Precision Agriculture

https://doi.org/10.1007/s11119-019-09656-8

Improving nitrogen assessment with an RGB camera

across uncertain natural light from above‑canopy
measurements

Bayu Taruna Widjaja Putra1 · Peeyush Soni2 

© Springer Science+Business Media, LLC, part of Springer Nature 2019

Abstract
The farming activities in developing countries are mostly conducted in daytime with vary-
ing intensities of natural light throughout the day. Also, the shade trees can further increase
the uncertainty of natural light exposure on plants. This research proposes an appropri-
ate method to standardize index values obtained from an RGB digital camera for assess-
ing biophysical properties, especially nitrogen content. Nutrient content in plants is an
important factor that characterizes plant yields and health. Determining the status of plant
nutrients often requires field observation. The conventional laboratory methods and remote
sensing applications (i.e. satellite, airborne and spectrometer) are still expensive. Also,
weather and field condition significantly affect the quality of measurement results. The
use of consumer-grade digital cameras has been explored as an alternative low-cost tool
for non-scientific end users; however, the use of a camera for above-canopy measurement
is severely constrained by unfavorable weather condition coupled with limited time avail-
able for the measurement that depends on the intensity of incident light and the condition
of plantation area. Furthermore, shade trees present in plantation areas reduce the quality
of measurement results. By using this newly proposed method, measurement accuracy is
improved and the potential use of Red, Green, and Blue (RGB) cameras during daytime is
explored. Since many studies showed that the Hue index was a potential tool for estimat-
ing biological properties, this study used exposure value (EV) to adjust the digital number
(DN) and Hue index to observe the potential of calibrated and standardized DN and Indices
for estimating greenness of Robusta coffee plants.

Keywords  Broadband greenness · Standardization · Above-canopy measurement ·

Chlorophyll · Nitrogen content · Coffee plant

* Peeyush Soni
soni@iitkgp.ac.in
1
Faculty of Agricultural Technology, Jember University, Jember 68121, East Java, Indonesia
2
Department of Agricultural and Food Engineering, Indian Institute of Technology Kharagpur,
Kharagpur 721302, India

13
Vol.:(0123456789)
 Precision Agriculture

Introduction

Estimating biophysical properties like chlorophyll and nitrogen content of vegetation using
digital cameras may produce different results, depending on the time of measurement.
Even when measurements conducted on the same day, the value of indices may notably
vary according to the time of the day, thus causing erroneous estimation of plant biophysi-
cal properties. Nitrogen (N) management is a key to increasing the sustainability (Masunga
et  al. 2016; Raun and Johnson 1999) of crop production. Both excessive and inadequate
fertilization of N are common problems in Indonesian coffee production. Excessive N ferti-
lization increases pollution and lowers economic return, as well as the quality and quantity
of yields. Inadequate fertilization could make plants more vulnerable to pests and diseases,
thereby reducing production quality and quantity.
Precision agriculture using aerial remote sensing (Aasen et  al. 2015; Jia et  al. 2004;
McNeil et al. 2016; Toth and Jóźków 2015) improves crop productivity by supplying plant-
specific inputs. However, Robusta coffee plantations are the exception for aerial remote
sensing implementation, because they are situated in hilly areas, covered by shade trees,
and contain plants of different growth stages within an area. The use of ground-based sen-
sors is usually recommended to address these problems. The ground-based commercial
nitrogen sensors such as C ­ ropSpec®, Green ­Seekers® and Crop ­Circle® are popularly used
for cereal plants. These active nitrogen sensors are not affected by ambient light intensity
and could be mounted on a tractor or a modified platform, provide information similar to
sensors on aerial vehicles and satellites, and overcome some of the logistical limitations
of remote sensing. However, these tools are still expensive for the farmers/smallholders in
developing countries.
A potential method to overcome these limitations is using consumer-grade digital cam-
eras. Recent studies have successfully assessed the biophysical properties of vegetation
through the direct-leaf measurement using digital cameras (Rigon et al. 2016; Vesali et al.
2015, 2016; Widjaja Putra and Soni 2018). In these studies, the Hue index is shown to
have offered better accuracy than other vegetation indices (VIs). However, compared with
direct-leaf measurement, the above-canopy measurement is more efficient, but it practi-
cally lacks accuracy in fields having shade trees and under uncertain natural light. Without
calibration and standardization of the digital number (DN) or VI for prevailing incident
light intensity, it loses accuracy. A DN of a digital camera image is obtained by extracting
the value of every pixel of each RGB band. Then, these two or three extracted DNs of each
RGB band can be incorporated as indices. Unlike a spectrometer that can be standardized
with the reflectance value through a white panel (spectralon), a digital camera needs to be
calibrated and standardized in different ways. Few studies are reported on how to standard-
ize the incoming light intensity towards the available VIs value obtained from digital cam-
eras through above-canopy measurement. Long-term use of calibration and standardization
models of digital cameras is difficult to achieve, in part because of frequent product upgra-
dation by camera manufacturers. Several studies suggest ways to eliminate the effect of
uncertain illumination using digital cameras. Bourgeon et al. (2016) offered a technique for
radiometric calibration under natural illumination using a color-board checker. Sakamoto
et  al. (2010, 2012) showed the potential of calibrating the DN with relative light inten-
sity (RLI) to eliminate the effect of uncertain illumination, like the gamma factor which
has a non-linear relationship between DN and the intensity of incident light. Gamma is a
non-linear operation used to shifting and scaling the luminance in image (Reinhard et al.
2001). Although these findings show the potential use of the consumer-grade camera for

