Field Crops Research 210 (2017) 71–80

Contents lists available at ScienceDirect

Field Crops Research

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

Dynamic monitoring of NDVI in wheat agronomy and breeding trials using MARK
an unmanned aerial vehicle
⁎
T. Duana,b, S.C. Chapmana,c, Y. Guob, B. Zhenga,
a
CSIRO Agriculture and Food, Queensland Biosciences Precinct, 306 Carmody Road, St Lucia 4067, QLD, Australia
b
College of Resources and Environmental Sciences, China Agricultural University, Beijing 100193, China
c
School of Agriculture and Food Sciences, The University of Queensland, Gatton 4343, QLD, Australia

A R T I C L E I N F O A B S T R A C T

Keywords: While new technologies can capture high-resolution normalized difference vegetation index (NDVI), a surrogate
High-throughput for biomass and leaf greenness, it is a challenge to efficiently apply this technology in a large breeding program.
Plot segmentation Here we validate a high-throughput phenotyping platform to dynamically monitor NDVI during the growing
Unmanned aerial vehicle season for the contrasting wheat cultivars and managements. The images were rapidly captured (approximately
Image processing
1 ha in 10 min) by an unmanned aerial vehicle (UAV) carrying a multi-spectral camera (RedEdge) at low altitude
Vegetative index
(30–50 m, 2–5 cm2 pixel size). NDVIs for individual plots were extracted from the reflectance at Red and Near
Infrared wavelengths represented in a reconstructed and segmented ortho-mosaic. NDVI measured by UAV and
RedEdge camera were strongly correlated with those measured by hand held GreenSeeker (R2 = 0.85) but were
offset with UAV readings about 0.2 units higher and more compressed. The high-throughput phenotyping
platform captured the variation of NDVI among cultivars and treatments (i.e. irrigation, nitrogen and sowing).
During the growing season, the NDVI approached saturation around flowering time (∼0.92), then gradually
decreased until maturity (∼0.35). Strong correlations were found between image NDVI around flowering time
and final yield (R2 = 0.82). Given that the image NDVI includes signals from background (soil and senescenced
leaves), ground cover from a high resolution hand-held camera was used to adjust the NDVI from UAV. This
slightly increased the correlation between adjusted NDVI and yield (R2 = 0.87). The high-throughput pheno-
typing platform in this study can be used in agronomy, physiology and breeding to explore the complex in-
teraction of genotype, environment and management. Data fusion from ground and aerial sampling improved
the accuracy of low resolution data to integrate observations across multiple scales.

1. Introduction
Monitoring the growth of wheat within a season is essential to de-
cision making in both precision farming and in large breeding programs
(Magney et al., 2016a; Tester and Langridge, 2010). Low altitude re-
mote sensing provides a practical technology to monitor the crop ca-
nopy status at large scales, particularly in agronomic experiments
where space, resource and time constraints limit manual sampling.
Strong associations have been demonstrated between spectral vegeta-
tion indices and attributes of crop growth and development and are
being more frequently assessed with the development of new instru-
ments (Brown and de Beurs, 2008; Erdle et al., 2011; Huang et al.,
2014; Kyratzis et al., 2015; Cabrera-Bosquet, 2011).
The normalized difference vegetation index (NDVI) which is the
differenced ratio of reflectance in the red and near-infrared wavelength
(Tucker, 1979), is widely used in both research and commercial
agronomy applications (Erdle et al., 2011; Guo et al., 2016; Inman
et al., 2008; Lopes and Reynolds, 2012; Marti et al., 2007). The original
use of NDVI was prompted by the motivation to indirectly predict grain
yield using bands available from space with Landsat satellite data (Aase
and Siddoway, 1981; Tucker et al., 1980). The vegetation index NDVI is
well correlated with leaf area index (LAI) and is more sensitive to
changes in the crop canopy when the LAI is low (i.e. during the early
stage), with the signal becoming saturated when the crop canopy closes
(Inman et al., 2008; Ma et al., 2001; Marti et al., 2007). Some studies
showed that the yield estimated from NDVI had a strong relationship
with grain yield in wheat (Magney et al., 2016a; Raun et al., 2001).
NDVI also has been used to estimate crop growth status based on the
different patterns of reflection of green organs and soil in wheat and
other cereals (Lopresti et al., 2015; Mekliche et al., 2015; Morgounov
et al., 2014).
Temporal ground-level measurements of NDVI have provided a

⁎
Corresponding author.
E-mail addresses: duantaohao@126.com (T. Duan), scott.chapman@csiro.au (S.C. Chapman), yan.guo@cau.edu.cnc (Y. Guo), bangyou.zheng@csiro.au (B. Zheng).

