Rangeland Ecology & Management 73 (2020) 642–648

Contents lists available at ScienceDirect

Rangeland Ecology & Management

journal homepage: www.elsevier.com/locate/rama

Quantifying the Dynamics of Livestock Distribution by Unmanned Aerial

Vehicles (UAVs): A Case Study of Yak Grazing at the Household Scale

Yi Sun

a , b , Shuhua Yi a , b , ∗, Fujiang Hou

c , Dongwen Luo

d , Junqi Hu

b , Zhaoye Zhou

e

a Institute of Fragile Eco-Environment, Nantong University, 999 Tongjing Road, Nantong, Jiangsu 226007, China b School of Geographic Science, Nantong University, 999 Tongjing Road, Nantong, Jiangsu 226007, China c State Key Laboratory of Grassland Agro-ecosystems, College of Pastoral Agriculture Science and Technology, Lanzhou University, Lanzhou, 730 0 0 0 Gansu, China d AgResearch Ltd, Grasslands Research Centre, Palmerston North, New Zealand e School of Civil Engineering, Lanzhou University of Technology, Lanzhou 730050, China

a r t i c l e i n f o

Article history:

Received 19 June 2019

Revised 15 May 2020

Accepted 28 May 2020

Keywords:

the Qinghai-Tibetan Plateau

dispersion

spatio-temporal distribution

grazing

FragMAP

a b s t r a c t

Managed grazing is the most extensive land use worldwide. Information about the spatio-temporal distri-

bution of livestock is critical for grassland ecosystem management. However, no direct and cost-effective

method exists to monitor livestock distribution under natural conditions. We proposed a practical method

that uses unmanned aerial vehicles (UAVs), and tested it at a typical household pasture on the Qinghai-

Tibetan Plateau. We monitored and analyzed the hourly spatial distribution of each yak ( Bos mutus ). We

found that (1) yak information extracted by our independently developed software was sufficiently accu-

rate ( P > 0.05); (2) yak density (frequency/area/period) decreased with increasing distance from campsites

in the early growing season, while the pattern reversed in the late growing season; (3) dispersion index

(DI, m

2 /yak) of the yak herd exhibited a tendency to decrease in the morning and then to increase in the

afternoon within a day; from June to September DI decreased in both the morning and afternoon, but

was similar at noon. The monitoring method we developed in this study characterized the spatial distri-

bution of the whole yak herd dynamically with high frequency, resolution, and efficiency. We conclude

that this method is suitable for long-term monitoring of animal behavior, and for studying relationship

between the distribution of pastoral livestock and resource availability, which will be beneficial for grass-

land sustainable management.

© 2020 The Society for Range Management. Published by Elsevier Inc. All rights reserved.

I

(

s

p

t

n

o

s

d

a

‘

o

s

(

l

Q

u

h

p

t

g

i

m

2

d

L

d

h

1

ntroduction

Livestock grazing is the most widespread land use on Earth

Wang et al., 2019 ). Herbivores influence grassland at a range of

patial and temporal scales due to heterogeneous foraging, tram-

ling and excretion ( Senft et al., 1987 ). It is essential to quan-

ify the spatio-temporal variations of herbivore distributions under

atural conditions to understand the nature, magnitude, and extent

f explicit interactions between the distribution of pastoral live-

tock and resource availability ( Butt, 2010 ), and to be able to pre-

ict how herbivores will respond to environmental change ( Bowler

nd Benton, 2005 ).

The Qinghai-Tibetan Plateau (QTP) is referred to as the

third pole’ because of its importance as an ecoregion. Grazing

f livestock on the QTP has occurred for thousands of years

∗ Corresponding author at: Institute of Fragile Eco-Environment, Nantong Univer-

ity, 999 Tongjing Road, Nantong, Jiangsu 226007, China.

E-mail address: yis@ntu.edu.cn (S. Yi).

o

m

(

o

ttps://doi.org/10.1016/j.rama.2020.05.004

550-7424/© 2020 The Society for Range Management. Published by Elsevier Inc. All righ

Xin et al., 2011 ; Feng et al., 2013 ). In the 1980s, the Range-

and Contract Responsibility System (RCRS) was introduced to the

TP and subsequently, the household became the basic unit of

tilization of alpine grassland ( Cao et al., 2013 ). Under house-

old management, livestock graze during the daytime but are

enned into a campsite overnight, resulting in a radial grazing pat-

ern ( Fernándezgiménez et al., 2018 ; Hou et al., 2003 ). Yak ( Bos

runiens ), ca. 13 × 10 6 , are the most important grazing livestock an-

mals on the QTP, as they provide meat, milk and income for the

ajority of people living in this region ( Harris, 2010 ; Xin et al.,

011 ). They largely depend on the native grassland to survive un-

er the management pattern of the household ( Cao and Du, 2011 ;

ong et al., 1999 ).