13
Precision Agriculture

crop monitoring, detailed field assessment is required to evaluate such methods for plant
monitoring from above-canopy measurement under uncertain natural light.
The main objectives of this study were to evaluate the use of a consumer-grade camera
for estimating biophysical properties such as chlorophyll and N contents of Robusta coffee
plants under varying natural light conditions. The use of VI by incorporating multiband is
still considered not optimal in estimating biophysical properties such as chlorophyll and
N content, under uncertain natural light conditions. Hence, exploring the VIs that can be
implemented in various measurement methods, both direct-leaves and above canopy meas-
urements is considered worthy. Further, influence of using different DNs such as calibrated
and standardized, is compared. Considering the importance of calibrated and standardized
DNs obtained from digital camera applications, this study aimed at the following specific
objectives. First, it evaluated the proposed methods that incorporate the EV to standardize
the available VIs for the different natural light condition without any additional sensors.
Second, it examined the DNs (both calibrated and non-calibrated) used in the Hue index
for estimating leaf greenness of Robusta coffee for estimating plant N content across differ-
ent natural light conditions. Lastly, it evaluated the VIs obtained from a commercial-grade
camera at different measurement levels (direct-leaf and above-canopy measurements).

Materials and methods

This study comprised of two sets of experiments. The first experiment evaluated the VI of
colors panel under different natural light, which aimed to determine relationships between
EV and VI of a consumer-grade camera. The second experiment assessed the chlorophyll
and N level of Robusta coffee plants by identifying the greenness level from the above-
canopy measurement.

Panels measurement

To test the sensitivity of VI across different illumination levels, a set of RGB cameras
(Canon IXUS 160) were installed on a platform aboveground (Fig.  1) and placed in an
open field. Cameras were aimed at the nadir position towards the object at various distances
ranging from 0.5 to 1.5 m. This study focused on both green and yellow colors (Seculine
ProDisk II with white balance filter). The camera configuration with auto-white balance,
auto-focus, and auto-exposure time was used in both panel and field measurements. A total
of 120 images were saved with 20 megapixels resolution, in JPEG (3864 × 5152 pixels)
format.
In this experiment, the two panels represented the greenness of leaves, where the green-
ness could be an indicator for estimating the chlorophyll and nitrogen content of the veg-
etation (corresponding to Fig. 3). These panels were used as references for further in situ
measurement of plant leaf greenness and to standardize the VI.