http://dx.doi.org/10.1016/j.fcr.2017.05.025
Received 5 March 2017; Received in revised form 17 May 2017; Accepted 31 May 2017
0378-4290/ Crown Copyright © 2017 Published by Elsevier B.V. All rights reserved.
T. Duan et al.
more robust and objective approach to indirectly estimate the stress
status of wheat (Lopes and Reynolds, 2012) and in-season requirements
for nitrogen fertilisation (Raun et al., 2002). At the same time, NDVI has
been calibrated to estimate the nitrogen content, aboveground nitrogen
uptake, and nitrogen efficiency of crops (Erdle et al., 2011; Foster et al.,
2016; Samborski et al., 2015). The close relationship of NDVI with crop
physiological attributes means that NDVI can also explain (or be con-
founded with) a diversity of other factors, e.g. moisture, nitrogen and
growth stage (Edwards et al., 2015; Foster et al., 2016; Marti et al.,
2007). NDVI has also been used as a surrogate for visual scoring of spot
blotch disease in wheat (Kumar et al., 2016). Hence, when using NDVI,
researchers need to be sure of which factors in the experiment are being
best represented by the measurement.
The common technology for NDVI acquisition is of two types: 1)
tracking seasonal phenology over a wide variety of environments from
space using satellite data (Lopresti et al., 2015; Pantazi et al., 2016;
Zhang et al., 2016) and 2) field or plot-level estimation to guide the
actual production for crops through hand-held sensors (Inman et al.,
2008; Lopes and Reynolds, 2012; Magney et al., 2016a). The resolution
of NDVI information from satellite data typically ranges from 5 to 30 m
pixels and is appropriate for field or regional level monitoring, but is
unsuitable in field breeding and agronomy trials given limitations of
accuracy and real-time monitoring (Guo et al., 2016; Perry et al., 2014;
Tattaris et al., 2016). Ground-level hand-held sensors or cameras have
greater accuracy, but are limited in both time and space resolution,
with data collection that can suffer from subjective measurement bias
and equipment interference (Schirrmann et al., 2016). Specialised
ground-based platforms allow collection of finer spatial and temporal
resolutions using adjusted NDVI sensors and spectral reflectance sen-
sors (Magney et al., 2016b; Soudani et al., 2012). Given the growing
demand for high-throughput phenotyping to support crop breeding,
there is greater interest to develop rapid and non-destructive technol-
ogies for high-throughput phenotyping (Chapman et al., 2014; Deery
et al., 2014; Tattaris et al., 2016).
The accuracy of NDVI measurement is largely influenced by the
biophysical characteristics of the canopy and growth environment (e.g.
vegetation cover, biomass, plant and soil moisture) and effect of the
measuring equipment (satellite drift, calibration uncertainties and at-
mospheric conditions) factors (Gutman, 1999). NDVI trends can also
change in response to environmental conditions, e.g. temperature and
water regime (Crusiol et al., 2016; Forkel et al., 2013). Sensor view
angle, solar angle, radiometric calibration and soil background also
influence reflectance from the crop canopy, so the selection of the op-
timal instruments and measurement conditions for monitoring and
predicting crop growth parameters (Mulla, 2013) is crucial. In large
breeding programmes, precise non-destructive biomass estimates could
be useful in selection, particularly if they are quick, cheap and easy to
perform.
Fully or largely automated capabilities have contributed to crop
improvement through a combination of modern technologies including
genetic engineering, robotics and imaging (Araus and Cairns, 2014;
Chapman et al., 2014; Sharma et al., 2015). Biomass estimation for
barley has been derived using the crop surface models created from the
Unmanned aerial vehicles (UAV) RGB aerial imagery (Bendig et al.,
2014), and technological advances such as autonomous mission plan-
ning has increased interest in their application in precision agriculture
(Khot et al., 2016; Rasmussen et al., 2016; Zhang and Kovacs, 2012).
For crop improvement efforts to continue to increase yield potential,
the connection of genotypes and phenotypes with high efficiency is
needed. An Unmanned Aerial Vehicle with autonomous flight control
can be used to undertake remote sensing tasks as quantitative or qua-
litative information about an object without physical contact (Chapman
et al., 2014; Tattaris et al., 2016).
For phenotype analysis at the plot-level based on an UAV image set,
a major challenge is to manage and extract plot level data from these
massive image datasets. A plot scale automatic segmentation algorithm
72
Field Crops Research 210 (2017) 71–80
has been developed for ground cover estimation based on the UAV
images (Duan et al., 2017). This method retrieves data from original
images, using the orthomosaic and imputed camera positions to identify
the plots in each image. These original data have not been blurred or
mixed by reconstruction. Another important step is filtering of ‘useless
pixels’ to remove the soil and other background pixels from images (e.g.
based on partitioning clustering, Schirrmann et al., 2016). This allows
computation of NDVI based on the plant material alone (i.e. removing
effects of LAI), and fast efficient filtering is required.
The objectives of this study were to 1) develop a high-throughput
phenotyping workflow to estimate the NDVI in the plot–level through
an unmanned aerial vehicle and multiple spectral camera, 2) compare
the NDVI measured from UAV and a commercial hand held sensor, 3)
improve the accuracy of UAV NDVI through merging data from high
resolution hand-held camera, 4) characterise NDVI for contrasting
cultivars, managements and environments.
2. Materials and methods
This study applied a high-throughput method to monitor dynamic
changes of NDVI using a low altitude unmanned aerial vehicle (UAV)
platform and a multi-spectral camera in a field experiment with con-
trasting wheat cultivars and managements. NDVI estimated from UAV
was also adjusted by ground coverage obtained from the hand-held
camera.
2.1. Field experiments
Wheat experiments were conducted in 2015 at the experimental
station of Gatton Campus, the University of Queensland (27.50°S,
153.01°E). Contrasting canopy structures were established by two irri-
gation treatments (irrigation and rain-fed), two nitrogen treatments
(high and low), two sowing dates (normal and late) and eight cultivars.
The experimental field was 54 m wide and 161 m long, and split into 4
treatment blocks comprising 621 plots in total, each plot being 2 m (7
rows) wide and 7 m long. The four treatments refer to four irrigation
and nitrogen treatments (i.e. RLN for rain-fed and low nitrogen, RHN
for rain-fed and high nitrogen, ILN for irrigation and low nitrogen, IHN
for irrigation and high nitrogen, Fig. 1). Each treatment was split into
two sub-blocks for two sowing treatments. The normal and ‘late’ sowing
dates were 21st May and 22nd June, respectively. Eight cultivars were
selected to represent the contrasting traits of canopy structure (i.e.
maturity, tiller number, transpiration efficiency, water soluble carbo-
hydrate and protein content) with three replicates in each treatment.
The results of three key cultivars are discussed in this study as they
were sown at both dates (Suntop, medium-maturity cultivar; Hartog,
medium-maturity cultivar; Gregory, late-maturity cultivar). In each
sub-block, cultivars were randomized into three replicate blocks in a
column-row design by the R package DiGGer (Coombes, 2016). The
plant density was 150 plants m−2. Fertiliser was applied at sowing with
205 kg ha−1 for high nitrogen and 50 kg ha−1 for low nitrogen (Urea,
46% N) after measuring the pre-planting soil nitrogen being ca
32.3 kg ha−1 (0–60 cm), averaged for samples across the field about
one month prior to sowing. Irrigation was applied to all treatments at
sowing and one month after sowing for emergence and development at
the early stage (49 mm in total). Additional irrigations were applied
only for irrigation treatments when needed with 127 mm applied for
normal sowing and 152 mm for late sowing. Seasonal rainfall was
128 mm and 113.6 mm in the normal and late sowing, respectively.
Average daily temperatures across the season (June to Oct) were
24.2 °C for maximum and 8.7 °C for minimum. After maturity, the plots
were harvested to estimate the yield, with periodic sampling for bio-
mass and plant organ components through the season (not presented
here).
T. Duan et al. Field Crops Research 210 (2017) 71–80

(discontinued product), 3DR Robotic Systems, Palo Alta CA, USA) was
flown over the field with controlled flight pattern as substantial image
overlap is required to obtain a satisfactory reconstruction results (see
example flights here: https://phenocopter.csiro.au). Autonomous flight
plans (‘lawnmower’ designs) were constructed using Mission Planner
(open-source flight planning software for Pix Hawk autopilot, http://
planner.ardupilot.com/) to have substantial overlap (i.e. 70% forward
and 80% side) at flight heights of 30 or 50 m and with a flight speed of
3 m per second. The total flight time was approximately in 10 min. A
multi-spectral camera (Micasense RedEdge, Seattle, USA), was mounted
in landscape format. The camera captures 5 images with the same re-
solution (1280 × 800) for 5 bands (475 nm for Blue with 20 nm
bandwidth (calculated as full width half maximum bandwidth), 560 nm
for Green with 20 nm bandwidth, 668 nm for Red with 10 nm band-
width, 840 nm for Near Infrared (NIR) with 40 nm bandwidth, 717 nm
for Red Edge with 10 nm bandwidth, Fig. 2). The ground sample dis-
tance (GSD) or pixel size was approximately 1.8 cm for 30 m altitude
flights, and 3 cm for 50 m altitude flights. The image sets were captured
at 1 s interval with recorded GPS tags. To improve the accuracy of re-
flectance data, extra images were captured for a calibrated reflectance
panel at 1 m height immediately before and after each flight. The de-
tails of the UAV data acquisition are in Table 1.