Traditionally, the spatio-temporal distribution of herbivores is

etermined using both direct and indirect methods. Direct meth-

ds mainly include observation and biologging, while indirect

ethods depend on some evidence of animal activity (e.g., feces)

Lange, 1969 ; Mulero-Pázmány et al., 2015 ). However, these meth-

ds are often time and labor-intensive and characterized with low

ts reserved.

Y. Sun, S. Yi and F. Hou et al. / Rangeland Ecology & Management 73 (2020) 6 42–6 48 643

r

n

p

a

d

R

m

p

e

g

a

t

g

2

(

n

i

e

P

e

F

n

s

y

T

o

p

a

f

d

s

m

M

S

A

a

c

(

s

c

m

i

P

t

p

t

t

o

t

3

t

d

Y

o

t

t

y

Fig. 1. Georeferenced orthomosaic of the studied pasture. The triangle indicates the

position of pasture, red line indicates borders of the Qinghai-Tibetan Plateau, dotted

black line indicates borders of the pasture, and yellow circle represents campsite

area. (For interpretation of the references to color in this figure legend, the reader

is referred to the web version of this article.)

Fig. 2. The herd track monitor method based on unmanned aerial vehicle.

n

P

1

C

h

D

t

t

w

a

T

epresentativeness ( Mulero-Pázmány et al., 2015 ). For instance, the

umber of studied animals is restricted by number of tags de-

loyed (because of high cost of equipment and labor), and there

re undesirable effects on the animals during the capture and han-

ling ( Ding et al., 2007 ; Du et al., 2017 ; Murray and Fuller, 20 0 0 ;

utz and Hays, 2009 , e.g., Fig. S1). Feces monitoring is compro-

ised on the QTP because of the widespread collection of faecal

ats for fuel (Fig. S2) ( Long et al., 2008 ). Therefore, a new cost-

fficient, herd based and noninvasive monitoring method is ur-

ently needed.

The rapid development of unmanned aerial vehicles (UAVs) has

llowed their applications in many ecological studies, e.g., vegeta-

ion cover ( Chen et al., 2016 ), plateau pika ( Yi et al., 2016 ), above-

round biomass ( Zhang et al., 2018 ), and plant species ( Sun et al.,

018 ). However, few studies have used UAVs to monitor livestock

Rey et al., 2017 ). Based on a range of social surveys and prelimi-

ary data, we were granted permission to carry out livestock mon-

toring around a field research station with a licensed UAV ( Sun

t al., 2018 ). The price of common UAV (with RGB camera, DJI

hantom 3) was ∼1300 USD, easy to carry and operate, and robust

nough to work for a long time on the QTP ( Yi et al., 2016 , Fig. S3).

urthermore, the noise from the multi-rotor UAVs appeared to

ot disturb herbivores when flying heights exceeded 70 m. In this

tudy, we developed a practical method based on UAVs to monitor

ak herd distribution for a typical household pasture on the QTP.

he specific objectives were to (1) test the precision and feasibility

f the method; and (2) characterize spatio-temporal distribution

atterns of a livestock herd under natural conditions. Widespread

nd long-term monitoring of livestock with a high precision and

requent sampling method has the potential to improve our un-

erstanding of interactions between livestock distribution and re-

ource availability, and may be beneficial for improving manage-

ent of grassland ecosystems.

aterials and methods

tudy area

The field work was conducted during June-September 2016, at

zi, Gansu Province, China (101 °52 ′ 07.9 ′ ’E, 33 °24 ′ 24.1 ′ ’N; 3547 m

.s.l.) ( Fig. 1 ). Azi is in a humid region with a mean annual pre-

ipitation > 600 mm and mean annual air temperature of 1.1 °C Sun et al., 2015 ). The soil type at the study site is alpine meadow

oil, primarily Mat-Cryic Cambisols ( Gao and Li, 1995 ). The plant

ommunity type is mainly alpine meadow which is dominated by

onocotyledonous species, primarily Poaceae and Cyperaceae. Var-

ous dicotyledonous species are also common, e.g., Ranunculaceae ,

olygonaceae , Saxifragaceae , Asteraceae , Scrophulariaceae and Gen-

ianaceae ( Ma et al., 2010 ).

The study area consists of a ∼48.5 ha pasture of gentle topogra-

hy (slope < 5 °, Fig. 1 ) used for warm season grazing. It was con-

inuously grazed at moderate to heavy yak stocking rates for more

han 30 years. Because of disease, attacks by predators, the selling

f livestock, and the birth of calves, yak numbers of the herd fluc-

uated between 283 and 312 during the monitoring period (310,

12, 294 and 283 in June, July, August and September, respec-

ively). Yak are penned each night and let out to graze naturally

uring the daytime ( Fig. 1 ).

