See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/342761485

Integrated method for rice cultivation monitoring using Sentinel-2 data and
Leaf Area Index

Article in The Egyptian Journal of Remote Sensing and Space Science · July 2020
DOI: 10.1016/j.ejrs.2020.06.007

CITATIONS READS

38 1,628

8 authors, including:

Abdelraouf M Ali I. Yu. Savin

National Authority for Remote Sensing and Space Sciences V.V. Dokuchaev Soil Science Institute
42 PUBLICATIONS 767 CITATIONS 196 PUBLICATIONS 3,341 CITATIONS

SEE PROFILE SEE PROFILE

Anton Poddubsky Nasser H.S Saleh

Peoples' Friendship University of Russia National Authority for Remote Sensing and Space Sciences
24 PUBLICATIONS 167 CITATIONS 19 PUBLICATIONS 284 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Khaled Abubakr Ali Abutaleb on 11 July 2020.

The user has requested enhancement of the downloaded file.

 The Egyptian Journal of Remote Sensing and Space Sciences xxx (xxxx) xxx

Contents lists available at ScienceDirect

The Egyptian Journal of Remote Sensing and Space Sciences

journal homepage: www.sciencedirect.com

Integrated method for rice cultivation monitoring using Sentinel-2 data

and Leaf Area Index q
Abdelraouf M. Ali a,b,⁎, Igor Savin a, Anton Poddubskiy b, Mohamed Abouelghar b, Nasser Saleh b,
Khaled Abutaleb b,c,d, Mohammed El-Shirbeny b,e, Peter Dokukin a
a
Agrarian-Technological Institute of the Peoples’ Friendship University of Russia, ul. Miklukho-Maklaya 6, Moscow 117198 Russia
b
National Authority for Remote Sensing and Space Sciences, 23 Joseph Tito, El-Nozha Elgedidah, Cairo, Egypt
c
Institute for Soil, Climate & Water, ARC, Pretoria, South Africa
d
University of Witwatersrand, Johannesburg, South Africa
e
Arab Organization for Agricultural Development, Cairo Office, Egypt

a r t i c l e i n f o a b s t r a c t

Article history: it is necessary to apply a remote sensing-based system for rice cultivation assessment parallel with the
Received 6 May 2020 field measurements of the crop biophysical parameters. This study aims to map the rice cultivated areas
Revised 18 June 2020 and give an estimate for the expected yield (ton/ha) using Sentinel-2 satellite data. The study was carried
Accepted 25 June 2020
out in an experimental site in the Kafr El-Sheikh governorate with a total area of 3240 ha. The multi-
Available online xxxx
temporal Normalized Difference Vegetation Index (NDVI) extracted from nine Sentinel-2 imagery cover
the whole summer season. The supervised nearest neighborhood object-based classification method was
Keywords:
employed, resulting in a classification map with an overall accuracy of 95% and a kappa coefficient of
Sentinel-2
NDVI
0.93. Yield prediction was carried out by using an empirical yield prediction model using the NDVI
LAI and the Leaf Area Index (LAI). The LAI was calculated using the Surface Energy Balance Algorithm for
Rice Land (SEBAL) model and then validated against the measured LAI. the Mean Absolute Percentage Error
Yield (MPAE) was calculated to estimate the error between the measured and predicted LAI and yield. The
MPAE was found to be ±6.76% (i.e. ±0.28 m2/m2) with a high correlation between the measured and
the calculated LAI with a coefficient of determination (R2 = 0.94). While for the yield, the MPAE was
found to be ±6.53% (i.e. ±0.66 ton/ha) and R2 of 0.95. This method is applicable to estimate area and yield
of rice in the northern Nile delta in adequate time before harvest.
Ó 2020 National Authority for Remote Sensing and Space Sciences. Production and hosting by Elsevier
B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).

1. Introduction off; as available arable land is becoming scarce, and competition

for other land uses evolve. This has resulted in intensified practices
Rice is the main staple food crop for a considerable percentage for additional crop cycles and production, amplifying water utiliza-
of the global population besides its role in global biogeochemical tion, and management to enhance rice productivity.
processes (Change and Security, 2016). Rice agriculture may face In Egypt, rice is the dominant summer crop in the Nile Delta
significant challenges in the coming decades due to the increasing because of its low cultivation costs comparing with other summer
resource pressures, severe weather because of climate change, field crops such as maize and cotton. It is the main food for about
population growth, and economic development (FAO, 2018). Over 50% of Egyptians, especially in the Nile Delta and northern Egypt
one billion people depend on rice to support diets and livelihoods. (Ghidan and Yacout, 2016). The total rice area in Egypt is
At the same time, the total accumulated area of rice has tapered 510,649 ha in 2015 produced 326,429 tons, as reported by the Min-
istry of Agriculture (FAO, 2015). Rice is cultivated mainly in the
northern Nile Delta, where the soil is characterized by high salinity
Peer review under responsibility of National Authority for Remote Sensing and
Space Sciences.
because of seawater intrusion from the Mediterranean Sea (El-kiki,
⁎ Corresponding author at: National Authority for Remote Sensing and Space 2019). This situation forces large areas of the northern Nile Delta to
Sciences (NARSS) & Agrarian-Technological Institute of the Peoples’ Friendship cultivate rice in summer as the enormous amount of irrigation
University of Russia, ul. Miklukho-Maklaya 6, Moscow. water that is consumed by rice enables periodical soil leaching
E-mail addresses: raouf.shoker@narss.sci.eg (A.M. Ali), m.elshirbeny@narss.sci.eg
and sustains a low level of soil salinity. Kafr El-Sheikh governorate
(M. El-Shirbeny).