2.3. Mosaic reconstruction and segmentation

After capturing image sets using UAV and RedEdge camera, the
images were processed by the Atlas web platform (atlas.micasense.com)
to reconstruct each 5 band ortho-mosaic TIF file (e.g. Fig. 1B, 110 and
78 day after sowing (DAS) for normal and late sowing, respectively).
The processing includes adjustment of the image reflectance based on
the calibration panel images pre- and post-flight. The ortho-mosaic files
were segmented into individual plots following the method in Duan
et al. (2017). A script written in the R language (R Development Core
Fig. 1. The treatment layout of the field experiment at Gatton. The four blocks represent Team, 2012) was used to outline the experiment region using four
four treatments (A: RLN for rain-fed and low nitrogen, RHN for rain-fed and high ni- corners in the reconstructed ortho-mosaic (Fig. 1B) and segment field
trogen, ILN for irrigation and low nitrogen, IHN for irrigation and high nitrogen). See text into individual plots according to the experiment design. In order to
for details of experiment design. A low-resolution reconstructed mosaic in RGB colour
avoid the alley spacing and side effects between plots, the edges of each
shown as captured on 8th Sep 2015 (B). The four pink points and yellow lines are the field
corners and planted area (input parameters of segmentation step). The pink grids are the
segmented plot were trimmed by an adjustable input parameter (10%
segmented plots with 10% trims in each side. (For interpretation of the references to color in this study). All regions with destructive harvest were excluded for
in this figure legend, the reader is referred to the web version of this article.) further calculation of NDVI (Fig. 1B, regions within the pink grid were
plots without harvest).
2.2. Capturing images using multi-spectral camera on UAV
2.4. NDVI measurement and estimation
A low cost unmanned aerial vehicle platform (PhenoCopter,) was
used to capture the multiple band images in the field at seven time The NDVI was calculated as the normalized transformed ratio be-
points during the growing season following protocols developed by tween the reflectance measured at the red wavelength range and NIR
Chapman et al., 2014 (Table 1 ). The original Phenocopter was petrol- wavelength range using the following formula (Rouse et al., 1974):
powered with the autonomous control system built in-house by CSIRO NDVI = (NIR − RED)/(NIR + RED) (1)
(Merz and Chapman, 2011) and has since been replaced by a com-
mercial electric model. An unmanned aerial vehicle (Iris+ quadcopter In the original research, the Red wavelength range was 620–700 nm
and the NIR was 750–950 nm. The reconstructed 5 band mosaic TIF
provided the Red and NIR information with peaks at 668 nm and
Table 1 840 nm wavelengths, with 10 and 40 nm full width half maximum
The flight details for an unmanned aerial vehicle and RedEdge camera during wheat bandwidths, respectively (https://support.micasense.com/hc/en-us/
growing season in 2015. The total flight time is about 10 min which is limited by battery articles/225950667-RedEdge-Manual-Specifications). The NDVI can
life. GSD is ground sampling distance and averaged for all images in a flight. The slight
variation of GSD at the same flight altitude is caused by the fluctuation of drone altitude
be calculated using Eq. (1) for each pixel of the mosaic. For the in-
in the air. dividual plot segmentation result, the average raw NDVI (NDVI-R) was
calculated as mean and standard deviation for each plot.
Date Flight altitude (m) Image number GSD (cm) During the wheat growing season, NDVI data for the target plots
9th Aug 50 2850 3.0
were collected continuously once a week from the 3rd week after
1st Sep 50 2771 3.0 sowing using the GreenSeeker RT100® (NTech Industries, Inc., Ukiah,
8th Sep 50 3380 3.0 CA, abbreviated here as NDVI-G). The wavelength bands are in the
15th Sep 30 2421 1.8 visible (red, 660 nm) and infra-red (NIR, 770 nm) regions of the spec-
26th Sep 50 3000 3.1
trum, and utilise active sensors. The full width half maximum band-
4th Oct 50 1850 3.1
19th Oct 30 2356 1.8 widths are approximately 25 nm. Measurements were taken around
midday by passing the sensor over the subplots at a height of 0.5–0.7 m

73
T. Duan et al. Field Crops Research 210 (2017) 71–80

Fig. 2. An example image set was captured by RegEdge camera which was mounted in an unmanned aerial vehicle with landscape mode.The example images came from the flight on 1st
September 2015 at 50 m height.

above the canopy. The GreenSeeker collected data at 10 Hz and data

were averaged for each plot with collection of data being triggered
while traversing the plot along the row with the capturing a swathe of
about 0.6 m wide at the top of the canopy.
For ground cover (GC) estimation of the late sowing, photos (two
per plot) were taken at the same position at the northern end of each
target plot every week using the hand held camera (Canon EOS 550D,
2592 × 1728) at a height of about 0.6–0.7 m above the canopy, which
covered more than 80% of each plot. The ground cover for each plot
was calculated using an efficient vegetation segmentation algorithm
which was developed for digital plant images taken in the natural en-
vironment (Duan et al., 2017; Guo et al., 2013). When the capture times
of NDVI and ground coverage were not matched (the maximum dif-
ference of capture times were less than 1 week), a linear spline inter-
polation was used to fit the ground coverage into capture time of NDVI-
R.
Given the low resolution of images from RedEdge (GSD was 2 or
3 cm depending on flight height, Table 1), it was not possible to ac-
curately filter the ortho-mosaic into background and vegetation. An
alternative method was developed to use the high-resolution ground
coverage images as a threshold. In each plot, the adjusted NDVI (NDVI-
A) was calculated after filtering out pixels in which NDVIs were less
than a certain percentile (i.e. 1 – proportion of ground cover). Then
NDVI-A also was calculated as the mean and standard deviation for
each plot. In practice, we would normally use a high-resolution RGB
camera on the UAV rather than using a ground-level image. However,
we did not have access to this setup at the time.
Fig. 3. The Haun Index and major growth stages for three cultivars (Suntop, Hartog and
Gregory) in the field experiments (normal and late sowing) at high nitrogen and irrigation
3. Results treatment. The black dots at the bottom are the seven flight times (DAS, See Table 1 for
actual flight date). The arrows at the bottom are the major growth stages (i.e. DC31,
The observed Haun Index and Zadoks Score for the three cultivars flowering and maturity). The values of Haun index are averaged from 12 observations.
can be seen in Fig. 3. The Huan Index describes the cereal growth stage The error bars are standard deviation.
and is calculated from the number of expanded leaves and a decimal
score approximating the proportion of expanding leaves (Haun, 1973). low nitrogen (3.6 ± 0.5 t ha−1 and 3.6 ± 0.3 t ha−1 for normal and
The Zadoks Score is a standardised reference scale used to evaluate and late sowing, respectively) (Fig. 4). The similar yields between irrigation
measure the plant growth stage in cereals (Zadoks et al., 1974). The and rainfed treatments with high nitrogen were likely associated with
Haun Index linearly increased until flag leaf stage (9.7 ± 0.2 for relatively high rainfall in the early season (116.3 mm in May/June/
Suntop, 9.6 ± 0.4 for Hartog and 12.0 ± 0.3 for Gregory in normal July).
sowing; 10.0 ± 0.4 for Suntop, 10.3 ± 0.3 for Hartog and Seven flights were conducted by an UAV at the altitude 30–50 m
11.2 ± 0.5 for Gregory in late sowing; Fig. 3). The medium-maturity during growth seasons (Table 1). For the normal sowing, the dates of
cultivars, Suntop and Hartog, had similar phenology (i.e. 42.0 ± 0.0 samplings ranged from 80 DAS (between DC31 and flowering) to 151
and 49.0 ± 0.0 DAS (DAS) for DC31, 91.5 ± 2.1 and 88.0 ± 0.0 DAS (after maturity). For late sowing, the dates of samplings ranged
DAS for flowering, 138.0 ± 0.0 and 128.0 ± 1.4 DAS for maturity in from 48 DAS (before DC31) to 119 DAS (before maturity). Most flights
normal and late sowings, respectively). The development of late-ma- were taken at one week intervals depending on the weather conditions.
turity cultivar, Gregory, was slower in two sowing dates (i.e. 53 and 66 The NDVI were dynamically monitored at seven time points during
DAS for DC31, 102 and 95 DAS for flowering, 146 and 132 DAS for growth season through the UAV platform (e.g. Fig. 5A and C). The re-
maturity in normal and late sowing, respectively). The final yield of the constructed images of whole field were segmented into individual plots
high nitrogen plots (4.7 ± 0.2 t ha−1 and 4.9 ± 0.6 t ha−1 for for each flight. Visual results of raw NDVI indicated the obviously
normal and late sowing, respectively) was significantly higher than the spatial and temporal variation of NDVI in each plot (e.g. NDVI in the

