Available online at www.sciencedirect.

com

ScienceDirect
Procedia Environmental Sciences 33 (2016) 3 – 13

The 2nd International Symposium on LAPAN-IPB Satellite for Food Security and Environmental
Monitoring 2015, LISAT-FSEM 2015

Processing system of MODIS data for monitoring the changes of

paddy field
Muhammad Tajul Arifina,* Liyantonoa, Yudi Setiawanb, Alvin Fatikhunnadaa
a
Department of Agricultural Engineering and Technology, Faculty of Agricultural Technology, Bogor Agricultural University, Jl. Kamper,
Kampus IPB Darmaga, Bogor 16680, West Java, Indonesia
c
Center for Environmental Research (PPLH), Bogor Agricultural University, Jl. Lingkar Akademik, Kampus IPB Darmaga, Bogor 16680, West
Java, Indonesia

Abstract

Indonesia has a role of paddy fields as the main resources in agricultural activities. Rice is the staple food of more than half of the
Indonesia's population. It makes a high demand of rice in Indonesia. Meanwhile, Indonesia has experienced by the paddy field
conversion, so that the monitoring of the fields is necessary. An advance remote sensing technology is a powerful tool to provide
accurate and up-to-date information of dynamics change in the paddy fields immediately. The main objective of this study is to
develop a reliable system that can monitor the changes in paddy fields. The system was developed using the time-series of
MODIS datasets through a graphical user interface (GUI) of Matlab. MODIS product used is MOD13Q1, which take daily
vegetation index data to display seasonal dynamics pattern of rice fields. As the time series data has still contains a residual noise,
therefore a correction signal by wavelet coiflet transformation filter applied before the further analysis. The crop pattern
clustering was done by k-mean method to get the same group of cropping pattern, including the intensity and their changes. As
the results, a graphical user interface (GUI) in Matlab software is developed, which has the time of data processing faster than the
conventional data processing.
© 2016
© 2016TheTheAuthors.
Authors. Published
Published by Elsevier
by Elsevier B.V.is an open access article under the CC BY-NC-ND license
B.V. This
Peer-review under responsibility of the organizing committee of LISAT-FSEM2015.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the organizing committee of LISAT-FSEM2015
Keywords: Matlab GUI; MODIS; k-mean clustering; rice field; wavelet transformation

* Corresponding author. Tel.: +62-856-5944-1199

E-mail address: kreativitas.tajul@yahoo.com.

1878-0296 © 2016 The Authors. Published 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/).
Peer-review under responsibility of the organizing committee of LISAT-FSEM2015
doi:10.1016/j.proenv.2016.03.050
4 Muhammad Tajul Arifin et al. / Procedia Environmental Sciences 33 (2016) 3 – 13

1. Introduction

Indonesia has the role of paddy field as the main source of agricultural activities as well as rice as a staple food.
On the other hand, there is an imbalance between domestic consumption of rice, and the availability of rice in the
country. The average consumption of rice in Indonesia reached 130 kilograms / capita / year [1]. In fact, the average
world rice consumption is only 60 kg / capita / year [1]. Paddy field in Indonesia is only 13.5 million hectares with
productivity of about 6 tons / hectare [1]. This is certainly not comparable to the level of consumption. This
imbalance occurred because the number of farmers are diminishing, and decreasing at least the availability of paddy
fields. Paddy fields in Indonesia experienced the conversion of agricultural land to non-agricultural land.
Efforts to comply with sustainable food can be done by ensuring the availability of paddy fields. If the land
conversion occurs with great frequency and amount, it will threaten national food security. Therefore, it is necessary
to monitor paddy fields area, especially paddy in relation to the land use change. There is a need to improve the land
surface characterization supported by the satellite technology, so that information of paddy field change can be
obtained accurately. Paddy field monitoring through Geographic Information System (GIS) and remote sensing
technology have been widely used, including: to detect the dynamics change in the land [2] and to evaluate the
changes of cropping index [3].
The uses of high temporal satellite imagery are necessary to develop methodologies that utilize information on
daily variations of lands to detect the changes in paddy field. The MODIS (Moderate Resolution Imaging
Spectroradiometer) high temporal resolution data have been used operationally to monitor the paddy field in
Indonesia. Utilization of MODIS satellite imagery has been widely performed, such as Xiao et al [4] that utilized the
MODIS image MOD09A1 daily 8 to map rice field in 13 provinces in southern China, especially over some unique
physical characteristics at a large area and Setiawan et al [5] utilized the MODIS image based EVI temporal patterns
to classify land use into 25 classes, including some detail types in paddy fields. In addition, the MODIS can be used
to predict the dynamics of changes in cropping patterns and monitoring change of paddy field so as to obtain the
estimated 1-2 months.
This research goal is developing of MODIS processing system with a graphical user interface (GUI) Matlab for
time-series data for monitoring of changes of paddy field in Indonesia. MODIS acquisition system can be built with
a graphical user interface (GUI) Matlab for time-series data sets. The benefit of this research is the development of
the acquisition and pre-processing system of MODIS satellite image data accurately and continuously. The result
will support governments and agencies to monitor the paddy fields in Indonesia.