In‑situ measurement

This set of experiments builds on the study of Widjaja Putra and Soni (2017) which
enhanced potential use of cameras from direct-leaf to above-canopy measurement. For
this purpose, a custom platform for attaching the camera was developed (Fig. 1). Direct-
leaf measurement was conducted on the same plant within the same day for comparing

13
 Precision Agriculture

Fig. 1  Image capturing using digital camera with a platform (left). Green and yellow colors across different
illumination from over-exposure to under-exposure (right) (Color figure online)

above-canopy and direct-leaves measurements, while the destructive analysis of N was

done in the laboratory. The destructive analysis examined leaf tissues of each plant that
underwent the previous set of all non-destructive measurements. To minimize error in
identifying plant properties, only plants and leaves that were free from pests and dis-
eases were chosen.
Experiments were conducted at the Indonesia Coffee and Cocoa Research Institute
(ICCRI), Jember, Indonesia (8°15′24.6″S 113°36′45.1″E) located in East Java, Indone-
sia. Robusta coffee plants were chosen randomly from different growth stages, ranging
from 2 to 10 years across different field shading conditions.

Leaves greenness through canopy measurement

A total of 320 images were collected from 40 plants using a digital camera. Average
values from four repeating measurements were used to compute the VIs to represent
each plant’s chlorophyll and N concentrations. Images were captured during 08.00 to
16.00 h. The digital number (DN) of infrared, background exposure and exposure-value
(EV) tends to increase as the distance between the camera and object increases (Saka-
moto et  al. 2011b). In these experiments, the distance between camera and plant was
determined based on the camera’s ability to adequately capture canopy cover or leaf
area of the plant. Wilted leaves can make the canopy cover smaller and increase the
soil or background visibility. The targeted object area or region of interest (ROI) was
targeted within the camera frame. In this in situ measurement, the camera configuration
involved auto white balance, auto-focus, and automatic exposure time. Black cloth was
used as background to avoid error in extracting the image value due to foliage plant and
weeds. For the analysis, the canopy above the cloth background or plant leaves area was
selected as the ROI (Fig. 2).

13
Precision Agriculture

Fig. 2  a Above-canopy measurement images of Robusta coffee plants with different soil background. b
Manual selection of ROI of the canopy/leaves area

Direct‑leaf measurement and N laboratory analysis

For this measurement, leaf number-2 (in drought/rain-fed areas) and leaf number-3 (in irri-
gated areas) were chosen from selected plagiotropic branches counting from the apex, a
total of 15 leaves were subsequently collected and labeled for further destructive examina-
tion (Widjaja Putra and Soni 2018). According to ICCRI (1999), 15 leaves should be taken
from each Robusta coffee plant for laboratory measurement purposes. Thus, these leaves
were measured using direct-leaf measurement before the N laboratory analysis. A SPAD-
502 m was used in assessing Chlorophyll content through direct-leaf measurement. Direct
leaf measurement and laboratory test were subsequently used for data validation.
The N value was determined by destructive measurement following the Kjeldahl pro-
cedure (Muñoz-Huerta et  al. 2013). For sampling, 15 labeled leaves per plant were cho-
sen from previous experiments to generalize the N content in plants. The collected leaves
from each plant were stored using paper bags and brought to ICCRI laboratory for chemi-
cal analysis. The percentage nitrogen content in the leaves was estimated. Taken from 40
plants, a total of 600 leaves were measured in this experiment. Willson. (1985) categorized

13
 Precision Agriculture

the N content of Robusta coffee plants into four different critical levels, namely deficient,
subnormal, normal, and high with the percentage N contents (weight/weight) of: < 1.8%,
1.80–2.70%, 2.71–3.30%, and > 3.30% respectively.

