitude during the time a single video was made.
This made it difficult to collect each video, and subsequently each screenshot, consistently in terms
of angle and height. In addition, to make an orthomosaic of a given area, multiple overlapping
photos of the same area are needed and it is optimal to have a consistent amount of overlap among
15

the photos. However, for our orthomosaic images, we used screenshot images that were manually
taken from a computer, so the degree of overlap among photos varied; this potentially affected the
quality of the orthomosaic. Furthermore, the spacing of the cattle in the herd varied substantially
among time points so each orthomosaic required a different number of photos to orthorectify
between each 15-minute sampling time period. As a result of these factors, the resolution (i.e.,
pixel density) of each orthomosaic varied.
Orthomosaics are typically used to create composite images of still objects. However, many
individual animals in our research herd were active when we took screenshots used for creating
the orthomosaic. An active animal could be in different locations in overlapping photos of the
whole cattle herd. As a result, some animals are distorted or stretched in our orthomosaic images
of the cattle herds. This stretching and distortion made it difficult to tell where the penning tag was
truly positioned on and animal’s back, which would reduce the accuracy of our distance
measurements. However, only a small number of individual animals appeared distorted and/or
stretched appeared in our orthomosaics: In 19 of the orthomosaics, there was only 1 distorted
animal, in 8 orthomosaics there were 2, and in 5 orthomosaics, 3 distorted animals appeared. In
addition, the overall extent to which this affects the measurement accuracy is relatively small.
Generally, a distortion or stretch would have caused us to mark the center of the penning tag up to
an animal’s length away from the true position of the center of the penning tag.
In rare cases (2 out of the 76 orthomosaics), the same animal would appear twice in the same image
or an animal would be completely missing from the orthomosaic. This image distortion is an issue
that has also affected other studies in which UAVs were used to capture photos for animal
enumeration (Hodgson et al. 2016; Whitehead et al. 2014b); in these studies, photoshop was used
to manually stitch together multiple photos into one composite image that captured every animal
and as a result, some animals appeared distorted. However, in a recent study, Seymour et al. (2017)
used a UAV to capture photos of grey seal (Halichoerus grypus) colonies and generated
orthomosaics; the orthorectification of their images corrected for inconsistencies that could have
resulted from animals moving in between photos. In our study, it is possible that the distorted
animals weren’t corrected by orthorectification either because our cattle moved too fast between
photos or because there wasn’t enough overlap in the photos used for orthorectification. Flying the
UAV faster to take overlapping photos of the entire herd more quickly would likely reduce the
16

number of distorted animals that appear in orthomosaics since there would be less time for animals
to move between photos. Capturing photos at a higher frequency, for example, at sub-second
intervals (Seymour et al. 2017), may also reduce the number of distorted animals that appear in
orthomosaics since it would provide more data for computational software to calculate an average
location of a moving animal.
Various types of human error may also have influenced results. These include manually measuring
distances using ImageJ software. It is difficult to manually mark the exact spot of the cattle penning
tag on each individual on every orthomosaic using the ImageJ software. It is equally difficult to
mark the same spot on the fence posts and feeding troughs (i.e., our known reference lengths).
This ultimately affects the precision of the distance measurements. However, we have
demonstrated that the extent to which this affected the precision is small; in 11 different
orthomosaics that we tested, our measurements were consistent to 0.5 m. In addition, the mean (n
= 11) difference between the length of a reference object that we measured on ImageJ software
and the length of that reference object that we measured in the field was 0.9 ± 0.7 m (mean ±
standard deviation), which indicates that our distance measurements were accurate enough to make
reliable inferences on fine-scale social interactions between animals. Thus, we are reasonably
confident that our method to quantify cow-calf social interactions should allow us to make
inferences about different facets of the social behaviour of cattle, especially interactions between
cows and calves.