ak herd monitoring and positioning

From June to September in 2016, the yak herd was monitored

n 5–8 days of each month (i.e., the UAV followed the herd and

ook an individual aerial photograph every hour during the day-

ime, e.g., Fig. 2 and Table S1). During June, July and August,

ak only grazed in the pasture but occasionally grazed in the

eighbouring winter pasture during September. A commercial DJI

hantom 3 professional UAV equipped with a standard built-in

2 megapixel RGB camera (Phantom 3 Professional, DJI Innovation

ompany, China) was used to take aerial photographs of the yak

erd. More detailed parameters are listed in Fig. S3. In addition,

JI provides developers with a software development kit (SDK)

hat allows them to design specific flight routes and flight alti-

udes for specific purposes. Based on this SDK, the FragMAP soft-

are was developed to plan the flight routes and acquire UAV im-

ges automatically instead of through manual operation ( Yi, 2016 ).

he software has multiple flight modes that can be used for dif-

644 Y. Sun, S. Yi and F. Hou et al. / Rangeland Ecology & Management 73 (2020) 6 42–6 48

Fig. 3. Location of yak. (a) Original aerial photograph of yak herd; (b) trimmed aerial photograph of yak herd; (c) geometric correction of aerial photograph in ArcGIS;

(d) location of each yak by HerdCounter, which is an independently developed Java software using OpenCV library; (e) verification of yak locations in ArcGIS.

f

a

t

e

o

H

d

o

a

p

e

t

B

t

f

f

m

o

l

g

c

e

l

o

3

(

f

a

m

d

a

h

t

i

P

m

(

h

c

2

(

o

t

fi

m

o

a

t

c

T

l

t

F

d

o

p

i

b

n

i

i

t

t

(

a

I

m

l

o

i

e

a

t

i

a

D

p

g

erent research purposes. In this study, to correct each individual

erial photograph of the herd, the “Mosaic” flight mode was used

o acquire overlapping aerial photographs and generate a georef-

renced orthomosaic of the pasture ( Fig. 1 ). Briefly, corners 1–4

f the pasture were set manually on an iPad (carrier of FragMAP,

uawei M5, Shenzhen, China), then way points were automatically

etermined based on the specified fly height, speed, and interval

f shoot and overlap rate (height was set at 100 m in this study,

nd resolution was < 5 cm). During each flight, the UAVs took

hotographs at an interval of 3 s automatically ( Yi, 2017 ; Zhang

t al., 2018 ). We generated a georeferenced orthomosaic of the pas-

ure by Pix4Dmapper (Pix4D S.A., Lausanne, Switzerland) ( Fig. 1 ).

riefly, (1) the initial processing imported the aerial photographs

hat were taken by UAVs along “Mosaic” mode and completed the

ollowing tasks automatically: keypoints were extracted by identi-

ying specific features as keypoints in the images; keypoints were

atched for images that have the same keypoints; camera model

ptimization is accomplished by calibrating the internal (e.g., focal

ength) and external parameters (e.g., orientation) of the camera;

eolocation GPS/GCP is determined by locating the model if geolo-

ation information is provided. (2) the point cloud and mesh gen-

ration was built on the Automatic Tie Points by completing the fol-

owing tasks automatically: additional Tie Points are created based

n the Automatic Tie Points to acquire a Densified Point Cloud ; and a

D Textured Mesh can be created based on the Densified Point Cloud .

3) orthomosaic map creation was accomplished by completing the

ollowing tasks automatically: creation of a Digital Surface Model

nd compution of Volumes and Orthomosaics ; creation of the Ortho-

osaic was based on orthorectification (removing the perspective

istortions from the images) ( https://support.pix4d.com/hc/en-us/

rticles/115002472186- Processing- steps ).

The UAV took aerial photographs of the yak herd hourly at a

eight of 70–100 m, which provides a ground coverage of more

han 3 hectares. There were 12–14 aerial photographs in one mon-

toring day, which depends on the grazing schedule of the herder.

hotographs were taken directly overhead whenever possible to

inimize potential errors associated with geometric correction

Fig. 2 ). During the monitoring period, two researchers and the

erder counted the number of yak in the herd in the field.

The yak is an unique bovine that has survived on the QTP for

enturies under harsh alpine environmental conditions ( Long et al.,

008 ), and is black or piebold with a body size between 1 and 3 m

including calves and adults). It can occupy dozens of or thousands

f pixels within an UAV image (determined by the yak size, posi-

ion and height of UAV). Therefore, each yak can be easily identi-

ed on the UAVs image ( Figs. 2 and 3 ).