Vegetation indices

Data collected from the field experiment using non-destructive measurement were mainly
tested using exposure-value referenced vegetation indices for estimating chlorophyll and N
concentrations. Previous studies (Rigon et al. 2016; Vesali et al. 2015) demonstrate that the
Hue index provides better results than other vegetation indices in estimating plant green-
ness levels. In this study, Hue was examined using DN, calibrated DN (cDN), exposure
value-adjusted cDN (ev-cDN), exposure-value referenced DN (evDN), exposure-value ref-
erenced cDN (evcDN), and exposure-value referenced DN-adjusted (evDN).
Procedure suggested by Sakamoto et  al. (2010, 2012) are used in calibrating the DN
of JPEG images against relative light intensity (RLI) for plant monitoring. Thus, by using
this method, gamma characteristics of captured images can be corrected using a six-degree
polynomial equation (Eq.  1) and the need for a calibration panel and an additional sun
irradiance sensor to obtain radiometric calibration can be eliminated, like that available in
Micasense ­RedEdge® camera. The two equations (Sakamoto et al. 2011a, b) to obtain cDN
are expressed as following (Eqs. 1 and 2)

RLI = ax6 + bx5 + cx4 + dx3 + ex2 + fx (1)

cDN = αRLI + β (2)

−15 −12 −10
where x is the DN, a  = 2.540  × 10 , b = −  1.325 × 10 , c = 2.383 × 10 ,
d = − 1.374 × 10−8, e = 1.372 × 10−7, f = 6.217 × 10−5, α = 3658, β = 0.1045, and cDN is the
daily median value for daytime. The additional information for recognizing light intensity
of captured images are obtained from EV, as EV can be used for adjusting the cDN (Nguy-
Robertson et al. 2016) (Eqs. 3 and 4).
[ ]
EV = 2 ⋅ log2 (F) − log2 (T) − log2 ISO∕minISO (3)

ev-cDN = cDN × 2EV (4)

EV, F, T, ISO, and minISO correspond to the exposure value of each camera, relative
aperture, exposure time (f-stop), ISO sensitivity of the captured image, and minimum ISO
of the camera, respectively. These values are recorded in the JPEG format header of each
image. This camera has minimum daytime ISO values of 100.
Unlike some previous studies (Sakamoto et  al. 2011a, b, 2012) which applied EV for
DN adjustment, this study proposed three methods for adjusting the Hue index, namely
exposure value − DN (evDN), exposure value − cDN (evcDN) ratios (Eqs. 5 and 6) and
standardization of EV/VI against EV reference (Fig. 3). For the third method, the concept
is expressed in the flowchart (Fig. 3b). Overall, Hue index was examined using these pro-
posed methods. The VIs used in this study are listed in Table 1.
evDN = EV∕DN (5)

evcDN = EV∕cDN (6)

13
Precision Agriculture

Fig. 3  The procedure in standardizing EV/VI against EV reference; proposed concept (a). Flowchart of
standardizing the EV/VI (b)

Table 1  RGB camera-derived vegetation indices

Vegetation index Formula Reference

Hue [
60(G−B)
] Rigon et al. (2016), Vesali et al.
If max(R,G,B) = R, max(R,G,B)−min(R,G,B) (2015)
If max(R,G,B) = G, 120 + [ 60(B−R)
]
max(R,G,B)−min(R,G,B)
If max(R,G,B) = B, 240 + [
60(R−G)
]
max(R,G,B)−min(R,G,B)
cHue or ev-cHue R,G,B altered to ­cDNR, ­cDNG, ­cDNB or R,G,B altered (This study); modified from
to ­cDNR × 2EV, ­cDNG × 2EV, ­cDNB × 2EV Sakamoto et al. (2011b,
2012)
evHue R,G,B altered to EV/DNR, EV/DNG, EV/DNB (This study)
evcHue R,G,B altered to EV/cDNR, EV/cDNG, EV/cDNB (This study)
ev/Hue EV/Hue (This study)
ev/Hue-adjusted EV/Hue adjusted (This study)
ev/cHue-adjusted EV/cHue adjusted (This study)

Results and discussion

Performance of indices with colored panels

In total, 120 images were collected from both green and yellow colors and subsequently
analyzed. Calibrated and non-calibrated DNs of each color were used to estimate the Hue
index and EV/Hue ratios. Figure 4a shows the strong relationship between EV against EV/
cHue, and EV against EV/Hue of both green and yellow colors. However, the cHue index
varies in a narrower range than Hue. Several observations are noted from the results (cor-
responding to Fig. 3). First, the minimum EV of the camera for capturing the objects using
natural light can be identified, which is the intersection point between EV/HueGREEN and
EV/HueYELLOW or using calibrated-Hue (EV/cHueGREEN and EV/cHueYELLOW). Second,
the capabilities of calibrated and non-calibrated Hues to identify different intersect points,