Future Use of UAVs in Studying Cattle Behaviour
Using unmanned aerial vehicles to quantify social affiliations has many benefits over traditional
methods of direct observation. Consumer-grade and industry-grade UAVs are becoming
increasingly more affordable, as the popularity and production rate of the vehicles has increased
over the last decade. As a result, UAVs have been used for wildlife monitoring, ecological surveys,
and a plethora of industrial applications (Ge et al. 2016; Przyborski et al. 2015; Cajzek & Klanšek
2014; Moranduzzo & Melgani 2014). UAVs are likely more cost-effective for quantifying social
affiliations in cattle than use of traditional remote sensing technology and/or radio telemetry
devices. Only one UAV is required to monitor a whole herd, as opposed to providing tracking
devices for every single animal being studied. Furthermore, consumer-grade UAVs on the market
today are likely more reliable given that collars equipped with GPS, GNSS, and radio transmitters
17

typically require a high degree of expertise, have greater distance error, and have additional issues
such as damage by weather, animals themselves and battery limitations. Lastly, using UAVs
eliminates the potential risk of compromising animal welfare as they tend to be less invasive
(reviewed by Christie et al. 2016). Collars represent extra weight for individual animals to carry,
and may cause skin abrasions if they are not properly fitted. Though UAVs may initially appear to
provide a strong visual and auditory stimulus, we observed that the cattle in our study rapidly
habituated to the UAVs over a short period of time. During the habituation period of our study, we
noticed that cattle were observant of the UAVs when they flew overhead at a proximity of 10 m.
After a few days of habituation, the cattle showed no visually observable behavioural response to
the UAVs at an altitude of 10 m. Many species of animals show little or no behavioural response
to UAVs (reviewed by Christie et al. 2016), except in cases where the UAV is flown at close
proximity (for example, see Ditmer et al. 2015) A recent study has also demonstrated that cattle
(Bos taurus) do not seem to be disturbed by overhead UAVs (Mulero-Pazmany et al. 2015). These
researchers did not visually observe any signs of response in free-ranging cattle when they
surveyed at an altitude of 100 m. For domesticated cattle, no behavorial responses were reported
by Whitehead et al. (2014b), who captured images of cattle in a feedlot at an altitude of 100m.
Even at a closer altitude of 25 m, drones show no effect on the feeding behaviour of cattle between
habituated and non-habituated groups (Nyamuryekung'e et al. 2016).
In conclusion, we successfully developed a novel and non-invasive method to quantify cow-calf
social affiliations using unmanned aerial vehicles to take photos and then using photogrammetry
to orthorectify multiple photos of the cattle herds from various offline video-screenshots collected
by the UAVs. We have thus strongly supported our hypothesis that UAVs could be used to quantify
social affiliations of beef cattle in a field setting. This method allows us to determine distances
among all the individuals of the herd with a sufficient level of accuracy and precision to provide
meaningful data to study social behaviour in cattle. In future work, this method could help advance
our understanding of the organization of social groups and hierarchies within a herd, which are
formed to reduce aggressive behaviour among individuals and maintain group cohesion (Mendl &
Held 2001). In addition, this method allowed for capturing high-quality video aerially and at a very
close distance, which would be useful for observing other important cattle behaviours, such as
ruminating and nursing. Studying these behaviours in cattle is important for increasing our
understanding of animal welfare (de Oliveira et al. 2014) and would ultimately be beneficial for
18

managers in the beef cattle industry who are trying to develop better management practices. Using
UAVs to further our understanding of cattle social dynamics could facilitate decisions that would
optimize animal grouping, breeding, and maternal behaviour and/or grazing, any or all of which
could potentially increase overall cattle productivity on farms.