Aerial photographs of the yak herd were trimmed to the mini-

al size required to cover the yak herd ( Fig. 3 a and b). They were

rtho-rectified according to the georeferenced orthomosaic, using

t least 10 reference point in ArcGIS (10.2.2). To find enough con-

rol points, any fixed object was regarded as a control point, e.g., a

orner of fence, a bend in a river, a big stone, or telegraph poles.

he differences between the actual values and fitted values (calcu-

ated by a correction model based on each pair of points) were less

han 2 m (residuals value, computed automatically in ArcGIS, e.g.,

ig. 3 c). A Java-based software program was developed indepen-

ently (HerdCounter, hereafter) to identify and count yaks in the

rtho-rectified images ( Fig. 3 d). In brief, each yak in an image was

ositioned using OpenCV package which created black squares to

dentify the yak (at the left dialog box of HerdCounter, select the

reed, e.g., select “yak” here, and import the propositional pixel

umber of each herbivore, Fig. 3 d). The squares were selected us-

ng the number of pixels between specified minimum and max-

mum values. We determined the correctness of automatic iden-

ification by checking the square for each yak ( Fig. 3 d). The yaks

hat were missed or selected mistakenly were corrected manually

it is feasible to visually identify each yak in the high-resolution

erial photographs, which were taken at the height of 70–100 m).

n brief, we added squares for missed yak, or delete squares for

istakenly identified yak after choosing the “delete” button on the

eft of HerdCounter ( Fig. 3 d). The latitude and longitude (position)

f each yak was estimated based on its column and row number

n the images and the latitudes and longitudes of the corners of

ach ortho-rectified image. The positions of each yak were output

s point file. The accuracy was evaluated by putting the points on

he ortho-rectified image to compare each point and correspond-

ng yak ( Fig. 3 e). After this process, the data were used for further

nalysis.

istribution of yak density

Based on the considerations of long-term monitoring and the

attern of herbivore grazing in the pasture we studied (“Rectan-

le” was the only flight mode of FragMAP when this study started,

Y. Sun, S. Yi and F. Hou et al. / Rangeland Ecology & Management 73 (2020) 6 42–6 48 645

Table 1

Paired comparisons (paired t -test) between the yak numbers that were counted in

the field and extracted from UAVs aerial photographs.

Month Ground count UAV count p

June 310 309.7 ± 11.2 0.882

July 312 310.9 ± 5.8 0.241

August 294 294.6 ± 9.2 0.824

September 283 281.8 ± 5.7 0.232

Total 0.630

Y

i

(

e

E

i

Y

D

(

e

t

d

o

l

w

d

D

D

w

c

i

a

a

o

r

a

f

D

R

C

d

w

b

C

i

e

(

c

e

t

T

Fig. 4. Sampling methods of yak density (one day data). Points with different colors

indicate the positions of the yak herd at different monitoring time (once an hour

during the monitoring day). (For interpretation of the references to color in this

figure legend, the reader is referred to the web version of this article.)

D

i

i

R

a

R

d

(

w

t

i

d

t

D

c

c

(

H

d

i, 2017 ), ten rectangles (200 m × 100 m, 2 ha) were set randomly

n the pasture (“Rectangle” flight mode was detail introduced in Yi

2017 ) and Yi et al. (2016) ). Distances between the campsite and

ach resampling area were regarded as the straight-line distance.

ach of the yak that appeared within each rectangle were counted

n ArcGIS. Yak density was calculated as Eq. (1) .

ak density ( head / ha / d )

= all the that yak appeared / 2 / number of days monitored (1)

ispersion of yak herd

Based on the position of yak herd of specific monitor time

Fig. 5 a), the buffer of each yak was set with a radius of 100 m (av-

rage travelling distance/2, Ding et al., 2014 ) to simulate potential

raveling area ( Fig. 5 b). The area within 100 m radius from the bor-

er of potential traveling area was subtracted by the ‘Buffer’ tool

f ArcGIS. The remnant area (the parts of dark green and deep yel-

ow) was regarded as the area of yak distributed ( Fig. 5 c), which

as used as ‘area of yak distribution’ in the Eq. (2) . Dispersion in-

ex (DI) was calculated as Eq. (2) .

ispersion index (m

2 / head

)= area of yak distribution / yak number

(2)

ata analysis

A goodness-of-fit test (Shapiro-Wilk test, univariate procedure)

as used to test the normality of data. Yak numbers that were

ounted in the field and the ones that were extracted from UAV

mages were compared by paired t -test. Repeated Measure ANOVA

nd one-way ANOVA were applied to analyze dispersion indices

mong monitoring times and months, respectively. The coefficients

f determination (R

2 ) and its P values were used to evaluate the

egression between yak density and distance from the campsites,

nd dispersing indices among monitor times. Analyses were per-

ormed using the statistical computing software R (version 3.5.3; R

evelopment Core Team 2019).

esults

haracteristics of the UAV-based monitor method

The UAV-based method was able to monitor yak herd size and

istribution dynamically ( Figs. 2 –4 , and Table 1 ). The herd sizes

ere not significantly different between ground-counted and UAV-

ased methods during the monitoring period ( p > 0.05, Table 1 ).

onnected with bits of manual correction (it is feasible to visually

dentify each yak based in the high-resolution aerial photographs,

.g., Fig. 3 ), the automatic identification software was efficient

HerdCounter could identify ∼300 yaks within 1 s in this study on

ommon laptop, e.g., Thinkpad T480, Lenovo, Beijing, China), and

nsured sufficient accuracy for extracting information of yak dis-

ribution, including herd size and location of each yak ( Fig. 3 and

able 1 ).