13
 Precision Agriculture

Fig. 4  Relationships between EV and EV/Hue using colored panel and field measurements; Comparison
between Calibrated Hue and uncalibrated Hue using colored panel (a). The uncalibrated Hue (b) spread
wider than the calibrated Hue (c) (Color figure online)

which means that the minimum EV (obtained through the intersection point) of calibrated
Hue and non-calibrated Hue is different. Third, by plotting any samples obtained from field
measurement, the level of exposure condition and plants’ foliage color distribution can be
identified. For example, Fig.  4b, c show calibrated and non-calibrated samples obtained
from above canopy measurements. The samples which are close to the regression line of
green panel indicate more greenness than those that are close to the regression line of yel-
low panel.

Application in a coffee plantation

Effect of light incident at camera

EV is affected mainly by F, T and ISO. During the day, the F and ISO values tend to be
static (Sakamoto et al. 2011a, b). The influence of incoming light intensity at the time of
above-canopy measurement is compared using DNs and calibrated DNs to show the dis-
tribution of coffee plants in the plantation (Fig. 4b, c). In this study, the shade trees were
located in the managed/irrigated areas. Shade trees reduced the amount of incoming light
intensity in the field, indicated by low value of EV. However, yellow or yellowish plant
leaves are found in drought/rain-fed plantations with fewer or no shade, indicated by the
higher value of EV. Thus, plantation areas with insufficient shade trees increased light
intensity. Viewed from an agronomic perspective, evapotranspiration increases due to the
lack of shade trees, which is indicated by wilted plant leaves resulting from the lack of
water (Campanha et al. 2004). Water is needed to assist the absorption of nutrients by the
roots. Limited availability of water in the soil limits the absorption of nutrients, causing
nutritional deficiencies including N.

Performance of digital cameras with SPAD and N laboratory

The proposed method incorporating a consumer-grade camera as low-cost device for pre-
dicting chlorophyll and N content is optimized to represent greenness level of plant leaves.
The relationships between 40 Robusta coffee plants (Plant leaf analysis using Kjeldahl and
SPAD meter values) and proposed indices obtained from the camera were calculated and
compared. The indices comprising of calibrated DN and un-calibrated DN are shown in
Fig. 5. Hue and cHue values respectively showed substantial correlation with SPAD values
and N concentrations. Among calibrated and non-calibrated indices, the same patterns of

13
Precision Agriculture

Fig. 5  Relationships between SPAD and VIs (above), and relationships between N laboratory and VIs
(below)

regression, be it linear or polynomial, are likely to occur. In some indices like evHue and
EV/Hue, EV improved the correlation. Nevertheless, EV/HueAdjusted, which adjusts the Hue
index to same EV for each image (corresponding to Fig. 3), decreased the correlation. Rel-
evant information in the form of JPEG images needed in this study (EV/cHueAdjusted), like
DN information of particular ROI in different exposure. This is because JPEG images are
already refined than unprocessed format (RAW) (Lebourgeois et al. 2008).
The correlation results indicate that the regression models identify differences between
calibrated and non-calibrated indices. To ensure which model can be implemented for a
commercial-grade cameras in different measurement circumstances, like direct-leaf and
above-canopy measurements, the relationships between above-canopy and direct-leaf as
found in previous works (Widjaja Putra and Soni 2017, 2018) should be probed further.