19

LITERATURE CITED
Afek Y, Brand A. 1998. Mosaicking of orthorectified aerial images. Photogramm. Eng. Rem. S.
64(2): 115-125.
Anderson DM, Estell RE, Cibils AF. 2013. Spatiotemporal Cattle Data—A Plea for Protocol
Standardization. Positioning. 4(1): 115-136.
Anderson DM, Winters C, Estell RE, Fredrickson EL, Doniec M, Detweiler C, Rus D, James D,
Nolen B. 2012. Characterising the spatial and temporal activities of free-ranging cows from GPS
data. Rangeland J. 34(2): 149-161.
Anderson K, Gaston KJ. 2013. Lightweight unmanned aerial vehicles will revolutionize spatial
ecology. Front. Ecol. Environ. 11(3): 138-146.
Candiago S, Remondino F, Giglio M De, Dubbini M, Gattelli M. 2015. Evaluating Multispectral
Images and Vegetation Indices for Precision Farming Applications from UAV Images. Remote
Sens. 7(4): 4026-4047.
Carl GR, Robbins CT. 1988. The energetic cost of predator avoidance in neonatal ungulates: hiding
versus following. Can. J. Zool. 66(1): 239-246.
Cajzek R, Klanšek U. 2014. An Unmanned Aerial Vehicle for Multi-Purpose Task in Construction
Industry. J. Appl. Eng. Sci. 14(2): 314-327.
Chamoso P, Raveane W, Parra V, Gonzalez A. 2014. UAVs Applied to the Counting and
Monitoring of Animals. In: Ambient Intelligence - Software and Applications, Advances in
Intelligent Systems and Computing. Switzerland: Springer International Publishing. 71-80.
Christie KS, Gilbert SL, Brown CL, Hatfield M, Hanson L. 2016. Unmanned aircraft systems in
wildlife research: Current and future applications of a transformative technology. Front. Ecol.
Environ. 14(5): 241-251.
Daleszczyk K. 2004. Mother-calf relationships and maternal investment in European bison Bison
bonasus. Acta Theriol. 49(4): 555-566.
de Oliveira FA, Luna SPL, do Amaral JB, Rodrigues KA, Sant’Anna AC, Daolio M, Brondani JT.
2014. Validation of the UNESP-Botucatu unidimensional composite pain scale for assessing
postoperative pain in cattle. BMC Vet. Res. 10(1): 200.
Ditmer MA, Vincent JB, Werden LK, Tanner JC, Laske, TG, Iaizzo, PA, Garshelis DL, Fieberg
JR. 2015. Bears Show a Physiological but Limited Behavioural Response to Unmanned Aerial
Vehicles. Curr. Biol. 25(17): 2278-2283.
Drewe JA, Weber N, Carter SP, Bearhop S, Harrison XA, Dall SRX, McDonald RA, Delahay RJ.
2012. Performance of proximity loggers in recording intra- and inter-species interactions: A
laboratory and field-based validation study. PLoS One. 7(6): e39068.
Durban JW, Fearnbach H, Perryman WL, Leroi DJ. 2015. Photogrammetry of killer whales using
a small hexacopter launched at sea. J. Unmanned Veh. Syst. 3(3): 131-135.