istribution of yak density

Characteristics of yak density distributed on the pasture var-

ed among months ( Fig. 6 ). Yak density decreased with increas-

ng distance from the campsite in June and July ( Fig. 6 a and b,

2 June = 0.44, P = 0.037; R2 July = 0.67, P = 0.004), while it exhibited

tendency to increase first and then decrease in August ( Fig. 6 c,

2 Aug. = 0.67, P = 0.020). In September, yak density increased with

istance increasing from the campsite following a power function

Fig. 6 d, R2 Sep. = 0.50, P = 0.022). The R

2 of regression analyses

ere low, especially in June and September, indicating that maybe

here are some other factors affecting the distribution of yak. For

nstance, there change in watering areas because of changes in hy-

rology in June and September, although there is a river the runs

hrough the pasture.

ispersion pattern of yak herd

Within a day, the lowest DI of yak herd appeared at noon. The

ommon tendency was that DI decreased in the morning, and in-

reased in the afternoon during each the four monitored months

Fig. 7 ). Among months, there was no differences in DIs at noon.

owever, from June to September, there was a tendency for DI to

ecrease in both the morning and the afternoon ( Fig. 7 ).

646 Y. Sun, S. Yi and F. Hou et al. / Rangeland Ecology & Management 73 (2020) 6 42–6 48

Fig. 5. Area of yak herd distribution. (a) position of yak herd at 11:00 (yellow) and 14:00 (green) on August 11, 2019, (b) potential traveling area of yak herds, and (c)

distribution area of yak herd at 11:00 and 14:00, respectively. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of

this article.)

D

C

o

i

1

m

c

m

h

2

i

m

m

c

t

c

h

p

e

a

s

s

o

o

s

t

P

t

c

o

(

r

m

Y

i

a

c

t

e

(

y

d

a

d

d

a

s

S

i

s

f

s

s

b

m

s

s

T

o

D

a

b

i

t

t

t

m

s

I

q

h

m

t

c

i

i

e

y

iscussion

haracteristics of the UAV-based method in yak herd monitoring

There are numerous methods available to monitor distribution

f animals. For instance, visual observation, biologging, and mon-

toring the distribution of feces or tracks ( Du et al., 2017 ; Lange,

969 ; Mulero-Pázmány et al., 2015 ). Generally, one person can only

onitor several herbivores and it is difficult to mark their position

orrectly; the biologging method needs lots of collars, and further-

ore, it is very difficult to install the equipment on the whole yak

erd, especially if they are accustomed to open grazing ( Du et al.,

017 , e.g., Fig. S1). Different from the commonly utilized monitor-

ng methods that only focus on some individuals that make esti-

ates of animal activities from indirect evidence, the UAV-based

ethod is focused on the entire animal herd ( Figs. 2 and S1). Be-

ause of the high monitor frequency and precise data extraction,

he method we developed characterized animal distribution pre-

isely ( Fig. 3 and Table 1 ).

Recently, some studies were conducted to monitor wildlife or

erbivores with unmanned aircraft by “MOSAIC” or randomly sam-

ling flying ( Guo et al., 2018 ; Mulero-Pázmány et al., 2015 ; Su

t al., 2018 ). Their results showed that unmanned aircraft is suit-

ble to survey animals, because of high resolution and noninva-

ive survey technology ( Mulero-Pázmány et al., 2015 ). However,

uch “survey” flying patterns (e.g., east-west- and north-south-

riented transects of Mulero-Pázmány et al., 2015 ) do not focus

n the entire herbivore herd so that it is difficult to monitor live-

tock distribution dynamically. Furthermore, the operating cost of

raditional unmanned aircraft is usually very expensive ( Mulero-

ázmány et al., 2015 ). Compared to other UAV-based sensors (e.g.,

hermal infrared camera), the lightweight UAVs (integrated RGB

amera) are flexible, stable, robust, efficient, low cost and easy to

perate ( Yi, 2017 , Fig. S3). It is very easy to monitor specific targets

e.g., individuals or a herd). In particular, gregarious livestock are

elatively easy to monitor with high frequency in a cost-efficient

anner.

ak distribution at household scale-spatial dynamics

The distribution of grazing herbivores is a major issue fac-

ng animal and rangeland managers ( Bailey et al., 1996 ). Gener-

lly, within pastures utilized by households, there is a pattern of

oncentrated impacts around campsites that attenuate with dis-

ance from camps, because animals are gathered in campsites

ach night, and are herded out to free range during the day

Fernándezgiménez et al., 2018 ). Similarly, in the present study,

ak density decreased with increasing distance from the campsite

uring June and July, while in August more yak were distributed

round the intermediate areas, and in September more yak were

istributed around the furthest areas. The pattern of yak spatial

istribution indicated that the utilization of pasture by yak was

kin to a “rotational grazing” regime (i.e., the highest grazing pres-

ure moved from near to far areas from the campsite from June to

eptember, e.g., Fig. S4).