https://doi.org/10.1016/j.ejrs.2020.06.007
1110-9823/Ó 2020 National Authority for Remote Sensing and Space Sciences. Production and hosting by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Please cite this article as: A. M. Ali, I. Savin, A. Poddubskiy et al., , The Egyptian Journal of Remote Sensing and Space Sciences, https://doi.org/10.1016/j.
ejrs.2020.06.007
2 A.M. Ali et al. / Egypt. J. Remote Sensing Space Sci. xxx (xxxx) xxx

is located in the northern Nile Delta. It is one of the top rice produc- and between 30° 200 36.8300 and 31° 170 15.1600 East. It is bordered
ers in Egypt. It produces 13,749 tons from 104,578 ha as reported by the Mediterranean Sea and Burulus Lake from the North, Behira
in 2015 (USDA, 2015). Starting from 2018(Sayed et al., 2019), the governorate from the west, Dakhlia governorate from the east, and
Egyptian government launched new legislation to reduce the rice Gharbia governorate from the south, as shown in Fig. 1.
area to face the problem of the expected scarcity of irrigation water As a part of the Kafr El-Sheikh Governorate, field measurements
because of the construction of the Millennium dam on Ethiopia. of rice biophysical parameters and yield were carried out in the
This may affect summer crop composition in enormous areas of Sakha experimental station with a total area of approximately
Egypt besides its effect on the soil characteristics (Amer et al., 3240 ha.
2017). The current concern of the agricultural research community
is how to keep an adequate demand for rice production from less 2.2. Satellite data processing
rice cultivated area at the same time how to cope with the seawa-
ter intrusion that expected to increase due to the increase of the Cultivated areas in Nile delta is very dense and heterogeneous;
sea-level rise because of climate change. This relatively complex therefore, high spatial resolution satellite imagery (<= 20 m) is nec-
situation calls for an integrated system for rice crop monitoring essary to accurately isolate any vegetation cover under investiga-
where satellite imagery and ground measurements could be tion in the Nile delta. Identification of a specific crop in an
employed powerfully (Ahmad et al., 2014; Boschetti et al., 2017). intensively cultivated area requires multi-temporal satellite
One of the main components of rice crop monitoring is yield images (Ali et al., 2018a,2018b) that cover the whole growing sea-
estimation. Basically, the yield of the main cereal crops is a global son. Sentinel-2 satellite provides multispectral images with high
issue related directly to food security and human life (Zhang et al., temporal and three different spatial resolutions (Table 1).
2015; Liu et al., 2018). Crop yield estimation in adequate time Nine images were downloaded from the ESA’s Earth Observing
before harvest ensures more effective agricultural management System Data and Information System website. The selected images
policies. The ground-based method for crop yield estimation is were selected to cover the whole rice growing stages of the rice
commonly used worldwide; however, it is expensive and time- crop (preparation, emergency, tillering, flowering, repining, and
consuming. Therefore, applying an accurate, cost-effective, and harvesting) during the season of 2018 with no clouds cover the
rapid method for crop yield prediction at local and regional scales study area (Table 2). Images pre-processing stages such as atmo-
is one of the main objectives of remote sensing agricultural appli- spheric correction, layer stacking, and study area clipping were
cations (Ahmad et al., 2014; Dong et al., 2016; Mosleh, 2015; performed using ENVI software. The atmospheric correction was
Noureldin et al., 2013; Paper, 2015). carried out using the FLAASH module for Sentinel-2 images. This
The second component of the rice crop monitoring system is the module relies upon standard condition for unearthly brilliance at
crop biophysical parameters that are closely related to the crop a sensor pixel, L. The conditions of this module are as per the fol-
vigor and vegetative health. These parameters are linked to crop lowing Eq. (1) as indicated by (Frouins et al., 1997).
productivity and crop conditions. Remote sensing is a powerful


