estimating plant water stress and evapotranspiration using
TRANSCRIPT
Estimating Plant Water Stress and
Evapotranspiration using Very-High-
Resolution (VHR) UAV Imagery
Suyoung Park
Submitted in total fulfilment of the requirements of the degree of
Doctor of Philosophy
June 2018
Department of Infrastructure Engineering
The University of Melbourne
3
ABSTRACT
Adequate and timely irrigation of crops is critical to maximise productivity and water
use efficiency. Thus, having a capability of monitoring crop water status based on real-time
monitoring can promise precise irrigation scheduling in a decision-making strategy.
Unmanned aerial vehicle (UAV) remote sensing becomes a readily usable tool for
agricultural water management with high temporal and spatial resolutions. This research
estimates crop water stress and evapotranspiration (ET) by integrating very-high-resolution
UAV imagery into crop water stress model and surface energy balance model in the aspects
of crop water status and water use. Thermal infrared (TIR) and multispectral (MS) imagery
from UAV systems are used as inputs of surface temperature and vegetation canopy
structure with the aim to employ the least ground-based measurements. In the first part of
the research, an adaptive estimation method of crop water stress index (CWSI) is proposed
to determine automatic and crop-property-specific thresholds of reference boundary limits
(Twet and Tdry). In the second part of the research, diurnal changes of plant water stress are
explored by a series of UAV sensing at different times of the day. In the last part of the
research, tree-by-tree instantaneous ET rate is estimated based on remote sensing energy
balance (RSEB) model, presenting water losses of individual plant and assessing the intra-
field variability of ET. This research contributes towards the improvement of the precise and
rapid estimation of the plant water stress level by a quantitative index: crop water stress
index (CWSI) using TIR-based UAV sensing in order to assess the spatial variability of
water stress over heterogeneous agricultural fields. In addition, this research contributes to
the analysis of the tree-by-tree ET which quantifies the plant water use as well as the
assessment of the intra-field variability of ET. The results of the research demonstrate that:
1) the adaptive CWSI method showed strong relationships with stem water potential (ψstem)
and stomatal conductance (gs) as plant physiological measurements. A higher correlation to
ψstem and gs was obtained than the single reference CWSI; 2) Diurnal CWSIs showed
significant correlation to gs according to the irrigation levels at three different times of the
day (from morning to afternoon); and 3) the estimated ET was obtained with a strong linear
relationship with leaf transpiration (Tr).
4
5
DECLARATION
This is to certify that
1. The thesis comprises only my original work towards the degree of Doctor of
Philosophy except where indicated in the preface
2. Due acknowledgement has been made in the text to all other material used.
3. The thesis is fewer than 100,000 words in length, exclusive of tables, figures and
bibliographies.
Suyoung Park
6
7
PREFACE
All the works presented in the thesis are original and were conducted during my PhD
candidature (2014-2018) at the University of Melbourne. The research was supervised by
Associate Professor Dongryeol Ryu and Doctor Sigfredo Fuentes from the University of
Melbourne, and Doctor Hoam Chung from Monash University.
This thesis is based on the published works from my PhD research. The contents such as
research ideas, approaches, algorithms and figures have appeared previously in the following
publications. Chapter 4 is reformatted from the original paper published in Remote Sensing in
2017.
Publications
• Park, Suyoung, Ryu, D., Fuentes, S., Chung, H., Hernandez, E., and O'Connell,
M. (2017). Adaptive Estimation of Crop Water Stress in Nectarine and Peach
Orchards Using High-Resolution Imagery from an Unmanned Aerial Vehicle
(UAV). Remote Sensing, 9, 1-15.
• Park, Suyoung, Nolan, A., Ryu, D., Fuentes, S., Hernandez, E., Chung, H., and
O'Connell, M. (2015). Estimation of crop water stress in a nectarine orchard using
high-resolution imagery from unmanned aerial vehicle (UAV). In T. Weber, M. J.
McPhee, & R. S. Anderssen (Eds.), MODSIM2015, 21st International Congress on
Modelling and Simulation (pp. 1413 - 1419). Canberra ACT Australia: Modelling
and Simulation Society of Australia and New Zealand.
8
9
ACKNOWLEDGMENTS
I would like to thank my supervisors, A/Prof. Dongryeol Ryu, Dr. Sigfredo Fuentes
and Dr. Hoam Chung. A/Prof. Dongryeol Ryu as a primary supervisor has perfectly
guided my research, gave insightful suggestions on my research questions and
discussions, and always inspired and motivated me to conduct the research. Without his
continuous guidance, my PhD research would not be possible. I would like to express my
deepest appreciation and gratitude to Dr. Sigfredo Fuentes. His valuable advices
particularly on plant physiology and plant sensing led me to accomplish my PhD research.
I would also like to express my sincere appreciation and gratitude to Dr. Hoam Chung for
his valuable advices on many details of UAV remote sensing. With his constant
encouragement, I was able to devote myself to the research.
In addition, I would like to thank Prof. Andrew Western as the chair of my thesis
advisory committee. His valuable comments helped me to look into my research from
another perspective. Further, I am thankful to Dr. Mark O'Connell from DEDJTR, who
provided the experimental sites and gave valuable advices on the research, as well as Dr.
Lola Suarez who has encouraged my research. I wish to express my gratitude to my
friends Chih-chung Chou, Naveen Joseph, Yue Wang and all MESM group members.
Subsequent interactive discussions with them have helped me to broaden my knowledge
continually. Furthermore, I would like to express my deepest gratitude to Prof. Wooseok
Jo, who has been a tremendous mentor for me.
Finally, a special thanks to my parents and parents-in-law. They have always provided
me with endless love. Words cannot express how grateful I am to both of my parents for
all of the sacrifices that they have made on my behalf. My deepest thanks also go to my
younger brother and sister. They are my happiness, which anchors me to focus on my
research. Even though I tried to balance between study and family life, there was no
sufficient time and affordability for my family. Most of all, to my husband Junchul Kim,
who has given me a great support and immense encouragement to achieve my PhD study.
This thesis is dedicated to my lovely husband.
10
11
ABBREVIATIONS
AGL above ground level
CI confidence interval
CPU central processing unit
CV coefficient of variation
CWSI crop water stress index
DEM digital elevation model
DGPS differential global positioning system
DN digital number
ET evapotranspiration
FOV field of view
GAF ground artificial feature
GCP ground control point
GMM Gaussian mixture modelling
GPS global positioning system
GSD ground sample distance
HRMET high resolution mapping of evapotranspiration
HS hyperspectral
IRT infrared thermometry
LAI leaf area index
LE latent heat flux
METRIC mapping evapotranspiration with internalized calibration
MODIS moderate resolution imaging spectroradiometer
MS multispectral
NDVI normalised difference vegetation index
NDWI normalised difference water index
NIR near infrared
NWSB non water stress baseline
OSAVI optimized soil adjusted vegetation index
PA precise agriculture
12
PAR photosynthetically active radiation
PRI photochemical reflectance index
RANSAC random sample consensus
RDI regulated deficit irrigation
RDVI renormalised difference vegetation index
RS remote sensing
RSEB remote sensing energy balance
S-SEBI simple surface energy balance index
SC stomatal conductance
SD standard deviation
SEBAL surface energy balance algorithm
SEBM surface energy balance method
SIFT scale invariant feature transform
SR sub-regions
SVAT soil-vegetation-atmospheric transfer
SWP stem water potential
TCARI transformed chlorophyll absorption in reflectance index
TIFF tagged image file format
TIR thermal infrared
TSEB two source energy balance model
TT Tatura trellis
UAV unmanned aerial vehicle
VHR very-high-resolution
VI vegetation index
VIS visible
VISNIR visible and near infrared
VL vertical leader
VPD vapour pressure deficit
WARS wet artificial reference surface
13
TABLE OF CONTENTS
1.1 Background ............................................................................................................................. 1
1.2 Research Motivations, Objectives and Questions ................................................................. 7
1.2.1 Research Problems and Motivations ............................................................................... 7
1.2.2 Research Objectives and Questions ................................................................................ 8
1.3 Structure of the thesis .......................................................................................................... 10
Chapter 2 : Literature review ..................................................................................................... 11
2.1 Estimation of crop water stress using remotely sensed data ............................................... 11
2.2 Estimation of evapotranspiration based on remote sensing approaches ............................. 16
Chapter 3 : Calibration of multispectral and thermal infrared cameras on-board UAV ..... 21
3.1 Introduction ....................................................................................................................... 21
3.2 UAV field campaign ............................................................................................................ 24
3.2.1 Study site ....................................................................................................................... 24
3.2.2 Data acquisition ............................................................................................................ 24
3.3 Camera lens calibration ....................................................................................................... 30
3.4 Radiometric calibration ....................................................................................................... 31
3.5 Band co-registration ............................................................................................................ 33
14
3.6 Temperature retrieval .......................................................................................................... 35
3.7 Conclusion ........................................................................................................................... 36
Chapter 4 : Adaptive estimation of crop water stress in nectarine and peach orchards using
high-resolution imagery from an unmanned aerial vehicle (UAV) ....................................... 39
4.1 Introduction ........................................................................................................................ 39
4.2 Materials and Methods ...................................................................................................... 42
4.2.1 Study site description and experimental design........................................................ 42
4.2.2 Aerial thermal infrared imagery acquisition ............................................................. 45
4.2.3 Physiological data acquisition .................................................................................... 46
4.2.4 Aerial TIR image processing ..................................................................................... 47
4.2.5 Feature extraction ........................................................................................................ 47
4.2.6 Adaptive Crop Water Stress Index (CWSI) .............................................................. 48
4.3 Results ................................................................................................................................ 51
4.3.1 Canopy temperatures from sub-regions .................................................................... 51
4.3.2 Mapping adaptive CWSI ............................................................................................ 53
4.3.3 Validation of adaptive CWSI ..................................................................................... 54
4.4 Discussion .......................................................................................................................... 57
4.5 Conclusions ........................................................................................................................ 59
Chapter 5 : Relationship of CWSI-based plant water stress estimation with the data acquisition
times of the day ........................................................................................................................... 61
5.1 Introduction .......................................................................................................................... 61
15
5.2 Methodology ........................................................................................................................ 62
5.2.1 Study site description .................................................................................................... 62
5.2.2 Data acquisition ............................................................................................................ 64
5.2.3 TIR image processing and plant water stress modelling .............................................. 65
5.3 Results and discussion ......................................................................................................... 67
5.3.1 Relationship of CWSI with midday stem water potential ............................................ 67
5.3.2 Relationship of CWSI with diurnal plant water stress .................................................. 69
5.4 Conclusions .......................................................................................................................... 74
Chapter 6 : Estimation of evapotranspiration in peach orchards using very-high-
resolution imagery from an unmanned aerial vehicle (UAV) ........................................... 75
6.1 Introduction ....................................................................................................................... 75
6.2 Methods and Dataset ............................................................................................................ 78
6.2.1 Study area ...................................................................................................................... 78
6.2.2 Data acquisition ............................................................................................................. 80
6.2.3 RS data processing and ET modelling .......................................................................... 83
6.3 Results and discussion ......................................................................................................... 89
6.4 Conclusions .......................................................................................................................... 95
Chapter 7 : Summary and conclusions ........................................................................................ 97
REFERENCES ........................................................................................................................ 101
LIST OF TABLES
Table 3.1 Spectral bands of MS camera ...................................................................................... 26
Table 3.2 Specifications of MS camera ...................................................................................... 26
Table 3.3 Specifications of TIR camera ...................................................................................... 26
Table 3.4 The results of calibration parameters .......................................................................... 31
Table 3.5 Band alignment parameters ......................................................................................... 34
Table 4.1 Descriptions of sampled trees based on cultivar, training system and irrigation
treatment. ..................................................................................................................................... 45
Table 4.2 Classified sub-regions within the studied field based on crop cultivar and tree
training system ............................................................................................................................ 49
Table 4.3 Thermal values of canopy temperature using probability modelling.......................... 53
Table 5.1 Descriptions of irrigation level.................................................................................... 63
Table 5.2 Weather conditions at three times of data acquisition ................................................ 64
Table 5.3 Statistics of the extreme gs over all irrigation regimes at three times of the day ....... 72
Table 5.4 Statistics of the extreme CWSI over all irrigation regimes at three times of the day . 73
LIST OF FIGURES
Figure 3.1 Work flow of calibration process for UAV-borne imagery ....................................... 24
Figure 3.2 Multispectral camera (Mini MCA6, Tetracam) ......................................................... 25
Figure 3.3 Thermal camera (A65, FLIR) .................................................................................... 25
Figure 3.4 Reflectance targets ..................................................................................................... 27
Figure 3.5 Location of reflectance targets and vegetation sampling points (in red dot) ............. 28
Figure 3.6 Design of ground control target (GCP)...................................................................... 28
Figure 3.7 Distributions of ground control points ....................................................................... 29
Figure 3.8 Design of ground feature target ................................................................................. 30
Figure 3.9 Reprojected calibration target .................................................................................... 30
Figure 3.10 Radial distortion map of band 2 ............................................................................... 31
Figure 3.11 Spectral signatures of calibration targets in black, gray and white ......................... 32
Figure 3.12 Linear regression result of band 4 (650 nm) ............................................................ 33
Figure 3.13 Linear regression result of band 6 (850 nm) ............................................................ 33
Figure 3.14 Band to band registration: left) original image; right) registered image ................. 34
Figure 3.15 Raw signal TIR image ............................................................................................. 36
Figure 3.16 Temperature-retrieved TIR image ........................................................................... 36
Figure 4.1 Canopy systems of the nectarine/peach trees in the study site: (a) Vertical leader;
(b) Tatura Trellis ......................................................................................................................... 43
Figure 4.2 (a) Location of the study site, Stonefruit Field Laboratory orchard (Tatura,
Victoria. Australia); (b) Water deficit plots are presented in yellow, and control plots in blue.
T1-T8 represents each irrigation treatment and red dots represent the biophysical sampled tree
for each plot. Ground control point (GCP) targets in a black square with a white cross. ........... 44
Figure 4.3 (a) Example of the orthomosaic image; (b) extracted edges; (c) dilated edges; (d)
eliminated edges in the orthomosaic image ................................................................................ 47
Figure 4.4 Flow chart of the decision of adaptive reference temperatures based on sub-regions
(SR) ............................................................................................................................................. 50
Figure 4.5 (a) The orthomosaic image of surface temperature (°C) derived from thermal
imagery using UAV and classified sub-regions based on cultivars and canopy architectures;
(b) Mixture modelling in the entire site; (c) in the Nectarine trained to Vertical Leader ........... 52
Figure 4.6 (a) Adaptive CWSI map derived from UAV remote sensing using thermal infrared
images. Four dotted rectangles in red represent the area of deficit irrigation treatments and
four dotted rectangles in blue the full irrigation of control treatments.; (b) Example of CWSI
map depicting water deficit and control treatment groups for Nectarine VL (Vertical Leader);
(c) Examples of the pixel-level resolution of CWSI for T1 (in deficit) and T2 (in control)
sampled trees ............................................................................................................................... 54
Figure 4.7 (a) Relationship between stem water potential (ψstem) and CWSI of single reference
baseline; (b) relationship between ψstem and adaptive CWSI ................................................... 55
Figure 4.8 (a) Relationships between stomatal conductance (gs) and CWSI using the single
reference baseline method; (b) relationship between gs and CWSI calculated using the
adaptive method .......................................................................................................................... 56
Figure 4.9 Adaptive stem water potential map (ψstem; MPa) ....................................................... 56
Figure 5.1 (a) Study site location, Stonefruit Field Laboratory (Tatura, Victoria. Australia);
(b) Different levels of water deficit plots are presented in red (0 %), orange (20 %), green
(40 %), and blue (100 %) ............................................................................................................ 63
Figure 5.2 Flow of the CWSI estimation based on adaptive reference baselines ....................... 66
Figure 5.3 (a) Adaptive CWSI map derived from midday UAV remote sensing. The
rectangles in red represent the area of irrigation treatments; (b) Detailed CWSI map depicting
the experimental plots of W1 (20 % irrigation), W2 (0 %), W3 (40 %) and W4 (100 %) ......... 67
Figure 5.4 Relationship between stem water potential (ψstem) and CWSI at 12 h ....................... 68
Figure 5.5 Differences between canopy and air temperature at three times of the day in
control (100 %) and deficit (0 %) plots ....................................................................................... 69
Figure 5.6 Relationships between stomatal conductance (gs) and CWSI: (a) results acquired at
9 h; (b) 12 h; (c) 15 h................................................................................................................... 70
Figure 5.7 Relationships between CWSI estimated at three times of the day and the stomatal
conductance ................................................................................................................................. 73
Figure 6.1 Research flow of data processing and ET modelling ................................................ 78
Figure 6.2 (a) Location of the study site, Stonefruit Field Laboratory orchard (Tatura,
Victoria. Australia); (b) canopy structures in Vertical leader (left) and Tatura Trellis (right). .. 79
Figure 6.3 Installed sensors to measure net radiation and micrometeorological data................. 80
Figure 6.4 Field description: (left) radiometric grey targets and temperature targets, the details
of sampling trees and meteorological tower; (right) the location of targets and sample trees .... 82
Figure 6.5 Surface temperature of the study area........................................................................ 84
Figure 6.6 The iterative method of calculating the sensible heat flux in HRMET model .......... 88
Figure 6.7 Tree-row map excluding soil background using histogram-based method: (a)
surface temperature map (°C) and; (b) NDVI map ..................................................................... 90
Figure 6.8 Relationship between NDVI derived from UAV sensing and LAI retrieved from
PAR measurements on the reference trees .................................................................................. 90
Figure 6.9 Tree-by-tree aggregated results: (a) surface temperature map (°C) and; (b) LAI
map .............................................................................................................................................. 91
Figure 6.10 The estimated ET map (mm/hr): (left) the ET distribution over the orchards;
(right) the tree ET which excluded background soil ................................................................ 92
Figure 6.11 Comparison between the estimated ET and leaf transpiration on the sample trees . 93
1
Chapter 1 : Introduction
1.1 Background
Water scarcity is a critical factor that limits agricultural production. According to
projected climatic changes, agricultural drought periods over Australia can increase up to
20 % by 2030 (Glover et al., 2008). The effects of climate change on soil water balance and
plant transpiration have significant impact on crop productivity and quality under limited
water resources. Therefore, the efficient and adequate use of water in maximizing
agricultural productivity is becoming more critical. Many studies investigating the effect of
water deficit have revealed that the delicate water management based on soil-cultivar-
environmental specificity is one of the main factors that determines the quality and yield of
crops as well as water use efficiency (Osroosh et al., 2015). This is particularly seen in
high-valued horticultural crops, such as grapevine and olives. Thus, having the capability to
monitor water status (water stress) and losses (evapotranspiration) from crops on a regular
basis can assist irrigation managements in a decision-making strategy.
Monitoring water stress levels on crops
The demands of precise crop management have been rising to meet reliable supplies of
agricultural production as the world population and irrigated lands are increasing (Ihuoma
and Madramootoo, 2017). For suitable irrigation scheduling, monitoring water stress status
from crops is essential. However, timely detection of water stress on crop fields is a
challenging task since symptoms are visualized when crops are already in significant water
deficit stage. Monitoring water stress in crops as early as possible will be the key to keep
the crops in sustainably productive.
Many efforts to estimate water stress of crops have been made in agricultural science,
environmental engineering and biophysics. Early studies developed various indices for crop
water stress based on the energy balance among soil-plant-atmosphere using ground
measurements. In general, stomata from leaves start closing partially, when plants are in
water deficit state. It results in increases in leaf temperature due to reductions in
transpiration rate. Based on this phenomenon, Jackson et al. (1981) introduced a robust and
2
reliable indicator, namely crop water stress index (CWSI) which represents the normalised
water stress from plants ranging from 0 (non-water-stressed) to 1 (high-water-stressed)
which can be used as a basis for irrigation scheduling. In general, the relationship between
temperature difference of canopy and air and vapour pressure deficit (VPD) is described in
terms of the energy balance theory in environmental biophysics. CWSI represents the ratio
of actual evapotranspiration (ETa) to potential evapotranspiration (ETp) of the plant, since
plants without water stress transpire fully at the ETp rate whereas the ETa of plants in water
stress decreases below ETp. Calculation of CWSI requires ground-based measurements of
meteorological data (canopy temperature, air temperature, vapour pressure deficit, and
aerodynamic resistance) and radiometric data (incoming solar radiation). A reformulated
CWSI was proposed based on the difference between canopy temperature (Tc) to the
reference temperature at full transpiration (Twet), which is normalised by the temperature
difference between the non-transpiring leaf temperature (Tdry) and Twet as the upper and
lower reference temperatures, respectively (Jones, 1992). Consequently, estimation of
CWSI requires the measurement of canopy temperature targeting dominant foliage, which
conventionally relies on manual or continuous point measurements. During the past
decades, the canopy temperature of the ground data was measured with a hand-held
infrared thermometer, targeting the dominant foliage for studies. However, this method is
labour-intensive, costly and impractical when monitoring large areas. Furthermore,
agricultural fields commonly present spatially heterogeneous biophysical conditions, which
result in spatial variation in the different water stress levels among crops/trees and even
within a plant (Moran et al., 1994).
Since numerous data from various remote sensing (RS) platforms have become
available, several studies of plant water stress have been conducted using remote sensing
imagery as a major input, aiming to replace ground-based measurements, and to make it
applicable to larger areas (Leinonen and Jones, 2004; Cohen et al., 2005; Alchanatis et al.,
2010; Fuentes et al., 2012). The approaches introduced in the studies can be classified into
two categories according to RS imagery: 1) vegetation reflectance characteristic-based
approach, which relies mostly on visible (VIS) or near infrared (NIR) imagery; 2)
temperature-based approach, which uses thermal infrared (TIR) imagery as an input to
surface temperature values in CWSI estimations.
3
The reflectance characteristic approach is based on vegetation indices (VIs) to represent
canopy structures and greenness pigments affected by water stress on leaves. Numerous VI
s (> 100 indices) can be calculated by considering the different spectral wavelengths
obtained from the camera sensors (Xue and Su, 2017). The photochemical reflectance
index (PRI) can be an indirect indicator of water stress, since it is correlated with changes
in chlorophyll fluorescence and stomatal conductance under water stress conditions (Berni
et al., 2009b; Zarco-Tejada et al., 2012). The normalised difference vegetation index
(NDVI), the red-edge NDVI, and the optimized soil adjusted vegetation index (OSAVI)
can also be used to show the changes of canopy structures induced by water stress, since
those indices are correlated with leaf (or stem) water potential (Haboudane et al., 2002;
Gago et al., 2015). Overall, the above VIs are indirect tools to show water stress on leaves,
because changes of canopy structure or greenness caused by water stress are not shown
immediately by water stress, but by accumulated stress symptoms. This delay-in-response
issue of the VI-based methods makes it a sub-optimal tool for irrigation management.