Relationships between above canopy and direct‑leaf measurements using camera

In this step, Hue and cHue indices obtained from above-canopy measurement were evalu-
ated against Hue index obtained from the direct-leaf level. These results showed that the
use of Hues (above-canopy level) were not standardized, compared with SPAD and N con-
tents. These regression lines of cHue of above canopy measurement offer better applicabil-
ity with Hues of direct-leaf levels, which are identified using homogenous artificial lights
and custom chamber. Although these Hues provide higher correlation coefficients (Fig. 5),
it does not mean that these regressions can be used directly. Another measurement model
at the direct-leaf level should be considered. The use of cHue obtained from above-canopy
measurement showed better correlation in estimating chlorophyll and N contents than Hue.
The estimation of chlorophyll and N using corresponding values of above-canopy cHue
and direct-leaf Hue were consistent, although above-canopy Hue provides better correla-
tion with this parameter. By using cHue, the gap of the regression line between above-
canopy and direct-leaf measurements was minimized.
Therefore, standardization of index values needs to be considered for use at any meas-
urement level. Thus, uniformity of values can be obtained  (Fig.  6). In the same vein,

13
 Precision Agriculture

Fig. 6  Comparison between above-canopy and direct-leaf measurements using camera

Sakamoto et  al. (2011a, b, 2012) suggested that indices using calibrated DNs result in
stronger relationships with other sensors in obtaining biophysical parameters.

ROC analysis for evaluating the accuracy of N fertilization

In practice, coffee farmers in several countries resort to two sources of recommendations

when applying fertilizer, which includes using general recommendations and relying on the
analysis of leaf tissue (ICCRI 1999; Winston et al. 2005). However, the laboratory analysis
generates the most appropriate recommendation. In this research, direct-leaf measurement
provided a stronger correlation than above-canopy measurement in assessing N content.
To investigate the performance of suggested methods in estimating N content from
above-canopy measurement, a receiver operating characteristic (ROC) analysis was con-
ducted. In this analysis, the data were split into two event categories based on N laboratory
tests. The first category was applying the fertilizer when the amount of nitrogen was low
and deficient; and the second category was discontinuing the application of fertilizer when
the amount of nitrogen was normal and abundant. Since the use of calibrated DN in Hue
generated standardized results, the indices of cHue, evcHue, and EV/cHue were analyzed
using ROC. The area under the ROC curve (AUC) was used to assess the predictive perfor-
mance of proposed models for evaluation (Coudun et al. 2006). The results of ROC analy-
sis are shown in Fig. 7.

13
Precision Agriculture

Fig. 7  ROC curves and area under curve (AUC) values of indices (above) and sensitivity and specificity of
indices (below)

Figure 7 presents the accuracy of prediction models which are indicated by AUC values.
This information shows that the proposed indices denote reliable basis for deciding as to
how much N Fertilizer should be used for Robusta coffee plant.

Conclusion

Consumer grade digital cameras are an alternative tool for rapid monitoring of biophysical
properties of the plant but their application is severely constrained under varying natural
light condition and different measurement methods. This study has improved the potential
of using consumer-grade digital cameras for measuring chlorophyll and nitrogen contents
of broadleaf tropical Robusta coffee plants under uncertain natural light condition. The
DNs were calibrated using RLI and/or EV to improve the relationship with chlorophyll and
N, under uncertain natural light. It is confirmed that the use of cHue (comprising cDNs) for
above-canopy measurement under uncertain natural lights showed comparable results with
direct-leaf measurement using homogenous artificial light in assessing chlorophyll and
nitrogen contents than Hue using DNs. It is also shown that the minimum threshold of EV
camera can be determined by identifying intersect point of green and yellow indices of a
color panel; which is useful to determine the capability of the camera for use in a particular
range of exposure value under natural lights especially for agro-forestry measurement like
in coffee plantation with very dense shade trees. Thus, EV should be considered as indica-
tor of capturing the plant from above-canopy than timing of measurement alone.
The key conclusions of this study are listed as below.

13
 Precision Agriculture

1. Minimum threshold of EV camera can be determined by identifying the intersect point

of green and yellow indices of the color panel. EV should also be recorded along with
timing of measurement when capturing plants from above-canopy position.
2. Incorporating EV referenced DNs (i.e. EV/Hue, evHue, and evcHue) can improve the
correlation. However, EV standardization should be considered. Incorporating EV ref-
erenced DNs can improve the correlation.
3. The use of calibrated indices acquired using a consumer-grade camera from above-
canopy measurement under uncertain natural lights provides acceptable estimates of
Chlorophyll and Nitrogen contents.