Aq Bqe
mean to study crop biophysical, biochemical, and phonological L¼ þ þ La ð1Þ
1  qe S 1  qe S
parameters such as biomass (Gitelson, 2016: Verger et al., 2014),
Leaf Area Index (LAI) (Aboelghar et al., 2011, 2014; Ali and where: q is the pixel reflectance, qe is the typical pixel reflectance, S
Aboelghar, 2019: Liang, 2014), and chlorophyll content is the atmospheric spherical albedo, La is the backscattered radiance,
(Haboudane et al., 2002). These parameters are linked with remo- A and B are coefficients that rely upon barometrical and geometric
tely sensed data represented as vegetation indices (VIs). conditions.
Many studies analyzed the relationship between one vegetation Image stacking is a process of combining multi-layers into a sin-
index and one or more of these parameters. Some studies analyzed gle output file. Normalized Difference Vegetation Index was calcu-
the correlation between remotely sensed and measured parame- lated, and a stacked image file was created through stacking the
ters with crop yield. A linear correlation was found between nor- nine NDVI images using ENVI software.
malized difference vegetation index (NDVI) and grain yield
(Tucker and Holben, 1981), linear relationship addressed the corre- 2.3. Mapping area extent of rice crop
lation between yield and band ratios (NIR/ Green) and (NIR/ Red)
(Zhang et al.,2015). Absorbed photosynthetically active radiation This methodology depends on the following steps that will be
(PAR) was used as an estimator for crop yield. For homogeneous described in separate subsections.
large cultivated areas, Landsat and SPOT5 imagery with an ade-
quate spatial resolution of 30 m and 10 m were used for crop yield (i) Generating NDVI time series
prediction (Awad, 2019; Ahmad et al., 2014; Yang et al., 2004: (ii) Masking non-cultivated areas
Pradhan et al., 2014). (iii) Images segmentation and classification
The objective of the current study is to propose a remote (iv) Classification accuracy assessment and final map
sensing-based system for rice yield forecasting using satellite data production.
and ground measurements. The proposed models are applicable
more than a month before harvest in any rice production area of 2.3.1. Generating NDVI time series
the Nile delta region of Egypt. The normalized difference vegetation index was calculated
using the red and the near-infrared band with 10 m spatial resolu-
tions according to Eq. (2).
2. Material and methods qir  qr
NDVI ¼ ð2Þ
qir þ qr
2.1. Study area
where qir is the near-infrared band reflectance and qr is the red
The study was carried out at the Kafr El-Sheikh Governorate. band reflectance.
Kafr El-Sheikh is located in the northern part of the Nile Delta, Every single date was processed separately. Then the final out-
between Rosetta and Damietta Nile branches. The location of the puts were stacked in one file. This resulted in a continuous
Governorate is between 31° 300 7.5900 and 31° 90 58.0900 North sequence of NDVI temporal values for each pixel during the grow-

Please cite this article as: A. M. Ali, I. Savin, A. Poddubskiy et al., , The Egyptian Journal of Remote Sensing and Space Sciences, https://doi.org/10.1016/j.
ejrs.2020.06.007
 A.M. Ali et al. / Egypt. J. Remote Sensing Space Sci. xxx (xxxx) xxx 3

Fig. 1. The study area (Sakha Agriculture research station - Kafr El-Sheikh governorate).

ing season. The calculated NDVI was used to determine the spectral
Table 1
Sentinel-2 satellite bands name, Id, wavelength and spatial resolution. response of rice during different phonological stages.

Band Name Band Id Bandwidth Resolution

(mm) (m) 2.3.2. Masking non-cultivated areas
Coastal aerosol Band 1 443 60 The multi-temporal NDVI product was used to mask the non-
Blue Band 2 490 10 cultivated areas, where an NDVI maximum and standard deviation
Green Band 3 560 10
Red Band 4 665 10
values were used as a threshold. Pixels with maximum and stan-
Red Edge Band 5 705 20 dard deviation NDVI value less than 0.2 were masked as non-
Red Edge Band 6 740 20 cultivated areas.
Red Edge Band 7 783 20
Near InfraRed Band 8 842 10
Red Edge Band 8a 865 20 2.3.3. Images segmentation and classification
Water Vapour Band 9 940 60
SWIR-Cirrus Band 10 1375 60
Image segmentation is the process of grouping pixels with the
SWIR Band 11 1620 20 same or closer attributes together as an object. The multi-
SWIR Band 12 2190 20 resolution segmentation impeded in eCognition software was
applied on a stack of the nine NDVI layers and the four 10 m band
of August 15th, 2018. The reason to include these four bands is the
fact that this specific date represents the heading stage with the
Table 2
maximum NDVI values for the rice crop areas. The multi-
Sentinel-2 images selected dates.
resolution segmentation algorithm depends on the scale parame-
Scene ID Acquisition date ter, shape, compactness, and the weight associated with each layer.
36RTV 19-May-2018 Due to the small sizes of the rice fields in the study area, the value
36RTV 24-May-2018 of ten was used as the scale parameter, while all the layers were
36RTV 08-Jun-2018 weighted equally. The maximum and standard deviation for NDVI
36RTV 21-Jul-2018
36RTV 15-Aug-2018
values of rice cultivation were used as additional input factors to
36RTV 21-Jul-2018 maximize the accuracy of crop classification and isolation of rice
36RTV 25-Aug-2018 cultivations. Also, the non-cultivated area was used as an exclusion
36RTV 19-Sep-2018 layer to guide the classification and reduce the interferences with
36RTV 29-Sep-2018
the non-cultivated area dominated by weeds and grasses.

Please cite this article as: A. M. Ali, I. Savin, A. Poddubskiy et al., , The Egyptian Journal of Remote Sensing and Space Sciences, https://doi.org/10.1016/j.
ejrs.2020.06.007
4 A.M. Ali et al. / Egypt. J. Remote Sensing Space Sci. xxx (xxxx) xxx