As crops in water stress close stomata and reduce transpiration rate in leaves, the leaf
temperature becomes higher than non-stressed crops. This plant physiological effect shows
that the temperature-based approach (e.g., CWSI) is sensitive to the variability of water
status and can be a direct and early detector of water stress (Colaizzi et al., 2012; Ihuoma
and Madramootoo, 2017). Many remote-sensing CWSI approaches have focused on an
image-based analysis from near-ground or proximal platforms, assessing plant water status
at the crop level. These approaches use both VIS and TIR images to obtain the level of crop
water stress as a quantitative index, where the image co-registration method is applied to
delineate crop canopy pixels and to obtain crop temperature separately from soil
temperature (Möller et al., 2007). Recently, research on water stress detection with
unmanned aerial vehicles (UAVs) has been conducted, providing a platform to monitor
water status at field scale with higher spatial resolution (Berni et al., 2009b; Zarco-Tejada
et al., 2012; Bellvert et al., 2014; Poblete-Echeverría et al., 2016; Carlos Zúñiga et al.,
2017).
An approach of CWSI estimation employing solely TIR imagery was introduced in
order to eliminate the co-registration process with VIS imagery (Meron et al., 2010a). A
4
statistic technique was employed to obtain canopy-related temperatures and lower reference
temperature (Twet) in TIR imagery. Later, the approach was evaluated by the conventional
methods to determine Twet based on: 1) wet surface temperature; and 2) energy balance
(Rud et al., 2014). These studies presented similar correlations to ground-truth data.
Therefore, the statistical CWSI reduces the amount of time of field work and calculation
complexity of CWSI, since it does not require either any meteorological data for energy
balance methods or an equipment of wet surface. However, determining an optimal
threshold to discriminate leaves from non-leaf material in histogram analysis is difficult
since it is associated with site-cultivar specificity.
Estimating evapotranspiration (ET) on crops
Estimating evapotranspiration (ET) is becoming in higher demand to understand the
sophisticated process of water losses in crops and to assess agricultural water demand/use
more accurately. Since plant ET gives information on water uptake and losses to the
atmosphere and availability, it plays a key role to plan an optimal irrigation management.
Estimations of field-based ET using weather based approaches, surface energy balance
methods (SEBM) and soil moisture measurements have been investigated and utilized in
spot (point) measurements for several decades (Nouri et al., 2015). The field-based ET has
a limitation when expanding the ET values over spatially large and heterogeneous area. In
particular, it is not feasible to produce ET variations within the area (Nouri et al., 2015).
With the consideration of spatial variations in land features (mixture of vegetated and soil
area), remote sensing (RS) based ET has been identified as one of the most reliable and
feasible methods for estimating ET at regional scales (Kustas and Anderson, 2009).
Especially, distributed ET values in the form of map, in which ET values are presented in
spatial domain, make it possible to access the spatial variability over the heterogeneous
crop fields in precise water management. Satellite images have been used as major input for
ET estimation for the past decades (Xia et al., 2016). In general, four categories have been
proposed as RS based ET methods (Courault et al., 2005; Nouri et al., 2015): 1) empirical
direct method using the direct relationship between ET and the combined RS and
meteorological data; 2) residual method of surface energy balance model; 3) vegetation
index method where a potential or reference ET is calculated from ground measurements
5
and RS data is used to estimate crop specific factor (crop coefficient) and; 4) deterministic
methods based on Soil-Vegetation-Atmospheric Transfer (SVAT) model. Among the
different approaches, the residual method has been widely applied to estimate ET (or latent
heat flux, LE or λET), where ET is obtained as a residual in SEBM and while other energy
balance components (net radiation, sensible heat flux and soil heat flux) are estimated
based on the combined empirical and physical relationships (Kalma et al., 2008). The
residual method in SEBM employs remote sensing data as surface temperature input and
canopy structural characteristics derived from vegetation indices (e.g., normalised
difference vegetation index (NDVI) or leaf area index (LAI)).
The common techniques are the surface energy balance algorithm (SEBAL)
(Bastiaanssena et al., 1998), the simple surface energy balance index (S-SEBI) (Roerink et
al., 2000), the mapping evapotranspiration with internalized calibration (METRIC) (Allen
et al., 2007), and the two source energy balance model (TSEB) (Norman et al., 1995;
Kustas and Norman, 1999). The SEBAL, SEBI and METRIC require the presence of the
cold and hot references (pixels) within the image for estimating the ET. The two extreme
reference temperatures are allocated to set the limit of evaporative fraction (0 and 1) where
the latent heat flux (λET) is set to 0 in dry surface and sensible heat energy (H) to 0 in wet
surface, respectively (Bastiaanssena et al., 1998). On the other hand, the TSEB approach
does not require the extreme references within the image and partitions the sensible and
latent energy fluxes separately from canopy and soil in the one pixel (unit area). TSEB has
been proven as a reliable method to estimate E and T both for uniform and row-crop fields.
The common techniques were designed for estimating ET over a relatively large area
with the primary input images from satellites. The satellites have provided the visible and
near infrared (VISNIR) and thermal (TIR) bands with spatial resolution (30 m – 1.0 km),
e.g., Landsat 7 with VISNIR 30 m and TIR 60 m, Landsat 5 with 30 m and 120 m, and
MODIS with 0.5 km and 1 km. As demanding subfield-scale, even tree-level management
for a precise agriculture, the resolution of typical satellite data is, however, coarse to
provide fine information on plant water status and symptoms of vigoro and stress. In
addition, the temporal resolution is rather lower to capture a target area in a projected time
on demand, since typical satellites revisit cycle usually takes several days (Zipper and
6
Loheide Ii, 2014; Ortega-Farías et al., 2016; Xia et al., 2016). High resolution TIR and
VISNIR imageries from a modern UAV sensing system can achieve less than 10 cm spatial
resolution depending on the flight altitude and camera sensor specifications (Berni et al.,
2009b; Zarco-Tejada et al., 2012; Roy and Ophori, 2014; Rud et al., 2014). The high
resolution imagery from the UAV makes it possible to interpret tree-level variabilities of
crop water uptake and status such as ET and water stress symptom, since precise canopy
extractions from soil background is feasible. In the case of heterogeneous fields (e.g., fruit
tree crops, the intra-field characteristics separately from tree row and inter-row are hardly
shown in the imagery from low resolution platform (Alessandro et al., 2015). On this
subject, the TSEB has been widely applied to partition the energy fluxes from tree (canopy)
and soil in the mixture pixels. As one of ET methods using a high resolution imagery,
Zipper and Loheide Ii (2014) introduced a combined model of one-source and two-source
scheme based on a pixel-based ET estimate, called High Resolution Mapping of
Evapotranspiration (HRMET). HRMET is designed to partition the available energy (net
radiation) to canopy and soil in a pixel and calculates the sensible and latent heat flux
iteratively.
7
1.2 Research Motivations, Objectives and Questions
In this section, the research motivations are discussed based on current research
limitations and the research objectives are also addressed. The research questions are
defined in order to achieve the objectives.
1.2.1 Research Problems and Motivations
Presence of partial canopy and soil pixels
As introduced in the background, recent research on the CWSI and ET estimation has
been largely investigated using remote sensing data. However, a fundamental problem
associated with extracting canopy pixels (or temperatures) is the presence of mixed pixels
that represent partial coverages of both canopy and soil. The mixed pixels of thermal
imagery can cause significant biases in canopy temperature during midday when they are
included in the clusters of canopy pixels. Thus, the mixed pixels should be excluded to
reduce errors when determining canopy-related temperatures such as in the histogram-
based CWSI and ET estimation.
A single upper and lower reference
Typically, most methods of estimating the upper and lower reference temperatures
employ single values of Twet and Tdry as a reference. However, the reference value of Twet
and Tdry can change according to field circumstances. In the case of temperate tree crops,
orchards may contain several crop cultivars and canopy training systems even under the
same irrigation regime. Therefore, applying a single set of Twet and Tdry could result in
inaccurate and misleading CWSI, and do not reflect the actual water status, particularly in
the orchards with a combination of multi cultivars and different canopy structures.
The relationship of estimated plant water stress with data acquisi tion time of
the day
CWSI measured at midday is known to provide reliable information on daily plant
water stress. UAV remote sensing becomes a readily usable tool for precise agricultural
water management with high temporal and spatial resolutions. In terms of data acquisition
aspect, UAV sensing has also been limited to conduct at midday in most previous research
8
in order to obtain the highest thermal contrast between canopy temperature and air
temperature, or soil background temperature. However, the relationship of CWSI-based
plant water stress with data acquisition time of the day has not been investigated yet.
Very-high-resolution ET
Remotely sensed ET has been known as a reliable method for estimating and mapping
ET at regional scales. Satellite images which have been used in ET estimation are relatively
low spatial resolution (0.1 – 1.0 km) such as Landsat and MODIS compared to airborne
images. Recently, very-high-resolution (VHR) TIR and VISNIR images from UAV
platform become available. It provides centimetre resolution in a pixel, enabling precise
canopy (leaf) extractions from soil background. However, current research on RS ET
utilizing VHR imagery is still on its early stage since 2015 and has not been fully
investigated in site/cultivar specific aspects.
1.2.2 Research Objectives and Questions
Considering the research limitations and problems identified above, this research is
divided into three main parts, in which a method of estimating plant water stress is
addressed and plant ET is modelled, in the aspects of crop water status and water
consumption, respectively.
Firstly, this research proposes an adaptive method of mapping high-resolution plant
water stress at field scale using infrared thermography on-board UAV, showing realistic
spatial representation of water status within plants. The research aims to develop an
automated method to interpret very high resolution data to obtain site/cultivar/canopy
structure-specific water stress. The sub-objectives are: 1) to propose an adaptive
methodology to determine the upper and lower references based on feature extraction and
probability modelling and; 2) to provide a practical method of water stress mapping,
utilizing a small number of ground truth biophysical measurements.
Secondly, this research investigates the relationship of CWSI-based plant water stress
and data acquisition time of the day. The research explores diurnal changes of plant water
9
stress which are captured by a series of different UAV sensing times. The investigation
aims to evaluate the differences in sensitivities of thermal gradient and response that exist
within an orchard and the extent of differences in plant temperatures between well-irrigated
and water deficit orchard from morning to late afternoon.
Lastly, this research estimates tree-by-tree instantaneous ET rates based on a remote
sensing energy balance (RSEB) model, presenting water availability of individual plant
systematically for assessing the intra-field variability of ET at sub-field scale. The sub-
objectives are: 1) to estimate the energy balance components based on HRMET method
using VHR multispectral and thermal data from a UAV platform; and 2) to assess the intra-
field variability of tree-level ET derived from a tree segmentation method of representing
more realistic canopy structure per tree.
To achieve above objectives, the research addresses series of questions to be
investigated:
1. How can signal based-radiometric temperature captured by TIR camera be
accurately calibrated to retrieve actual temperature?
TIR sensing provides a radiometric temperature, emitted radiation from surface. The
radiometric temperature can be variable by view angle of TIR sensor, sensor
displacement (vignette), and surface emissivity. This means TIR-based leaf
temperature may not be representative of the absolute leaf temperature. Thus,
radiometric temperature should be calibrated.
2. How to improve current CWSI methods for accurate and reliable estimations?
a. Is it possible to employ a probability model to achieve an automatic decision
of upper and lower reference temperatures?
b. How do the characteristics of different cultivar and canopy training systems
effect the CWSI estimations?
10
c. What is expected in a new adaptive method of CWSI proposed in this
research? Will the adaptive method improve the performance for estimating
CWSI in terms of accuracy and efficiency?
3. Can CWSI-based plant water stress be estimated anytime during the day?
a. How can behaviours of water stress during the day be interpreted with
precise sensitivity?
b. How does the CWSI estimated in morning or late afternoon correspond to
the CWSI estimated at midday for overcoming the limitation of data
acquisition time?
4. Can the current RS ET model be applicable to VHR sensing data and produce
accurate ET?
a. How can VHR sensing data be applied to the current ET model?
b. How to exclude the soil background effect in VHR data for individual plant
ET?
c. How can a more realistic canopy structure be coupled in the model as an
input?
1.3 Structure of the thesis
In the remainder of this thesis, Chapter 2 reviews related studies with comparisons and
discussions. Chapter 3 describes the characteristics of VHR sensing data from a UAV
platform and image calibration methods which are carried out in the research. Chapter 4
introduces a novel method of CWSI estimations. Chapter 5 presents diurnal changes of
plant water stress, estimated at the different sensing times. Chapter 6 describes VHR ET
estimation and mapping based on individual tree. Lastly, Chapter 7 summarizes the detailed
findings, discussions and conclusions, and future works are suggested.
11
Chapter 2 : Literature review
This chapter reviews detailed background regarding various remote sensing methods
for plant water stress and evapotranspiration (ET). The review starts with the fundamental
research of crop water stress estimations in detail, then deals with ET models; applications
of remote sensing methods will be discussed based on relevant works in the aspects of their
differences and limitations.
2.1 Estimation of crop water stress using remotely sensed data
This chapter presents the related researches on plant water stress detection and
estimation using remote sensing methods. The methods are classified into two categories:
vegetation indices approaches in various spectral ranges and temperature-based approaches
using thermal infrared (TIR) range.
Visible-near infrared (VNIR) vegetation indices derived water stress
In general, plant pigments on leaves (e.g., xanthophyll and chlorophyll) absorb most
radiance in visible (VIS, 400 nm – 740 nm) wavelengths and reflect the absorbed radiance
in near infrared (NIR, 740 nm – 1200 nm) wavelengths. Water-stressed plants typically
show reduced reflectance in NIR due to a drop of photosynthetic absorbance (Kim et al.,
2011). Therefore, various methods for detecting water stress have been studied using
spectral vegetation indices combined VIS and NIR wavelength. In particular, with the
introduction of hyperspectral sensors that provide a large number of narrow bands, there
are more opportunities for analysing specific targeted wavelengths of vegetation and water.
Among the spectral indices utilizing narrow bands, the Normalized Difference Vegetation
Index (NDVI) with two very specific wavelengths of 750 nm and 705 nm, namely the Red
edge NDVI, showed a strong correlation (r = 0.94) with plant water stress (Kim et al.,
2011). The imagery from hyperspectral sensor, however, is very sensitive to ambient
illumination changes. For obtaining a valid dataset, the sensor requires frequent calibration
during data acquisition.
12
Photochemical Reflectance Index (PRI) has been used as an indicator of photosynthetic
pigment changes in leaves to predict crop water stress. In a case study performed in a citrus
orchard, PRI showed a high correlation with plant-based water stress indicator (leaf water
potential and stomatal conductance) (Zarco-Tejada et al., 2012). An improved PRI has
been introduced namely Normalised Photochemical Reflectance Index (PRInorm) by Berni
et al. (2009b), which employs a new PRI normalised by Renormalised Difference
Vegetation Index (RDVI) and a red edge index. PRInorm presented a higher relationship
with plant-based water stress indicator since it is sensitive to detect the xanthophyll
pigment changes as well as reduced leaf area caused by water stress (Gago et al., 2015).
Other vegetation indices have been examined to present the possibility of detecting
plant water stress. Optimized Soil Adjusted Vegetation Index (OSAVI) and Transformed
Chlorophyll Absorption in Reflectance Index (TCARI) are the canopy structural indices,
which have presented better performances, minimizing soil reflectance effect unlike NDVI
(Haboudane et al., 2002; Zarco-Tejada et al., 2012). A ratio (TCARI/OSAVI) was
introduced by Haboudane et al. (2002) which was designed for predicting chlorophyll
content accurately. The ratio has also shown to be sensitive to changes of canopy structure
cover induced by plant water stress (Zarco-Tejada et al., 2012; Gago et al., 2015).
A normalised difference water index (NDWI) that employs two narrow NIR
wavelengths (0.86 and 1.24 μm) for estimating vegetation liquid water by satellite image is
proposed by Gao (1996). NDWI is a ratio of reflectance at two NIR wavelengths,
maximizing the differences between green and dry vegetation affected by liquid water
molecules in canopies. There would be a limitation to apply NDWI to entire crop field,
since the effect of soil surface reflectance as a background is also included in NDWI and it
results in negative outcome of analysis in water status of vegetation.
VNIR-based vegetation indices (e.g., NDVI, TCARI, OSAVI, and PRI) are good
indicators of detecting plant water stress, especially in monitoring changes for long term
stress at regional scale. In the case of broadband NDVI and narrowband Red edge NDVI,
13
research showed that the reflectance changes were not significant in mild water-stressed
stage. Thus, the indices accommodated water stress symptoms better, after the moderate
water-stressed stage (Kim et al., 2011).
Temperature-based approach of estimating water stress
Canopy temperature-based methods have been acknowledged as a sensitive tool to
detect plant water stress (Cohen et al., 2005). Water stress effects on stomatal conductance
(SC) and leaf transpiration, resulting in increased canopy temperature. Instead of measuring
stomatal conductance directly using leaf gas exchange methods, the techniques using
infrared thermometry (IRT) to detect stomatal closure and estimate the conductance have
been introduced (Jones, 1999; Jones et al., 2002; Leinonen et al., 2006; Fuentes et al.,
2012). Direct measurements could be inconvenient when a large sampling volume of leaves
is necessary (Jones, 1999). The IRT concept is based on leaf temperature that tends to
increase when stomatal closure occurs, since it limits transpiration and hence, the cooling
effect of leaves. Based on this concept, Jones (1999) proposed an index ( 𝐼𝑔), which has a
direct linear relationship with stomatal conductance. The index (𝐼𝑔) has a sensitive
relationship with SC, even responding to relatively small variation of SC.
CWSI has been widely adopted as an indicator of plant water stress. CWSI theory is
based on the fact that leaf transpiration cools the canopy surface in non-water stressed
plant. The leaf transpiration drops and the canopy temperature increases in water-stressed
plant, as available water is depleted in root zone. Thus, CWSI is presented in a ratio of
actual evapotranspiration (ETa) to potential evapotranspiration (ETp) of plant (Jackson et
al., 1981). A linear relationship between canopy temperature (Tc), air temperature (Ta) and
vapour pressure deficit (VPD) was found (Idso et al., 1981) and developed an empirical
CWSI, which employs lower and upper baselines to estimate plant water stress as follows:
𝐶𝑊𝑆𝐼 = (𝑇𝑐−𝑇𝑎)−(𝑇𝑙𝑜𝑤𝑒𝑟−𝑇𝑎)
(𝑇𝑢𝑝𝑝𝑒𝑟−𝑇𝑎)−(𝑇𝑙𝑜𝑤𝑒𝑟−𝑇𝑎) (1)
14
where, 𝑇𝑙𝑜𝑤𝑒𝑟: lower baseline equitant to the temperature of non-water-stressed canopy;
and 𝑇𝑢𝑝𝑝𝑒𝑟: upper baseline equitant to the temperature of water-stressed canopy
The lower and upper baselines present the 𝑇𝑐 − 𝑇𝑎 at potential ET rates in non-water-
stressed and water-stressed plant respectively. Later, Jones (1992) proposed a reformulated
CWSI from the empirical CWSI as follows:
𝐶𝑊𝑆𝐼 = 𝑇𝑐− 𝑇𝑤𝑒𝑡
𝑇𝑑𝑟𝑦− 𝑇𝑤𝑒𝑡 (2)
where, 𝑇𝑤𝑒𝑡: the temperature of non-water-stressed canopy and; 𝑇𝑑𝑟𝑦: the temperature
of water-stressed canopy
In the reformulated CWSI, 𝑇𝑤𝑒𝑡 and 𝑇𝑑𝑟𝑦 represent the reference temperature at full
transpiration and at non-transpiring leaf temperature as the lower and upper reference
temperatures, respectively.
During the past several years, researches on water stress estimation have been
investigated in different crops and improved in the use of RS imagery. First of all, the
research adopting both VIS and TIR imagery has been attempted to estimate crop water
stress (Leinonen and Jones, 2004; Möller et al., 2007). For the calculation of pure crop
temperature excluding the soil temperature, the co-registration method of VIS and TIR
imagery has been accepted widely. This enables to distinguish canopy cover area and soil
surface in VIS image, and then compute the canopy cover temperature in TIR image, which
is co-registered with VIS image. Despite the usefulness of combined two types of imagery
data, it rather requires time-consuming work of co-registration.
Histogram-based approach has been introduced to separate the canopy temperature
from soil temperature using TIR imagery solely (Meron et al., 2003; Meron et al., 2010a).
The method employs the pixel histogram of TIR image and uses a constant threshold (e.g.,
the coldest 33 % of the histogram) to obtain canopy-related temperatures in the histogram.
The threshold represents the boundary of vegetation and soil distribution in the histogram,
and the value is determined by statistical and empirical methods. The study showed that the
15
process of CWSI estimation was facilitated and simplified, eliminating the need for image
processing with VIS image. However, the fine determination of threshold values to discard
non-canopy pixels in histogram analysis is required, depending on site specific crop types.
In addition, a fundamental problem associated with the histogram-based clustering of
canopy pixels and subsequent calculation of canopy temperatures is the presence of mixed
pixels that represent partial coverages of both canopy and soil. Due to the nature of RS
imagery, mixed pixels of canopy and non-canopy background (e.g., soil) exist along their
boundaries. The mixed pixels of thermal imagery can cause significant bias in canopy
temperature during midday when they are included in the clusters of canopy pixels.
Determination of Twet and Tdry for CWSI poses critical influence on the accuracy of
CWSI values as an indicator of crop water stress. The various methods have been proposed
and examined for accurate determination of lower and upper reference values. Twet can be
obtained by (1) the calculation of Non Water Stress Baseline (NWSB) in a linear regression
function between Tc – Ta and VPD (Jackson et al., 1981), (2) the temperature measurement
of wet artificial reference surface (WARS), acting as crops in abundant water status
regarded as the temperature of a fully transpiring crop (Meron et al., 2013), and (3) a
statistical method based on histogram analysis of canopy temperature (Rud et al., 2014;
Poblete-Echeverría et al., 2016). Rud et al. (2014) evaluated above three methods to
calculate Twet and subsequent CWSI, resulting that all forms of CWSI have similar
correlations with actual reference data. In other words, CWSI based on statistical analysis
does not require either meteorological data for energy balance method or an equipment of
WARS. This outcome enables time reduction of field-work and calculation complexity of
CWSI. As the upper reference, Tdry has been derived from (1) ancillary meteorological
variables based on isothermal radiation (Möller et al., 2007; Jones, 1999), (2) dry reference
leaf coated with petroleum jelly to prevent transpiration (Jones, 1999), and (3) an empirical
method by adding a fixed threshold (e.g., 5 °C) to Ta (Cohen et al., 2005; Irmak et al.,
2000; Rud et al., 2014; López-López et al., 2011). The determination of Twet and Tdry from
meteorological data or empirical method has been widely adopted. However, Twet and Tdry
change with weather condition and weather changes with time of day and location. Thus,
even for the same crop, different Twet and Tdry have been suggested from different places
(Yazar et al., 1999; Irmak et al., 2000).
16
Due to the raising need of cost-effective and fast turnaround RS data acquisition,
recent research on water stress detection using UAV has been conducted, providing
detailed information of plant water status at field scale with higher spatial resolution
(Berni et al., 2009b; Zarco-Tejada et al., 2012; Rud et al., 2014; Poblete-Echeverría et al.,
2016; Carlos Zúñiga et al., 2017). An effort of finding an optimal UAV flight time was
made in UAV-based CWSI research, showing that midday UAV sensing promises the
optimal data acquisition, avoiding shade effects on the surface (Bellvert et al., 2014). Such
UAV-based methods will be a promising tool to detect plant water status with the
integrated technology, combining advanced image analytic algorithms and improved
camera systems (Gago et al., 2015).