Usually, the interactions between livestock and vegetation are

ntricate ( Sun et al., 2015 ). Biotic factors such as species compo-

ition, plant morphology, productivity, and forage quality may af-

ect grazing distribution ( Bailey et al., 1996 ). In this study, a pos-

ible reason for the yak distribution pattern may be that in early

ummer pasture, biomass increased quickly and the higher palata-

ility of the regrowth forage in the vicinity of the campsites was

ore attractive to the yak ( Dyer et al., 1986 ). Along with the ces-

ation of forage growth in August, yak grazed farther from camp-

ites to achieve their daily intake requirements ( Ding et al., 2014 ).

herefore, the yak distribution may not only indicate the “effects

n grassland”, but may also be a “response to grassland condition”.

ispersal of yak herd-temporal dynamics

Animal behavior is a manifestation of gene-environment inter-

ctions, and serves as a mediator between anthropogenic distur-

ances and animals’ fitness ( Berger-Tal et al., 2011 ). As dispersal of

ndividuals is a central part of population dynamics, it is one of

he most discussed topics in management programs, especially in

he field of wildlife conservation ( Berger-Tal et al., 2015 ). However,

here are few studies on how dispersal relates to grazing manage-

ent. The major reason is that it is difficult to quantify the disper-

al of domestic herbivores in a herd ( Bowler and Benton, 2005 ).

n this study, the UAV-based method we developed allowed us to

uantify the dispersion of a domestic herbivore.

For the first time, we characterized dispersal dynamics of yak

erd with field-measured data ( Fig. 5 and 7 ). We found that yak

ore dispersed in the morning and afternoon than at noon, and

hat dispersion in both the morning and afternoon tended to de-

rease from June to September ( Fig. 7 ). The results are mainly sim-

lar to the traveling patterns (more travelling distance in the morn-

ng and afternoon) of yak on the QTP that were reported by Ding

t al. (2014) . One possible reason is that the selective foraging of

ak decreased from June to September. Shao et al. (2010) found

Y. Sun, S. Yi and F. Hou et al. / Rangeland Ecology & Management 73 (2020) 6 42–6 48 647

Fig. 6. Relationships between yak density and distance from campsite. (a) June, (b) July, (c) August and (d) September.

Fig. 7. Dispersion pattern of yak herd. (a) June, R 2 = 0.89, P < 0.001; (b) July, R 2 = 0.75, P = 0.001; (c) August, R 2 = 0.70, P = 0.009; (d) September, R 2 = 0.61, P = 0.024; different

capital letters indicate significant differences among months and different lowercase letters indicate significant differences among monitor times.

t

w

i

r

t

l

t

d

i

1

i

e

b

o

C

t

t

o

i

u

h

y

w

w

b

t

o

hat morphological and behavioral adaptations of yak for eating a

ide variety of plant species when the herbage mass and qual-

ty is very low. Usually, the protein concentration of the herbage

ises rapidly to a maximum in June, and declines slowly over

he following four months ( Ding et al., 2014 ). That is, yak were

ess selective as the growing season progressed, which decreased

he distance of individuals travelled so that the dispersion index

ecreased.

The distribution of free-grazing herbivores is a major issue fac-

ng animal and rangeland managers ( Bailey et al., 1996 ; Walker,

995 ). The dispersion index of his study can play an important role

n grassland ecosystem management, though more theoretical and

xperimental work need to be conducted to reveal relationships

etween dispersal of domestic herbivores herds and biotic and abi-

tic factors.

onclusions

The UAV-based relatively cost-effective and noninvasive moni-

oring method of this study was able to characterize the distribu-

ion dynamics of the yak herd we studied, which extends the study

f livestock behavior from individuals to the herd. The dispersion

ndex provided an indication of livestock distribution under nat-

ral conditions, and it may play an important role in predicting

ow livestock respond to environmental changes. The pattern of

ak spatial distribution indicated that the utilization of pasture

as akin to a “rotational grazing” regime from June to September,

hich is helpful in providing insights on interactive mechanisms

etween livestock grazing and grassland conditions. The informa-

ion generated by the study contributes to the sustainable devel-

pment of grassland ecosystems.

648 Y. Sun, S. Yi and F. Hou et al. / Rangeland Ecology & Management 73 (2020) 6 42–6 48

D

c

i

A

v

m

i

w

o

N

S

S

f

R

B

B

B

B

B

C

C

C

D

D

D

D

F

F

G

G

H

H

L

L

L

M

M

M

R

R

S

S

S

S

S

W

W

X

Y

Y

Z

eclaration of Competing Interest

The authors declare that they have no known competing finan-

ial interests or personal relationships that could have appeared to

nfluence the work reported in this paper.

cknowledgments

We are grateful to Professor Anthony David McGuire for his

aluable comments and time spent in English editing of the

anuscript; Jianguo Zhang, Baoping Meng, Jianjun Chen and Jiax-

ng Du assistance in GIS data processing and analysis. This work

as funded by grants from the National Key Research and Devel-

pment Program of China ( 2018YFC040 6 601 ), the Chinese National

atural Science Foundation Commission ( 31901393 ) and Natural

cience Foundation of Qinghai Province ( 2019-0301-ZJC-0265 ).