2. Methods

2.1. Data

To cover all area of Indonesia, the MODIS datasets used are: h27v08, h27v09, h 28v08, h28v09, h 29v08,
h29v09, h30v08, h30v09, h31v08, and h31v09, which collected from 2000-2015. However, as the case study, we
used two tiles of h28v09 and h29v09 in order to cover the Island of Java. The MODIS EVI embedded in MOD13Q1
has been used in [6] to monitor the changes of paddy field in Java. Satellite images were used in this study is the
type MOD13Q1 MODIS with composited 16-days and pixel size of 250 m. This MODIS datasets is available for
free and can be obtained through the website of NASA (http://earthexplorer.usgs.gov). The MOD13Q1 product
provides a 23 data per year at the same location. Observation of land surface continuously by the MODIS data was
applied to understand the distribution and dynamics pattern of paddy field, which then used to identify the detail
type of paddy field.
A topographic map of Indonesia (RBI) provided by the Geospatial Information Agency (BIG) was used as a basis
map to delineate the study area. This map categorizes the paddy field into two classes, namely irrigated and rainfed
paddy field.
 Muhammad Tajul Arifin et al. / Procedia Environmental Sciences 33 (2016) 3 – 13 5

2.2. Instrument

Matlab 2015R was used to develop the GUI in the data acquisition and processing. Quantum GIS Desktop 2.8.2
is used to correct the projection system and delineating the map.

2.3. SDLC

The method used to build the system was referred to MODIS development based on the System Development
Life Cycle (SDLC) [7]. The SDLC method includes some various phases as follows: investigation stage of the
system, system analysis, system design, system implementation and system maintenance. The flow of GUI Matlab
developed in this study can be seen in Fig. 1.

Fig. 1. Flow SDLC of making GUI Matlab

3. Results

3.1. System investigation

Regarding to the MODIS time-series datasets that processed manually, there are several software used in the
image processing, such as: MODIS re-projection tool (MRT), ERDAS Imagine, Excel, and Matlab, as can be seen in
Fig. 2. Consequently, we need a system that can process the MODIS images easily and quickly.
The MRT is used for reading the HDF data, and convert it into *.TIF format. In addition, MRT software is used
as an inter-scene mosaic of these HDF files. The time-series data in .TIF format are then stacked by software
ERDAS Imagine in *.img format. By using the ERDAS Imagine, this *.img .format is converted into ASCII type,
which then can be analyzed by the Matlab software. These ASCII data are in accordance Matlab format and then
processed for the wavelet transformation and classification k-mean. The coiflet order 4 app level 5 is selected in the
wavelet transformation procedure. The result of wavelet transformation was the de-noised pattern/signal, which then
classified using k-mean method. Classification results are then analyzed to obtain the class center of each. These
manual steps of MODIS processing above require 5 days to analyzing the area of Java.
6 Muhammad Tajul Arifin et al. / Procedia Environmental Sciences 33 (2016) 3 – 13

Fig. 2. MODIS image processing stage

3.2. System analysis

Systems analysis phase conducted in the development of Graphical User Interface (GUI) Matlab in order to
facilitate the processing operation MODIS and time savings. Based on the results of functional analysis system that
is built to integrate image processing systems into a single view/interface. Integrated process that consists of this
work as follows:
1. Data reading and stacking
2. Projection and clipping
3. The wavelet transform
4. Classification k-mean