2.2 Estimation of evapotranspiration based on remote sensing
approaches
Evapotranspiration (ET) is the most difficult, but necessary component to quantify in
hydrological balance cycle. Quantification of water loss from vegetation surface is a
fundamental requirement in irrigation scheduling at the local management with respect to
climate change in the global assessment (Nouri et al., 2015). In precise agriculture (PA),
plant ET measurement becomes important to understand the sophisticated process of water
consumption in crops, thus it assists to scheme a timely and optimal water management.
For several decades, field-based ET methods have been widely utilized for predicting
plant ET, which are mainly classified into weather based approaches, surface energy
balance methods (SEBM) and soil moisture measurements (Allen et al., 1998; Allen, 2000;
Zhao-Liang et al., 2009). The field-based methods are performed mainly by spot
measurements and have provided ET at local and regional scale by extrapolating technique.
Therefore, the field-based ET has a limitation when extrapolating the ET values over
spatially large area that contains micrometeorological changes. In addition, it is challenging
to access the inter- and intra-field variability of ET through the field-based approaches.
Remote sensing (RS) based-ET estimation has been introduced and improved with
various sensors. Satellite images have been used as major inputs of ET estimation for the
17
last decade (Xia et al., 2016). RS based approach has been known as one of the reliable and
efficient methods for estimating ET at regional scales (Kustas and Anderson, 2009). As RS
has a capability of capturing the features of mixed landscape in partially vegetated surface,
RS based-approach is beneficial to estimate ET in heterogeneous surface in contrast to the
field-based ET. In particular, RS based-approach produces ET estimation as a map, which
presents ET values in pixels and shows the spatial variability over mixed landscape.
In general, four categories were proposed as RS methods for estimating ET namely: 1)
empirical direct method; 2) residual method; 3) inference method where a potential (or
reference) ET is calculated from ground measurements and RS data is used to estimate crop
coefficient and; 4) deterministic method based on Soil-Vegetation-Atmospheric Transfer
(SVAT) model (Courault et al., 2005; Calcagno et al., 2007; Nouri et al., 2015). In the
empirical direct method, ET is estimated directly from a simplified relationship between
ET and the combined RS data (e.g., TIR) and meteorological data (e.g., air temperature).
For example, a simple direct method was introduced and analysed, which relies on the
assumption that daily ET has a direct relationship with the difference of surface and air
temperature (Ts − Ta) at midday as shown in Equation 3 (Jackson et al., 1977; Seguin and
Itier, 1983).
𝐸𝑇𝑑𝑎𝑖𝑙𝑦 = 𝑅𝑛,𝑑𝑎𝑖𝑙𝑦 + 𝐴 − 𝐵(𝑇𝑠,𝑚𝑖𝑑𝑑𝑎𝑦 − 𝑇𝑎,𝑚𝑖𝑑𝑑𝑎𝑦) (3)
where, 𝐸𝑇𝑑𝑎𝑖𝑙𝑦: daily ET; 𝑅𝑛,𝑑𝑎𝑖𝑙𝑦: daily net radiation; 𝑇𝑠,𝑚𝑖𝑑𝑑𝑎𝑦: surface temperature
measured at midday; 𝑇𝑎,𝑚𝑖𝑑𝑑𝑎𝑦: air temperature measured at midday and; A and B:
coefficients
The method assumes soil heat flux is near zero (𝐺𝑑𝑎𝑖𝑙𝑦 = 0) as daily sum and the ratio
𝐻 𝑅𝑛⁄ is the constant value during the day. A and B are coefficients, determined site-
specifically as calibration parameters. The method showed up to 10 % accuracy of ET
estimation when the calibration parameters were empirically well-calculated at a local
scale. Another empirical direct method has been explored based on the relationship
between NDVI and Ts (Moran et al., 1994; Carlson et al., 1995). The method relies on the
fact that the amount of vegetation (e.g., NDVI) affects the ET quantity, and ET is
18
associated with surface temperature. In other words, as vegetation coverage increases, more
ET is expected, and the surface temperature tends to be lower. The method showed that the
relationship between surface temperature and NDVI is strong, but varies in seasonal
changes (Carlson et al., 1995; Yuan and Bauer, 2007).
The residual method uses the surface energy balance model (SEBM) and has been
widely utilized to estimate ET, where ET is obtained as a residual (latent heat flux, LE or
λET) in SEBM and while other energy balance components (net radiation, sensible heat
flux and soil heat flux) are estimated based on the combined empirical and physical
relationships as follows (Su, 2002; Kalma et al., 2008):
𝜆𝐸𝑇 = 𝑅𝑛 − 𝐻 − 𝐺 (4)
Where, 𝜆𝐸𝑇: latent heat flux (W/m2); 𝑅𝑛: net radiation at the surface (W/m2); 𝐻:
sensible heat flux to the air (W/m2) and; 𝐺: soil heat flux (W/m2)
The residual method in SEBM employs remote sensing data to obtain surface
temperature and canopy structural characteristics derived from vegetation indices (Yuei-An
and Kumar Kar, 2014) and; to estimate Rn and G (Bastiaanssena et al., 1998). Extensive
researches have been explored using different techniques: 1) Surface Energy Balance
Algorithm for Land (SEBAL) (Bastiaanssena et al., 1998); 2) Surface Energy Balance
System (SEBS) (Su, 2002), Surface Energy Balance Index (SEBI) , Simplified Surface
Energy Balance Index (S-SEBI) (Roerink et al., 2000) and; 3) Two Source Energy Balance
(TSEB) (Norman et al., 1995; Kustas and Norman, 1999). First of all, SEBAL is designed
to estimate ET at regional scale with minimum ground data. Surface temperature, incoming
solar radiation, NDVI, and threshold values (wet and dry) are required for the model. The
threshold values are two extreme reference temperatures inside the target area which are
allocated to set the limit in dry surface (λET = 0) and wet surface (H = 0). Based on the
concept of wet and dry thresholds to sensible heat flux calculation, SEBS algorithm
estimates evaporative fraction and turbulent fluxes using RS data (e.g., Ts, albedo, NDVI,
emissivity), meteorological data (Ta, air pressure, relative humidity, wind speed), and
radiation data. SEBI and S-SEBI have been developed to estimate ET from RS data,
19
following after SEBAL. S-SEBI model is a simplified model and requires only RS data
without any further information, because the method determines wet and dry thresholds by
the minimum and maximum reflectance of temperature in the target image. Thus, the
method can show reliable result, when wet and dry surface extremes exist in the target area.
On the other hand, TSEB does not require the surface extreme within the image. As a dual-
source model, it partitions the sensible and latent energy fluxes separately from canopy and
soil. The TSEB has been proven as a reliable method to estimate E and T both for
vegetated area and bare soil (Xia et al., 2016). In particular, the model shows better ET
estimation than general single-source model in bare soil. In the case of vegetated surface,
however, it could be a heavy process in terms of operational ET estimation due to complex
parameterizations compared to single-source model (French et al., 2015; Xia et al., 2016).
For the past decades, satellite imagery has been a primary input to estimate ET over a
relatively large area. In general, the spatial resolution of satellite TIR imagery is coarser
than 60 m (e.g., Landsat 7) and the revisit cycle takes several days (e.g., 16 days of Landsat
8). The current UAV sensing has provided very-high-resolution (VHR) imagery down to
sub-centimetre spatial resolution, depending on the flight altitude and camera sensor
specifications (Berni et al., 2009b; Zarco-Tejada et al., 2012; Roy and Ophori, 2014; Rud
et al., 2014). In addition, UAV sensing can capture a target area in a projected time on
demand with high temporal resolution (Zipper and Loheide Ii, 2014; Ortega-Farías et al.,
2016). Furthermore, the operation of UAV sensing is cost-effective and the onboard
camera system (e.g., TIR, MS, hyperspectral sensor) can be selective according to the
research purpose (Brenner et al., 2017). With the rapid growing attraction and interest
towards UAV sensing, several researches on ET estimation have been investigated with the
present energy balance models during the past few years (Zipper and Loheide Ii, 2014;
Hoffmann et al., 2015; Ortega-Farías et al., 2016; Xia et al., 2016; Brenner et al., 2017).
However, capability and limitation of UAV sensing is still being figured out in the aspects
of data consistency of acquisition, software development of data processing and quality
assessment of data. Subsequently, research on energy fluxes estimation from UAV sensing
is still in its early stage and needs to be further investigated and validated (Hoffmann et al.,
2015; Brenner et al., 2017).
20
21
Chapter 3 : Calibration of multispectral and thermal infrared
cameras on-board UAV
3.1 Introduction
This chapter first discusses the overview of the existing camera calibration methods for the
use of UAV systems. Then it describes the proposed calibration methods and experimental
results conducted in this research in the following sub-chapters. The geometric calibration is
performed to calculate camera orientations and distortion parameters. Reflectance and
temperature parameters are obtained in radiometric calibrations in order to improve the quality
of the multispectral (MS) and thermal infrared (TIR) images acquired from the UAV. In
addition, as a post-processing step, the spatial co-registration method (simply, called band to
band registration) is introduced.
Recently, UAV platforms become popular in scientific image acquisition with various
sensors such as MS, hyperspectral (HS) and TIR cameras due to their large potentials in
agricultural applications such as orchard mapping (Johnson, 2003), quantitative monitoring of
agricultural crop (Lelong et al., 2008), rangeland monitoring (Rango, 2009), vegetation vigour
monitoring (Berni et al., 2009a; Dunford et al., 2009) and estimating leaf specific-pigment e.g.,
leaf carotenoid content (Zarco-Tejada et al., 2013).
In the case of MS and TIR sensors, non-metric digital cameras are generally used in UAV
payload systems and are widely used in photogrammetric applications. These digital cameras
require geometric and radiometric calibrations for determining their orientations and improving
the quality of their imagery (Douterloigne et al., 2009; Pérez et al., 2012; Kedzierski and
Wierzbicki, 2015). Although the use of multiple sensors at the same time increases the remote
sensing capability of UAV systems, the integration of multi-sensors provides challenges in the
spatial co-registration of multiple bands and different image sensors.
Since the camera calibration is one of the main topics in remote sensing (Hastedt and
Luhmann, 2015), many studies have introduced various camera calibration techniques over the
past years. With regards to the method of camera calibration, the aspects of the complex but
22
precise calibration were discussed (Mason et al., 1997), while simple calibration techniques
were introduced (Karras and Mavrommati, 2002). Moreover, several methods for the camera
calibration using pattern recognition techniques have been addressed (Douskos et al., 2007;
Grammatikopoulos et al., 2007; Honkavaara et al., 2006; Remondino and Fraser, 2006; Zhang
et al., 2010; Zhang, 2000).
Geometric calibration
In general, low-cost digital cameras have serious issues of geometric instability and limited
accuracy. Therefore, the geometric calibration is required to improve the quality of acquired
images.
The intrinsic property of a camera is described by several parameters: focal length (f);
principal point (xp, yp), pixel size (sx, sy); lens distortion coefficients (in general, represented as
the term of ki and pi) and others (e.g., affinity) (Luhmann, 2006). The principal point represents
the centre of the projection of the image (Clarke et al., 1998) and the lens distortion is defined
by radial distortion ( ki) and tangential distortion (pi) (Williams, 2005). Radial lens distortion is
presented in the form of barrel distortion while tangential distortion is very small in magnitude
(Fraser, 1997).
Radiometric calibration
The radiometric calibration is based on the radiometric behaviour of the captured imagery in
the different regions of the spectrum in which data has been recorded (Del Pozo et al., 2014).
This behaviour relies on the characteristics of the camera sensors as well as the weather
conditions (Biggar et al., 2003). An empirical line approach has been introduced and a vicarious
calibration model based on the empirical approach has been proposed, comparing magnitudes to
field-based measurements (López et al., 2011; Moran et al., 2001). As laboratory calibration is
not always valid under the field conditions such as illumination change, the vicarious
calibration model that takes into account the flight conditions by using natural surfaces or
artificial targets is broadly used (Honkavaara et al., 2009). As the result of the calibration,
23
physical quantities (or, reflectance) can be identified in units of radiance (Wm−2sr−1nm−1)
(or, %) for all pixels.
Mosaicking and co-registration
To generate a geo-referenced orthoimage from multi-sensors such as various MS and TIR
images, a co-registration or mosaicking process is often required as in the post-processing.
Mosaicking is conducted to merge consecutive images to produce a large single orthomosaic.
Co-registration is to match images from different sensors in pixels or in geo-referenced
coordinates. The co-registration methods have been employed to combine MS and TIR imagery
taken from UAVs (Bendig et al., 2012; Berni et al., 2009a; Bryson et al., 2013). However, there
is still a lack of discussion related to the co-registration accuracy in their experiments.
There are also various methods for mosaicking UAV imagery (Laliberte, 2008; Laliberte et
al., 2010; Mitch et al., 2010; Turner et al., 2014b; Turner et al., 2012). Nowadays, UAV-
oriented photogrammetric software packages provide an automatic mosaic generation from
UAV imagery, such as PhotoScan (Agisoft LLC, Russia) and Pix4D (Pix4D SA, Switzerland).
Apart from commercial software, some researchers have developed the mosaicking software
and utilized it in the research (Turner et al., 2014a) using the existing image-based matching
algorithms such as the Scale Invariant Feature Transform (SIFT) (Lowe, 1999) matching. To
reduce false matches which cause alignment errors, a Random Sample Consensus (RANSAC)
algorithm (Fischer and Bolles, 1981) is widely used.
In this chapter, MS and TIR camera calibration methods and their test results using imagery
acquired in UAV experiments are briefly described as a preliminary step for the following core
chapters (Chapters 4, 5 and 6). The aim of these experiments is to eliminate errors contained in
imagery and to increase the accuracy and quality of MS and TIR imagery. The sub-objectives
are: 1) to calculate camera internal orientation and distortion parameters using geometric
calibration; 2) to obtain reflectance in each band of MS imagery using empirical radiometric
calibration; 3) to co-register MS bands using spatial transformation method; and 4) to compute
calibrated temperature per pixel from TIR imagery. Based on these objectives, the work-flow is
designed as shown in Figure 3.1.
24
Figure 3.1 Work flow of calibration process for UAV-borne imagery
3.2 UAV field campaign
3.2.1 Study site
The experiment was conducted on 18th of October 2014 at Mace wheat site (80 m × 170 m)
at the University of Melbourne Dookie campus, Victoria, Australia. Mace wheat variety was
applied in three treatments: variable nitrogen, control and diseased plots with three repetitions
of each treatment. The nitrogen plots were applied by a standard level of nitrogen dose (100 kg
N/ha) with the zero nitrogen level plots acting as a control for the trial. The disease plots were
inoculated naturally; all other plots were sprayed with fungicide to avoid infection. However,
natural disease did not appear at the time.
3.2.2 Data acquisition
3.2.2.1 Aerial imagery acquisition
A multispectral (Mini MCA6, Tetracam) camera and a thermal infrared (A65, FLIR) camera
were carried by the UAV platform (Figure 3.2 and Figure 3.3). Three flights were conducted at
25
midday with a cloudless cover to minimise shadows and illumination changes between flights.
The UAV flew at a speed of 2 ms-1 at an altitude of 90 m above ground level (AGL) with a
sufficient percentage of forward and side image overlap (60 – 80 %) in order for feature
matching and mosaicking in post processing.
Figure 3.2 Multispectral camera (Mini MCA6, Tetracam)
Figure 3.3 Thermal camera (A65, FLIR)
The MS camera consists of six discrete bands which capture the reflectance at blue, green,
red and near-IR spectral bands as shown in Table 3.2. The spatial resolution is 1280 × 1024
pixels with a focal length of 9.6 mm. The specifications of MS camera are presented in Table
3.2. The TIR camera captures temperature with spectral range of 7.5 – 13 µm, a spatial
resolution of 640 × 512 pixels, a focal length of 9 mm, and a field of view of 69 ° (H) × 56 °
(V) as summarised in Table 3.3.
26
Table 3.1 Spectral bands of MS camera
Channel Filter(nm) Band width
0 530 10
1 470 10
2 570 10
3 650 10
4 760 10
5 850 10
Table 3.2 Specifications of MS camera
Parameter Value
Num. of Channels 6
Weight 0.7 kg
Image resolution 1280 × 1024
Radiometric resolution 10 bits
Speed 1.3 fps
Pixel size 5.2 μm
Focal length 9.6 mm
Table 3.3 Specifications of TIR camera
Parameter Value
Object temperature range -40 °C to 160 °C
Focal Plane Array Uncooled VOX microbolometer
Weight 0.2 kg
Image resolution 640 × 512
FOV 25 °(H) × 20 °(V)
Image frequency 9 Hz
Pixel size 17 μm
Focal length 25 mm
27
3.2.2.2 Ground-based measurements
Ground data were collected as close as possible to the time of aerial data acquisition and
were used to correlate spectral measurement to biophysical parameters. Field spectral
measurements on calibration targets were collected using a spectroradiometer (Fieldspec, ASD
Inc.). Radiometric calibration was conducted to retrieve the reflectance from radiance at sensor
in each camera. This calibration process makes it possible to calculate the precise vegetation
indices based on reflectance values. For the radiometric calibration, three reflectance targets in:
white, gray and black (Figure 3.4), water body, and vegetation were used for an empirical line
calibration. The empirical line calibration method was used to fit multispectral digital number
(DN, pixel value) value to field level reflectance. The measurements for each target were
carried out at least three times to apply the averaged data to fit DN to reflectance. Figure 3.5
shows the field design for the deployment of reflectance targets and vegetation sampling points
(in red dot), measured by the spectroradiometer.
Figure 3.4 Reflectance targets
28
Figure 3.5 Location of reflectance targets and vegetation sampling points (in red dot)
Targets of Ground Control Points (GCPs) were designed for image spatial resolution as well
as visibility of all the bands of MS camera and TIR camera. In particular, the target material is
made of aluminium sheet which can be detected as a dark target in the TIR image due to low
emissivity characteristics. Figure 3.6 shows the design of GCP target.
Figure 3.6 Design of ground control target (GCP)
29
The location and the number of GCPs in the test field were chosen according to the field of
view (FOV) of MS and TIR camera. Total 24 GCPs were installed in the study site and
surveyed using a Differential GPS (DGPS) with 3 cm standard deviation error. Figure 3.7
shows the deployment of 24 GCPs and surveying results.
Figure 3.7 Distributions of ground control points
Since the test site was a homogenously vegetated field covered by wheat, there were not
many distinctive features for the automatic matching procedure of the bundle adjustment.
Therefore, two types of artificial features, coated by aluminium foil, were designed and
randomly distributed at the site to assist the matching (Figure 3.8).
30
Figure 3.8 Design of ground feature target
3.3 Camera lens calibration
Calibration of each camera of MS sensor was conducted to remove the lens distortion as
well as to calculate focal length and principal point of each camera. A chess-board target (750 ×
330 mm) was designed and captured by each camera with 20 different roll and pitch angles to
produce convergent images. An open source tool, Camera Calibration Toolbox (Matlab,
Mathworks Inc., Matick, MA. USA), was adopted to calculate internal orientation and distortion
parameters. In Figure 3.9, target images of the master camera (band 1) are shown which were
re-projected by the calibration result.
Figure 3.9 Reprojected calibration target
Table 3.4 shows the calibrated parameters of MS sensor with 6 bands. Radial distortion
coefficients were chosen up to the second order to be able to use the third degree of polynomial
equation. The degree of lens distortion was the highest in band 4, while the distortion of band 2
was the least. Figure 3.10 illustrates the radial distortion map of band 2 camera.
31
Table 3.4 The results of calibration parameters
Camera k1 k2 p1 p2 f x y
Band 1 -0.09941 -0.09852 -0.00611 0.00113 1938.57 741.36 461.84
Band 2 -0.05935 -0.03630 -0.00634 0.01392 1895.72 726.25 447.85
Band 3 -0.06930 -0.26784 -0.0116 0.01404 1952.36 776.47 442.32
Band 4 -0.18046 0.43793 -0.00579 -0.00752 1908.98 715.76 455.62
Band 5 -0.15228 -0.05768 -0.01205 -0.00041 1899.24 701.08 459.22
Band 6 -0.16713 0.07478 -0.0097 0.00313 1931.01 699.78 470.84
Figure 3.10 Radial distortion map of band 2
3.4 Radiometric calibration
For the radiometric calibration, various features such as calibration targets, water body and
natural vegetation surface, were used and their radiometric values were measured by a
spectroradiometer. The DN values of each image were obtained in accordance with the same
locations to the filed measurements. The three graphs in Figure 3.11 show spectral signatures of
calibration targets, which were acquired in the field measurement and used as inputs of
radiometric calibration. The black and gray targets show consistent reflectance values with 0.06
32
and 0.33 over the spectral wavelength of the MS sensor (400 nm – 900 nm). The reflectance of
the white target was fluctuated in two green bands and the reflectance value was saturated.
Thus, the white target was excluded in the radiometric calibration.
Figure 3.11 Spectral signatures of calibration targets in black, gray and white
An empirical line calibration using linear regression model was carried out to calculate the
first order coefficients to convert DN to reflectance at each sensor. Figure 3.12 and Figure 3.13
show the results of linear regression between DN and reflectance of band 4 (650 nm, red) and
33
band 6 (850 nm, NIR). Those red and NIR bands are commonly used to calculate vegetation
indices (e.g., NDVI), and should be calibrated to obtain accurate indices.
Figure 3.12 Linear regression result of band 4 (650 nm)
Figure 3.13 Linear regression result of band 6 (850 nm)
3.5 Band co-registration
In post processing, band to band registration was conducted, since the MS camera sensor
consists of six different sensors which are placed in different locations of the camera’s main
body. Band to band registration is a procedure that aligns all six sensors on a plane, resulting in
position-aligned bands for colour composition map or thematic map. In this experiment, band 1
was set to the master band as a reference plane. Other bands were assigned to slave bands, then
34
all slave bands were aligned to the master. To align all the sensors, the method of affine
transformation was used and parameters (scale, rotation and translation) were considered.
Firstly, scale and rotation factor between the master and the slave sensors were measured. Then,
images corrected by scale and rotation (namely, scale- and rotation-free images) were produced
using an image processing software (PixelWrench2, Tetracam). In the next step, each x and y
translation factors between the master and the slave cameras were measured. Table 3.5 shows
the results of band to band alignment parameters.
Table 3.5 Band alignment parameters
Band no. Rotation Scaling Trans X Trans Y
Master 1 - - - -
Slave 2 -0.325 1.007772 -6 -5
Slave 3 -0.988 1 0 -1
Slave 4 -0.834 1.001717 -2 0
Slave 5 -0.065 1.001717 -2 -2
Slave 6 -0.415 1.005168 -4 -5
After co-resignation, master and slave images were integrated into multi-page TIFF format
using PixelWrench2. True colour composition images were produced as a result of band to band
registration, showing that the edges of objects (e.g., black and white target) in the image were
correctly matched (Figure 3.14).
Figure 3.14 Band to band registration: left) original image; right) registered image
35
3.6 Temperature retrieval
Temperature calibration of TIR images using the measured temperature of the ground target
was carried out to retrieve temperatures of surface. TIR images were obtained in 14-bit raw
format of signal-based values emitted from the objects. The raw TIR images were processed
with the customized code in Matlab R2014b (Mathworks Inc.) and converted to 14-bit
temperature-based data using the manufacturer (FLIR Systems) register values. To retrieve
accurate temperatures from the raw signal data, the temperature values of calibration targets
such as water body, bare soil, and vegetation were measured with a hand-held infrared
thermometer concurrently with the UAV operation over the experimental site. Equation 3.1 and
Equation 3.2 are the formulas of one-point calibration for retrieving the adjusted surface
temperature provided by FLIR systems (FLIRSystems, 2013).