upplementary materials

Supplementary material associated with this article can be

ound, in the online version, at doi: 10.1016/j.rama.2020.05.004 .

eferences

ailey, D. , Gross, J. , Laca, E. , Rittenhouse, L. , Coughenour, M. , Swift, D. , L. Sims, P. ,

1996. Mechanisms that result in large herbivore grazing distribution patterns. J.Range Manag. 49, 386–400 .

erger-Tal, O. , Blumstein, D. , Carroll, S. , Fisher, R. , Mesnick, S. , A Owen, M. , Saltz, D. ,St. Clair, C. , Swaisgood, R. , 2015. A systematic survey of the integration of be-

havior into wildlife conservation and management. Conserv. Biol. 30, 291–295 .

erger-Tal, O. , Polak, T. , Oron, A. , Lubin, Y. , Kotler, B. , Saltz, D. , 2011. Integrating an-imal behavior and conservation biology: a conceptual framework. Behav. Ecol.

22, 236–239 . owler, D. , Benton, T. , 2005. Causes and consequences of animal dispersal strategies:

Relating individual behaviour to spatial dynamics. Biol. Rev. 80, 205–225 . utt, B. , 2010. Pastoral resource access and utilization: Quantifying the spatial and

temporal relationships between livestock mobility, density and biomass avail-

ability in southern Kenya. Land Degrad. Dev. 21, 520–539 . ao, J.J. , Yeh, E.T. , Holden, N.M. , Qin, Y. , Ren, Z.W. , 2013. The roles of overgrazing, cli-

mate change and policy as drivers of degradation of China’s grasslands. NomadPeople 17, 82–101 .

ao, J.J. , Du, G.Z. , 2011. Differential benefits of multi- and single-household grasslandmanagement patterns in the Qinghai-Tibetan Plateau of China. Hum. Ecol. 39,

217–227 . hen, J.J. , Yi, S.H. , Qin, Y. , Wang, X.Y. , 2016. Improving estimates of fractional vege-

tation cover based on UAV in alpine grassland on the Qinghai-Tibetan Plateau.

Int. J. Remote Sens. 37, 1922–1936 . ing, L. , Long, R. , Yang, Y. , Xu, S. , Wang, C. , 2007. Behavioural responses by yaks in

different physiological states (lactating, dry or replacement heifers), when graz- ing natural pasture in the spring (dry and germinating) season on the Qing-

hai-Tibetan plateau. Appl. Anim. Behav. Sci. 108, 239–250 . ing, L.M. , Wang, Y.P. , Brosh, A. , Chen, J.Q. , Gibb, M. , Shang, Z. , Guo, X. , Jiandui, M. ,

Zhou, J. , Wang, H.C. , Qiu, Q. , Long, R.J. , 2014. Seasonal heat production and en-

ergy balance of grazing yaks on the Qinghai-Tibetan plateau. Anim. Feed Sci.Technol. 198, 83–93 .

u, W. , Yan, T. , Chang, S. , Wang, Z. , Hou, F. , 2017. Seasonal hogget grazing as a po-tential alternative grazing system for the Qinghai-Tibetan plateau: weight gain

and animal behaviour under continuous or rotational grazing at high or lowstocking rates. Rangel. J. 39, 329–339 .

yer, M.I. , Deangelis, D.L. , Post, W.M. , 1986. A model of herbivore feedback on plant

productivity. Math. Biosci. 79, 171–184 .

ernándezgiménez, M.E. , Allington, G.R.H. , Angerer, J. , Reid, R.S. , Jamsranjav, C. ,Ulambayar, T. , Hondula, K. , Baival, B. , Batjav, B. , Altanzul, T. , 2018. Using an inte-

grated social-ecological analysis to detect effects of household herding practices on indicators of rangeland resilience in Mongolia. Environ. Res. Lett. 13, 075010 .

eng, B.F. , Zhao, X.Q. , Dong, Q.M. , Xu, S.X. , Zhao, L. , Cao, J.H. , 2013. The effect of f eedsupplementing and processing on the live weight gain of Tibetan sheep during

the cold season on the Qinghai-Tibetan Plateau. J. Anim. Vet. Adv. 12, 208–211 . ao, Y. , Li, M. , 1995. Soil regionalization of the Qingahi-Xizang Plateau. Mt. Res. 13,

203–211 .

uo, X. , Shao, Q. , Li, Y. , Wang, Y. , Wang, D. , Liu, J. , Fan, J. , Fan, Y. , 2018. Applicationof UAV remote sensing for a population census of large wild herbivores-taking

the headwater region of the Yellow River as an example. Mt. Res. 10, 1041 . arris, R.B. , 2010. Rangeland degradation on the Qinghai-Tibetan plateau: A review

of the evidence of its magnitude and causes. J. Arid Environ. 74, 1–12 . ou, F.J. , Li, G. , Yang, F.G. , 2003. Grazing behavior of Gansu wapiti ( Cervus elaphus

kansuensis ) in summer and winter on the alpine grasslands of Qilianshan Moun-

tain. Acta Ecol. Sin. 23, 1807–1815 . ange, R.T. , 1969. The piosphere: sheep track and dung patterns. J. Range Manag.