After segmentation, the supervised nearest neighborhood algo- moderate agreement = (0.40) to (0.60), good agreement = (0.60)
rithm (with manual sample selection) was implemented for to (0.80), very good agreement = (0.80) to (1.00). During the field-
object-based image classification. The collected ground data were work, 770 points (102 fields) representing the used classes were
divided into two groups. The first group contains 70% of the collected collected using a hand-held global positioning system (GPS).
points, and it was used for training. While the other 30% was used for Fig. 2 below shows the spatial distribution of the collected points.
accuracy assessment. The collected points were transformed to
polygons by overlaying these points on the NDVI composite, and 2.4. LAI estimation
the field boundaries corresponding to these points were digitized
using the on-screen digitizing method. The training data includes Leaf Area Index (LAI) is the ratio of the total area of all leaves on
representative samples for three different classes (rice, maize, and a plant to the ground area represented by the plant. It is an indica-
cotton). For this technique, we considered setting the convergence tor of biomass and canopy resistance. LAI was computed using sur-
threshold of 0.85 as recommended by (Kalacska et al., 2018; Ali face energy balance (SEBAL., 2002) algorithm of land Eq. (6):
et al.,2018b, 2018a) with the infinite number of iterations (iterations

0:069  SAVI
continue until reaching the convergence threshold). LAI ¼ ln =0:9 ð5Þ
0:59

2.3.4. Classification accuracy assessment where; SAVI is the soil adjusted vegetation, index calculated from
Accuracy assessment was conducted using the Kappa coeffi- Eq. (6) using a value of 0.5 for L.
cient and the error matrix to represent each class’s accuracy and
 
the overall accuracy. This analysis is a distinct multivariate proce- ðNir  RedÞ
SAVI ¼ þ ð1 þ L ð6Þ
dure used to determine classification accuracy from the error ðNIR þ REd þ LÞ
matrix. It generates a Kappa coefficient that has a probable range
from 0 to 1 Eq. (3), according to Congalton (1991). 30% of random
2.5. Rice yield map
points for the whole study test areas were set aside for accuracy
assessment These points were overlaid on the respective seg-
After masking the spatial distribution of the rice fields, the rice
mented images in ARCGIS. The segments with the points were
yield was calculated as a function of LAI and NDVI. Using the tem-
selected and the respective attribute tables joined. The ‘‘new”
poral profile of NDVI of the rice crop extracted from sentinel-2
object shapefile is converted to a raster, which was then converted
imagery (the peak greenness of rice which corresponding to the
to thematic layer in Envi software. The thematic layer was then
heading stage) was determined to be August 15th NDVI layer
imported to eCognition as training or test areas (TTA) file and used
(Fig. 3). Moreover, as mentioned before, the LAI was also estimated
for the accuracy assessment.
and validated using the data correspond to the same date.
Pr Pr The rice yield map was estimated according to reported and
N i¼1 xij  i¼1 ðxi  xj Þ
K¼ P ð3Þ tested the empirical statistical model by (Noureldin et al., 2013),
N2  ri¼1 ðxi  xj Þ
as described in Eq. (7). The model was applied one month before
There is one probable clarification of Kappa Coefficient: poor harvesting. This model was generated and validated in the same
agreement = less than (0.20), fair agreement = (0.20) to (0.40), study area.

Fig. 2. Distribution of ground control point for image classification and accuracy assessment.

Please cite this article as: A. M. Ali, I. Savin, A. Poddubskiy et al., , The Egyptian Journal of Remote Sensing and Space Sciences, https://doi.org/10.1016/j.
ejrs.2020.06.007
 A.M. Ali et al. / Egypt. J. Remote Sensing Space Sci. xxx (xxxx) xxx 5

Fig. 3. Phonological stages of a typical rice crop and corresponding NDVI values.

Fig. 4. Spatial distribution of LAI and yield points as collected from the field.

Y ¼ 3:110 þ 1:684  ðLAIÞ þ 12:458  ðNDVIÞ ð7Þ 2.6. LAI and yield validations

where Y is the estimated yield, while (LAI): is the leaf area index, it Error measurement is usually used to compare models and
was calculated from Eq. (5), and NDVI: is the Normalize Difference test model accuracy. Many statistical methods were well known
vegetation index, and it is calculated from the peak greenness of for error measurements, such as the root square mean error
rice crop growth stages. (RSME), the mean absolute error (MAE), and the mean absolute

Please cite this article as: A. M. Ali, I. Savin, A. Poddubskiy et al., , The Egyptian Journal of Remote Sensing and Space Sciences, https://doi.org/10.1016/j.
ejrs.2020.06.007
6 A.M. Ali et al. / Egypt. J. Remote Sensing Space Sci. xxx (xxxx) xxx

Fig. 5. NDVI spatial distribution over the growing season of 2018.

percentage error (MAPE). In this study, the MAPE was used to test of the view rings). It is not possible to restrict either the azimuth or
and validate the LAI, and the yield used models (Eq. (8)). The the zenith view angles, as described by (Ecology et al., 2003).
MAPE has the advantage of scale independence and gives the Five measurements were taken in each field and then averaged
error as a percentage from the actual value, so it becomes easy to one. These measurements were used to validate the calculated
to compare and test models. MAPE provides more of a standard- LAI using the SEBAL equation. The yield validation was carried
ized error measure. out using data from 29 farms after harvest. The obtained data
was the total yield per field; then it was averaged as ton/ha for
 ! each field. The spatial distribution of both LAI and yield measure-
1 X  Ay  F y 
MAPE ¼    100 ð8Þ ments is represented in Fig. 4.
n Ay 

3. Results and discussion

where MAPE is the mean absolute percentage error, n is the sample
size Ay and F y are the actual and forecasted yield, respectively. 3.1. Generating NDVI time series
On August 15th, a total number of 26 rice fields were visited to
measure the LAI using the hand-held Ceptometer lP-80. This device The spatial distribution of NDVI over the study area is illus-
measures the LAI of small plots, gapped, and non-uniform canopies trated in Fig. 5. The NDVI values ranged from 0.1 to 0.83 through-
and easily by restricting the view in terms of both azimuth (with out the season. Not only is the value varying but also the spatial
view restricting caps) and zenith angles (by excluding one or more extent is increasing throughout the growing season as the crop

Please cite this article as: A. M. Ali, I. Savin, A. Poddubskiy et al., , The Egyptian Journal of Remote Sensing and Space Sciences, https://doi.org/10.1016/j.
ejrs.2020.06.007
 A.M. Ali et al. / Egypt. J. Remote Sensing Space Sci. xxx (xxxx) xxx 7

Fig. 6. Map of cultivated and non-cultivated land.