𝑂𝑎𝑑𝑗𝑢𝑠𝑡𝑒𝑑 = 𝑆 −𝑅
𝑒(𝐵/𝑇𝑘𝑛𝑜𝑤𝑛) − 1 (3.1)
𝑇𝑎𝑑𝑗𝑢𝑠𝑡𝑒𝑑 (𝑖𝑛 𝐾𝑒𝑙𝑣𝑖𝑛) =𝐵
log(𝑅
𝑆− 𝑂𝑎𝑑𝑗𝑢𝑠𝑡𝑒𝑑+1)
(3.2)
where S is a 14-bit raw signal value, T an object temperature, R a constant for converting
flux to temperature derived from Planck’s constant h, B a constant derived from Boltzmann’s
constant and Planck’s constant h, O an offset (signal to radiance). The temperature of the
calibration target was used to calculate the adjusted offset parameter (𝑂𝑎𝑑𝑗𝑢𝑠𝑡𝑒𝑑), and then
adjusted temperature was obtained.
As a result, a raw TIR image and a temperature-retrieved image are presented in Figure 3.15
and Figure 3.16, respectively.
36
Figure 3.15 Raw signal TIR image
Figure 3.16 Temperature-retrieved TIR image
3.7 Conclusion
This chapter described the technical methods of image calibrations and their test results. The
calibration experiment was conducted by UAV-based imagery and field-based measurements.
Various targets (e.g., chess-board, reflectance target, GCP, GAF and temperature target) were
designed and used to conduct the calibration. Fieldworks such as GPS survey, GCP installation,
targets deployment, spectral measurements were carried out to accommodate with camera
specifications and UAV flight plan. After acquiring UAV-borne images, post-processing was
37
carried out to calibrate the images in geometric and radiometric aspects. The empirical line
method was applied to accomplish the radiometric calibration. Radiometric coefficients were
estimated using relationships between spectrum measurements of reflectance targets (and
natural feature targets) and corresponding DN values of images. Throughout the radiometric
calibration process, it allowed to extract quantitative data from the imagery acquired from UAV
with the determination of the radiometric calibration parameters. Some challenges, however,
appeared in radiometric calibration process. The white calibration target was saturated up to
maximum DN value due to high reflective characteristic.
In another post-processing, band co-registration was conducted in order to eliminate the
band misalignment. As a result, single mosaic image with multi-layers was produced in multi
TIFF. Nevertheless, there exists a limitation related to the presented band registration method.
Since the method is based on 2D planar geometry, the identified parameters are more or less
affected by the camera geometry in 3D space. In other words, when UAV-borne images are
taken with significant variations of flight height or tilted angle, parameters for the band
registration should be re-calibrated. As a future work, co-registration methods to accommodate
3D geometry (e.g., bore-sight calibration) will be investigated in order to calculate invariant
parameters under any circumstance.
Lastly, since the research was conducted by numerous series of images from multiple
sensors installed on UAVs, all imagery data required significant pre- and post- processing
procedures. In particular, TIR images often changed with different temperature ranges due to
histo-adjustment of brightness of image when captured. Thus, manual image-by-image
corrections were often required in the research. A feature-matching-based method is considered
as a future work to adjust the temperature changes of consecutive TIR images systematically.
38
39
Chapter 4 : Adaptive estimation of crop water stress in
nectarine and peach orchards using high-resolution
imagery from an unmanned aerial vehicle (UAV)
4.1 Introduction
According to projected climatic changes, agricultural drought periods over Australia
can increase up to 20 % by 2030 (Glover et al., 2008). Climate change effects on soil water
balance and plant transpiration can impact crop productivity and quality significantly. Both
quality and yield of crops can be enhanced by adequate and timely irrigation based on real-
time monitoring of water status without increasing the level of agricultural water used.
Such improvements in water use efficiency include changing irrigation frequency to match
crop water requirement that maximizes yield and quality. As an efficient indicator of crop
water status, crop water stress index (CWSI) has been introduced based on the difference
between foliage and air temperature (Idso et al., 1981; Jackson et al., 1981). Jones (1992)
proposes a reformulated CWSI by the difference between canopy temperature (Tcanopy) to
the reference temperature or threshold temperature at full transpiration (Twet), which is
normalised by the temperature difference between the non-transpiring leaf temperature
(Tdry) and Twet as the upper and lower reference temperatures, respectively. Consequently,
estimations of CWSI require the measurement of canopy temperature targeting dominant
foliage, which conventionally relies on manual or continuous point measurements.
However, the ground-based handheld thermography approach is labour-intensive, costly
and impractical when monitoring large areas, furthermore agricultural fields commonly
feature spatially heterogeneous biophysical conditions and spatial variation exists in water
stress levels between crops/trees and even within a plant (Moran et al., 1994).
Since numerous data from various remote sensing platforms have become available,
several studies of crop water stress have been conducted using remote sensing images as a
major input, aiming to simplify the calculation of CWSI, to replace ground-based
measurements, and to make it applicable to larger areas (Leinonen and Jones, 2004;
Alchanatis et al., 2010; Fuentes et al., 2012). Furthermore, research on water stress
detection with unmanned aerial vehicles (UAV) has been recently conducted, providing a
40
platform to monitor water status at field scale with higher spatial resolution (Berni et al.,
2009b; Zarco-Tejada et al., 2012; Bellvert et al., 2014). In addition, research on combining
remote sensing data and ground-based data in vineyards has been also carried out to predict
water status at temporal and spatial basis (Acevedo-Opazo et al., 2010; Acevedo-Opazo et
al., 2013).
Many remote-sensing CWSI approaches have focused on an image-based analysis from
near-ground or proximal platforms, assessing plant water status at the crop level. These
approaches use both visible and thermal infrared images to obtain the level of crop water
stress as a quantitative index, where the image co-registration method is applied to
delineate crop canopy pixels and to obtain crop temperature separately from soil
temperature (Möller et al., 2007). The canopy-related temperature is derived from the
thermal image, in which non-canopy temperature such as soil is masked out by the co-
registered visible image based on colour segmentation. Despite the advantages in obtaining
canopy pixels, the co-registration is often time-consuming. In order to eliminate the co-
registration process and consequently to facilitate the estimation of stress index, a
histogram approach using only thermal images has been proposed. In the approach, an
individual histogram of each thermal image is analysed to distinguish between canopy-
related pixels and non-canopy pixels by empirical or statistical methods (Meron et al.,
2010b; Rud et al., 2014). However, a fundamental problem associated with the histogram-
based clustering of canopy pixels and subsequent calculation of canopy temperatures is the
presence of mixed pixels that represent partial coverages of both canopy and soil. Due to
the nature of raster imagery, mixed pixels of canopy and non-canopy background such as
soil exist along their boundaries. The mixed pixels of thermal imagery can cause significant
bias in canopy temperature during midday when they are included in the clusters of canopy
pixels. Thus, the mixed pixels should be excluded to reduce errors in the histogram-based
analysis when determining canopy-related temperatures.
In addition to the delineation of canopy pixels, determination of Twet and Tdry for CWSI
poses critical influence on the accuracy of CWSI values as an indicator of crop water
stress. Determination of Twet can be made by several methods such as (1) non water stress
41
baseline (NWSB) by a linear regression function between Tcanopy – air temperature (Tair)
and vapour pressure deficit (VPD) (Jackson et al., 1981), (2) wet artificial reference surface
(WARS), acting as crops in abundant water status regarded as the temperature of a fully
transpiring crop (Meron et al., 2013), and (3) a statistical method based on histogram
analysis of canopy temperature (Rud et al., 2014; Poblete-Echeverría et al., 2016). Tdry can
be derived from (1) ancillary meteorological variables based on isothermal radiation
(Möller et al., 2007; Jones, 1999), (2) dry reference leaf coated with petroleum jelly
(Vaseline) to prevent transpiration (Jones, 1999), (3) an empirical method by adding a fixed
threshold (such as 5 °C ~ 7 °C) to Tair (Cohen et al., 2005; Irmak et al., 2000; Rud et al.,
2014; López-López et al., 2011), and (4) a statistical method based on histogram analysis
of canopy temperature. Rud et al. (2014) compared three methods to calculate Twet – from
WARS, a method based on energy balance, and a statistical histogram approach – and
showed that all CWSI values have similar correlations with actual reference data. Since
CWSI based on statistical analysis does not require either any meteorological data for
energy balance method or equipment for WARS, it reduces the time related to field work
and the CWSI calculation complexity. The statistical histogram approach however would
provide a valid range of CWSI values only when representative wet and dry references are
available in the scene. For example, if all the crops are under either appropriate water
supply or water stressed, the statistical approach alone can yield a biased result in the
CWSI calculation. Although the method proposed here is not applicable to an area of
similar levels of water stress (uniform temperatures), it still remains a practical method to
estimate CWSI across an area where canopies representing different water stress levels
exist. In addition, the limitations associated with the representative reference temperature
could be alleviated using a minimum number of ground truth data (i.e., stem water potential
or stomatal conductance). Since the biased results are shifted and scaled from true results,
the general water stress model can be adjusted by a simple relationship between CWSI and
the ground truth data.
Most estimation of the upper and lower reference temperatures employs single values of
Twet and Tdry as a reference. However, the reference value of Twet and Tdry can change with
field circumstances. In the case of temperate tree crops, orchards may contain several crop
42
cultivars and canopy training systems even under the same irrigation regime. Therefore,
applying a single set of Twet and Tdry could result in inaccurate and misleading CWSI which
do not reflect the actual water status, particularly in the orchards with a combination of
multi cultivars and different canopy structures.
Thus, this study proposes a new method to estimate CWSI. The main goal is to develop
an automated method to interpret very high resolution data to obtain water stress indices at
field scale by using UAV-borne sensing at near real-time window, for early detection of
water stress in crops. This method can be used to guide optimal irrigation management and
to regulate crop quality and yield. The specific objectives are: 1) to develop an adaptive
methodology to obtain Twet and Tdry based on feature extraction and probability modelling.
The proposed method does not require any artificial references or meteorological data to
determine the reference baselines, since it is based on classification of sub-regions which
have the same property in an entire field, employing automated thresholding of Twet and
Tdry for each sub-region and; 2) to provide a practical method which can show realistic
representation of the water stress map of plants at high-resolution, utilizing a small number
of ground truth biophysical measurements.
4.2 Materials and Methods
4.2.1 Study site description and experimental design
4.2.1.1 Site description
The site is a nectarine/peach orchard (0.97 ha, 138 m × 70 m), located near Tatura,
Victoria, Australia (36°26'08"S, 145°16'13"E, 114 m AMSL) and administrated by the
Department of Economic Development, Jobs, Transport and Resources (DEDJTR). The
region has a temperate climate with average annual rainfall of approximately 480 mm.
Annual average reference crop evapotranspiration is approximately 1190 mm (Allen et al.,
1998). A nectarine (Prunus persica (L.) Batsch cv. Autumn Bright) and a peach (Prunus
persica (L.) Batsch cv. August Flame) were planted in the winter of 2013. The experiment
was carried out on the 22nd of February 2015. The nectarine and peach were at full leaf-up
and late maturity phenological stage (i.e. stage 3 of fruit growth). Both cultivars are spaced
43
at inter-row distance of 4.5 m and 1.0 m between trees with rows orientated in a north –
south direction on a fine sandy loamy Shepparton soil. Trees are trained to two canopy
systems: vertical leader and Tatura trellis (Figure 4.1). For the Tatura trellis trees, the
canopy dimension of each leader was approximately 1.85 m high and 0.5 m wide. For
vertical leader trees, canopy dimensions were approximately 2.0 m high and 0.8 m wide.
(a)
(b)
Figure 4.1 Canopy systems of the nectarine/peach trees in the study site: (a) Vertical
leader; (b) Tatura Trellis
44
(a)
(b)
Figure 4.2 (a) Location of the study site, Stonefruit Field Laboratory orchard (Tatura,
Victoria. Australia); (b) Water deficit plots are presented in yellow, and control plots
in blue. T1-T8 represents each irrigation treatment and red dots represent the
biophysical sampled tree for each plot. Ground control point (GCP) targets in a black
square with a white cross.
4.2.1.2 Experiment design
The study site contained eight experimental plots (Figure 4.2), where control and deficit
treatments were applied as irrigation regime (Table 4.1). A drip irrigation system was used
along the tree lines spaced at every 0.5 m with an emitter flow rate of 1.6 Lh-1. Irrigation
amount was based on a crop water use (crop evapotranspiration, ETc) model (Allen et al.,
1998) in four control plots and the remainder of orchards. Four deficit plots were imposed
45
by withholding irrigation through valves for five days prior to the UAV sensing and
ground-based measurements. Each plot contained at least 20 trees.
Table 4.1 Descriptions of sampled trees based on cultivar, training system and
irrigation treatment.
Plot Crop Cultivar Tree Training Treatment
T1
Nectarine 1
Vertical Leader Deficit
T2 Vertical Leader Control
T3 Tatura Trellis Deficit
T4 Tatura Trellis Control
T5
Peach 2
Tatura Trellis Deficit
T6 Tatura Trellis Control
T7 Vertical Leader Control
T8 Vertical Leader Deficit
1 Prunus persica (L.) Batsch cv. Autumn Bright, 2 Prunus persica (L.) Batsch cv.
August Flame
4.2.2 Aerial thermal infrared imagery acquisition
4.2.2.1 Aerial images
A multi-rotor unmanned aircraft (S900 manufactured by DJI) was used to carry payload
with a thermal infrared (TIR) camera (A65, FLIR Systems, Inc.) to measure canopy
temperature. The TIR camera was integrated with an on-board computer and GPS for geo-
tagging. The TIR camera captures temperature with spectral range of 7.5 – 13 µm, thermal
sensitivity of < 0.05 °C at +30 °C, a spatial resolution of 640 × 512 pixels, a focal length of
25 mm, and a field of view of 25 ° (H) x 20 ° (V). Images were acquired across the entire
study site at nadir view angle at midday with a cloudless sky to minimize shadows casted
by the crop canopy and to capture the period of peak ETc and likely symptoms of water
stress. The UAV was flying at a speed of 2 ms-1 at an altitude of 90 m above ground level
(AGL) with the footprint of 39.2 m × 31.3 m and ground sample distance (GSD) of 6.12
cm, considering sufficient image overlaps for photogrammetric processing.
46
4.2.2.2 Ground targets
Regular surface patterns of orchards can cause an ill-conditioned image matching
between consecutive images due to either lack of key points (interesting points) or
mismatched points. Thus, two types of ground target, ground control point (GCP) and
ground artificial feature (GAF), were designed for a reliable photogrammetric bundle
adjustment with a large number of consecutive images. In particular, the GCP target was
made of an aluminum sheet with a size of 0.6 x 0.6 m. Since aluminum material has a low
emissivity, the target appears distinctively as a cold object in thermal imagery. In total, 24
GCPs were distributed in the site (Figure 4.2) and surveyed by a differential GPS (DGPS)
with higher than 3 cm positional accuracy. The GAF targets were made of aluminum
coated cardboard with various shapes (triangle, trapezoid, and rectangle), and were
randomly distributed across the site at known locations and configurations in order to
enhance image matching quality in consecutive TIR images.
4.2.3 Physiological data acquisition
4.2.3.1 Leaf temperature and gas exchange measurements
Leaf temperature for the calibration of TIR images was measured on five sunlit leaves
and five shaded leaves per plot (n = 10 leaves × 8 plots = 80 observations), using a
handheld infrared thermometer (TN410LCE, ZyTemp, Radiant Innovation Inc.) within a
40-minute time window during the UAV image acquisition. Stomatal conductance (gs, mol
m-2 sec-1) and transpiration rate (E, mmol m-2 sec-1) were measured on three mature and
fully expanded sunlit leaves, corresponding to the fifth leave from the tip of a main shoot,
per plot (n = 3 leaves × 8 plots = 24 observations) using a portable photosynthesis system
(LI-6400, LI-COR Inc., Lincoln, Nebraska, USA) during midday (solar noon) at the same
time of aerial image acquisition.
4.2.3.2 Stem water potential measurements
Stem water potential (ψstem, MPa) was measured on two fully expanded shaded leaves
per plot (n = 2 leaves × 8 plots = 16 observations) during midday using a Scholander
47
pressure chamber (Model 3000, Soil Moisture Equipment Co., Santa Barbara, CA, USA).
The leaves were selected from branches near the main trunk and were covered with an
aluminum foil bag for a minimum of one hour of equilibration time, prior to the
measurement of ψstem.
4.2.4 Aerial TIR image processing
All TIR images were captured in a 14-bit raw signal format of signal-based values
emitted from the objects. The raw images were converted to 16-bit temperature-based
images by using a customized code written in Matlab R2014b (Mathworks Inc., Matick,
MA. USA). A one-point calibration method (FLIRSystems, 2013) was used to retrieve the
adjusted surface temperature using an averaged canopy temperature of a sampled tree,
which was measured with the handheld infrared thermometer concurrently with the UAV
operation. All consecutive images were processed via aerial image triangulation with the
geo-tagged flight log and the GCPs by using a photogrammetric software (PhotoScan,
Agisoft LLC, Russia). Digital elevation model (DEM) was generated based on the point
cloud which is a set of matching points between overlapping images. Then, a georeferenced
orthomosaic was built using the DEM as the surface parameter in the software.
4.2.5 Feature extraction
(a)
(b)
(c)
(d)
Figure 4.3 (a) Example of the orthomosaic image; (b) extracted edges; (c) dilated
edges; (d) eliminated edges in the orthomosaic image
48
Mapping CWSI requires extraction of canopy pixels to sample canopy temperature
values excluding soil and other non-leaf materials. Inclusion of non-canopy temperature
can result in significant errors in the interpretation of CWSI. However, the extracted
canopy pixels may contain ‘edge pixels’ that feature mixed canopy-soil temperature values.
The edge pixels typically show higher temperature than pure canopy pixels during midday
periods. Tree rows were oriented north-south in the study site, the majority of edges ran in
corresponding direction in the orthomosaic image (Figure 4.3.a). An edge detection method,
combined Sobel and Canny, was applied to exclude ambiguous mixed pixels in the canopy-
soil boundary in Matlab R2014b (Figure 4.3.b). Sobel and Canny are designed to detect the
gradient changes maximally to edges in vertical and horizontal directions (Raman and
Himanshu, 2009). Other common existing edge methods such as Prewitt and Roberts can
be also applicable, however the combined Sobel and Canny outperformed the other options
in detecting tree boundaries along the rows in the orthomosaic image. The detected edges
were then dilated for more conservative exclusion of the mixed pixels, which were
distributed up to six pixels across the edges, along the direction of thermal gradient (Figure
4.3.c). Finally, the pixels of edges were masked out from the original orthomosaic image
(Figure 4.3.d).
4.2.6 Adaptive Crop Water Stress Index (CWSI)
As mentioned above, orchards may have a number of cultivars and tree training systems
that have an impact on thermal responses. Applying a single set of Twet and Tdry over the
entire area may result in non-representative water stress indices due to an inaccurate
normalised span of CWSI. For instance, the range of CWSI can be shifted, shrunk, or
extended resulting in an inconsistent indicator of the actual crop water status. In this study,
an adaptive scheme that utilizes multiple sets of reference Twet and Tdry for heterogeneous
field conditions is proposed. The proposed method is based on dividing the entire field into
sub-regions which can be classified with the same properties (Table 4.2).
49
Table 4.2 Classified sub-regions within the studied field based on crop cultivar and
tree training system
Sub-region Crop cultivar Tree Training
1. N_VL Nectarine Autumn Bright (N) Vertical Leader (VL)
2. N_TT Nectarine Autumn Bright (N) Tatura Trellis (TT)
3. P_TT Peach August Flame (P) Tatura Trellis (TT)
4. P_VL Peach August Flame (P) Vertical Leader (VL)
4.2.6.1 CWSI calculation
The CWSI algorithm applied in the study was suggested by Jones (1992), which can be
represented as follows:
𝐶𝑊𝑆𝐼 = 𝑇𝑐𝑎𝑛𝑜𝑝𝑦 − 𝑇𝑤𝑒𝑡
𝑇𝑑𝑟𝑦 − 𝑇𝑤𝑒𝑡
(4.1)
where Tcanopy is canopy temperature from the aerial TIR image, Twet the temperature of a
fully transpiring leaf or lower reference, and Tdry is the temperature of a non-transpiring
leaf, also considered as upper reference.
4.2.6.2 Adaptive 𝑻𝒘𝒆𝒕 and 𝑻𝒅𝒓𝒚
A specific statistical analysis is proposed to estimate adaptive thresholds of Twet and Tdry
based on sub-regions (Figure 4.4). Firstly, a temperature histogram is generated from TIR
image subset of each sub-region. This study assumes that Twet can be taken from the coldest
part of the histogram from TIR image (Rud et al., 2014) and Tdry is also the temperature of
a non-transpiring leaf, which can be derived from the highest part of the histogram.
50
Figure 4.4 Flow chart of the decision of adaptive reference temperatures based on
sub-regions (SR)
The histograms feature distinctive bimodal and normal density distributions,
representing vegetated pixels and soil pixels. The temperature differences of two features
are obvious, corresponding to a range up to 30 – 35 °C at midday. The Gaussian mixture
modelling (GMM) is fit to the temperature distribution to objectively cluster canopy/soil
pixels and to estimate representative Twet and Tdry values for canopy. The GMM assigns
each observation to a cluster by maximizing the posterior probability, and composes of k
multivariate normal density components, where k is the number of clusters (e.g., k = 2 for
bimodal distribution). The higher-temperature component of Gaussian distribution,
representing non-canopy pixels is eliminated to exclude soil background effects in the
Gaussian distribution. Then, an automated threshold of Twet and Tdry for each sub-region is
determined by the critical values of 99 % confidence interval limits from the lower-
temperature component of Gaussian distribution.
51
4.3 Results
4.3.1 Canopy temperatures from sub-regions
The entire study site, except for the four water deficit plots (Figure 4.2) was submitted
to the same irrigation level based on daily water needs calculated using meteorological
data. Canopy temperature from each sub-region, however, was characterized by two major
features: cultivar type and canopy structure determined by the tree training system.Figure
4.5.b–e show the distributions of surface temperature (mainly canopy and soil) from each
sub-region. The histograms featured distinctive bimodal distribution, representing
vegetation (23 °C – 38 °C) and soil background (38 °C – 57 °C) due to apparent
temperature differences up to 34 °C (Figure 4.5.a). The temperature distribution of soil
represents a mixture of dry grass and bare soil, whereas that of vegetation shows actively
growing and transpiring canopies.
Due to the bimodal distribution, the GMM technique with two univariate components
was fitted to the histograms, featuring different mean and standard deviation for each sub-
region. The outputs of GMM fit to canopy temperatures are presented in Table 4.3. Both
canopy temperatures of N_TT and P_TT were distributed with relatively high standard
deviations (SDs) of 2.5 °C and 2.9 °C, respectively, whereas the SDs of N_VL and P_VL
were in the range of 1.9 °C and 1.5 °C, respectively. The canopy structure of TT is in Y-
shape and wider than VL. Consequently, the structure of TT causes the canopy
temperatures to spread wider due to greater differences in the leaf energy balance from
different vertical positions and lower leaf area density within the canopies. Another
interesting feature was the lower mean value of P_VL (27.7 °C), compared to the mean
values of all the other sub-regions (approximately 30 °C) with a narrower spread in the
temperature distribution. The different mean and SD values for the histograms indicate that
the distribution of canopy temperatures can vary between different cultivar and tree
architecture even under similar irrigation levels. Thus, a concept of adaptive reference
baselines determined by region-specific probability modelling is proposed in this study to
standardize the different temperature features for CWSI based on the cultivars and the tree
architecture types.