74
T. Duan et al.
high nitrogen and irrigation treatment under late sowing for three
cultivars Suntop, Hartog and Gregory in Fig. 5A). The average raw
NDVI (NDVI-R) of the three cultivars ranged from 0.89 ± 0.03 to
0.94 ± 0.02 from 48 to 78 DAS, respectively, then gradually reduced
to 0.69 ± 0.10 at 119 DAS (Fig. 5B). Variation of NDVI was observed
in the same plot apparently associated with slightly variable plant es-
tablishment (Fig. 5B). The outliers of boxplot mainly came from the
signals of background pixels (e.g. soils and senescence leaves). After
adjusting NDVI by ground coverage (NDVI-A), the variations of NDVI
were reduced in each plot with slightly greater NDVI, especially in the
later stages (0.9 ± 0.02 to 0.94 ± 0.01 at 48–78 DAS respectively,
then 0.76 ± 0.06 at 119 DAS, Fig. 5D). NDVI of Hartog decreased
faster than for Suntop, despite their similar flowering date (Fig. 1), with
the reduction in Suntop over time being more similar to the later
flowering Gregory.
Combined across all dates, there were good relationships between
the NDVIs estimated from raw images (NDVI-R) and measured by hand-
held GreenSeeker (NDVI-G, Fig. S1) (normal sowing data: R2 = 0.85,
y = 0.30 + 0.82x, N = 180, **P < 0.01. Late sowing data:
R2 = 0.84, y = 0.27 + 0.82x, N = 234, **P < 0.01, Fig. 6). Weaker
correlations and flatter slopes were found at the flight 2 (103 and 71
DAS for normal and late sowing, respectively) and 3 (78 and 110 DAS
for normal and late sowing, respectively) which had higher ground
coverage and higher flight altitude. Compared with GreenSeeker on the
ground, the lower resolution of multiple spectral camera cannot sample
the small patch of soils from higher altitude which results in a higher
NDVI-R with smaller range. Note also that the NDVI-R values were
about 0.3 units greater than the NDVI-G values, so while they are highly
correlated over a range, the range was smaller for NDVI-R (ca. 0.4–0.95
units) than for NDVI-G (ca. 0.25–0.9).
NDVI in each plot was estimated over time using the mosaic images
(Fig. 7). For NDVI-R, all pixel values were averaged to represent NDVI
in each plot, then the mean and standard deviation were estimated for
each treatment. NDVI-R followed the similar trend and reached the
maximum values around the flowering time (0.91 ± 0.03 and
0.92 ± 0.02 for the normal and late sowing respectively, Fig. 7), then
gradually decreased to maturity (0.23 ± 0.02 and 0.47 ± 0.16 for the
normal and late sowing, respectively, Fig. 7). The maximum NDVI-R
under high nitrogen treatments was slightly higher than those under
low nitrogen treatments around flowering time (NDVI-R: 0.91 ± 0.15
75
Field Crops Research 210 (2017) 71–80
Fig. 4. The final yield measurement for three cultivars (Suntop,
Hartog and Gregory), two irrigation treatments (irrigation (I) and
rain-fed (R)), two nitrogen treatments (low (LN) and high (HN) ni-
trogen), and two sowing dates (normal and late sowing). The error
bars indicate the standard deviation from 3 replicates. The measuring
dates were 21th Oct and 5th Nov for the normal and late sowing,
respectively.
and 0.88 ± 0.10 for high and low nitrogen, respectively, Fig. 7). There
was a larger difference at maturity, which was not statistically sig-
nificant when averaged across treatments (0.46 ± 0.15 and
0.36 ± 0.10 for high and low nitrogen, respectively, Fig. 7), although
it was clear that the late-sown IHN treatment had a higher significantly
greater NDVI at the last two flights in all three cultivars. The genotypic
variations of NDVI-R gradually increased from flowering time to ma-
turity (0.92 ± 0.02, 0.91 ± 0.03 and 0.92 ± 0.02 for Suntop,
Hartog and Gregory at flowering time, respectively. 0.34 ± 0.10,
0.32 ± 0.09 and 0.39 ± 0.09 for Suntop, Hartog and Gregory at
maturity, respectively).
The dynamic growth of plot-level ground cover was extracted from
images taken by a handhold RGB camera (Fig. 8 upper). There were 10
useful time points for the ground cover estimation for late sowing (for
44–119 DAS). Note that there was a lower GC value around 58 DAS due
to temporary pesticide damage of the canopy from a spray just after the
observations on 51 DAS. The ground cover adjusted NDVI estimation
(NDVI-A) was considered as the soil background has been filtered
(Fig. 8 lower). During the period of higher GC (over 0.7 during about
50–100 DAS, Fig. 8 upper), the difference between NDVI-R and NDVI-A
was small for IHN, ILN, RHN and RLN, respectively (NDVI-A:
0.93 ± 0.01, 0.88 ± 0.04, 0.91 ± 0.21 and 0.88 ± 0.04, Fig. 8
lower; NDVI-R: 0.92 ± 0.02, 0.83 ± 0.07, 0.90 ± 0.03 and
0.85 ± 0.06, Fig. 7 late sowing). However, as GC decreased (under 0.7
over 100 DAS, Fig. 5 upper), there was a difference between NDVI-A
(0.81 ± 0.06, 0.66 ± 0.09, 0.62 ± 0.16 and 0.58 ± 0.14, Fig. 5
lower) and NDVI-R (0.78 ± 0.09, 0.57 ± 0.10, 0.53 ± 0.19 and
0.47 ± 0.16 Fig. 7, late sowing) for IHN, ILN, RHN and RLN respec-
tively.
Across these diverse environments and cultivars, NDVI had strong
correlations with wheat yields for both methods of NDVI estimation.
The correlations gradually decreased from flowering time to maturity
(Fig. S2). The flight near the flowering time, when crop status is asso-
ciated with establishment of grain number, was used to compare the
correlations of two methods (i.e. NDVI-R and NDVI-A at 90 DAS for late
sowing). The NDVI-A showed a stronger correlations than NDVI-R with
yield (R2 = 0.87 and 0.82 for NDVI-A and NDVI-R at late sowing,
Fig. 9).
T. Duan et al. Field Crops Research 210 (2017) 71–80