22, 396–400 . ong, R.J. , Apori, S.O. , Castro, F.B. , Ørskov, E.R. , 1999. Feed value of native forages of

the Tibetan Plateau of China. Anim. Feed Sci. Technol. 80, 101–113 . ong, R.J. , Ding, L.M. , Shang, Z.H. , Guo, X.H. , 2008. The yak grazing system on the

Qinghai-Tibetan plateau and its status. Rangel. J. 30, 241–246 .

a, M. , Zhou, X. , Wang, G. , Ma, Z. , Du, G. , 2010. Seasonal dynamics in alpinemeadow seed banksalong an altitudinal gradient on the Tibetan Plateau. Plant

Soil 336, 291–302 . ulero-Pázmány, M. , Barasona, J.Á. , Acevedo, P. , Vicente, J. , Negro, J.J. , 2015. Un-

manned Aircraft Systems complement biologging in spatial ecology studies. Ecol. Evol. 5, 4 808–4 818 .

urray, D.L. , Fuller, M.R. , 20 0 0. A critical review of the effects of marking. In: Boi-

tani, L., Fuller, T.K. (Eds.), Research Techniques in Animal Ecology: Controversiesand Consequences. Columbia University Press, New York, USA, pp. 15–65 .

ey, N. , Volpi, M. , Joost, S. , Tuia, D. , 2017. Detecting animals in African Savanna withUAVs and the crowds. Remote Sens. Environ. 200, 341–351 .

utz, C. , Hays, G.C. , 2009. New frontiers in biologging science. Biol. Lett. 5, 289–292 .enft, R.L. , Coughenour, M.B. , Bailey, D.W. , Rittenhouse, L.R. , Sala, O.E. , Swift, D.M. ,

1987. Large herbivore foraging and ecological hierarchies. Bioscience 37,

789–799 . hao, B. , Long, R. , Ding, Y. , Wang, J. , Ding, L. , Wang, H. , 2010. Morphological adap-

tations of yak ( Bos grunniens ) tongue to the foraging environment of the Qing-hai-Tibetan Plateau. J. Anim. Sci. 88, 2594–2603 .

u, X.K. , Dong, S.K. , Liu, S.L , Cracknell, A.P. , Zhang, Y. , Wang, X.X. , Liu, G.H. , 2018.Using an unmanned aerial vehicle (UAV) to study wild yak in the highest desert

in the world. Int. J. Remote Sens. 39, 5490–5503 .

un, Y. , Angerer, J.P. , Hou, F.J. , 2015. Effects of grazing systems on herbage massand liveweight gain of Tibetan sheep in Eastern Qinghai-Tibetan Plateau, China.

Rangel. J. 37, 181–190 . un, Y. , Yi, S. , Hou, F. , 2018. Unmanned aerial vehicle methods makes species com-

position monitoring easier in grasslands. Ecol. Indic. 95, 825–830 . alker, J. , 1995. Viewpoint: grazing management and research now and in the next

millennium. J. Range Manag. 48, 350–357 . ang, L. , Delgado-Baquerizo, M. , Wang, D. , Isbell, F. , Liu, J. , Feng, C. , Liu, J. , Zhong, Z. ,

Zhu, H. , Yuan, X. , Chang, Q. , Liu, C. , 2019. Diversifying livestock promotes mul-

tidiversity and multifunctionality in managed grasslands. Proc. Natl. Acad. Sci. 116, 6187–6192 .

in, G.S. , Long, R.J. , Guo, X.S. , Irvine, J. , Ding, L.M. , Ding, L.L. , Shang, Z.H. , 2011. Bloodmineral status of grazing Tibetan sheep in the northeast of the Qinghai-Tibetan

Plateau. Livest. Sci. 136, 102–107 . i, S.H. , 2017. FragMAP: a tool for long-term and cooperative monitoring and anal-

ysis of small-scale habitat fragmentation using an unmanned aerial vehicle. Int.

J. Remote Sens. 38, 2686–2697 . i, S.H. , Chen, J. , Qin, Y. , Xu, G. , 2016. The burying and grazing effects of plateau

pika on alpine grassland are small: a pilot study in a semiarid basin on theQinghai-Tibet Plateau. Biogeosciences 13, 6273–6284 .

hang, H. , Yi, S. , Li, C. , Yu, Q. , Wang, Y. , 2018. Estimation of grassland canopy heightand aboveground biomass at the quadrat scale using Unmanned Aerial Vehicle.

Remote Sens. Basel 10, 851 .