52
(a)
(b)
(c)
(d)
(e)
(f)
Figure 4.5 (a) The orthomosaic image of surface temperature (°C) derived from
thermal imagery using UAV and classified sub-regions based on cultivars and
canopy architectures; (b) Mixture modelling in the entire site; (c) in the Nectarine
trained to Vertical Leader
53
Table 4.3 Thermal values of canopy temperature using probability modelling
Sub-region p 1 =0.5
(°C) Mean (°C)
SD 2
(°C) 99 % CI 3 (°C)
Entire 37.93 29.64 2.45 23.31 35.97
1_N_VL 37.87 30.17 1.93 25.20 35.13
2_N_TT 38.26 30.46 2.51 23.98 36.94
3_P_TT 39.57 30.47 2.94 22.89 38.05
4_P_VL 34.19 27.72 1.45 23.98 31.46
1 probability, 2 standard deviation, 3 confidence interval
4.3.2 Mapping adaptive CWSI
Figure 4.6.a shows the estimated CWSI map based on adaptive of Twet and Tdry, showing
the water status variability over the orchard. In Figure 4.6.b, details of water deficit and
control treatment groups in Nectarine VL are presented. The CWSI values estimated in all
four water stressed groups were significantly higher (CWSI = 0.79) than CWSI in the
control groups (CWSI = approx. 0.37). Figure 4.6.c shows the magnified pixel values of
CWSI (6.12 cm GSD).
54
(a) (b) (c)
Figure 4.6 (a) Adaptive CWSI map derived from UAV remote sensing using thermal
infrared images. Four dotted rectangles in red represent the area of deficit irrigation
treatments and four dotted rectangles in blue the full irrigation of control treatments.;
(b) Example of CWSI map depicting water deficit and control treatment groups for
Nectarine VL (Vertical Leader); (c) Examples of the pixel-level resolution of CWSI
for T1 (in deficit) and T2 (in control) sampled trees
4.3.3 Validation of adaptive CWSI
Midday ψstem and gs measurements were used to provide the relationships with the
adaptive CWSI. In addition, a single Twet and Tdry as used in conventional methods were
derived from the VPD formula (Paltineanu et al., 2013; Roy and Ophori, 2014) in order to
compare them with the proposed adaptive method. The GAF targets were placed next to the
eight sampled trees, where ψstem and gs were measured, in order to identify the tree
locations in orthomosaic image and CWSI map. CWSI values of the sampled trees were
55
extracted, aggregated, and averaged from the CWSI map. The ψstem measurements in the
sampled trees showed good negative relationship with the adaptive CWSI (Figure 4.7.b).
Specifically the relationship was significant with determination coefficients (R2) of 0.72.
Similarly, a strong relationship was found between gs measurements and the adaptive
CWSI with R2 of 0.82 (Figure 4.8.b). The CWSI estimated using the single reference
temperature (baseline), however, exhibited a weaker correlation between ψstem and gs
(Figure 4.7.a and Figure 4.8.a). In particular, the T8 tree showed the largest bias among the
sampled trees in the single reference CWSI method. The CWSI value sampled from the T8
was calculated as 0.29, which represents no-water-stress condition, although T8 was in fact
in the water deficit treatment group (of P_VL) and the ground ψstem measurement of T8
indicated water deficit condition (−2.91 MPa). Overall, the average canopy temperature of
P_VL was lower than the values in the other treatment groups. When examined separately,
the mean temperature of T8 (28.5 °C) was higher than the mean value of T7 (26.0 °C) from
the control plot in the same crop canopy system (P_VL). However, the mean canopy
temperature of T8 was in a similar range with those of control-group trees in the other sub-
regions. The results imply that the reference temperature values for CWSI need to be
adjusted with plant cultivars and/or tree training system in order to make CWSI a general
predictor of the plant water stress (i.e., SWP) across different plants and management
types. In other words, the proposed adaptive thresholds of Twet and Tdry would be required
for the horticultural fields that contain multiple plant cultivars and tree training systems.
(a)
(b)
Figure 4.7 (a) Relationship between stem water potential (ψstem) and CWSI of single
reference baseline; (b) relationship between ψstem and adaptive CWSI
56
(a)
(b)
Figure 4.8 (a) Relationships between stomatal conductance (gs) and CWSI using the
single reference baseline method; (b) relationship between gs and CWSI calculated
using the adaptive method
Figure 4.9 Adaptive stem water potential map (ψstem; MPa)
57
4.4 Discussion
TIR imagery can be used to extend a small number of the ground measurements to
spatially distributed measure of crop water status at the field scale effectively. However,
delineating plant canopy pixels that are not influenced by the soil background and sampling
reference temperature values automatically is not a trivial task. In this work, an edge
detection and elimination approach is employed to exclude ambiguous mixture pixels, and
the Gaussian mixture model (GMM) is applied to automatically sample reference canopy
temperature. The lower-temperature GMM parameters are then used to estimate the
reference (cold and hot) temperature values at each sub-region. The adaptive thresholds
proposed are obtained at 0.5 and 99.5 percentiles of temperature distributions from TIR
imagery to specify the Twet and Tdry. According to the GMM results, the variability of
canopy temperatures between different combination of cultivar and the tree architecture
was evident. Characterizing the whole study site with a single set of reference temperature
values leads to biased and non-representative CWSI results in case the field is composed of
multiple combinations of cultivar and tree architecture. The adaptive CWSI proposed can
automatically estimate proxy plant water stress status consistent with the SWP,
standardizing temperature distributions into an integrated span of CWSI.
The proposed method is carried out based on the assumption that there exist a wide
range of water stress levels in the field resulting in representative canopy temperature
values for stressed and non-stressed plants. This assumption can be justified by assuming
that, even under the same irrigation regime, temperature variations caused by different
water availabilities for individual crops could occur by external factors such as soil
properties, terrain elevation, and uneven irrigation management. However, deviation from
the assumed condition can limit the application of the method used in this work.
The proposed method of CWSI poses a few limitations when it is to be applied to all
possible conditions of water status in crop fields. A hypothetical example is when the
distribution of field canopy temperature features a narrow range, indicating that the actual
crop water status of all crops is at a similar level. In such case, the estimated CWSI would
still arbitrarily allocate a range of water stress levels within the narrow range of actual crop
58
water status. This is in fact a fundamental limitation of the CWSI-based method proposed.
However, the drifting CWSI range can be anchored in advance with the help of auxiliary
data such as the air temperature. Tdry can be predicted by Tair + threshold temperature (e.g.,
5 °C) in the empirical method (Cohen et al., 2005; Irmak et al., 2000; Rud et al., 2014;
López-López et al., 2011)). If the estimated Tdry in GMM distribution is lower than Tdry
predicted by Tair, it would indicate non-severe crop water stress. On the contrary, if the
range of canopy temperature distribution is very narrow and close to Tdry predicted by Tair,
it would mean that most crops are under water deficit condition.
In addition, remotely estimated CWSI by itself is not a direct measure of actual crop
water status, since its mean level and dynamic range can vary with plant cultivars and
environmental factors. Nevertheless, the estimated CWSI can be converted to more widely
accepted water stress measures, such as ψstem and gs, when a small number of the
measurements are taken concurrently with the TIR imagery. Moreover, SWP (ψstem) has
been widely accepted as an indicator of crop water status (Scholander et al., 1965). Thus,
many crop growers and physiological researchers have determined water status of crops by
ψstem using the pressure chamber method, particularly for grapevines and stone fruits
(ChonÉ et al., 2001; Naor, 2005). However, even though the SWP-based water stress
estimation is a reliable method, it poses limitations due to resources (labour, equipment)
required and hence limits spatial coverage when applied at field scale. However, the SWP-
based water stress mapping can be extended to cover large areas when it is combined with
the adaptive CWSI map via the CWSI vs. ψstem relationship. The SWP map in Figure 4.9 is
created by utilizing the CWSI vs. ψstem model shown in Figure 4.7.b. The map shows
spatially explicit and detailed map of the SWP as a quantitative indicator for crop water
stress.
This study was carried out with one flight campaign in the late summer before leaf
senescence. A single flight campaign is not sufficient to explain water stress variations at
different crop phenological stages. However, canopy cover of stonefruit crops in terms of
fractional radiation interception or leaf area index (LAI) does not change much after full
leaf-up (late spring) until commencement of leaf senescence (late autumn). The rapid
59
change in cover occurs early in spring, when evaporative demand is low, and the soil water
profile is typically full after winter rainfall in Vitoria, Australia. The likelihood of crop
water stress during the spring period is therefore low, with growers having the option to
irrigate if dry soil conditions prevail. Nevertheless, further experiments are necessary to
explore crop water stress at different phenological stages.
4.5 Conclusions
This study proposed for the first time an adaptive CWSI calculation method using two
main approaches: 1) pure canopy extraction by edge detection and statistical analysis of the
distribution of surface temperatures; and 2) the adaptive and systematic determination of
lower (wet) and upper (dry) references, obtaining the 99 % CI in the distribution of canopy
temperatures in each sub-region. A strong linear relationship between the adaptive CWSI
and the ψstem (and gs) was obtained in this work. The present approach can potentially
provide a feasible method of assessing real water stress of plants with high spatial
resolution to assess effectively its variability at the plant and field scale for precision
irrigation purposes. As a future work, further research on various crop fields and different
crop phenological stages will be investigated to make the present method applicable to
general cases.
Acknowledgments: This research was supported by the Innovation Seed Fund for
Horticulture Development grant from the University of Melbourne and Department of
Economic Development, Jobs, Transport and Resources (DEDJTR), Victoria and ARC
LIEF Grant (LE130100040). The use of the experimental stonefruit orchard as a resource
for the study is acknowledged, with financial support from Horticulture Innovation
Australia and DEDJTR (project SF12003: ‘Increased stone fruit profitability by
consistently meeting market expectations’).
60
61
Chapter 5 : Relationship of CWSI-based plant water stress
estimation with the data acquisition times of the day
5.1 Introduction
The new method of CWSI estimation (Adaptive CWSI) proposed and evaluated in Chapter
4 was implemented in a case study to assess the relationship between CWSI-based plant water
stress and the time of the day for data acquisition.
The diurnal cycle of plant water status is important to understand the plant behavior of water
uptake and to detect water stress sensitivity. Plant physiological parameters such as predawn
ψleaf and midday ψstem are known to detect plant water stress at early stages (Davies and Lakso,
1979; Williams and Araujo, 2002; Bhusal et al., 2018).
CWSI is a reliable indicator of plant water stress for irrigation management practices. In
general, CWSI has been commonly measured (or calculated) at midday, since midday CWSI is
known to be accurate and sensitive to obtain maximum daily plant water stress (Testi et al.,
2008; Bellvert et al., 2014). Thus, the use of UAV sensing has also been limited to be
conducted at midday in previous researches to obtain the highest thermal contrast between
canopy temperature and air temperature, or the minimum shade effects casted by canopy,
although UAV remote sensing becomes a readily usable tool for precise agricultural water
management with high temporal and spatial resolutions. There have been only a few attempts to
investigate the diurnal changes of CWSI including TIR data acquired in the morning using
UAV-based remote sensing (Gonzalez-Dugo et al., 2013; Bellvert et al., 2014; Micol et al.,
2015). However, for these studies either the TIR image processing failed due to low contrast of
surface temperatures, or results of CWSI estimation were inaccurate due to a narrow range of
upper and lower boundary limits in the morning hours.
With regards to the recent technology advances of photogrammetric software and TIR
sensor, TIR imagery can nowadays be processed even with the low contrast temperature present
within the scene. In addition, the method of Adaptive CWSI is not dependent on meteorological
data, which is required for the energy balance method or empirical methods to obtain the lower
62
and upper baselines, since adaptive CWSI determines the lower and upper baselines by
statistical probability modelling; therefore, it enables to estimate plant water stress level even
with the small range of boundary limits such as those present during morning time.
This chapter explores diurnal changes of plant water stress which are captured by a series of
UAV remote sensing at different times of the day. The aim of the research was to accurately
interpret the diurnal behaviour of plant water stress with measured plant parameters and
estimated CWSIs. The plant physiological parameters such as ψstem and gs are measured on
plants in different irrigation regimes to investigate the diurnal variations of water-stressed plants
from morning to late afternoon. CWSIs estimated at three times of the day are assessed with
plant parameters and the relationships between the CWSIs and the data acquisition time are
presented.
5.2 Methodology
5.2.1 Study site description
The experimental site is a nectarine orchard (0.7 ha, 150 m × 45 m), located near Tatura,
Victoria, Australia (36°26’08” S, 145°16’13” E, 114 m AMSL). The site is the Stonefruit Field
Laboratory, administered by the Department of Economic Development, Jobs, Transport, and
Resources (DEDJTR). The annual average reference crop evapotranspiration is 1190 mm and
the annual precipitation is 480 mm in the region. The site has a single cultivar: nectarine
(September Bright) with Open Tatura canopy system (2.0 m × 1.0 m size). The nectarine was in
fruit maturity phenological stage with full leaf-up. The Open Tatura (OT) is V shape with two
branches, designed to increase light interception for high yields. The trees are planted in a
north-south direction on a fine, sandy, loamy, Shepparton soil. The tree- and inter-row spacing
is 1.0 m and 4.5 m, respectively.
63
(a) (b)
Figure 5.1 (a) Study site location, Stonefruit Field Laboratory (Tatura, Victoria.
Australia); (b) Different levels of water deficit plots are presented in red (0 %),
orange (20 %), green (40 %), and blue (100 %)
The study site contained four experimental plots (Figure 5.1), where the different irrigation
levels were applied as control and water deficit treatments as shown in Table 5.1.
Table 5.1 Descriptions of irrigation level
Plot no. Cultivar Treatment Irrigation level
W1_20
Nectarine
Deficit 20 %
W2_0 Deficit 0 %
W3_40 Deficit 40 %
W4_100 Control 100 %
64
The site was drip-irrigated via an emitter with flow rate of 1.6 Lh-1, and the irrigation was
managed based on a crop water use (crop evapotranspiration) model (Allen et al., 1998). Each
plot contained three trees, where the level of irrigation amount was controlled through valves of
the emitter.
5.2.2 Data acquisition
Three UAV field campaigns were conducted on the 19th of January 2017. The aerial thermal
infrared (TIR) imagery was acquired from morning to afternoon distributed at: 9 h, 12 h, and 15
h local time. The experimental day was with a clear sky, moderate winds (0.6 m sec-1), air
temperature (30.6 °C) and relative humidity (26.7 %) at midday from the nearby (< 200 m)
meteorological station. The weather condition of each time of UAV campaign is shown in Table
5.2.
Table 5.2 Weather conditions at three times of data acquisition
Acquisition time
of data
Air
temperature
Relative
humidity
Wind
speed
9 h 26.8 °C 34.7 % 0.9 m s-1
12 h 30.6 °C 26.7 % 0.6 m s-1
15 h 33.2 °C 18.8 % 1.2 m s-1
A thermal infrared (TIR) camera (A65, FLIR Systems, Inc., Wilsonville, OR, USA) was
integrated with a GPS in an on-board CPU for geo-tagging and all gears were mounted to a
multi-rotor UAV platform (S1000, DJI, Shenzhen, China). TIR images were captured in the
spectral wavelength of 7.5 – 13 µm, a spatial resolution of 640 × 512 pixels, a focal length of 25
mm, and a FOV of 25° (H) × 20° (V).
All aerial images from TIR sensor were taken at the short time window (< 10 minutes) to
capture the homogenous features during the flight over the site. The TIR camera was mounted
to a high performance gimbal to enable the images to be acquired at nadir view from the UAV.
The UAV flew at an altitude of 90 m above ground level (AGL) to capture images with over
65
80 % forward and 40 % side overlap by an autonomous flight plan. The footprint of TIR image
is 39 m × 31 m with a ground sample distance (GSD) of 6 cm.
During the UAV sensing, two types of calibration targets were deployed at the site: (1)
temperature calibration target for TIR images; (2) ground control point (GCP) and ground
artificial feature (GAF) for image orthomosaicking. The water body and rubber plates were
deployed for calibration targets of TIR images as cold and hot features with high emissivity.
The specific GCP and GAF targets made of aluminium materials were designed for performing
an accurate and robust orthomosaicking during the image processing.
With each UAV flight, the measurements of crop physiological data were carried out at the
irrigation treatment plots at the same time of aerial image acquisition. Gas exchange
measurements using a photosynthesis system (LI-6400, LI-COR Inc., Lincoln, Nebraska, USA)
were conducted on fully expanded leaves to obtain stomatal conductance (gs, mmol m-2 sec-1) at
9 h, 12 h and 15 h. Leaf temperature was measured on both sunlit and shaded leaves using a
thermometer (TN410LCE, ZyTemp, Radiant Innovation Inc.). Midday stem water potential
(ψstem, MPa) was measured on two fully expanded shaded leaves per tree in each plot using a
Scholander pressure chamber (Model 3000, Soil Moisture Equipment Co., Santa Barbara, CA,
USA). The leaves were selected from branches near the main trunk and were covered with an
aluminium foil bag for a minimum of one hour of equilibration time, prior to the measurement
of ψstem.
5.2.3 TIR image processing and plant water stress modelling
Based on the method of Adaptive CSWI estimation proposed in Chapter 4, all aerial TIR
images were processed to estimate plant water stress at each time: 9 h, 12 h and 15 h.
The raw signal-formatted TIR images were converted to temperature-based images using a
one-point calibration method (FLIRSystems, 2013). When correlating the signal to the
temperature, calibration targets made of rubber sheets and water body were used to refer the
actual temperatures as hot and cold features. The target temperatures were measured with TIR
snapshot by a handheld thermal imaging camera (T640, FLIR Systems, Inc., Wilsonville, OR,
66
USA) and a handheld thermometer concurrently with the UAV flight. All calibrated TIR images
were stitched into a georeferenced image by using photogrammetric software (PhotoScan,
Agisoft LLC, Russia).
For the extraction of pure canopy pixels from the imagery, the mixed pixels (edges) along
the boundary between canopy and soil pixels were detected by the combined Sobel and Canny
technique in Matlab R2014b. The detected edges were dilated for more conservative exclusion
of the mixed pixels, and removed from the original orthomosaic image.
Figure 5.2 Flow of the CWSI estimation based on adaptive reference baselines
The temperature histogram was generated from the edge-free TIR image to analyse the
temperature distribution statistically. Figure 5.2 describes the process flow of CWSI estimations
in the research. Since the study site has a single cultivar (nectarine) with the same canopy
structure (Open Tatura), the site was treated as one property. Thus, the classification of sub-
regions was not required at the site. The Gaussian mixture modelling (GMM) was fit to the
temperature distribution to cluster canopy and soil pixels and to estimate representative Twet and
Tdry values for the CWSI estimation. The higher-temperature component (2nd distribution) of the
histogram distribution, representing soil pixels was removed to exclude non-canopy effects.
67
Then, the adaptive threshold of Twet and Tdry was determined at the values of 99 % confidence
interval (CI) limits from the Gaussian modelling.
5.3 Results and discussion
5.3.1 Relationship of CWSI with midday stem water potential
(a) (b)
Figure 5.3 (a) Adaptive CWSI map derived from midday UAV remote sensing. The
rectangles in red represent the area of irrigation treatments; (b) Detailed CWSI map
depicting the experimental plots of W1 (20 % irrigation), W2 (0 %), W3 (40 %) and
W4 (100 %)
68
Figure 5.3.a shows the estimated CWSI map based on adaptive of Twet and Tdry derived from
midday UAV flight. The water stress variability over the orchard is visually presented in the
CWSI map. Figure 5.3.b shows the details of four experimental plots in water deficit and
control treatment groups: W1_20, W2_0, W3_40 and W4_100.
CWSI values of trees from each experimental plot were extracted and averaged from the
CWSI map to result in the representative CWSI value for each plot. The CWSI values in all
four irrigation groups were strongly correlated negatively with midday ψstem with determination
coefficients (R2) of 0.83 (Figure 5.4). In W1_20 and W2_0 plots, water stress treatments were
applied, subsequently the measured ψstem of each plot indicated water deficit condition with the
similar value of −2.6 MPa and −2.8 MPa, respectively. The ψstem values for W3_40 and
W4_100 were also shown in close range with the values of −1.7 MPa and −1.6 MPa, which
represent non-water-stress condition. Similar results were obtained for the estimated CWSI. The
CWSIs in W1_20 and W2_0 were estimated in the similar values (approx. 0.66 and 0.60,
respectively) and higher compared to in W3_40 and W4_100 (approx. 0.49 and 0.44,
respectively).
Figure 5.4 Relationship between stem water potential (ψstem) and CWSI at 12 h
69
5.3.2 Relationship of CWSI with diurnal plant water stress
The differences of canopy and air temperature (Tc – Ta) were also measured at three times of
the day. Figure 5.5 shows Tc – Ta of control plot (W4_100) and deficit plot (W2_0) from
morning to afternoon. In both irrigation treatment plots, Tc – Ta changed considerably during the
day, starting from positive and becoming negative Tc – Ta. In the morning (9 h), canopy
temperature was higher than air temperature since the canopy was becoming warm. At midday,
canopy temperature was becoming cooler than air in control plot. This is considered due to
transpiration cooling. In the case of deficit plot, Tc – Ta was becoming zero, indicating higher
canopy temperature than control plot, which was derived from water stress. Overall, water
deficit plot resulted in higher canopy temperature than irrigated plot during three measurements,
related to normal diurnal depression.
Figure 5.5 Differences between canopy and air temperature at three times of the day
in control (100 %) and deficit (0 %) plots
Stomatal conductance (gs) measurements were used to provide the relationships with the
estimated CWSIs at the different time of the day as shown in Figure 5.6.
70
(a)
(b)
(c)
Figure 5.6 Relationships between stomatal conductance (gs) and CWSI: (a) results
acquired at 9 h; (b) 12 h; (c) 15 h
71
Similarly with midday ψstem, a strong relationship was found between gs measurements and
the adaptive CWSI at 9 h, 12 h, and 15 h. Specifically, the relationship of each time was
significant with determination coefficients (R2) of 0.92, 0.77 and 0.86, respectively.
From the 9 h results (Figure 5.6.a), the measured gs from each plot showed a gradual
increase of values according to the irrigation level with gs values from 18.4 (mmol m-2 sec-1) at
W2_0 to 39.8 (mmol m-2 sec-1) at W4_100. Similar but inversed results were found for the
estimated CWSI, showing that the CWSI values gradually dropped with higher irrigation level.
The canopy temperature difference from the irrigation extreme plots (W2_0 and W4_100) was
2.6 °C. A CWSI research, which employs the VPD and air temperature to calculate Twet and Tdry
(or baselines), has shown that the CWSI calculation with morning data was challenging and
failed to determine the baseline due to a narrow range of temperature differences between the
upper and lower baselines (e.g., 2 °C) (Gonzalez-Dugo et al., 2013). As Adaptive CWSI
method relies on statistical modelling of canopy temperatures, the proposed method was able to
obtain Twet and Tdry in the small range of temperature differences. In Figure 5.6.b and Figure
5.6.c, the relationships presented were derived from midday and afternoon. Both CWSIs were
significantly correlated with gs measurements.