Fig. 5. The dynamic changes of NDVI during growth season for 3 cultivars (Suntop, Hartog and Gregory) (to demonstrate the method clearly, we show results for one replicate in high
nitrogen and irrigation treatment in late-sowing trial). The sub-figures in the left show NDVI in each pixel (A and B). The boxplots on the right show the variation of NDVI (C and D). The
NDVIs are estimated using two methods (i.e. values in raw images (A and C), values from pixels after removing background (B and D)). See texts for details about adjusted NDVI using
ground coverage.

Fig. 6. Comparison between the GreenSeeker measurement NDVI (NDVI-G) and the raw NDVI (NDVI-R). There are 414 values totally: 180 matching values (5 time points from 80 to
128 days after sowing (DAS) at normal sowing) and 234 matching values (7 time points from 48 to 119 DAS at late sowing). Colours mean different flights.

76
T. Duan et al.
Fig. 7. Dynamic changes of NDVI which are estimated from raw ortho-mosaic (NDVI-R) constructed from aerial photos taken on seven dates. The field experiments included three
cultivars (Suntop, Hartog and Gregory), two irrigation treatments(irrigation (I) and rain-fed (R)), two nitrogen treatments (low (LN) and high (HN) nitrogen), and two sowing dates
(normal and late sowing). The error bars indicate the standard deviation from 3 replicates.
4. Discussion
4.1. Characterization of the field crop growth using photography
In agriculture, several technologies have been used to capture
images for high throughtput phenotyping, e.g. UAV, ground based ve-
hicle and hand-hold camera (Deery et al., 2014; Lopresti et al., 2015;
Schirrmann et al., 2016). Hand-held cameras have been used in con-
trolled environments and field scale under the natural conditions to
evaluate green leaf coverage (Sharma et al., 2015), LAI (Mora et al.,
2016) and NDVI (Schirrmann et al., 2016) in the field scale and plant
height, leaf number and leaf morphology in the plant scale (Duan et al.,
2016; Paproki et al., 2012; Pound et al., 2014). Compared with a
handheld camera, the gound based vehicle provides faster phenotyping
in the field (Deery et al., 2014; Kitchen et al., 2010), but is often limited
by canopy height and when fields are hard to access due to rainfall or
irrigation (Tattaris et al., 2016). So it is time-consuming and restricted
for image acquisition in each individual plot using a handheld camera
or ground based vehicle in the large breeding or management experi-
ment (e.g. over 600 individual plots in our experiment design, Fig. 1,
and thousands of plots in breeding trials), especially when we want to
monitor the dynamic growth between different genotypes and treat-
ments in the field. The Unmanned Aerial Vehicles (UAV) provides low
cost and greater flexibility in flight scheduling (Torres-Sánchez et al.,
2015), making it a useful tool for farmers and researchers to monitor
crops at the field scale (Chapman et al., 2014; Zarco-Tejada et al.,
77
Field Crops Research 210 (2017) 71–80
2014). In addition, the UAV can undertake autonomous flights at low
altitudes with large overlaps, which permit the acquisition of images
with sub-centimetre spatial resolution and the generation of ortho-
mosaic (Fig. 1B) and 3D reconstructions for multiple types of cameras
(Chapman et al., 2014; Sankaran et al., 2015).
4.2. Significance of dynamic monitoring of NDVI’s development for wheat
The NDVI are indicators of the combined effects of chlorophyll
concentration, canopy leaf area and yield (Erdle et al., 2011). The es-
timation of the NDVI can be used as a reference index for the dynamic
monitoring of the biomass change during the growth season of wheat.
Here we developed an efficient workflow to dynamically monitor the
NDVI change over the wheat growing season (Fig. 3).
NDVI estimates are influenced by many factors, such as measure-
ment time, sensors and environmental conditions (Crusiol et al., 2016),
and there is no one absolutely accurate measurement method for NDVI
estimation. Improved precision would also contribute to applications
for field management (Foster et al., 2016). For image based methods,
most published NDVI include soil and leaf due to the low resolution of
the hyperspectral cameras (Torres-Sánchez et al., 2013). In theory, re-
moving soil background from images can provide more accurate results
for some indices. In this study, as the relative low resolution of RedEdge
camera, the ground coverages were estimated from separate sets of
images which were taken at ground level by a hand held visual camera
based on an efficient vegetation segmentation algorithm (Guo et al.,
T. Duan et al. Field Crops Research 210 (2017) 71–80

Fig. 8. For the late sowing experiment, the dynamic changes of NDVI which are estimated from raw images but adjusted with ground coverage (NDVI-A). The NDVI-A is adjusted
according to the ground coverage (top row) estimated from the images with a hand-held camera for each target plot. The field experiments were conducted for three cultivars (Suntop,
Hartog and Gregory), two irrigation treatments(irrigation (I) and rain-fed (R)), two nitrogen treatments (low (LN) and high (HN) nitrogen).The error bars indicate the standard deviation
from 3 replicates.

2013). The ground cover adjusted NDVI (NDVI-A) at about flowering
time and the final yield showed a relatively higher correlation
(R2 = 0.87) than the raw NDVI (NDVI-R, R2 = 0.82) after minimizing
the influence of the soil (Figs. 9, S2). Further, this method showed that
the higher NDVI-R values in the late-sowing IHN treatment (Fig. 7)
were not only a result of differences in ground cover (Fig. 8). In prac-
tice, it is also possible to use a high-resolution camera on the UAV ra-
ther than to take photos on the ground. This was done by Chapman
et al., 2014; but in this experiment the UAV can only carry one camera
at a time and was dedicated to the RedEdge for this work.
The integration of remote sensing image data with crop models
represents a potential extension to this work for monitoring crop
growth status and estimating regional crop yields, such as the wheat
regional prediction technique based on coupling the WheatGrow and
PROSAIL models (Zhang et al., 2016). The complex interaction of
genotype (G), environment (E) and management (M) requires selection
specific trait in the target population environment (Cooper and
Hammer, 1996; Wilson et al., 2015). Data from the method provided
here can be used to parameterise a crop model to improve prediction of
the impact of NDVI on seasonal growth, water use, biomass and yield.