Fig. 7. Spatial distribution of different crops in the study area.

Please cite this article as: A. M. Ali, I. Savin, A. Poddubskiy et al., , The Egyptian Journal of Remote Sensing and Space Sciences, https://doi.org/10.1016/j.
ejrs.2020.06.007
8 A.M. Ali et al. / Egypt. J. Remote Sensing Space Sci. xxx (xxxx) xxx

Table 3
The error matrix for the accuracy assessment.

Cotton Rice Maize Total

Cotton 33 3 0 36
Rice 0 129 0 129
maize 0 6 60 66
Total 33 138 60 231

Fig. 8. Spatial distribution of rice area.

ber. The high temporal access to NDVI data provides more details
regarding the agriculture season. Data represented in Fig. 5 reflect
that the different stages of the growing season also provide details
about areas left abundant during the growing season.

3.2. Masking the non-cultivated areas

Using the NDVI maximum and the standard deviation, non-

cultivated areas were masked, as shown in Fig. 6. The maximum
NDVI and standard deviation values used to discriminate between
cultivated and non-cultivated areas were 0.2 throughout the sea-
son. An approximation of 33.5% of the study area was found to
be non-cultivated fields (1084.62 ha). This area includes the urban
and built-up areas found in the study area as it has an NDVI value
below zero.

3.3. Spatial distribution of rice area

The object-based classification method using the NDVI time ser-

Fig. 9. Correlation between satellite-based LAI and field-measured. ies was accurate to discriminate between different crops and dis-
tinguishes the spatial distribution of the rice crop in the study
area. Fig. 7 presents the final map for area classification. The rice
grows until August 15th, which is the peak greenness stage (head- cultivated areas occupied 52.4% of the study area (1699.08 ha).
ing stage). After this stage, the plant goes to mature, and the NDVI This followed by the area cultivated by maize, which found to
starts again to decline, which starts late August and early Septem- occupy 10.1% (327.35 ha), and cotton come at last with 4%

Please cite this article as: A. M. Ali, I. Savin, A. Poddubskiy et al., , The Egyptian Journal of Remote Sensing and Space Sciences, https://doi.org/10.1016/j.
ejrs.2020.06.007
 A.M. Ali et al. / Egypt. J. Remote Sensing Space Sci. xxx (xxxx) xxx 9

Fig. 10. Map of Sentinel-2 based LAI.

(128.95 ha). The object-based classification depending on the NDVI The SEBAL estimated LAI has MAPE of 6.76% (0.28 m2/m2) which
threshold cross-matched the phonology responses of cotton, rice, is considered a small bias in the model. In a study by (Campos-
and maize, achieved high accuracy to filter out the rice area pre- Taberner et al., 2017) using Sentinel-2 data and the PROSAIL model
cisely. It revealed strong high accuracy with an overall accuracy in Spain, Italy, and Greece a higher RMSE was found (0.56, 0.82,
of 95% and a kappa coefficient of 0.92, as described on the Table 3. and 0.77 m2/m2) respectively. While (Setiyono et al., 2018)
These results are also similar to (Kaufman et al., 1997) other stud- reported LAI estimated from MODIS and Sentinel-1 SAR data using
ies: (Huang, 2002) used Landsat-derived vegetation indices, such a logistic function over eight different rice production provinces in
as NDVI, EVI, and (Mosleh et al., 2015) used MODIS vegetation Vietnam with RMSE ranging from 0.11 to 0.57 m2/m2. In a different
indices time series, and (Ali et al., 2018a,2018b) used Sentinel-2 study carried in Spain and Italy to estimate LAI from Landsat8 and
to estimate winter crops in Egypt. We found that the study site SPOT5 using the PROSAIL inverted model, (Campos-Taberner et al.,
remained under rice cultivation extensively. 2016) reported a RMSE of (0.39 and 0.38 m2/m2) for Landsat and
The ArcGIS software was used to mask the spatial distribution
of the area under rice cultivation the classified image by assigning
the value of 1 to objects with rice as the class name and 0 to every-
where else. Masking aims to generate a map to be used in the yield
mapping (Fig. 8).