In the aspect of diurnal variation of plant water stress, morning gs showed less variation than
midday and afternoon gs over the all irrigation regimes (0, 20, 40 and 100 % irrigation) as
shown in Table 5.3. The distribution of gs, which is the difference of extreme gs in 0 % and
100 %, was becoming significantly wider from morning to midday, then, remaining at the
similar scale (slightly wider) from midday to afternoon. This indicates that all gs were highly
correlated water stressed levels, but had the different range of gs from morning to afternoon.
72
Table 5.3 Statistics of the extreme gs over all irrigation regimes at three times of the
day
Acquisition
time
SD of
extreme gs
(mmol m-2 sec-1)
Mean of
extreme gs
(mmol m-2 sec-1)
Size of
extreme gs
(%)
9 h 9.35 28.61 32.7
12 h 17.46 32.48 53.8
15 h 18.98 30.46 62.3
SD: Standard Deviation
In the water stressed plots, the gs of W1_20 and W2_0 responded at the rates of 18 – 24
(mmol m-2 sec-1) at 9 h. The gs of W1_20 and W2_0 slightly decreased and remained at 13 – 24
(mmol m-2 sec-1) at 12 h. Then, gs of W1_20 and W2_0 considerably dropped to 11 – 17 (mmol
m-2 sec-1) at 15 h. The response of gs in water stressed plants can indicate that stomatal
conductance decreases due to the stomatal closure and less transpiration during daytime.
In the case of control plot (W4_100), gs showed higher rates than other irrigation plots at
three times of the day. Particularly, midday gs was obtained at the highest rate, 53 (mmol m-2
sec-1).
Overall, midday and afternoon gs were measured at the similar range, while morning gs
showed less conductance rate than the other time. In water deficit plots, midday and afternoon
gs were responded at lower rates than morning gs, indicating that the water-stressed plant
behaves with more limited stomatal conductance during daytime.
73
Table 5.4 Statistics of the extreme CWSI over all irrigation regimes at three times of
the day
Acquisition
time
SD of
extreme CWSI
Mean of
extreme CWSI
Size of
extreme CWSI
9 h 0.088 0.43 20.5 %
12 h 0.098 0.55 17.9 %
15 h 0.093 0.53 17.5 %
SD: Standard Deviation
Figure 5.7 Relationships between CWSI estimated at three times of the day and the
stomatal conductance
The estimated CWSIs taken three times of the day showed significant correlation with gs
according the irrigation levels. Unlike the variation of the extreme gs along the day, CWSIs
were presented in the similar range of the extreme values at each flight time as shown in Table
5.4. The morning CWSI was slightly lower (0.11 of CWSI) than midday and afternoon CWSIs
like morning gs (Figure 5.7 and Table 5.4). However, the size of CWSI range was obtained at
the similar scale at each time of the day. This could imply that Adaptive CWSIs enable to
separate different water stressed plants with high sensitivity and consistent scale from morning
to afternoon of the same day. Nevertheless, the range of Adaptive CWSI can be biased in terms
of absolute quantity, due to the limitation of statistical histogram approach, as described in
74
Chapter 4. Since the bias can be adjusted by a simple relationship between CWSI and the level
of irrigation, it still remains a practical method to estimate CWSI across an area during the day.
5.4 Conclusions
This study analysed the diurnal changes of plant water stress using: 1) plant physiological
data: stomatal conductance (gs); and 2) estimated CWSI derived from TIR UAV sensing at
three different times of the day: 9 h, 12 h and 15 h. The study tested the performance of the
method proposed to estimate diurnal behaviour of water stress in the different irrigation regime.
The estimation was performed by adopting the method of Adaptive CSWI. The significant
relationship between CWSI and the gs (and ψstem) was found at every flight times of the day in
this work. Particularly, morning CWSI was also sensitive to plant water stress. Thus, the
adaptive CWSI method showed a capability to interpret plant water stress level even with the
small range of boundary limits in the morning. Particularly, morning CWSI presented a relative
consistency with midday and afternoon CWSI. Thus, the present research can potentially
promise any-daytime CWSI estimations of the day; from late morning to late afternoon, since
CWSI can be estimated with low contrast of surface temperatures. Subsequently, the method
can assist to extend the time-window of UAV remote sensing during the day. As a future
research, various crop-cultivar-site- and time-specific experiments will remain for further
works.
75
Chapter 6 : Estimation of evapotranspiration in peach
orchards using very-high-resolution imagery
from an unmanned aerial vehicle (UAV)
6.1 Introduction
Drought and water shortage have been increased and resulted in declined productions in
irrigated agriculture in Australia (Qureshi et al., 2013). Therefore, efficient water use as
agricultural resources requires a more precise and adequate irrigation management with the
limited availability of water. In addition, having the capability of estimating crop water use via
evapotranspiration (ET) can assist irrigation managements on a regular basis in terms of a
decision-making strategy. In case of high valued crops such as grapevines and olives, various
studies have investigated the effect of deficit irrigation including regulated deficit irrigation
(RDI) and found out that the delicate water management based on soil-cultivar-environmental
specificity is one of the main factors for increasing the quality and yield of crops, as well as
water use efficiency (Roberto et al., 2005; Buesa et al., 2017).
Due to the importance of monitoring plant water consumption, ET is one of the most critical
key factors to estimate water loss from vegetation surface as well as a fundamental requirement
in irrigation scheduling at the local management with respect to climate change in the global
assessment (Nouri et al., 2015). For instance, estimating ET on crops becomes more in demand
in planning an optimal irrigation management due to increasing water scarcity which limits
agricultural production. Moreover, plant ET provides quantitative information for analysing the
sophisticated process of water consumption in precise agriculture (PA). However, estimating
accurate ET still remains a significant challenge due to its complex task.
In recent decades, remote sensing (RS) based-ET estimation has been introduced
increasingly in agricultural studies since satellite images have been broadly applied in ET
estimation. The RS-based ET has been known as the most reliable and efficient methods for
estimating ET especially at regional scales (Kustas and Anderson, 2009). This is mainly
because RS provides a high capability of capturing the features of heterogeneous surface in
partially vegetated surface as well as considering the heterogeneity of plant species. In addition,
76
RS based ET approaches produce a map, representing ET values in pixel by pixel as well as
showing the spatial variability over mixed landscape.
With regards to RS based ET estimation, in general, the existing approaches can be
categorized into four methods: empirical direct method, residual method, inference method and
deterministic method.
In the empirical direct method, ET estimation is performed from the correlation between ET
and the combined RS data such as TIR, and between ET and meteorological data such as air
temperature. For instance, a simple direct method was introduced, which relies on the
assumption that daily ET has a direct relationship with the difference of surface and air
temperature at midday (Jackson et al., 1977; Seguin and Itier, 1983). The method requires the
empirical calibration parameters at a local scale. Based on the relationship between NDVI and
surface temperature, another empirical direct method has also been proposed (Moran et al.,
1994; Carlson et al., 1995). The method depends on the fact that NDVI affects the ET quantity
and ET is affected by surface temperature. For example, as vegetation coverage increases, more
ET is expected, and consequently the surface temperature tends to be cool. As a ramification of
the method, the strong relationship between surface temperature and NDVI was found, but the
relationship varies in seasonal changes (Carlson et al., 1995; Yuan and Bauer, 2007).
The residual method has been widely adopted to obtain ET, which is based on SEBM. In the
model, ET is obtained as a residual (latent heat flux) while the other components - net radiation,
sensible heat flux and soil heat flux - are also estimated (Su, 2002; Kalma et al., 2008). The
residual method in SEBM uses RS data to calculate surface temperature and canopy structural
characteristics (Yuei-An and Kumar Kar, 2014), and to estimate net radiation and soil heat flux
(Bastiaanssena et al., 1998).
In the aspects of different techniques related to the residual method, firstly, Surface Energy
Balance Algorithm for Land (SEBAL) estimates ET at regional scale with minimum ground
data which includes surface temperature, incoming solar radiation, NDVI, and wet and dry
threshold values (Bastiaanssena et al., 1998). The threshold values are two extreme reference
temperatures inside the target area which are allocated to set the limit to non-evapotranspiration
77
in dry surface and non-sensible heat flux in wet surface. Secondly, Surface Energy Balance
System (SEBS) estimates evaporative fraction and turbulent fluxes using RS data,
meteorological data and radiation data (Su, 2002). Thirdly, Surface Energy Balance Index
(SEBI) and Simplified SEBI (S-SEBI) have been developed to estimate ET from RS data
(Roerink et al., 2000). S-SEBI only relies on RS data without any further inputs since wet and
dry thresholds are determined simply by using the minimum and maximum reflectance values
of temperature in the target image. Thus, S-SEBI can produce reliable results when wet and dry
surface extremes exist in the area. Lastly, Two Source Energy Balance (TSEB) has been
introduced to partition the sensible and latent heat fluxes separately from canopy and soil
(Norman et al., 1995; Kustas and Norman, 1999). TSEB does not consider the surface extreme
within the image and shows better ET estimation than general single-source model in bare soil
and partially vegetated surface (Xia et al., 2016). It could be a heavy process compared to
single-source model in vegetated surface in terms of operational ET estimation (French et al.,
2015; Xia et al., 2016).
As demanding as high spatial and temporal resolutions, UAV remote sensing is becoming
more popular, since it provides very-high-resolution (VHR) imagery down to sub-centimetre
spatial resolution (Berni et al., 2009b; Zarco-Tejada et al., 2012; Roy and Ophori, 2014; Rud et
al., 2014). In addition, UAV sensing can capture a target area in a projected time on demand
(Zipper and Loheide Ii, 2014; Ortega-Farías et al., 2016). Furthermore, the operation of UAV
sensing is cost-effective and the on-board camera system can be assembled according to the
research purpose (Brenner et al., 2017).
Several studies on UAV-based high resolution ET have been investigated with the current
energy balance models during the past few years (Zipper and Loheide Ii, 2014; Hoffmann et al.,
2015; Ortega-Farías et al., 2016; Xia et al., 2016; Brenner et al., 2017). The research of UAV-
based ET estimation is still in its early stage and needs to be further investigated and validated
(Hoffmann et al., 2015; Brenner et al., 2017).
In this chapter, tree-by-tree energy balance components are estimated based on HRMET
surface energy balance model using VHR MS and TIR imagery from a UAV. The main
objective is to assess the intra-field variability of ET over a drip-irrigated peach orchard at sub-
78
field scale. The tree-level analysis is conducted to produce tree-level water use. The approach of
the research consists of five steps (Figure 6.1): 1) data acquisition of VHR imagery and ground
measurements; 2) RS data processing to produce surface temperature and canopy characteristics
variables of ET modelling; 3) tree-by-tree modelling to identify individual plant and allocate the
representative spatial variable to each tree; 4) calculation of energy balance components in the
ET model and mapping instantaneous ET rate and; 5) evaluation of the estimated ET with
ground measurements.
Figure 6.1 Research flow of data processing and ET modelling
6.2 Methods and Dataset
6.2.1 Study area
The experiment was carried out on 19th of January 2017 at the peach and nectarine orchards
(0.95 ha, 140 m × 58 m) near Tatura, Victoria, Australia (36°26’08” S, 145°16’13” E, 114 m
AMSL), administrated by the Department of Economic Development, Jobs, Transport, and
79
Resources (DEDJTR) (Figure 6.2). The climate of Tatura is temperate. The annual rainfall is
approximately 480 mm and the annual average ETc (reference crop evapotranspiration) is 1190
mm. The site has two cultivars: a peach (Prunus persica (L.) Batsch cv. August Flame); and a
nectarine (Prunus persica (L.) Batsch cv. Autumn Bright). The nectarine and peach were 3 – 4
years old, late maturity phenological stage at full leaf-up. Each cultivar is trained to two canopy
structures: Vertical leader (VL) with 2.0 m × 0.8 m size and Tatura Trellis (TT) with 1.9 m ×
1.0 m size. TT is in Y shape with two arm branches, since the architecture of TT is designed to
maximize the leaf surface which exposes and captures the sun for enhancing the productivity
(Elkins, 2002). The trees were planted in a north-south direction on a fine, sandy, loamy,
Shepparton soil. The tree spacing is 1.0 m tree and the distance of an inter-row is approximately
4.5 m. The region was drip-irrigated via an emitter and the irrigation amount was calculated by
ETc model (Allen et al., 1998).
(a)
(b)
Figure 6.2 (a) Location of the study site, Stonefruit Field Laboratory orchard (Tatura,
Victoria. Australia); (b) canopy structures in Vertical leader (left) and Tatura Trellis
(right).
80
6.2.2 Data acquisition
6.2.2.1 Energy balance and Micrometeorological data
A tower was installed for measuring both net radiation and micrometeorological data near
the centre of the study area (Figure 6.3). A net radiometer with 4 components (NR01,
Hukseflux Thermal Sensors B.V., Delft, Netherlands) measured net radiation,
incoming/outgoing short wave, and long wave radiations at 2.6 m AGL (0.6 m above tree).
Temperature and humidity (HMP155, Vaisala Corporation, Helsinki, Finland) sensor was
installed above the tree. A wind monitor (Wind Monitor-AQ, R.M. Young Company, Michigan,
USA) was placed at the top of the tower at 4 m AGL and measured wind speed and direction.
Two sets of soil heat flux plates (HFP01-L, Hukseflux Thermal Sensors B.V., Delft,
Netherlands) were installed with the soil temperature probes (TCAV, Campbell Scientific, Inc.,
UT, USA) and soil moisture reflectometer (CS616, Campbell Scientific, Inc., UT, USA). One
set was placed in the middle of the inter-row, the other set in the tree-row. The different
placements accommodated to capture a gradient of soil heat fluxes from underneath tree (mostly
shaded) to inter-row (non-vegetated).
Figure 6.3 Installed sensors to measure net radiation and micrometeorological data
81
6.2.2.2 Crop Physiological data
As mentioned above, the tower for measuring net radiation, meteorological data, and soil
heat flux was installed in the tree row. The measurement of crop physiological data was carried
out at the four adjacent trees in the same tree-row. Leaf temperature was measured in both sunlit
and shaded leaves using a handheld thermometer (TN410LCE, ZyTemp, Radiant Innovation
Inc.). Gas exchange measurements using a photosynthesis system (LI-6400, LI-COR Inc.,
Lincoln, Nebraska, USA) were taken over fully expanded and measured leaves to obtain
stomatal conductance (gs, mol m-2 sec-1) and transpiration rate (E, mmol m-2 sec-1).
Photosynthetically active radiation (PAR) was measured above and under the canopy for light
interception by the canopy using a ceptometer (Sunfleck, model SF-80, Decagon Devices Inc.,
Washington, DC).
6.2.2.3 UAV campaigns
UAV campaign was conducted on the day of clear sky with moderate wind (0.6 m sec-1), air
temperature (30.6 °C), and relative humidity (26.7 %). A thermal infrared (TIR) camera (A65,
FLIR Systems, Inc., Wilsonville, OR, USA) and multispectral (MS) camera (RedEdge,
MicaSense, Seattle, WA, USA) were integrated with a GPS in an onboard CPU for geo-tagging
and all gears were mounted to an UAV platform (S1000, DJI, Shenzhen, China).
All aerial images from TIR and MS sensors were taken at the short time window (< 15
minutes) to capture the homogenous features at midday over the site. TIR images are captured
in the of 7.5 – 13 µm, a spatial resolution of 640 × 512 pixels, a focal length of 25 mm, and a
FOV of 25 °(H) × 20 °(V). MS images were acquired concurrent with TIR sensing. The MS
camera consists of five discrete bands in the blue (475 nm), green (560 nm), red (668 nm), red
edge (717 mm), and near infrared (NIR) (840 nm). A spatial resolution is 1280 × 960 pixels
with a focal length of 5.5 mm.
Both cameras were mounted to a high performance gimbal to enable the images to be
acquired at nadir view from the UAV. The UAV sensing was conducted at solar noon for
capturing the period of high ET and for minimizing shadows cast by the tree. The UAV flew at
82
an altitude 90 m above ground level (AGL) to capture images with over 80 % forward and 40 %
side overlap by an autonomous flight plan. The footprint of TIR image is 39 m × 31 m with a
ground sample distance (GSD) of 6 cm. MS image has the footprint of 108 m × 81 m and GSD
of 8 cm.
During the UAV sensing, two types of ground targets were deployed at the site: (1)
radiometric calibration target for TIR and MS images; (2) ground control point (GCP) ground
artificial feature (GAF) for image processing. The water body and rubber plates were used for
calibration targets of TIR images as those are cold and hot features. Tarps (3 m × 3 m) in three
different reflectance rates (6 %, 12 %, and 33 %) were placed for the calibration of MS images
(Figure 6.4). The specific GCP and GAF targets were designed for performing an accurate and
robust orthomosaicking during the image processing. Those targets were made of aluminium
sheet with the consideration of spatial resolution and spectral characteristic of both TIR and MS
images.
Figure 6.4 Field description: (left) radiometric grey targets and temperature targets,
the details of sampling trees and meteorological tower; (right) the location of targets
and sample trees
83
6.2.3 RS data processing and ET modelling
6.2.3.1 Aerial imagery processing
When processing the very high-resolution imagery, a post calibration process is carried out
for producing reliable inputs which are one of key factors for the accurate ET estimate. TIR
images were recorded in a raw format which consists of the signal-based values from the
surface temperatures. A one-point calibration method (FLIRSystems, 2013) was applied to
convert the raw images to temperature-based images by using a customized code written in
Matlab R2014b (Mathworks Inc., Matick, MA. USA). When correlating the signal to the
temperature, calibration targets made of rubber sheets and water body were used to refer the
actual temperatures as hot and cold features. The target temperatures were measured with TIR
snapshot by a handheld thermal imaging camera (T640, FLIR Systems, Inc., Wilsonville, OR,
USA) and a handheld thermometer concurrently with the UAV flight. In case of MS imagery,
the radiometric calibration was performed to retrieve the reflectance values in each band. The
reflectance from three uniform tarps was correlated to the digital number (DN) values from the
MS imagery, and the linear regression relationship was applied to convert all DN values to the
reflectance values. All consecutive TIR and MS images were mosaicked into each
georeferenced image by using a photogrammetric software (PhotoScan, Agisoft LLC, Russia).
Figure 6.5 shows the mosaicked TIR image.
84
Figure 6.5 Surface temperature of the study area
6.2.3.2 LAI estimation
The ET model used in the research requires LAI as one of the spatial variables to present the
canopy characteristics. In many researches, spectral vegetation indices (e.g., NDVI, simple
ratio, and reduced simple ratio) have been largely used to link to the LAI estimation, showing
that NDVI is highly correlated with LAI and the relationship can be adopted in various forms of
regression model (e.g., general exponential, simple linear, or higher curve model) (Stenberg et
al., 2004; Steltzer and Welker, 2006; Fan et al., 2009). The exponential regression method was
employed to retrieve LAI from remote sensed NDVI in the research. The reference LAI values
were determined using PAR measurements above and under the canopy for light interception by
the canopy in the four sample trees. The measured-PAR of each tree consists of transmitted and
scattered radiations through the canopy and within the canopy, respectively. A radiative model
of transmitting and scattering was introduced by Norman and Jarvis (1975) and simplified
(DecagonDevicesInc., 2001) as follows:
85
𝜏 = exp {𝐴(1 − 0.47𝑓𝑏)𝐿 ((1 − 1 2𝐾⁄ )𝑓𝑏 − 1⁄ )} (6.1)
𝐾 = 1/2 cos 𝜃 (6.2)
where, 𝜏: canopy transmittance, 𝐴: canopy absorption, 𝑓𝑏: fraction of direct radiation, 𝐿:
LAI, 𝐾: coefficient of canopy extinction, and 𝜃: sun zenith angle
The 𝜏 is a ratio of the transmitted PAR to the incident PAR, and the 𝑓𝑏 is the fraction of
incident PAR radiation, which is also called as beam fraction. 𝐾 is the coefficient of canopy
extinction and can be simplified with the sun zenith angle (𝜃). The PAR-based LAI in the
reference tree was retrieved by inverting the radiative model (Equation 6.1). Then, the LAI map
over the study site was estimated by the exponential regression model with NDVI map
generated from MS imagery of the UAV sensing.
6.2.3.3 Tree segmentation
In above two sections, surface temperature and LAI map were generated into the raster
imagery with geo-referenced coordinates. For the tree-by-tree analysis, feature segmentation
technique was used in the research. First of all, the non-canopy background pixels were
excluded in both temperature and LAI imagery using histogram method. A threshold value was
determined to separate the canopy and non-canopy distributions in the histogram. For the
accurate separation, edges along the boundary between canopy and background were removed
using edge detection method, since edges were the mixed canopy-soil properties and showed
generally higher temperature and lower LAI (or NDVI) than pure canopy pixels. The details of
the edge detection method used in the research were presented in Chapter 4.2.5. Second, pixels,
which belong to the individual trees, were classified and grouped systematically by the
informative field descriptions: the plant spacing and row spacing distances. The grouped pixels
were aggregated and averaged as a representative value of each tree by using a customized code
written in Matlab R2017b (Mathworks Inc., Matick, MA. USA).
86
6.2.3.4 RSEB algorithm in ET modelling
High resolution mapping of evapotranspiration (HRMET) model was implemented to
estimate energy balance components, subsequently ET rate in the research. HRMET is a pixel-
based surface energy balance model and was developed to incorporate high resolution RS data.
Advantages of the model can be described as it does not depend on wet and dry reference
features to calculate turbulent fluxes in the imagery. Most of input variables in the model can be
obtained from RS data, and the model does not require a heavy process and is executed with
minimized inputs as a feasible tool to adapt an applicable irrigation scheduling. The algorithm
of the model follows the residual method of general surface energy balance model as in
Equation 6.3:
𝜆𝐸𝑇 = 𝑅𝑛 − 𝐻 − 𝐺 (6.3)
where, 𝜆𝐸𝑇: latent heat flux (W/m2); 𝑅𝑛: net radiation at the surface (W/m2); 𝐻: sensible
heat flux to the air (W/m2) and; 𝐺: soil heat flux (W/m2)
The instantaneous net radiation is calculated to separate canopy and soil components based
on two-source model (Norman et al., 1995) as follows:
𝑅𝑛 = 𝑆𝑖𝑛 − 𝑆𝑜𝑢𝑡 + 𝐿𝑖𝑛 − 𝐿𝑜𝑢𝑡 (6.4)
𝑅𝑛 = 𝑓𝑐𝑅𝑛,𝑐 + (1 − 𝑓𝑐)𝑅𝑛,𝑠 (6.5)
𝑅𝑛,𝑐 = (1 − 𝛼𝑐)𝑅𝑛 + 𝜀𝑐𝐿𝑖𝑛 − 𝐿𝑜𝑢𝑡,𝑐 (6.6)
𝑅𝑛,𝑠 = (1 − 𝛼𝑠)𝑅𝑛 + 𝜀𝑠𝐿𝑖𝑛 − 𝐿𝑜𝑢𝑡,𝑠 (6.7)
𝐿 = 𝜀𝜎𝑇4 (6.8)
where, 𝑆𝑖𝑛, 𝑆𝑜𝑢𝑡, 𝐿𝑖𝑛, 𝐿𝑜𝑢𝑡 is incoming and outgoing shortwave and incoming and outgoing
longwave (W/m2), respectively. 𝑓𝑐 is fractional cover derived from LAI using exponential
87
relationship. The subscript 𝑐 and 𝑠 imply canopy and soil. 𝛼 and 𝜀 represent the surface albedo
and emissivity. 𝛼 was taken from Goodwin et al. (2004). 𝜎 is the constant of the Stefan-
Boltzmann, 𝑇 is the surface temperature in Kelvin.