20

Ey-Chmielewska H, Chruściel-Nogalska MFB. 2015. Photogrammetry and Its Potential
Application. Adv. Clin. Exp. Med. 24(4): 737-741.
Finger A, Patison KP, Heath BM, Swain DL. 2014. Changes in the group associations of freeranging beef cows at calving. Anim. Prod. Sci. 54(3): 270-276.
Fisher DO, Blomberg SP, Owens IPF. 2002. Convergent Maternal Care Strategies in Ungulates
and Macropods. Evolution. 56(1): 167-176.
Ge L, Li X, Ng AH. 2016. UAV for mining applications: A case study at an open-cut min and a
longwall mine in New South Wales, Australia. Geoscience and Remote Sensing Symposium
(IGARSS). IEEE 2016.
Grant RJ, Albright JL. 2001. Effect of animal grouping on feeding behaviour and intake of dairy
cattle. J. Dairy Sci. 84: E156-E163.
Handcock RN, Swain DL, Bishop-Hurley GJ, Patison KP, Wark T, Valencia P, Corke P, O’Neill
CJ. 2009. Monitoring animal behaviour and environmental interactions using wireless sensor
networks, GPS collars and satellite remote sensing. Sensors. 9(5): 3586-3603.
Hirata M, Nakagawa M, Funakoshi H, Iwamoto T, Otozu W, Kiyota D, Kuroki S, Fukuyama K.
2003. Mother – young distance in Japanese Black cattle at pasture. J. Ethol. 21(2): 161-168.
Hodgson A, Kelly N, Peel D. 2013. Unmanned aerial vehicles (UAVs) for surveying Marine
Fauna: A dugong case study. PLoS One. 8(11): 1-15.
Hodgson A, Peel D, Kelly N. 2017. Unmanned Aerial Vehicles (UAVs) for surveying marine
fauna: assessing detection probability. Ecol. Appl. 27 (4):1253-1267.
Hodgson JC, Baylis SM, Mott R, Herrod A, Clarke RH. 2016. Precision wildlife monitoring using
unmanned aerial vehicles. Sci. Rep. 6: 22574.
von Keyserlingk MAG, Weary DM. 2007. Maternal behaviour in cattle. Horm. Behav. 52(1): 106113.
Klemas VV. 2015. Coastal and Environmental Remote Sensing from Unmanned Aerial Vehicles:
An Overview. J. Coast. Res. 31(5): 1260-1267.
Koh LP, Wich SA. 2012. Dawn of drone ecology: low-cost autonomous aerial vehicles for
conservation. Trop. Conserv. Sci. 5(2): 121-132.
Laliberte AS, Herrick JE, Rango A, Winters C. 2010. Acquisition, Orthorectification, and Objectbased Classification of Unmanned Aerial Vehicle (UAV) Imagery for Rangeland Monitoring.
Photogramm. Eng. Rem. S. 76(6): 661-672.
Lewis JS, Rachlow JL, Garton EO, Vierling LA. 2007. Effects of habitat on GPS collar
performance: using data screening to reduce location error. J. Appl. Ecol. 44(3): 663-671.
Linchant J, Lisein J, Semeki J, Lejeune P, Vermeulen C. 2015. Are unmanned aircraft systems
(UASs) the future of wildlife monitoring? A review of accomplishments and challenges. Mammal
Rev. 45(4): 239-252.

21

Matese A, Toscano P, Di Gennaro SF, Genesio L, Vaccari FP, Primicerio J, Belli C, Zaldei A,
Bianconi R, Gioli B. 2015. Intercomparison of UAV, aircraft and satellite remote sensing
platforms for precision viticulture. Remote Sens. 7(3): 2971-2990.
Mendl M, Held S. 2001. Living in groups: an evolutionary perspective. In: Keeling LJ, Gonyou
HW, editors. Social behaviour in farm animals. Cambridge (UK): CABI Publishing. 7-36.
Moranduzzo T, Melgani F. 2014. Monitoring structural damages in big industrial plants with UAV
images. Geoscience and Remote Sensing Symposium (IGARSS). IEEE 2014.
Moreland EE, Cameron MF, Angliss RP, Boveng PL. 2015. Evaluation of a ship-based unoccupied
aircraft system (UAS) for surveys of spotted and ribbon seals in the Bering Sea pack ice. J.
Unmanned Veh. Syst. 3(3): 114-122.
Mulero-Pázmány M, Barasona JÁ, Acevedo P, Vicente J, Negro JJ. 2015. Unmanned Aircraft
Systems complement biologging in spatial ecology studies. Ecol. Evol. 5(21): 4808-4818.
Mulero-Pázmány M, Stolper R, Van Essen LD, Negro JJ, Sassen T. 2014. Remotely piloted
aircraft systems as a rhinoceros anti-poaching tool in Africa. PLoS One. 9(1): e83873.
Nyamuryekung’e S, Cibils AF, Estell RE, Gonzalez AL. 2016. Use of an unmanned aerial vehicle
- Mounted video camera to assess feeding behaviour of raramuri criollo cows. Rangeland Ecol.
Manag. 69(5): 386-389.
O’Neill CJ, Bishop-Hurley GJ, Williams PJ, Reid DJ, Swain DL. 2014. Using UHF proximity
loggers to quantify male-female interactions: A scoping study of estrous activity in cattle. Anim.
Reprod. Sci. 151(1): 1-8.
Pajares G. 2015. Overview and Current Status of Remote Sensing Applications Based on
Unmanned Aerial Vehicles (UAVs). Photogramm. Eng. Rem. S. 81(4): 281-329.
Patison KP, Swain DL, Bishop-Hurley GJ, Robins G, Pattison P, Reid DJ. 2010. Changes in
temporal and spatial associations between pairs of cattle during the process of familiarisation.
Appl. Anim. Behav. Sci. 128(1): 10-17.
Phillips CJC, Rind MI. 2001. The Effects on Production and Behaviour of Mixing Uniparous and
Multiparous Cows. J. Dairy Sci. 84(11): 2424-2429.
Przyborski M, Szczechowski B, Szubiak W, Szulwic J, Widerski T. 2015. Photogrammetric
Development of the Threshold Water at the Dam on the Vistula River in Wloclawek From
Unmanned Aerial Vehicles (UAV). In: SGEM2015 Conference Proceedings. 18-24.
Schein MW, Fohrman MH. 1955. Social dominance relationships in a herd of dairy cattle. Br. J
Anim. Behav. 3(2): 45-55.
Seymour A, Dale J, Hammill M, Halpin P, Johnston D. 2017. Automated detection and
enumeration of marine wildlife using unmanned aircraft systems (UAS) and thermal imagery. Sci.
Rep. 7.
Sowell BF, Mosley JC, Bowman JGP. 1999. Social behaviour of grazing beef cattle : Implications
for management. J. Anim. Sci. 77(E-Suppl): 1-6.