4.3. Application of high throughput phenotyping in breeding programs and

crop models 5. Conclusion

NDVI is closely correlated to an adaptive trait termed ‘stay-green’
during the period of grain filling which has been empirically included in
visual selection of breeding lines (Pinto et al., 2016). Past research
showed that stay-green in spring wheat can be determined by NDVI
independently from phenology through about 300 advanced lines under
different environments (Lopes and Reynolds, 2012; Christopher et al.,
2016). For trials of several 1000 plots (> 900) genotypes, only
10–15 min flights with a UAV platform can collect all the image in-
formation of the whole field (Chapman et al., 2014). In this paper, over
600 individual plots (Fig. 1) were segmented for computation of plot
values.
The NDVI was dynamically monitored using a UAV platform and
multi-spectral camera during growing season for the contrasting wheat
cultivars and management. The workflow in this study is suitable to
capture images of a large breeding and agronomy trial and extract
phenotypes in plot scale. NDVI measured by UAV and camera in the air
was strongly correlated with those values measured by hand-held
sensor (e.g. GreenSeeker), with the advantage of being collected much
more quickly. NDVI around the flowering period had the higher cor-
relation with final yield after adjusting with ground coverage, which
indicates how data fusion from multiple sources can provide more in-
sight into adaptation mechanisms of crops.