The incoming and outgoing shortwave radiations are calculated separately from canopy and
soil using the radiation extinction model in the Equation 6.4 – 6.7 and the Stefan-Boltzmann
law equation (Equation 6.8). Then, soil heat flux is approximated by 35 % of the net radiation
reaching the soil based as follows (Norman et al., 1995):
𝐺 = 0.35𝑅𝑛,𝑠 (6.9)
The sensible heat flux, H, is computed as follows:
𝐻 = 𝜌𝑎𝑐𝑝(𝑇𝑠 − 𝑇𝑎) (г𝐻𝑎 + г𝑒𝑥)⁄ (6.10)
where, 𝜌𝑎: the molar density of air (mol m-3), 𝑐𝑝: the specific heat of air (J mol-1 C-1), г𝐻𝑎:
the aerodynamic resistance to heat transport (s m-1), and г𝑒𝑥: the excess resistance to heat
transport (s m-1).
Aerodynamic and excess resistances are calculated in each pixel iteratively using the
methods from Campbell (2000) and Norman and Becker (1995), respectively. The model
employs an iterative convergence method to calculate H, since both H and г𝐻𝑎 are co-dependent
variables. The level of H effects on the aerodynamic stability, subsequently also on the
aerodynamic resistance to heat transport (Bastiaanssena et al., 1998). The iterative method of H
is described in a flow chart in Figure 6.6. The method does not require the wet and hot extreme
pixels due to the iterative calculation of the turbulent heat fluxes.
88
Figure 6.6 The iterative method of calculating the sensible heat flux in HRMET
model
When the convergence criteria are achieved, which is less than 0.1 % change between the
consecutive iterations, the latent heat flux of each pixel was computed using the updated
sensible heat flux in the Equation 6.3.
89
6.3 Results and discussion
The study area consists of two canopy systems in each nectarine and peach tree: Vertical
leader (VL) and Tatura Trellis (TT) as shown in Figure 6.2. At the time of UAV flight over the
study area, the canopy temperature, captured from TIR sensing, showed spatial variations with
regards to each canopy system and cultivar. In particular, the first three rows were planted with
nectarine of VL canopy and represented lower temperature (29.8 °C) compared to the average
temperature of other tree-tows (31.7 °C), showing approximately 2 °C temperature difference.
Figure 6.7.a shows the intra-field spatial variability of surface temperatures without soil
background in an enhanced contract colormap. The distributed canopy temperatures over the
field were visualized; the temperature of nectarine with VL was apparently represented cooler
than that of others. The entire study field was under full-irrigation regime. The coefficient of
variation (CV) of canopy temperature was 5.3 % over the field. Thus, this result indicated that
the canopy temperatures (Tc) varied from the different canopy structures and hence, would
result in the variability of ET rate. The spatially distributed NDVI was generated by using VHR
red and NIR imagery from the UAV sensing, shown in Figure 6.7.b. The figure presents canopy
NDVI, where soil NDVI was classified and excluded using histogram-based method for the tree
segmentation process in the next step. Canopy NDVI also showed spatial variability with 4.5 %
of CV, which was slightly lower than the CV of canopy temperature during the UAV campaign.
The canopy NDVI in the reference sample trees were correlated to PAR-based LAI with an
exponential relationship with determination coefficients (R2) of 0.9 as shown in Figure 6.8. LAI
over the field was retrieved using the regression method. Since the UAV data acquisition was
carried out to capture instantaneous features over the field at the matured growth stage, NDVI
values from the reference trees indicated the similar value (approximately 0.75). In general
cases, the broad distributions of NDVI and LAI are required to model the relationship
accurately. However, the research assumes the retrieved LAI over the field follows the range of
the reference PAR-LAI with an ignorable error, since NDVI values from all trees were in the
narrow distribution and were similar to NDVI values from the reference trees.
90
(a) (b)
Figure 6.7 Tree-row map excluding soil background using histogram-based method:
(a) surface temperature map (°C) and; (b) NDVI map
Figure 6.8 Relationship between NDVI derived from UAV sensing and LAI retrieved
from PAR measurements on the reference trees
91
Figure 6.9 shows the results of the tree-by-tree segmentation both in surface temperature
and LAI map. The orchards consisted of 12 rows and 120 trees were planted in each row.
According to the field descriptions, tree and inter-row spacing are 1.0 m and 4.5 m distance,
respectively. Using the field descriptions and geo-referenced positioning of each row and tree,
the section of each tree was segmented. Then, the mean value of surface temperature and LAI
were assigned to each tree section as a representative tree value. The results provided better
analytic maps to feature the individual plants and to interpret the variability between plants,
enabling the analysis of tree-by-tree water losses.
(a) (b)
Figure 6.9 Tree-by-tree aggregated results: (a) surface temperature map (°C) and; (b)
LAI map
The energy balance components including latent heat flux were estimated in the HRMET
model and the hourly ET rate is calculated over the study field as shown in Figure 6.10. The
estimated ET in the reference trees indicated approximately 0.62 mm h-1. Relatively higher ET
was observed in the first three rows (nectarine with VL) and in the northern part of tree rows,
92
presenting 0.76 mm h-1 rate at the most. Since the study field was as a small-sized orchard (< 1
ha) and UAV data acquisition time was less than 15 minutes, the meteorological variables such
as incoming shortwave radiation, wind speed and vapour pressure were regarded to be
consistent across the field during the UAV sensing. Thus, the different ET rates along the trees
were determined, which were mainly derived by the differences of canopy temperature and LAI
(hence, vegetation fractional cover), as the similar patterns were confirmed in surface
temperature map and NDVI (or LAI) map (Figure 6.9).
Figure 6.10 The estimated ET map (mm/hr): (left) the ET distribution over the
orchards; (right) the tree ET which excluded background soil
93
The estimated ET was compared with the leaf transpiration rate (mmol m-2 sec-1) measured
by the gas exchange analyser (LI-6400, LI-COR Inc., Lincoln, Nebraska, USA) which was
conducted on the three reference trees as shown in Figure 6.11.
Figure 6.11 Comparison between the estimated ET and leaf transpiration on the
sample trees
The estimated ET showed a strong relationship (R2: 0.9) with the average value of leaf
transpiration. The magnitude of relationship could be over-estimated, as the dataset consists of a
small number of measurements. Nevertheless, the result can indicate that the estimated canopy
ET was correlated with the ground-measurement of leaf transpiration.
Although a direct measurement of ET such as eddy covariance was not available as the
validation data in the research, the crop ET (𝐸𝑇𝑐) value of the study field was compared with
the estimated ET in order to evaluate the method. 𝐸𝑇𝑐 was acquired by using reference crop ET
(𝐸𝑇𝑜) and the basal crop coefficient (𝐾𝑐𝑏) adjusted for tree size as follows:
𝐸𝑇𝑐 = 𝐾𝑐𝐸𝑇𝑜 (6.11)
where, 𝐾𝑐 is the crop coefficient and can be divided into two coefficients: soil evaporation
coefficient (𝐾𝑒) and basal crop coefficient (𝐾𝑐𝑏) which refer to the contribution of soil
evaporation and crop transpiration in modelling crop ET, respectively (Allen et al., 1998). 𝐾𝑐𝑏
94
has been considered to improve the daily crop water use in irrigated row crops. Thereby, 𝐸𝑇𝑐 is
expressed as follows:
𝐸𝑇𝑐 = (𝐾𝑒 + 𝐾𝑐𝑏)𝐸𝑇𝑜 (6.12)
A research has been conducted to determine 𝐾𝑐𝑏 over the same study site of peach orchards
at Tatura, Victoria (Goodwin et al., 2006). Therefore, this research adopted the site-specific
crop coefficient (1.52) from the finding of Goodwin et al. (2006).
As 𝐸𝑇𝑐 indicates the daily evapotranspiration, the estimated hourly ET from the UAV
sensing was up-scaled to the daily value using the extrapolation method (Chávez et al., 2008) as
follows:
𝐸𝑇𝑑 = (𝐿𝐸𝑖/𝑅𝑛,𝑖)𝑅𝑛,𝑑(𝑐𝑓/𝜆𝑣𝜌𝑤) (6.13)
where, 𝐿𝐸𝑖 and 𝑅𝑛,𝑖 are instantaneous latent heat flux and net radiation and obtained from
the estimated energy components. 𝑅𝑛,𝑑 is the mean net radiation of daily (24 hours) and
measured from the net radiometer. 𝑐𝑓 is a unit conversion factor of time. 𝜆𝑣 and 𝜌𝑤 are latent
heat of vaporization and water density, respectively.
As a result, the extrapolated daily ET was obtained as 5.50 (mm d-1), whereas the crop ET
was calculated as 6.35 (mm d-1). The difference of two daily ET values showed as 0.85 (mm d-
1). Considering that the extrapolated ET can include the estimation errors (e.g., ranging from
0.25 to 1.17 mm d-1) depending on the extrapolation methods and crop types (Chávez et al.,
2008), the estimated daily ET value can be interpreted to be within a range of expected ET,
showing a close value of the crop ET. Although it was challenging to evaluate the results
thoroughly due to the absence of sufficient validation data such as the directly measured ET or
multi-seasonal UAV data, the estimated results were compared with leaf transpiration and the
daily crop ET for showing that the method has a potential possibility to estimate tree-by-tree ET
with intra-field variability and to accommodate to VHR imagery. The research method can be
supported by further experiments and needs to be explored with different
sites/cultivars/phenological stages.
95
6.4 Conclusions
This research examined the estimation of ET using very-high-resolution (VHR)
multispectral and thermal imagery (GSD < 8 cm) derived from the UAV sensing. The energy
balance components were estimated based on HRMET surface energy balance model. The tree-
by-tree analytic maps were produced by the systematic feature segmentation method based on
pure canopy extraction and statistical analysis of the distribution of surface temperatures and
LAI. A strong linear relationship between the estimated ET and the leaf transpiration was
obtained in this work. The estimated ET presented a close value to the crop ET over the study
site. The proposed approach can potentially provide a practical method of assessing the intra-
field variability of tree-by-tree ET at sub-field scale for the irrigation scheduling. As a future
work, further research on VHR ET estimation will be investigated along with various RSEB
models and different field sites.
96
97
Chapter 7 : Summary and conclusions
The aim of the research was to estimate plant water stress and evapotranspiration (ET) by
integrating very-high-resolution (VHR) imagery into crop water stress model and surface
energy balance model. Thermal infrared (TIR) and multispectral (MS) imagery from unmanned
aerial vehicles (UAVs) system was used as inputs of surface temperature and vegetation canopy
structure. The proposed methodology relatively required the least ground-based measurements
compared to the existing methods. The major results and findings are presented as follows.
New adaptive approach of CWSI estimation
• The research focused on presenting the plant water stress level by a quantitative index:
crop water stress index (CWSI), conducted by TIR-based UAV remote sensing in order
to assess the spatial variability of water stress over the area. The new method of CWSI
estimation was proposed to estimate CWSI by addressing an adaptive approach in two
aspects.
• Firstly, the research found that the behaviour of canopy temperature was different from
cultivar type and canopy training structure over the region even in the same irrigation
regime. General CWSI methods, employing single reference boundary (simply, Twet and
Tdry), could not accommodate to such behaviour. Thus, the research proposed multiple
reference boundaries based on classifying the region into sub-regions. In the research,
sub-region was defined as an area which contained the same crop property. Allowing
multiple sets of Twet and Tdry, sub-region specific CWSI was achieved in the research.
• Secondly, a statistical approach combined with feature extraction was proposed to
determine automatic thresholds of Twet and Tdry. The research hypothesised that Twet and
Tdry can be obtained at two extremes in the distribution of canopy temperatures under the
assumption of the presence of water-stressed and non-water-stressed plants in the region.
Mixed pixels of canopy and non-canopy were discarded to obtain pure canopy
distribution in the temperature histogram, by adopting the method of edge detection and
extraction. The edge-free distributions allowed separation of canopy and soil
98
temperatures by applying the bimodal Gaussian mixture modelling. The threshold of Twet
and Tdry was determined automatically at 99 % confident interval limits from the canopy
temperature distribution in the modelling.
• The adaptive CWSI was evaluated by presenting strong relationships with plant
physiological measurements: stem water potential (ψstem) and stomatal conductance (gs).
In addition, the adaptive CWSI was compared with a conventional CWSI method
(simply, single reference CWSI), which use single Twet and Tdry by non-water-stress-
baseline derived from VPD and Tc−Ta. The adaptive CWSI showed a higher correlation
to ψstem and gs than the single reference CWSI. The estimated CWSI values were
presented as an analytic map, which enabled to interpret the distributed plant water stress
in high resolution, and to assess spatial variability within the region.
Extended research of the adaptive CWSI
• The research also explored daytime CWSIs and plant physiological parameter (gs) in
water deficit irrigation regime as extended research of adaptive CWSI. Daytime CWSI
was acquired by TIR-UAV sensing at three different times of the day: 9 h, 12 h, and 15 h.
The magnitude of gs changed linearly according to the plant water stress level at three
times of measurements. Midday and afternoon gs were measured at the similar range,
while morning gs showed less conductance rate than the other time.
• In water deficit irrigation plots, midday and afternoon gs responded at lower rates than
morning gs, showing that the water-stressed plants behave with more limited stomatal
conductance during daytime. The estimated CWSIs showed significant correlation with gs
according to the irrigation levels at every flight times of the day in the research. In
particular, morning CWSI was estimated at a reliable level to interpret plant water stress.
The research found that the estimated CWSIs were incorporated according to magnitude
of gs during the day.
• By presenting the relationships between estimated CWSI and plant physiological
measurements taken three times of the day, the research could potentially promise any-
99
daytime CWSI estimation of the day and the extended optimal flight time of UAV
sensing.
Quantifying the plant water use
• The research concentrated on quantifying the plant water use as the form of
evapotranspiration (ET) conducted by TIR-MS-based UAV remote sensing in order to
present the spatial variability of ET. The research estimated ET by using high resolution
mapping of evapotranspiration (HRMET), one of the current surface energy balance
models.
• In order to obtain accurate parameters of the ET model, several steps of pre- and post-
processing were conducted in the research. TIR imagery was calibrated to retrieve actual
surface temperature and MS imagery was radiometrically corrected for producing canopy
structural index. TIR and MS orthoimages were co-registered with geo-reference.
• Photosynthetically active radiation (PAR)-based leaf area index (LAI) was estimated to
obtain the canopy cover parameter of ET model. PAR-based LAI was calculated by a
radiative model of transmitting and scattering radiations. Spatially distributed LAIs were
generated by the exponential relationship with NDVI map.
• Individual trees were segmented systematically by the information of tree- and inter-
spacing. As a result, the estimated ET presented the tree-level ET for interpreting the
variability between plants and allowing the analysis of tree-by-tree water losses.
• The estimated ET showed a strong relationship with measured leaf transpiration. Daily
crop ET was compared with the estimated ET. The result was laid in the expected ET
range, showing a close value to the crop ET.
To sum up, the evaluation results in this research showed that: 1) the adaptive CWSI
method showed strong relationships with stem water potential (ψstem) and stomatal conductance
(gs) as plant physiological measurements. A higher correlation to ψstem and gs was obtained than
100
the single reference CWSI; 2) Diurnal CWSIs showed significant correlation to gs according to
the irrigation levels at three different times of the day (from morning to afternoon); and 3) the
estimated ET was obtained with a strong linear relationship with leaf transpiration (Tr).
This research contributes towards the improvement of the precise and rapid estimation of
the plant water stress level by the adaptive method of CWSI using TIR-based UAV sensing in
order to assess the spatial variability of water stress over heterogeneous agricultural fields. In
addition, this research contributes to the analysis of the tree-by-tree ET which quantifies the
plant water use as well as the assessment of the intra-field variability of ET using TIR-MS-
based UAV sensing.
The main research was conducted by one day UAV campaign in summertime. The main
study site was nectarine and peach orchards as a heterogeneous field. Thus, various crop-
cultivar-site- and time-specific experiments will remain for further research. Future research
will show to what extent the presented approaches are applicable in the area of precise
agricultures.
101
REFERENCES
Acevedo-Opazo, C., Tisseyre, B., Taylor, J. A., Ojeda, H. and Guillaume, S. (2010) A
model for the spatial prediction of water status in vines ( Vitis vinifera L.) using high
resolution ancillary information. Precision Agriculture, 11, 358-378.
Acevedo-Opazo, C., Valdés-Gómez, H., Taylor, J. A., Avalo, A., Verdugo-Vásquez, N.,
Araya, M., Jara-Rojas, F. and Tisseyre, B. (2013) Assessment of an empirical spatial
prediction model of vine water status for irrigation management in a grapevine field.
Agricultural Water Management, 124, 58-68.
Alchanatis, V., Cohen, Y., Cohen, S., Moller, M., Sprinstin, M., Meron, M., Tsipris, J.,
Saranga, Y. and Sela, E. (2010) Evaluation of different approaches for estimating and
mapping crop water status in cotton with thermal imaging. Precision Agriculture, 11.
Alessandro, M., Piero, T., Salvatore Filippo Di, G., Lorenzo, G., Francesco Primo, V.,
Jacopo, P., Claudio, B., Alessandro, Z., Roberto, B. and Beniamino, G. (2015)
Intercomparison of UAV, Aircraft and Satellite Remote Sensing Platforms for Precision
Viticulture. Remote Sensing, 7, 2971.
Allen, R. G. (2000) Using the FAO-56 dual crop coefficient method over an irrigated
region as part of an evapotranspiration intercomparison study. Journal of Hydrology, 229,
27-41.
Allen, R. G., Pereira, L. S., Raes, D., Smith, M. and others (1998) Crop
Evapotranspiration: Guidelines for computing crop water requirements. F. A. O. Irrig. Drain.
Pap., 300, D05109.
Allen, R. G., Tasumi, M. and Trezza, R. (2007) Satellite-Based Energy Balance for
Mapping Evapotranspiration with Internalized Calibration (METRIC)-Model. ASCE
American Society of Civil Engineers, United States, pp. 380.
102
Bastiaanssena, W. G. M., Menentia, M., Feddesb, R. A. and Holtslagc, A. A. M. (1998)
A remote sensing surface energy balance algorithm for land (SEBAL). 1. Formulation.
Elsevier Science Division, Netherlands, pp. 198.
Bellvert, J., Zarco-Tejada, P., Girona, J. and Fereres, E. (2014) Mapping crop water
stress index in a 'Pinot-noir' vineyard: comparing ground measurements with thermal remote
sensing imagery from an unmanned aerial vehicle. Precision Agriculture, 15, 361-376.
Bendig, J., Bolten, A. and Bareth, G. (2012) Introducing a low-cost mini-UAV for
thermal-and multispectral-imaging. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci, 39,
345-349.
Berni, J. A., Zarco-Tejada, P. J., Suárez, L. and Fereres, E. (2009a) Thermal and
narrowband multispectral remote sensing for vegetation monitoring from an unmanned aerial
vehicle. IEEE Transactions on Geoscience and Remote Sensing, 47, 722-738.
Berni, J. A. J., Zarco-Tejada, P. J., Sepulcre-Cantó, G., Fereres, E. and Villalobos, F.
(2009b) Mapping canopy conductance and CWSI in olive orchards using high resolution
thermal remote sensing imagery. Remote sens. environ., 113, 2380-2388.
Bhusal, N., Bhusal, S. J. and Yoon, T.-M. (2018) Comparisons of physiological and
anatomical characteristics between two cultivars in bi-leader apple trees (Malus × domestica
Borkh.). Scientia Horticulturae, 231, 73-81.
Biggar, S. F., Thome, K. J. and Wisniewski, W. (2003) Vicarious radiometric calibration
of EO-1 sensors by reference to high-reflectance ground targets. IEEE Transactions on
Geoscience and Remote Sensing, 41, 1174-1179.
Brenner, C., Thiem, C. E., Wizemann, H.-D., Bernhardt, M. and Schulz, K. (2017)
Estimating spatially distributed turbulent heat fluxes from high-resolution thermal imagery
acquired with a UAV system. International Journal Of Remote Sensing, 38, 3003-3026.
103
Bryson, M., Johnson-Roberson, M., Murphy, R. J. and Bongiorno, D. (2013) Kite aerial
photography for low-cost, ultra-high spatial resolution multi-spectral mapping of intertidal
landscapes. PloS one, 8, e73550.
Buesa, I., Perez, D., Castel, J., Intrigliolo, D. S. and Castel, J. R. (2017) Effect of deficit
irrigation on vine performance and grape composition of Vitis vinifera L. cv. Muscat of
Alexandria. Australian Journal of Grape and Wine Research, 23, 251-259.
Calcagno, G., Mendicino, G., Monacelli, G., Senatore, A. and Versace, P. (2007)
Distributed Estimation of Actual Evapotranspiration through Remote Sensing Techniques.
Kluwer Academic Publishers, Netherlands, pp. 125.
Campbell, G. S. (2000) An introduction to environmental biophysics, Springer.
Carlos Zúñiga, E., Lav, R. K., Sindhuja, S. and Pete, W. J. (2017) High Resolution
Multispectral and Thermal Remote Sensing-Based Water Stress Assessment in Subsurface
Irrigated Grapevines. Remote Sensing, 9, 961.
Carlson, T. N., Gillies, R. R. and Schmugge, T. J. (1995) An interpretation of
methodologies for indirect measurement of soil water content. Netherlands, pp. 191.
Chávez, J. L., Neale, C. M. U., Prueger, J. H. and Kustas, W. P. (2008) Daily
evapotranspiration estimates from extrapolating instantaneous airborne remote sensing ET
values. Irrigation Science, 27, 67-81.
ChonÉ, X., van Leeuwen, C., Dubourdieu, D. and GaudillÈRe, J. P. (2001) Stem Water
Potential is a Sensitive Indicator of Grapevine Water Status. Ann. Bot., 87, 477-483.
Clarke, T., Wang, X. and Fryer, J. (1998) The principal point and CCD cameras. The
Photogrammetric Record, 16, 293-312.
Cohen, Y., Alchanatis, V., Meron, M., Saranga, Y. and Tsipris, J. (2005) Estimation of
leaf water potential by thermal imagery and spatial analysis. J. Exp. Bot., 56, 1843-1852.
104
Colaizzi, P. D., Kustas, W. P., Anderson, M. C., Agam, N., Tolk, J. A., Evett, S. R.,
Howell, T. A., Gowda, P. H. and O’Shaughnessy, S. A. (2012) Two-source energy balance
model estimates of evapotranspiration using component and composite surface temperatures.
Advances in Water Resources, 50, 134-151.
Courault, D., Seguin, B. and Olioso, A. (2005) Review on estimation of
evapotranspiration from remote sensing data: From empirical to numerical modeling
approaches. Irrigation & Drainage Systems, 19, 223-249.
Davies, F. S. and Lakso, A. N. (1979) Diurnal and Seasonal Changes in Leaf Water
Potential Components and Elastic Properties in Response to Water Stress in Apple Trees.