22

Swain DL, Bishop-Hurley GJ. 2007. Using contact logging devices to explore animal affiliations:
Quantifying cow-calf interactions. Appl. Anim. Behav. Sci. 102(1): 1-11.
Trac [Internet]. 2010a. Tilt Displacement. [Cited 2017 Aug 30] Available from:
https://trac.osgeo.org/ossim/wiki/tilt_displacement
Trac [Internet]. 2010b. Relief Displacement. [Cited 2017 Aug 30] Available from:
https://trac.osgeo.org/ossim/wiki/relief_displacement
UAF (University of Alaska Fairbanks) News and Information [Internet]. 2015. [cited 2017 May
2]. Available from: http://news.uaf.edu/ researchers-study-sea-otters-unmanned-aircraft.
Veissier I, Lamy D, Le Neindre P. 1990. Social-behaviour in domestic beef-cattle when yearling
calves are left with the cows for the next calving. Appl. Anim. Behav. Sci. 27(3): 193-200.
Verhoeven G, Doneus M, Briese C, Vermeulen F. 2012. Mapping by matching: A computer
vision-based approach to fast and accurate georeferencing of archaeological aerial photographs. J.
Archaeol. Sci. 39(7): 2060-2070.
Vermeulen C, Lejeune P, Lisein J, Sawadogo P, Bouche P. 2013. Unmanned Aerial Survey of
Elephants. PLoS One. 8(2): e54700.
Waiblinger S, Boivin X, Pedersen V, Tosi MV, Janczak AM, Visser EK, Jones RB. 2006.
Assessing the human-animal relationship in farmed species: A critical review. Appl. Anim. Behav.
Sci. 101(3): 185-242.
Whitehead K, Hugenholtz CH, Myshak S, Brown O, LeClair A, Tamminga A, Barchyn TE,
Moorman B, Eaton B. 2014a. Remote sensing of the environment with small unmanned aircraft
systems (UASs), part 1: a review of progress and challenges. J. Unmanned Veh. Syst. 2(3): 69-85.
Whitehead K, Hugenholtz CH, Myshak S, Brown O, LeClair A, Tamminga A, Barchyn TE,
Moorman B, Eaton B. 2014b. Remote sensing of the environment with small unmanned aircraft
systems (UASs), part 2: scientific and commercial applications. J. Unmanned Veh. Syst. 2(3): 86102.
Wilson AM, Barr J, Zagorski M. 2017. The feasibility of counting songbirds using unmanned
aerial vehicles. Auk. 134(2): 350-362.
Zhang C, Kovacs JM. 2012. The application of small unmanned aerial systems for precision
agriculture: A review. Precis. Agric. 13(6): 693-712.

23