Acknowledgements  The research was supported by the General Directorate of Higher Education (DIKTI-
Indonesia) and the Asian Institute of Technology (Thailand). The authors would also like to acknowledge
the support received from Indonesian Coffee and Cocoa Research Institute (ICCRI) for assisting field data
collection.

References
Aasen, H., Burkart, A., Bolten, A., & Bareth, G. (2015). Generating 3D hyperspectral information with
lightweight UAV snapshot cameras for vegetation monitoring: From camera calibration to qual-
ity assurance. ISPRS Journal of Photogrammetry and Remote Sensing, 108, 245–259. https​://doi.
org/10.1016/j.isprs​jprs.2015.08.002.
Bourgeon, M.-A., Paoli, J.-N., Jones, G., Villette, S., & Gée, C. (2016). Field radiometric calibration of a
multispectral on-the-go sensor dedicated to the characterization of vineyard foliage. Computers and
Electronics in Agriculture, 123, 184–194. https​://doi.org/10.1016/j.compa​g.2016.02.019.
Campanha, M. M., Santos, R. H. S., de Freitas, G. B., Martinez, H. E. P., Garcia, S. L. R., & Finger, F. L.
(2004). Growth and yield of coffee plants in agroforestry and monoculture systems in Minas Gerais,
Brazil. Agroforestry Systems, 63(1), 75–82. https​://doi.org/10.1023/B:AGFO.00000​49435​.22512​.2d.
Coudun, C., Gégout, J.-C., Piedallu, C., & Rameau, J.-C. (2006). Soil nutritional factors improve models of
plant species distribution: an illustration with Acer campestre (L.) in France. Journal of Biogeography,
33(10), 1750–1763. https​://doi.org/10.1111/j.1365-2699.2005.01443​.x.
ICCRI. (1999). Field School of Integrated Pest Management—Field Guide 2: Coffee Commodity. Material
Set of CRASH COURSE: Book Edition 1. Jember, Indonesia: Indonesian Coffee and Cocoa Research
Institute.
Jia, L., Buerkert, A., Chen, X., Roemheld, V., & Zhang, F. (2004). Low-altitude aerial photography for
optimum N fertilization of winter wheat on the North China Plain. Field Crops Research, 89(2–3),
389–395. https​://doi.org/10.1016/j.fcr.2004.02.014.
Lebourgeois, V., Bégué, A., Labbé, S., Mallavan, B., Prévot, L., & Roux, B. (2008). Can commercial digital
cameras be used as multispectral sensors? A crop monitoring test. Sensors, 8(11), 7300–7322. https​://
doi.org/10.3390/s8117​300.
Masunga, R. H., Uzokwe, V. N., Mlay, P. D., Odeh, I., Singh, A., Buchan, D., et al. (2016). Nitrogen miner-
alization dynamics of different valuable organic amendments commonly used in agriculture. Applied
Soil Ecology, 101, 185–193. https​://doi.org/10.1016/j.apsoi​l.2016.01.006.
McNeil, B. E., Pisek, J., Lepisk, H., & Flamenco, E. A. (2016). Measuring leaf angle distribution in broad-
leaf canopies using UAVs. Agricultural and Forest Meteorology, 218–219, 204–208. https​://doi.
org/10.1016/j.agrfo​rmet.2015.12.058.
Muñoz-Huerta, R. F., Guevara-Gonzalez, R. G., Contreras-Medina, L. M., Torres-Pacheco, I., Prado-Oli-
varez, J., & Ocampo-Velazquez, R. V. (2013). A review of methods for sensing the nitrogen status
in plants: Advantages, disadvantages and recent advances. Sensors, 13(8), 10823–10843. https​://doi.
org/10.3390/s1308​10823​.
Nguy-Robertson, A. L., Buckley, E. M. B., Suyker, A. S., & Awada, T. N. (2016). Determining factors that
impact the calibration of consumer-grade digital cameras used for vegetation analysis. International
Journal of Remote Sensing. https​://doi.org/10.1080/01431​161.2016.11990​61.
Raun, W. R., & Johnson, G. V. (1999). Improving nitrogen use efficiency for cereal production. Agronomy
Journal, 91(3), 357–363.