Physiologia Plantarum, 46, 109.
DecagonDevicesInc. (2001) Decagon AccuPar Ceptometer Operating Manual. Decagon
Devices Inc, Pullman, WA, USA.
Del Pozo, S., Rodríguez-Gonzálvez, P., Hernández-López, D. and Felipe-García, B.
(2014) Vicarious radiometric calibration of a multispectral camera on board an unmanned
aerial system. Remote Sensing, 6, 1918-1937.
Douskos, V., Kalisperakis, I. and Karras, G. (2007) Automatic calibration of digital
cameras using planar chess-board patterns. In Proceedings of the 8th Conference on Optical,
pp. 9-12.
Douterloigne, K., Gautama, S. and Philips, W. (2009) Fully automatic and robust UAV
camera calibration using chessboard patterns. In 2009 IEEE International Geoscience and
Remote Sensing Symposium, Vol. 2, pp. II-551-II-554.
Dunford, R., Michel, K., Gagnage, M., Piégay, H. and Trémelo, M. L. (2009) Potential
and constraints of Unmanned Aerial Vehicle technology for the characterization of
Mediterranean riparian forest. International Journal of Remote Sensing, 30, 4915-4935.
105
Elkins, R. B. (2002) Effect of training system and rootstock on growth and productivity
of Golden Russet® Bosc pear trees. Acta horticulturae.
Fan, L., Gao, Y., Brück, H. and Bernhofer, C. (2009) Investigating the relationship
between NDVI and LAI in semi-arid grassland in Inner Mongolia using in-situ
measurements. Springer Science + Business Media, Austria, pp. 151.
Fischer, M. A. and Bolles, R. C. (1981) A Paradigm for Model Fitting with Applications
to Image Analysis and Automated Cartography. Comm. ACM, 24, 381-395.
FLIRSystems (2013) User's manual FLIR Ax5 series. FLIR Systems, Wilsonville, OR,
USA.
Fraser, C. S. (1997) Digital camera self-calibration. ISPRS Journal of Photogrammetry
and Remote sensing, 52, 149-159.
French, A. N., Hunsaker, D. J. and Thorp, K. R. (2015) Remote sensing of
evapotranspiration over cotton using the TSEB and METRIC energy balance models.
Remote Sensing of Environment, 281.
Fuentes, S., Bei, R., Pech, J. and Tyerman, S. (2012) Computational water stress indices
obtained from thermal image analysis of grapevine canopies. Irrig. Sci., 30, 523-536.
Gago, J., Douthe, C., Coopman, R. E., Gallego, P. P., Ribas-Carbo, M., Flexas, J.,
Escalona, J. and Medrano, H. (2015) UAVs challenge to assess water stress for sustainable
agriculture. Agricultural Water Management, 153, 9-19.
Gao, B. C. (1996) NDWI - A normalized difference water index for remote sensing of
vegetation liquid water from space. Remote Sensing of Environment, 58, 257-266.
Glover, J., Johnson, H., Lizzio, J., Wesley, V., Hattersley, P. and Knight, C. (2008)
Australia’s crops and pastures in a changing climate – Can Biotechnology Help?, Australian
Government Bureau of Rural Sciences, Canberra.
106
Gonzalez-Dugo, V., Zarco-Tejada, P., Nicolás, E., Nortes, P., Alarcón, J., Intrigliolo, D.
and Fereres, E. (2013) Using high resolution UAV thermal imagery to assess the variability
in the water status of five fruit tree species within a commercial orchard. Precision
Agriculture, 14, 660-678.
Goodwin, I., Whitfield, D. M. and Connor, D. J. (2004) Relationship between peach tree
transpiration and effective canopy cover. Acta horticulturae, 283-289.
Goodwin, I., Whitfield, D. M. and Connor, D. J. (2006) Effects of tree size on water use
of peach (Prunus persica L. Batsch). Irrigation Science, 24, 59-68.
Grammatikopoulos, L., Karras, G. and Petsa, E. (2007) An automatic approach for
camera calibration from vanishing points. ISPRS journal of photogrammetry and remote
sensing, 62, 64-76.
Haboudane, D., Miller, J. R., Tremblay, N., Zarco-Tejada, P. J. and Dextraze, L. (2002)
Integrated narrow-band vegetation indices for prediction of crop chlorophyll content for
application to precision agriculture. Elsevier Science, United States, pp. 416.
Hastedt, H. and Luhmann, T. (2015) Investigations on the quality of the interior
orientation and its impact in object space for UAV photogrammetry. International archives
of photogrammetry and remote sensing, 40, 321.
Hoffmann, H., Nieto, H., Jensen, R., Guzinski, R., Zarco-Tejada, P. J. and Friborg, T.
(2015) Estimating evapotranspiration with thermal UAV data and two source energy balance
models. Hydrology & Earth System Sciences Discussions, 12, 7469.
Honkavaara, E., Ahokas, E., Hyyppä, J., Jaakkola, J., Kaartinen, H., Kuittinen, R.,
Markelin, L. and Nurminen, K. (2006) Geometric test field calibration of digital
photogrammetric sensors. ISPRS Journal of Photogrammetry and Remote Sensing, 60, 387-
399.
107
Honkavaara, E., Arbiol, R., Markelin, L., Martinez, L., Cramer, M., Bovet, S.,
Chandelier, L., Ilves, R., Klonus, S. and Marshal, P. (2009) Digital airborne
photogrammetry—A new tool for quantitative remote sensing?—A state-of-the-art review
on radiometric aspects of digital photogrammetric images. Remote Sensing, 1, 577-605.
Idso, S. B., Jackson, R. D., Pinter, J. P. J., Reginato, R. J. and Hatfield, J. L. (1981)
Normalizing the stress-degree-day parameter for environmental variability. Agric. Meteorol.,
24, 45-55.
Ihuoma, S. O. and Madramootoo, C. A. (2017) Recent advances in crop water stress
detection. Computers and Electronics in Agriculture, 141, 267-275.
Irmak, S., Haman, D. Z. and Bastug, R. (2000) Determination of Crop Water Stress
Index for Irrigation Timing and Yield Estimation of Corn. Agron. J., 92, 1221-1227.
Jackson, R. D., Idso, S. B., Reginato, R. J. and Pinter, P. J., Jr. (1981) Canopy
temperature as a crop water stress indicator Wheat. Water Resources Research, 17, 1133-
1138.
Jackson, R. D., Reginato, R. J. and Idso, S. B. (1977) Wheat canopy temperature: a
practical tool for evaluating water requirements. Water Resources Research (USA), 13, 651.
Johnson, L. F. (2003) Collection of Ultra High Spatial and Spectral Resolution Image
Data Over California vineyards with a Small UAV. Proceedings of the International
Symposium on Remote Sensing of Environment, CONF 30, 663.
Jones, H. G. (1992) Plants and microclimate : a quantitative approach to environmental
plant physiology, Cambridge University Press, New York, USA.
Jones, H. G. (1999) Use of infrared thermometry for estimation of stomatal conductance
as a possible aid to irrigation scheduling. Agricultural and Forest Meteorology, 95, 139-149.
108
Jones, H. G., Stoll, M., Santos, T., De Sousa, C., Chaves, M. M. and Grant, O. M. (2002)
Use of infrared thermography for monitoring stomatal closure in the field: application to
grapevine. Journal of Experimental Botany, 53, 2249-2260.
Kalma, J. D., McVicar, T. R. and McCabe, M. F. (2008) Estimating Land Surface
Evaporation: A Review of Methods Using Remotely Sensed Surface Temperature Data.
Surveys in Geophysics, 29, 421-469.
Karras, G. E. and Mavrommati, D. (2002) Simple calibration techniques for non-metric
cameras. International archives of photogrammetry remote sensing and spatial information
sciences, 34, 39-46.
Kedzierski, M. and Wierzbicki, D. (2015) Review: Radiometric quality assessment of
images acquired by UAV’s in various lighting and weather conditions. Measurement, 76,
156-169.
Kim, Y., Glenn, D. M., Park, J., Ngugi, H. K. and Lehman, B. L. (2011) Hyperspectral
image analysis for water stress detection of apple trees. Computers and electronics in
agriculture, 77, 155-160.
Kustas, W. and Anderson, M. (2009) Advances in thermal infrared remote sensing for
land surface modeling. Agricultural and Forest Meteorology, 149, 2071-2081.
Kustas, W. P. and Norman, J. M. (1999) Evaluation of soil and vegetation heat flux
predictions using a simple two-source model with radiometric temperatures for partial
canopy cover. Agricultural and Forest Meteorology, 94, 13-29.
Laliberte, A. S. (2008) A Procedure for Orthorectification of Sub-Decimeter Resolution
Imagery Obtained with an Unmanned Aerial Vehicle (UAV). Proceedings of the Annual
Conference - American Society for Photogrammetry and Remote Sensing, 1, 405.
Laliberte, A. S., Herrick, J. E., Rango, A. and Winters, C. (2010) Acquisition,
Orthorectification, and Object-based Classification of Unmanned Aerial Vehicle (UAV)
109
Imagery for Rangeland Monitoring. Photogrammetric Engineering & Remote Sensing, 76,
661-672.
Leinonen, I., Grant, O. M., Tagliavia, C. P. P., Chaves, M. M. and Jones, H. G. (2006)
Estimating stomatal conductance with thermal imagery. Plant, Cell & Environment, 29,
1508-1518.
Leinonen, I. and Jones, H. G. (2004) Combining thermal and visible imagery for
estimating canopy temperature and identifying plant stress. J. Exp. Bot., 55, 1423-1431.
Lelong, C., Burger, P., Jubelin, G., Roux, B., Labbé, S. and Baret, F. (2008) Assessment
of Unmanned Aerial Vehicles Imagery for Quantitative Monitoring of Wheat Crop in Small
Plots. Sensors, 8, 3557.
López-López, R., Ramírez, R. A., Sánchez-Cohen, I., Bustamante, W. O. and González-
Lauck, V. (2011) Evapotranspiration and Crop Water Stress Index in Mexican Husk
Tomatoes (Physalis ixocarpa Brot). In Evapotranspiration-From Measurements to
Agricultural and Environmental Applications InTech, Jiutepec, México.
López, D. H., García, B. F., Piqueras, J. G. and Alcázar, G. V. (2011) An approach to the
radiometric aerotriangulation of photogrammetric images. ISPRS journal of photogrammetry
and remote sensing, 66, 883-893.
Lowe, D. G. (1999) Object recognition from local scale-invariant features. In
Proceedings of the Seventh IEEE International Conference on Computer Vision, Vol. 2, pp.
1150-1157 vol.1152.
Luhmann, T. (2006) Close range photogrammetry : principles, methods and applications,
Dunbeath : Whittles, 2006.
Mason, S., Rüther, H. and Smit, J. (1997) Investigation of the Kodak DCS460 digital
camera for small-area mapping. ISPRS Journal of Photogrammetry and Remote Sensing, 52,
202-214.
110
Meron, M., Sprintsin, M., Tsipris, J., Alchanatis, V. and Cohen, Y. (2013) Foliage
temperature extraction from thermal imagery for crop water stress determination. Precision
Agriculture, 14, 467-477.
Meron, M., Tsipris, J. and Charitt, D. (2003) Remote mapping of crop water status to
assess spatial variability of crop stress. Wageningen, the Netherlands:; Wageningen
Academic, pp. 405.
Meron, M., Tsipris, J., Orlov, V., Alchanatis, V. and Cohen, Y. (2010a) Crop water
stress mapping for site-specific irrigation by thermal imagery and artificial reference
surfaces. Precision Agriculture, 11, 148-162.
Meron, M., Tsipris, J., Orlov, V., Alchanatis, V. and Cohen, Y. (2010b) Crop water
stress mapping for site-specific irrigation by thermal imagery and artificial reference
surfaces. Precision Agriculture, 11, 148-162.
Micol, R., Cinzia, P., Chiara, C., Michele, M., Lorenzo, B., Sergio, C., Stefano, A. and
Roberto, C. (2015) Discriminating Irrigated and Rainfed Maize with Diurnal Fluorescence
and Canopy Temperature Airborne Maps. ISPRS International Journal of Geo-Information,
Vol 4, Iss 2, Pp 626-646 (2015), 626.
Mitch, B., Alistair, R., Fabio, R. and Salah, S. (2010) Airborne vision‐based mapping
and classification of large farmland environments. Journal of Field Robotics, 27, 632-655.
Möller, M., Alchanatis, V., Cohen, Y., Meron, M., Tsipris, J., Naor, A., Ostrovsky, V.,
Sprintsin, M. and Cohen, S. (2007) Use of thermal and visible imagery for estimating crop
water status of irrigated grapevine. J. Exp. Bot., 58, 827-838.
Moran, M. S., Bryant, R., Thome, K., Ni, W., Nouvellon, Y., Gonzalez-Dugo, M., Qi, J.
and Clarke, T. (2001) A refined empirical line approach for reflectance factor retrieval from
Landsat-5 TM and Landsat-7 ETM+. Remote Sensing of Environment, 78, 71-82.
111
Moran, M. S., Clarke, T. R., Inoue, Y. and Vidal, A. (1994) Estimating crop water deficit
using the relation between surface-air temperature and spectral vegetation index. Remote
Sensing of Environment, 49, 246-263.
Naor, A. (2005) Irrigation scheduling of peach-deficit irrigation at different phenological
stages and water stress assessment. In Acta horticulturae Santiago, Chile, pp. 339-350.
Norman, J. M. and Becker, F. (1995) Terminology in thermal infrared remote sensing of
natural surfaces. Netherlands, pp. 153.
Norman, J. M. and Jarvis, P. G. (1975) Photosynthesis in Sitka spruce (Picea sitchensis
(Bong.) Carr.). 5. Radiation penetration theory and a test case. Journal of Applied Ecology
(UK), v. 12, 839.
Norman, J. M., Kustas, W. P. and Humes, K. S. (1995) Source approach for estimating
soil and vegetation energy fluxes in observations of directional radiometric surface
temperature [Agric. For. Meteorol., 77 (1995) 263-293]. Netherlands, pp. 297.
Nouri, H., Beecham, S., Anderson, S., Hassanli, A. M. and Kazemi, F. (2015) Remote
sensing techniques for predicting evapotranspiration from mixed vegetated surfaces. Urban
Water Journal, 12, 380-393.
Ortega-Farías, S., Ortega-Salazar, S., Poblete, T., Kilic, A., Allen, R., Poblete-
Echeverría, C., Ahumada-Orellana, L., Zuñiga, M. and Sepúlveda, D. (2016) Estimation of
Energy Balance Components over a Drip-Irrigated Olive Orchard Using Thermal and
Multispectral Cameras Placed on a Helicopter-Based Unmanned Aerial Vehicle (UAV).
Remote Sensing, 8.
Osroosh, Y., Troy Peters, R., Campbell, C. S. and Zhang, Q. (2015) Automatic irrigation
scheduling of apple trees using theoretical crop water stress index with an innovative
dynamic threshold. Computers and Electronics in Agriculture, 118, 193-203.
112
Paltineanu, C., Septar, L. and Moale, C. (2013) Crop Water Stress in Peach Orchards and
Relationships with Soil Moisture Content in a Chernozem of Dobrogea. Journal of Irrigation
& Drainage Engineering, 139, 20-25.
Pérez, M., Agüera, F. and Carvajal, F. (2012) Digital camera calibration using images
taken from an unmanned aerial vehicle. The International Archives of the Photogrammetry,
Remote Sensing and Spatial Information Sciences, Vol XXXVIII-1/C22, Pp 167-171 (2012),
167.
Poblete-Echeverría, C., Sepulveda-Reyes, D., Ortega-Farias, S., Zuñiga, M. and Fuentes,
S. (2016) Plant water stress detection based on aerial and terrestrial infrared thermography: a
study case from vineyard and olive orchard. In Acta horticulturae Brisbane, Australia, pp.
141-146.
Qureshi, M. E., Hanjra, M. A. and Ward, J. (2013) Impact of water scarcity in Australia
on global food security in an era of climate change. Food Policy, 38, 136-145.
Raman, M. and Himanshu, A. (2009) Study and Comparison of Various Image Edge
Detection Techniques. Int. J. Image Process., 3, 1-11.
Rango, A. (2009) Unmanned aerial vehicle-based remote sensing for rangeland
assessment, monitoring, and management. Journal of Applied Remote Sensing, 3.
Remondino, F. and Fraser, C. (2006) Digital camera calibration methods: considerations
and comparisons. International Archives of Photogrammetry, Remote Sensing and Spatial
Information Sciences, 36, 266-272.
Roberto, T., Riccardo, d. A., Giovanni, M. and Arturo, A. (2005) The effect of deficit
irrigation on seasonal variations of plant water use in Olea europaea L. Plant and Soil, 139.
Roerink, G. J., Su, Z. and Menenti, M. (2000) S-SEBI: A Simple Remote Sensing
Algorithm to Estimate the Surface Energy Balance. Netherlands, pp. 147.
113
Roy, S. and Ophori, D. (2014) Estimation of crop water stress index in almond orchards
using thermal aerial imagery. J. Spat. Hydrol., 12, 29-29.
Rud, R., Cohen, Y., Alchanatis, V., Levi, A., Brikman, R., Shenderey, C., Heuer, B.,
Markovitch, T., Dar, Z., Rosen, C., Mulla, D. and Nigon, T. (2014) Crop water stress index
derived from multi-year ground and aerial thermal images as an indicator of potato water
status. Precision Agriculture, 15, 273-289.
Scholander, P. F., Hammel, H. T., Bradstreet, E. D. and Hemmingsen, E. A. (1965) Sap
Pressure in Vascular Plants. Science, 148, 339-346.
Seguin, B. and Itier, B. (1983) Using midday surface temperature to estimate daily
evaporation from satellite thermal IR data. International Journal of Remote Sensing, 4, 371.
Steltzer, H. and Welker, J. M. (2006) Modeling the effect of photosynthetic vegetation
properties on the NDVI-LAI relationship. Ecological Society of America, United States, pp.
2765.
Stenberg, P., Rautiainen, M., Manninen, T., Voipio, P. and Smolander, H. (2004)
Reduced simple ratio better than NDVI for estimating LAI in Finnish pine and spruce stands.
The Finnish Society of Forest Science, Finland, pp. 3.
Su, Z. (2002) The Surface Energy Balance System (SEBS) for estimation of turbulent
heat fluxes. Hydrology and Earth System Sciences, Vol 6, Iss 1, Pp 85-100 (2002), 85.
Testi, L., Goldhamer, D. A., Iniesta, F. and Salinas, M. (2008) Crop water stress index is
a sensitive water stress indicator in pistachio trees. Irrigation Science, 26, 395-405.
Turner, D., Lucieer, A., Malenovský, Z., King, D. and Robinson, S. (2014a) Spatial Co-
Registration of Ultra-High Resolution Visible, Multispectral and Thermal Images Acquired
with a Micro-UAV over Antarctic Moss Beds. Remote Sensing, 6, 4003.
114
Turner, D., Lucieer, A. and Wallace, L. (2014b) Direct Georeferencing of Ultrahigh-
Resolution UAV Imagery. IEEE Transactions on Geoscience and Remote Sensing, 52, 2738-
2745.
Turner, D., Lucieer, A. and Watson, C. (2012) An Automated Technique for Generating
Georectified Mosaics from Ultra-High Resolution Unmanned Aerial Vehicle (UAV)
Imagery, Based on Structure from Motion (SfM) Point Clouds. Remote Sensing, 4, 1392.
Williams, K. (2005) Digital Vertical Aerial Camera System for High-resolution Site
Inspections in Conservation Easment Monitoring.
Williams, L. E. and Araujo, F. J. (2002) Correlations among Predawn Leaf, Midday
Leaf, and Midday Stem Water Potential and their Correlations with other Measures of Soil
and Plant Water Status in Vitis vinifera. American Society for Horticultural Science, United
States, pp. 448.
Xia, T., Kustas, W. P., Anderson, M. C., Alfieri, J. G., Gao, F., McKee, L., Prueger, J.
H., Geli, H. M. E., Neale, C. M. U., Sanchez, L., Alsina, M. M. and Wang, Z. (2016)
Mapping evapotranspiration with high-resolution aircraft imagery over vineyards using one-
and two-source modeling schemes. Hydrology and Earth System Sciences, Vol 20, Iss 4, Pp
1523-1545 (2016), 1523.
Xue, J. and Su, B. (2017) Significant Remote Sensing Vegetation Indices: A Review of
Developments and Applications. Journal of Sensors, 2017, 17.
Yazar, A., Howell, T. A., Dusek, D. A. and Copeland, K. S. (1999) Evaluation of crop
water stress index for LEPA irrigated corn. Irrigation science.
Yuan, F. and Bauer, M. E. (2007) Comparison of impervious surface area and
normalized difference vegetation index as indicators of surface urban heat island effects in
Landsat imagery. Remote Sensing of Environment, 106, 375-386.
115
Yuei-An, L. and Kumar Kar, S. (2014) Evapotranspiration Estimation with Remote
Sensing and Various Surface Energy Balance Algorithms—A Review. Energies, 7, 2821-
2849.
Zarco-Tejada, P. J., González-Dugo, V. and Berni, J. A. J. (2012) Fluorescence,
temperature and narrow-band indices acquired from a UAV platform for water stress
detection using a micro-hyperspectral imager and a thermal camera. Remote sens. environ.,
117, 322-337.
Zarco-Tejada, P. J., Guillén-Climent, M. L., Hernández-Clemente, R., Catalina, A.,
González, M. R. and Martín, P. (2013) Estimating leaf carotenoid content in vineyards using
high resolution hyperspectral imagery acquired from an unmanned aerial vehicle (UAV).
Agricultural and Forest Meteorology, 171-172, 281-294.
Zhang, W., Jiang, T. and Han, M. (2010) Digital camera calibration method based on
PhotoModeler. In Image and Signal Processing (CISP), 2010 3rd International Congress on,
Vol. 3 IEEE, pp. 1235-1238.
Zhang, Z. (2000) A flexible new technique for camera calibration. IEEE Transactions on
pattern analysis and machine intelligence, 22, 1330-1334.
Zhao-Liang, L., Ronglin, T., Zhengming, W., Yuyun, B., Chenghu, Z., Tang, B.,
Guangjian, Y. and Xiaoyu, Z. (2009) A Review of Current Methodologies for Regional
Evapotranspiration Estimation from Remotely Sensed Data. Sensors, 9, 3801-3853.
Zipper, S. C. and Loheide Ii, S. P. (2014) Using evapotranspiration to assess drought
sensitivity on a subfield scale with HRMET, a high resolution surface energy balance model.
Agricultural and Forest Meteorology, 197, 91-102.
Minerva Access is the Institutional Repository of The University of Melbourne
Author/s:
Park, Suyoung
Title:
Estimating plant water stress and evapotranspiration using very-high-resolution (VHR) UAV
imagery
Date:
2018
Persistent Link:
http://hdl.handle.net/11343/216508
File Description:
Thesis: Estimating Plant Water Stress and Evapotranspiration using Very-High-Resolution
(VHR) UAV Imagery
Terms and Conditions:
Terms and Conditions: Copyright in works deposited in Minerva Access is retained by the
copyright owner. The work may not be altered without permission from the copyright owner.
Readers may only download, print and save electronic copies of whole works for their own
personal non-commercial use. Any use that exceeds these limits requires permission from
the copyright owner. Attribution is essential when quoting or paraphrasing from these works.