78
T. Duan et al.
Fig. 9. Linear regressions between measured yield and NDVI estimation (the raw (NDVI-
R) and the ground cover adjusted (NDVI-A)). There were four environments (IHN, ILN,
RHN and RLN) and three cultivars (Suntop, Hartog and Gregory) for normal sowing. The
NDVI estimation came from the days that most close to the flowering period during the
flight dates for the late sowing. The ground cover was estimated based on the hand hold
camera images for the target plots.
Acknowledgements
This research was supported by CSIRO, with TD receiving a scho-
larship funded by the China Scholarship Council, and with additional
travel and research activities (BZ, SC) supported by the CREST Program
“Knowledge Discovery by Constructing AgriBigData”.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in the
online version, at http://dx.doi.org/10.1016/j.fcr.2017.05.025.
References
Aase, J.K., Siddoway, F.H., 1981. Assessing winter wheat dry matter production via
spectral reflectance measurements. Remote Sens. Environ. 11, 267–277. http://dx.
doi.org/10.1016/0034-4257(81)90025-0.
Araus, J.L., Cairns, J.E., 2014. Field high-throughput phenotyping: the new crop breeding
frontier. Trends Plant Sci. 19, 52–61. http://dx.doi.org/10.1016/j.tplants.2013.09.
008.
Bendig, J., Bolten, A., Bennertz, S., Broscheit, J., Eichfuss, S., Bareth, G., 2014. Estimating
biomass of barley using crop surface models (CSMs) Derived from UAV-based RGB
imaging. Remote Sens. 6, 10395–10412. http://dx.doi.org/10.3390/rs61110395.
Brown, M.E., de Beurs, K.M., 2008. Evaluation of multi-sensor semi-arid crop season
parameters based on NDVI and rainfall. Remote Sens. Environ. 112, 2261–2271.
http://dx.doi.org/10.1016/j.rse.2007.10.008.
Cabrera-Bosquet, L., 2011. NDVI as a potential tool for predicting biomass, plant nitrogen
content and growth in wheat genotypes subjected to different water and nitrogen
conditions. Cereal Res. Commun. 39, 147–159. http://dx.doi.org/10.1556/CRC.39.
2011.1.15.
Chapman, S.C., Merz, T., Chan, A., Jackway, P., Hrabar, S., Dreccer, M.F., Holland, E.,
Zheng, B., Ling, T.J., Jimenez-Berni, J., 2014. Pheno-copter: a low-altitude, autono-
mous remote-sensing robotic helicopter for high-throughput field-based phenotyping.
Agronomy 4, 279–301. http://dx.doi.org/10.3390/agronomy4020279.
Christopher, J.T., Christopher, M.J., Borrell, A.K., Fletcher, S., Chenu, K., 2016. Stay-
green traits to improve wheat adaptation in well-watered and water-limited en-
vironments. J. Exp. Bot. 67, 5159–5172.
Coombes, N., 2016. DiGGer: DiGGer Design Generator Under Correlation and Blocking.
Cooper, M., Hammer, G.L., 1996. Plant Adaptation and Crop Improvement. CAB
International.
79
Field Crops Research 210 (2017) 71–80
Crusiol, L.G.T., Carvalho, J. de F.C., Sibaldelli, R.N.R., Neiverth, W., Rio, A. do, Ferreira,
L.C., Procópio, S. de O., Mertz-Henning, L.M., Nepomuceno, A.L., Neumaier, N.,
Farias, J.R.B., 2016. NDVI variation according to the time of measurement, sampling
size, positioning of sensor and water regime in different soybean cultivars. Precis.
Agric. 1–21. http://dx.doi.org/10.1007/s11119-016-9465-6.
Deery, D., Jimenez-Berni, J., Jones, H., Sirault, X., Furbank, R., 2014. Proximal remote
sensing buggies and potential applications for field-based phenotyping. Agronomy 4,
349–379. http://dx.doi.org/10.3390/agronomy4030349.
Duan, T., Chapman, S.C., Holland, E., Rebetzke, G.J., Guo, Y., Zheng, B., 2016. Dynamic
quantification of canopy structure to characterize early plant vigour in wheat geno-
types. J. Exp. Bot. 67, 4523–4534. http://dx.doi.org/10.1093/jxb/erw227.
Duan, T., Zheng, B., Guo, W., Ninomiya, S., Guo, Y., Chapman, S.C., 2017. Comparison of
ground cover estimates from experiment plots in cotton, sorghum and sugarcane
based on images and ortho-mosaics captured by UAV. Funct. Plant Biol. 44, 169–184.
http://dx.doi.org/10.1071/FP16123.
Edwards, D., Oldroyd, G., Kyratzis, A., Skarlatos, D., Fotopoulos, V., Vamvakousis, V.,
Katsiotis, A., 2015. Agriculture and climate change – adapting crops to increased
uncertainty (AGRI 2015) investigating correlation among NDVI index derived by
unmanned aerial vehicle photography and grain yield under late drought stress
conditions. Procedia Environ. Sci. 29, 225–226. http://dx.doi.org/10.1016/j.proenv.
2015.07.284.
Erdle, K., Mistele, B., Schmidhalter, U., 2011. Comparison of active and passive spectral
sensors in discriminating biomass parameters and nitrogen status in wheat cultivars.
Field Crops Res. 124, 74–84. http://dx.doi.org/10.1016/j.fcr.2011.06.007.
Forkel, M., Carvalhais, N., Verbesselt, J., Mahecha, M.D., Neigh, C.S.R., Reichstein, M.,
2013. Trend change detection in NDVI time series: effects of inter-annual variability
and methodology. Remote Sens. 5, 2113–2144. http://dx.doi.org/10.3390/
rs5052113.
Foster, A.J., Kakani, V.G., Mosali, J., 2016. Estimation of bioenergy crop yield and N
status by hyperspectral canopy reflectance and partial least square regression. Precis.
Agric. 1–18. http://dx.doi.org/10.1007/s11119-016-9455-8.
Guo, W., Rage, U.K., Ninomiya, S., 2013. Illumination invariant segmentation of vege-
tation for time series wheat images based on decision tree model. Comput. Electron.
Agric. 96, 58–66. http://dx.doi.org/10.1016/j.compag.2013.04.010.
Guo, L., An, N., Wang, K., 2016. Reconciling the discrepancy in ground- and satellite-
observed trends in the spring phenology of winter wheat in China from 1993 to 2008.
J. Geophys. Res. Atmos. 121, 1027–1042. http://dx.doi.org/10.1002/
2015JD023969.
Gutman, G.G., 1999. On the use of long-term global data of land reflectances and vege-
tation indices derived from the advanced very high resolution radiometer. J.
Geophys. Res. Atmos. 104, 6241–6255. http://dx.doi.org/10.1029/1998JD200106.
Haun, J.R., 1973. Visual quantification of wheat development1. Agron. J. 65, 116–119.
http://dx.doi.org/10.2134/agronj1973.00021962006500010035x.
Huang, K., Liu, Z., Yang, L., 2014. Evaluation of winter wheat productivity in Huang-
Huai-Hai region by multi-year graded MODIS-NDVI. Trans. Chin. Soc. Agric. Eng. 30,
153–161. http://dx.doi.org/10.3969/j.issn.1002-6819.2014.02.020.
Inman, D., Khosla, R., Reich, R., Westfall, D.G., 2008. Normalized difference vegetation
index and soil color-based management zones in irrigated maize. Agron. J. 100,
60–66. http://dx.doi.org/10.2134/agrojnl2007.0020.
Khot, L.R., Sankaran, S., Carter, A.H., Johnson, D.A., Cummings, T.F., 2016. UAS ima-
ging-based decision tools for arid winter wheat and irrigated potato production
management. Int. J. Remote Sens. 37, 125–137. http://dx.doi.org/10.1080/
01431161.2015.1117685.
Kitchen, N.R., Sudduth, K.A., Drummond, S.T., Scharf, P.C., Palm, H.L., Roberts, D.F.,
Vories, E.D., 2010. Ground-based canopy reflectance sensing for variable-rate ni-
trogen corn fertilization. Agron. J. 102, 71–84. http://dx.doi.org/10.2134/
agronj2009.0114.
Kumar, S., Röder, M.S., Singh, R.P., Kumar, S., Chand, R., Joshi, A.K., Kumar, U., 2016.
Mapping of spot blotch disease resistance using NDVI as a substitute to visual ob-
servation in wheat (TriticumTriticum aestivum L.). Mol. Breed. 36, 1–11. http://dx.
doi.org/10.1007/s11032-016-0515-6.
Kyratzis, A., Skarlatos, D., Fotopoulos, V., Vamvakousis, V., Katsiotis, A., 2015.
Investigating correlation among NDVI index derived by unmanned aerial vehicle
photography and grain yield under late drought stress conditions. Procedia Environ.
Sci. 29, 225–226. http://dx.doi.org/10.1016/j.proenv.2015.07.284.
Lopes, M.S., Reynolds, M.P., 2012. Stay-green in spring wheat can be determined by
spectral reflectance measurements (normalized difference vegetation index) in-
dependently from phenology. J. Exp. Bot. 63, 3789–3798. http://dx.doi.org/10.
1093/jxb/ers071.
Lopresti, M.F., Di Bella, C.M., Degioanni, A.J., 2015. Relationship between MODIS-NDVI
data and wheat yield: a case study in northern buenos aires province, argentina. Inf.
Process. Agric. 2, 73–84. http://dx.doi.org/10.1016/j.inpa.2015.06.001.
Ma, B.L., Dwyer, L.M., Costa, C., Cober, E.R., Morrison, M.J., 2001. Early prediction of
soybean yield from canopy reflectance measurements. Agron. J. 93, 1227–1234.
http://dx.doi.org/10.2134/agronj2001.1227.
Magney, T.S., Eitel, J.U.H., Huggins, D.R., Vierling, L.A., 2016a. Proximal NDVI derived
phenology improves in-season predictions of wheat quantity and quality. Agric. For.