3.4. LAI estimation and validation

According to (Noureldein et al., 2013), two variables are

required for yield prediction. The first variable is the LAI. As dis-
cussed in the methodology, the SEBAL equation is used to calculate
LAI based on the Sentinel-2 data. The MAPE of the satellite-based
estimated LAI was found to be 6.76% of the field measured LAI. This
means that the Sentinel-2 based LAI is ±6.76 of the actual mea-
sured LAI (±0.28 m2/m2). The measured LAI ranged between 1.83
and 7.26 m2/m2, with an average of 4.1 m2/m2. The correlation
between actual LAI and predicted LAI on the peak greenness period
of the rice crop was very high, with R2 = 0.96, as illustrated in Fig. 9.
The LAI in the study area ranged between 2.5 and 7 in the peek
greenness period (heading stage), as shown in Fig. 10. The varying
range of LAI could be attributed to the different varieties, and cul-
tivation dates. Fig. 11. Correlation between Sentinel-2 based yield and actual yield.

Please cite this article as: A. M. Ali, I. Savin, A. Poddubskiy et al., , The Egyptian Journal of Remote Sensing and Space Sciences, https://doi.org/10.1016/j.
ejrs.2020.06.007
10 A.M. Ali et al. / Egypt. J. Remote Sensing Space Sci. xxx (xxxx) xxx

Fig. 12. Spatial distribution of rice yield map for Sakha station.

(0.51 and 0.47 m2/m2) for SPOT5. The correlation found between 4. Conclusion
the estimated LAI and the in situ LAI (R2 = 0.94) was found to agree
with (Setiyono et al., 2018; Campos-Taberner et al.,2017, 2016). This study implemented a new methodology to estimate the
spatial distribution of the rice area and rice yield. This methodol-
ogy was applied and tested under Egypt intensive agriculture con-
3.5. Rice yield estimation and validation
ditions at Sakha station. Multi-temporal NDVI derived from
sentinel-2 satellite imagery was found to be very important in crop
The rice yield was calculated according to (Noureldein et al.,
discrimination. The multi-resolution segmentation using the NDVI
2013) based on Sentinel-2 driven LAI, and NDVI was validated
time series over the growing season was found to be very useful in
against the actual yield, and the MAPE was found to be 6.53%. It
detecting field boundaries, which assisted in crop identification
means that the predicted value is ±6.53% of the actual yield (i.e.,
and mapping with an accuracy of 95%. The SEBAL model predicated
±0.66 ton/ha). The actual yield ranged between 6 and 13 tons/ha
LAI with a MAPE of 6.76% conducted to study the spatial distribu-
with an average of 10 tons/ha. The predicted yield was also found
tion of the rice area. Using NDVI and LAI derived from Sentinel-2
to have a high correlation with the actual yield with R2 = 0.89, as
satellite data predicted rice yield with an average error of 0.66
illustrated in Fig. 11.
ton/ha. The NDVI at the heading stage was also found to be repre-
Fig. 12 shows the spatial distribution of the Sentinel-2 based
sentative of yield quantity. Finally, we recommend using this
yield. The predicated yield ranged between 4.5 and 12.5 ton/ha
method with high accuracy to estimate the rice area and the yield
with an average of 10 tons/ha. The results presented in Fig. 12
on delta Egypt because of the fragmentation of agriculture field
show that the estimated rice yield map was spatially variable
crops.
among the sampling sites, Rice yield map was carried out using
the analysis of NDVI multi-temporal has been used in many studies
using NDVI from NOAA/AVHRR and MODIS images (Qian et al., Author contributions
2019.; Mosleh et al., 2015).
The correlation between the estimated and the in situ measured All the authors substantially contributed to this article. A.M.A,
yield was found to be very high (R2 = 0.95) with a MAPE of 6.53% M.A., N.S, M.A.E, and K.A conceptualized the study and developed
(i.e. 0.66 ton/ha). This error was found to agree with what is the methodology. The satellite imagery were analyzed by A.M.A,
reported by the different studies. A percentage of 5% and 8% (0.3 N.S, M.A.E, and K.A. M.A, K.A, A.M.A and P.D Accomplished data
and 0.46 ton/ha) was reported by (Campos-Taberner et al., 2016) analysis and wrote a draft of the manuscript. A.P, I. S., K.A, A.M.A
for the spring and summer season, respectively in the case of and P. D contributed to reviewing and editing the manuscript.
MODIS data against the reported yield. While (Maki et al., 2017)
have reported a higher percentage ranging between 3.5% and Funding
121.4% (i.e. 0.14 to 1.84 ton/ha) using the LAI derived from the
Constellation of small Satellites for the Mediterranean basin Obser- The authors would like to declare that the funding of the study
vation (OSMO-SkyMed) data. has been supported by the National Authority for Remote Sensing

Please cite this article as: A. M. Ali, I. Savin, A. Poddubskiy et al., , The Egyptian Journal of Remote Sensing and Space Sciences, https://doi.org/10.1016/j.
ejrs.2020.06.007
 A.M. Ali et al. / Egypt. J. Remote Sensing Space Sci. xxx (xxxx) xxx 11