13
Precision Agriculture

Reinhard, E., Adhikhmin, M., Gooch, B., & Shirley, P. (2001). Color transfer between images. IEEE Com-
puter Graphics and Applications, 21(4), 34–41. https​://doi.org/10.1109/38.94662​9.
Rigon, J. P. G., Capuani, S., Fernandes, D. M., & Guimarães, T. M. (2016). A novel method for the esti-
mation of soybean chlorophyll content using a smartphone and image analysis. Photosynthetica, 1–9.
https​://doi.org/10.1007/s1109​9-016-0214-x.
Sakamoto, T., Gitelson, A. A., Nguy-Robertson, A. L., Arkebauer, T. J., Wardlow, B. D., Suyker, A. E., et al.
(2012). An alternative method using digital cameras for continuous monitoring of crop status. Agricul-
tural and Forest Meteorology, 154–155, 113–126. https​://doi.org/10.1016/j.agrfo​rmet.2011.10.014.
Sakamoto, T., Gitelson, A. A., Wardlow, B. D., Arkebauer, T. J., Verma, S. B., Suyker, A. E., et al. (2011a).
Application of day and night digital photographs for estimating maize biophysical characteristics. Pre-
cision Agriculture, 13(3), 285–301. https​://doi.org/10.1007/s1111​9-011-9246-1.
Sakamoto, T., Shibayama, M., Kimura, A., & Takada, E. (2011b). Assessment of digital camera-derived
vegetation indices in quantitative monitoring of seasonal rice growth. ISPRS Journal of Photogramme-
try and Remote Sensing, 66(6), 872–882. https​://doi.org/10.1016/j.isprs​jprs.2011.08.005.
Sakamoto, T., Shibayama, M., Takada, E., Inoue, A., Morita, K., Takahashi, W., et  al. (2010). Detecting
seasonal changes in crop community structure using day and night digital images. Photogrammetric
Engineering & Remote Sensing, 76(6), 713–726. https​://doi.org/10.14358​/PERS.76.6.713.
Toth, C., & Jóźków, G. (2015). Remote sensing platforms and sensors: A survey. ISPRS Journal of Photo-
grammetry and Remote Sensing, 115, 22–36. https​://doi.org/10.1016/j.isprs​jprs.2015.10.004.
Vesali, F., Omid, M., Kaleita, A., & Mobli, H. (2015). Development of an android app to estimate chloro-
phyll content of corn leaves based on contact imaging. Computers and Electronics in Agriculture, 116,
211–220. https​://doi.org/10.1016/j.compa​g.2015.06.012.
Vesali, F., Omid, M., Mobli, H., & Kaleita, A. (2016). Feasibility of using smart phones to estimate chloro-
phyll content in corn plants. Photosynthetica, 1–8. https​://doi.org/10.1007/s1109​9-016-0677-9.
Widjaja Putra, B. T., & Soni, P. (2017). Evaluating NIR-Red and NIR-Red edge external filters with digital
cameras for assessing vegetation indices under different illumination. Infrared Physics and Technol-
ogy, 81, 148–156. https​://doi.org/10.1016/j.infra​red.2017.01.007.
Widjaja Putra, B. T., & Soni, P. (2018). Enhanced broadband greenness in assessing Chlorophyll a and b,
Carotenoid, and Nitrogen in Robusta coffee plantations using a digital camera. Precision Agriculture,
19(2), 238–256. https​://doi.org/10.1007/s1111​9-017-9513-x.
Willson, C. (1985). Coffee: botany, biochemistry and production of beans and beverage. In M.
N. Clifford & K. C. Willson (Eds.), (pp. 135–156). Boston, MA: Springer US. https​://doi.
org/10.1007/978-1-4615-6657-1_6.
Winston, E., Op de Laak, J., Marsh, T., Lempke, H., & Chapmen, K. (2005). Arabica coffee manual for Lao.
RAP Publication 2005/08. Bangkok, Thailand: FAO Regional Office for Asia and the Pacific.

Publisher’s Note  Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

13