Meteorol. 217, 46–60. http://dx.doi.org/10.1016/j.agrformet.2015.11.009.
Magney, T.S., Vierling, L.A., Eitel, J.U.H., Huggins, D.R., Garrity, S.R., 2016b. Response of
high frequency Photochemical Reflectance Index (PRI) measurements to environ-
mental conditions in wheat. Remote Sens. Environ. 173, 84–97. http://dx.doi.org/10.
1016/j.rse.2015.11.013.
Marti, J., Bort, J., Slafer, G. a., Araus, J. l., 2007. Can wheat yield be assessed by early
measurements of Normalized Difference Vegetation Index? Ann. Appl. Biol. 150,
253–257. http://dx.doi.org/10.1111/j.1744-7348.2007.00126.x.
Mekliche, A., Hanifi-Mekliche, L., Aïdaoui, A., Gate, P.H., Bouthier, A., Monneveux, P.H.,
T. Duan et al.
2015. Grain yield and its components study and their association with normalized
difference vegetation index (NDVI) under terminal water deficit and well-irrigated
conditions in wheat (Triticum durum Desf. and Triticum aestivum L.). Afr. J.
Biotechnol. 14, 2142–2148. http://dx.doi.org/10.4314/ajb.v14i26.
Merz, T., Chapman, S., 2011. Autonomous unmanned helicopter system for remote sen-
sing missions in unknown environments. In: Proceedings of the International
Conference on Unmanned Aerial Vehicle in Geomatics (UAV-g). Zurich, Switzerland,
ISPRS.
Mora, M., Avila, F., Carrasco-Benavides, M., Maldonado, G., Olguín-Cáceres, J., Fuentes,
S., 2016. Automated computation of leaf area index from fruit trees using improved
image processing algorithms applied to canopy cover digital photograpies. Comput.
Electron. Agric. 123, 195–202. http://dx.doi.org/10.1016/j.compag.2016.02.011.
Morgounov, A., Gummadov, N., Belen, S., Kaya, Y., Keser, M., Mursalova, J., 2014.
Association of digital photo parameters and NDVI with winter wheat grain yield in
variable environments. Turk. J. Agric. For. 38, 624–632. http://dx.doi.org/10.3906/
tar-1312-90.
Mulla, D.J., 2013. Twenty five years of remote sensing in precision agriculture: key ad-
vances and remaining knowledge gaps. Biosyst. Eng. 114, 358–371. http://dx.doi.
org/10.1016/j.biosystemseng.2012.08.009.
Pantazi, X.E., Moshou, D., Alexandridis, T., Whetton, R.L., Mouazen, A.M., 2016. Wheat
yield prediction using machine learning and advanced sensing techniques. Comput.
Electron. Agric. 121, 57–65. http://dx.doi.org/10.1016/j.compag.2015.11.018.
Paproki, A., Sirault, X., Berry, S., Furbank, R., Fripp, J., 2012. A novel mesh processing
based technique for 3D plant analysis. BMC Plant Biol. 12, 63–75. http://dx.doi.org/
10.1186/1471-2229-12-63.
Perry, E.M., Morse-McNabb, E.M., Nuttall, J.G., O’Leary, G.J., Clark, R., 2014. Managing
wheat from space: linking MODIS NDVI and crop models for predicting australian
dryland wheat biomass. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 7,
3724–3731. http://dx.doi.org/10.1109/JSTARS.2014.2323705.
Pinto, R.S., Lopes, M.S., Collins, N.C., Reynolds, M.P., 2016. Modelling and genetic dis-
section of staygreen under heat stress. Theor. Appl. Genet. 1–20. http://dx.doi.org/
10.1007/s00122-016-2757-4.
Pound, M.P., French, A.P., Murchie, E.H., Pridmore, T.P., 2014. Automated recovery of
three-dimensional models of plant shoots from multiple color images. Plant Physiol.
166, 1688–1698. http://dx.doi.org/10.1104/pp.114.248971.
R Development Core Team, 2012. R: A Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna, Austria.
Rasmussen, J., Ntakos, G., Nielsen, J., Svensgaard, J., Poulsen, R.N., Christensen, S.,
2016. Are vegetation indices derived from consumer-grade cameras mounted on
UAVs sufficiently reliable for assessing experimental plots? Eur. J. Agron. 74, 75–92.
http://dx.doi.org/10.1016/j.eja.2015.11.026.
Raun, W.R., Solie, J.B., Johnson, G.V., Stone, M.L., Lukina, E.V., Thomason, W.E.,
Schepers, J.S., 2001. In-season prediction of potential grain yield in winter wheat
using canopy reflectance. Agron. J. 93, 131–138. http://dx.doi.org/10.2134/
agronj2001.931131x.
Raun, W.R., Solie, J.B., Johnson, G.V., Stone, M.L., Mullen, R.W., Freeman, K.W.,
Thomason, W.E., Lukina, E.V., 2002. Improving nitrogen use efficiency in cereal
grain production with optical sensing and variable rate application. Agron. J. 94,
815–820. http://dx.doi.org/10.2134/agronj2002.8150.
Rouse Jr., J.W., Haas, R.H., Schell, J.A., Deering, D.W., 1974. Monitoring vegetation
systems in the great plains with Erts. NASA 351, 309–317.
Samborski, S.M., Gozdowski, D., Walsh, O.S., Lamb, David. W., Stępień, M., Gacek, E.S.,
Drzazga, T., 2015. Winter wheat genotype effect on canopy reflectance: implications
for using NDVI for in-season nitrogen topdressing recommendations. Agron. J. 107,
80
Field Crops Research 210 (2017) 71–80
2097. http://dx.doi.org/10.2134/agronj14.0323.
Sankaran, S., Khot, L.R., Espinoza, C.Z., Jarolmasjed, S., Sathuvalli, V.R., Vandemark,
G.J., Miklas, P.N., Carter, A.H., Pumphrey, M.O., Knowles, N.R., Pavek, M.J., 2015.
Low-altitude, high-resolution aerial imaging systems for row and field crop pheno-
typing: a review. Eur. J. Agron. 70, 112–123. http://dx.doi.org/10.1016/j.eja.2015.
07.004.
Schirrmann, M., Hamdorf, A., Garz, A., Ustyuzhanin, A., Dammer, K.-H., 2016. Estimating
wheat biomass by combining image clustering with crop height. Comput. Electron.
Agric. 121, 374–384. http://dx.doi.org/10.1016/j.compag.2016.01.007.
Sharma, B., Ritchie, G.L., Rajan, N., 2015. Near-remote green: red perpendicular vege-
tation index ground cover fraction estimation in cotton. Crop Sci. 55, 2252–2261.
http://dx.doi.org/10.2135/cropsci2014.09.0625.
Soudani, K., Hmimina, G., Delpierre, N., Pontailler, J.-Y., Aubinet, M., Bonal, D., Caquet,
B., de Grandcourt, A., Burban, B., Flechard, C., Guyon, D., Granier, A., Gross, P.,
Heinesh, B., Longdoz, B., Loustau, D., Moureaux, C., Ourcival, J.-M., Rambal, S., Saint
André, L., Dufrêne, E., 2012. Ground-based Network of NDVI measurements for
tracking temporal dynamics of canopy structure and vegetation phenology in dif-
ferent biomes. Remote Sens. Environ. 123, 234–245. http://dx.doi.org/10.1016/j.
rse.2012.03.012.
Tattaris, M., Reynolds, M.P., Chapman, S.C., 2016. A direct comparison of remote sensing
approaches for high-throughput phenotyping in plant breeding. Crop Sci. Hortic. 7,
1131–1139. http://dx.doi.org/10.3389/fpls.2016.01131.
Tester, M., Langridge, P., 2010. Breeding technologies to increase crop production in a
changing world. Science 327, 818–822. http://dx.doi.org/10.1126/science.1183700.
Torres-Sánchez, J., López-Granados, F., De Castro, A.I., Peña-Barragán, J.M., 2013.
Configuration and specifications of an unmanned aerial vehicle (UAV) for early site
specific weed management. PLoS One 8, 58210–58224. http://dx.doi.org/10.1371/
journal.pone.0058210.
Torres-Sánchez, J., López-Granados, F., Serrano, N., Arquero, O., Peña, J.M., 2015. High-
throughput 3-D monitoring of agricultural-tree plantations with unmanned aerial
vehicle (UAV) technology. PLoS One 10, 1–20. http://dx.doi.org/10.1371/journal.
pone.0130479.
Tucker, C.J., Holben, B.N., Elgin, J., McMurtrey, I., 1980. Relationship of spectral data to
grain yield variation. Photogramm. Eng. Remote Sens. 46, 657–666 (19800051130).
Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring
vegetation. Remote Sens. Environ. 8, 127–150. http://dx.doi.org/10.1016/0034-
4257(79)90013-0.
Wilson, P.B., Rebetzke, G.J., Condon, A.G., 2015. Of growing importance: combining
greater early vigour and transpiration efficiency for wheat in variable rainfed en-
vironments. Funct. Plant Biol. 12, 1107–1115. http://dx.doi.org/10.1071/FP15228.
Zadoks, J.C., Chang, T.T., Konzak, C.F., 1974. A decimal code for the growth stages of
cereals. Weed Res. 14, 415–421. http://dx.doi.org/10.1111/j.1365-3180.1974.
tb01084.x.
Zarco-Tejada, P.J., Diaz-Varela, R., Angileri, V., Loudjani, P., 2014. Tree height quanti-
fication using very high resolution imagery acquired from an unmanned aerial ve-
hicle (UAV) and automatic 3D photo-reconstruction methods. Eur. J. Agron. 55,
89–99. http://dx.doi.org/10.1016/j.eja.2014.01.004.
Zhang, C., Kovacs, J.M., 2012. The application of small unmanned aerial systems for
precision agriculture: a review. Precis. Agric. 13, 693–712. http://dx.doi.org/10.
1007/s11119-012-9274-5.
Zhang, L., Guo, C.L., Zhao, L.Y., Zhu, Y., Cao, W.X., Tian, Y.C., Cheng, T., Wang, X., 2016.
Estimating wheat yield by integrating the WheatGrow and PROSAIL models. Field
Crops Res. 192, 55–66. http://dx.doi.org/10.1016/j.fcr.2016.04.014.