and Space Science (NARSS), Egypt, 5-100 Project (RUDN University) Ecology, F., Unit, E., Team, P., 2003. Champenoux, ‘‘Ground-based measurements of
leaf area index : a review of methods, instruments and current controversies. J.
and Agrarian-Technological Institute of the Peoples’ Friendship
Exp. Bot. 54 (392), 2403–2417.
University of Russia. Food and Agriculture Organization of the United NationsRome, (FAO_2018).
Food and Agriculture Organization of the United NationsRome, (FAO-2015)
Declaration of Competing Interest Frouins, R., Kaufman, Y. J., Tanrc, D., Gordon, H. R., Nakajima, T., Lenoble, J.,
et al.,1997. Passive remote sensing of tropospheric correction for the aerosol
effect aerosol and atmospheric, 102(97), 815–830.
The authors declare that they have no known competing finan- Ghidan W., Yacout, M., 2016. Genetic Variability among Egyptian Rice Genotypes
cial interests or personal relationships that could have appeared (Oryza sativa L.) for Their Tolerance to Cadmium Genetic Variability among
Egyptian Rice Genotypes (Oryza sativa L.) for Their Tolerance to Cadmium. J.
to influence the work reported in this paper. Appl. Life Sci. Int., 4(2): 1-9.
Gitelson, A., 2016. Remote Sensing Estimation of Crop Biophysical Characteristics at
Acknowledgments Various Scales. Transactions of the ASAE. 46 (4), 1235–1246.
Haboudane, D., Miller, J.R., Tremblay, N., Zarco-tejada, P.J., Dextraze, L., 2002.
Integrated narrow-band vegetation indices for prediction of crop chlorophyll
The authors would like to thank the National Authority for content for application to precision agriculture., 81: 416–426, https://ipad.fas.
Remote Sensing and Space Science (NARSS) for funding the field usda.gov/highlights/2016/01/Egypt/Index.htm.
Huang, J., 2002. Rice yield estimation using remote sensing and simulation model. J.
survey and remote sensing work. The authors would like to Agri- Zhejiang Univ. Sci. 3 (4), 461.
culture Research center, Sakha Station, Kafr elsheikh, Egypt for Kalácska, M., 2018. Calibration and assessment of seasonal changes in leaf area
field data sample analysis. The authors would like to thank 5-100 index of a tropical dry forest in different stages of succession. Tree Physiol. 25
(6), 733–744.
Project (RUDN University) and Agrarian-Technological Institute of
Kaufman, Y.J., Wald, A.E., Remer, L.A., Gao, B., Li, R., Flynn, L., 1997. The MODIS 2 . 1-
the Peoples’ Friendship University of Russia for supporting the m Channel — Correlation with Visible Reflectance for Use in Remote Sensing of
research activities. Aerosol. IEEE Trans. Geosci. Remote Sens., 35(5): 1286–1298.
Liang, S., Schaepman, M.E., Kneubuehler, M., 2014. Remote Sensing Signatures:
measurements, modeling and applications CHAPTER 10 Remote sensing
signatures: Measurements, modelling and applications. Advances in
References Photogrammetry, Remote Sensing and Spatial Information Sciences: 2008
ISPRS Congress Book – Li, Chen & Baltsavias (eds) 2014.
Liu, T., Liu, X., Liu, M., Wu, L., 2018. Evaluating heavy metal stress levels in rice based
Aboelghar, M., Arafat, S., Yousef, M.A., El-shirbeny, M., Naeem, S., 2011. Using SPOT
on remote sensing phenology. Sensors (Switzerland) 18 (3).
data and leaf area index for rice yield estimation in Egyptian Nile delta, Egypt. J.
Maki, M., Sekiguchi, K., Homma, K., Hirooka, Y., Oki, K., 2017. Estimation of rice yield
Remote Sens. Sp. Sci. 14, 81–89.
by SIMRIW-RS, a model that integrates remote sensing data into a crop growth
Aboelghar, M., Ali, A., Arafat, S., 2014. Spectral wheat yield prediction modeling
model. J. Agric. Met. 73, 2–8.
using SPOT satellite imagery and leaf area index. Arab. J. Geosci. 7 (2), 465–474.
Mosleh, M.K.K., 2015. Use of GIS and Remote Sensing in Mapping Rice Areas and
https://doi.org/10.1007/s12517-012-0772-6.
Forecasting Its Production at Large Geographical Extent. Sensors 15 (1), 769–
Ahmad, I.; Ghafoor, A.; Bhatti, M. I.; Akhtar, I. H.; Ibrahim, M., 2014. Satellite Remote
791.
Sensing and GIS-based Crops Forecasting & Estimation System in Pakistan.
Mosleh, M.K., Hassan, Q.K., Chowdhury, E.H., 2015. Application of remote sensors in
Conference: Crop Monitoring for Improved for Security, FAO-UN.At: Vientiane,
mapping rice area and forecasting its production: A review. Sensors
Lao People’s Democratic Republic. Volume: RAP PUBLICATIO 2014/28.
(Switzerland) 15 (1), 769–791.
Ali, A.M., Aboelghar, M., 2019. Comparative Analysis of Different Methods of Leaf
Noureldin, N.A., Aboelghar, M.A., Saudy, H.S., Ali, A.M., 2013. Rice yield forecasting
Area Index Estimation of Strawberry under Egyptian Condition. Int. J. Adv.
models using satellite imagery in Egypt. Egypt. J. Remote Sens. Sp. Sci. 16 (1),
Remote Sens. GIS 8 (1), 2963–2970.
125–131.
Ali, A.M., Aboelghar, M.A., El-shirbeny, M.A., Salem, N.H., 2018a. Comparative
Paper, T., 2015. Remote Sensing And Its Application In Agricultural Pest. Journal of
analysis of some winter crops area estimation using landsat-8 and sentinel-2
Agriculture and Environment 2015, 43–61.
satellite imagery. Futur. Food J. Food, Agric. Soc. 6 (2).
Pradhan, S., Bandyopadhyay, K., Sahoo, R.N., Sehgal, V.K., 2014. Predicting Wheat
Ali, A.M., El-Shirbeny, M.A., Saleh, N.H.S., 2018b. Comparative analysis of some
Grain and Biomass Yield Using Canopy Reflectance of Booting Stage. J. Indian
winter crops area estimation using landsat-8 and sentinal-2 satellite imagery.
Soc. Remote Sens. 42 (4), 711–718.
Asian J. Agric. Biol. 6 (2), 189–197.
Qian, Yonglan, Zhengwei Yang, Liping Di, Shahinoor Rahman, Zhenyu Tan, Lei Xue,
Amer, M.H., Abd-El-Hafez, S.A., Abd-El-Ghany M.B., 2017. Water-Saving In Irrigated
Feng Gao, Eugene Genong Yu, and Xiaoyang Zhang. n.d., 2019 ‘‘Crop Growth
Agriculture in Egypt, Book, LAP LAMBERT Academic Publishing, 2017.
Condition Assessment at County Scale Based on Heat-Aligned Growth Stage.
Awad, M.M., 2019. Toward Precision in Crop Yield Estimation Using Remote Sensing
Remote Sens., 11(20), 2439.
and Optimization Techniques. Agriculture 9 (3), 54.
Sayed, H., Abdelaal, A., Thilmany, D., 2019. Grains Production Prospects and Long
Boschetti, M., Busetto, L., Manfron, G., Laborte, A., Asilo, S., Pazhanivelan, S., Nelson,
Run Food Security in Egypt. Sustainability 11 (16), 1–17.
A., 2017. PhenoRice: A method for automatic extraction of spatio-temporal
SEBAL, 2002 ‘‘Surface Energy Balance Algorithms for Land Idaho Implementation,”
information on rice crops using satellite data time series. Remote Sens. Environ.
Advanced Training and Users Manual, August Version 1.0 page 20
https://doi.org/10.1016/j.rse.2017.03.029.
Setiyono, T.D., Quicho, E.D., Gatti, L., Campos-Taberner, M., Busetto, L.,
Campos-Taberner, M., García-Haro, F.J., Camps-Valls, G., Grau-Muedra, G., Nutini, F.,
Collivignarelli, F., García-Haro, F.J., Boschetti, M., Khan, N.I., Holecz, F., 2018.
Busetto, L., Katsantonis, D., Stavrakoudis, D., Minakou, C., Gatti, L., Barbieri, M.,
Spatial Rice Yield Estimation Based on MODIS and Sentinel-1 SAR Data and
Holecz, F., Stroppiana, D., Boschetti, M., 2017. Exploitation of SAR and Optical
ORYZA Crop Growth Model. Remote Sens. 10, 293.
Sentinel Data to Detect Rice Crop and Estimate Seasonal Dynamics of Leaf Area
Tucker, C.J., Holben, B.N., 1981. Remote Sensing of Total Dry-Matter Accumulation
Index. Remote Sens. 9, 248.
in Winter Wheat. Remote Sens. Environ. 11, 171–189.
Campos-Taberner, M., García-Haro, F.J., Camps-Valls, G., Grau-Muedra, G., Nutini, F.,
Verger, A., Vigneau, N., Chéron, C., Gilliot, J., Comar, A., Baret, F., 2014. Remote
Crema, A., Boschetti, M., 2016. Multitemporal and multiresolution leaf area
Sensing of Environment Green area index from an unmanned aerial system over
index retrieval for operational local rice crop monitoring. Remote Sens. Environ.
wheat and rapeseed crops. Remote Sens. Environ. 152, 654–664.
187, 102–118.
Yang, P.; Tan, G.X.; Zha, Y.; Shibasaki, R.; Planning, R.; Indices, V., 2004. Integrating
Change C.; Security F. 2016. the state of food and agriculture. (Website: http://
Remotely Sensed Data With an Ecosystem Model to Estimate Crop Yield in
www.fao.org/state-of-food-agriculture/en/).
North China. Commission VII, WG VII/2,:1988-2003.
Congalton, R.G., 1991. A review of assessing the accuracy of classifications of
Zhang, Geli, Xiao, Xiangming, Dong, Jinwei, Kou, Weili, Jin, Cui, Qin, Yuanwei, Zhou,
remotely sensed data. Remote Sens. Environ. 37 (1), 35–46.
Yuting, Wang, Jie, Menarguez, Michael Angelo, Biradar, Chandrashekhar, 2015.
Dong, J., Xiao, X., Menarguez, M.A., Zhang, G., Qin, Y., Thau, D., Moore, B., 2016.
Mapping Paddy Rice Planting Areas through Time Series Analysis of MODIS
Mapping paddy rice planting area in northeastern Asia with Landsat 8 images,
Land Surface Temperature and Vegetation Index Data. ISPRS J. Photogrammet.
phenology-based algorithm and Google Earth Engine. Remote Sens. Environ.
Remote Sens. https://doi.org/10.1016/j.isprsjprs.05.011.
https://doi.org/10.1016/j.rse.2016.02.016.

Please cite this article as: A. M. Ali, I. Savin, A. Poddubskiy et al., , The Egyptian Journal of Remote Sensing and Space Sciences, https://doi.org/10.1016/j.
ejrs.2020.06.007
View publication stats