3.3. System design

Graphical user interface (GUI) built with graphical objects such as buttons pushbutton, edit, sliders, text, combo,
axis axes, as well as menus and others for use in the treatment process. The GUI is used to read HDF and built it into
*.TIF stacking matrix, MODIS projected to WGS-84, clipping the province, wavelet transformation and
classification k-mean. Such processes are divided into several sections in the MODIS datasets processing according
to the user needs. Matlab GUI design can be seen in Fig. 3.
 Muhammad Tajul Arifin et al. / Procedia Environmental Sciences 33 (2016) 3 – 13 7

Fig. 3. Design of GUI Matlab

3.4. System implementation

GUI Builder is a graphical user interface (GUI) that built through graphical objects such as buttons/pushbutton,
edit, sliders, text, combo, and axis as well as menus. The GUI Matlab was developed including the data reading,
projection, wavelet transformation and k-mean classification. This development of GUI Matlab refers to the step by
step process in the previous study [5]. The interface of GUI Matlab is given by Fig. 4.

Fig. 4. Interface of GUI Matlab

8 Muhammad Tajul Arifin et al. / Procedia Environmental Sciences 33 (2016) 3 – 13

As shown by Fig. 4, some function of each part in GUI Matlab as follows: 1) data reading dan stacking, 2)
projection dan area subset, 3) wavelet transform, 4) K-means classification and 5) map display.
Reading data on a GUI Matlab from the raw data of HDF format (hierarchical) is converted into a data matrix of
4800 x 4800 data size for each tile and then cut it into data size of 800 x 800 (or 1600 x 1600) pixel according to the
size of the province. Each data then is stacked, which results the subsequent data with 353 datasets during 15 years.
Projection in the interface is used to adjust the georeferenced map of BIG, which all data projected into WGS-84
projection system. Reprojection MODIS datasets to the WGS-84 is performed using QGIS GDAL. Moreover, subset
or clipping is used to border the size of the pixels according to the provincial area. The subset can be applied based
on both paddy fields area and province boundary.
In this study, the wavelet transform coiflet was applied to filter the temporal pattern of MODIS VI. This step is
required since the temporal pattern of MODIS EVI contains a residual noise caused by atmospheric disturbances, as
it cannot be removed effectively by the algorithm MODIS VI. According to [8] wavelet transform (WT) coiflet
order 4 gave an excellent result to determine the phenological stage as well as the rice growing period. In addition,
the WT allows the determination of the characteristics of the area of rice phenology. According [9] basically, the
idea of the wavelet transform (WT) is the decomposition (process of changing into a simpler form) signal on a
different spatial time or a set time scale basic functions timescale. The disturbances are filtered by the wavelet
transform, to reduce interference that is not erased in temporal data.
ାஶ
ͳ ‫ݔ‬െܾ
ܹ݂ሺܽǡ ܾሻ ൌ  න ݂ሺ‫ݔ‬ሻͳ ᫔ ൬ ൰ ݀‫ݔ‬ (1)
ିஶ ξܽ ܽ

where is the scale parameter, b is the parameter shift and a shift parameter is the mother wavelet. In the form of
discrete, two parameters of a and b is given as follows:

ሺܽǡ ܾሻ ൌ ሺʹ௝ ǡ ʹ௝ ݇ሻ (2)

The j and k is an integer. Wavelet is a wave local in time or space and satisfies orthogonal conditions. Because the
wavelet has a compact support, which means that the value to 0 outside the specified time interval. Time component
time-series data can be maintained for wavelet transformation. The level 4 of coiflet app was selected in the wavelet
transformation.
Input data in Classification k-mean is the TIF data input in the WGS-84 projection. The Matlab will then generate
the id-center and the pixel pitch class. The id-center and pixel class graph can be viewed in GUI application. In GUI
Matlab, several tiles of HDF files should be collected from 2000 to 2015. The acquisition readout system is based on
the sequence data in a certain folder. In the system, we provided several folders including: de-noised, geo, tiles, k-
mean result, processing, and projection-and-clip. The de-noised folder contains some files of filtered raw data. The
geo folder is a folder for reference, which indicates file of 800, 1600 pixel size or selected province, e.g. West Java.
Meanwhile, the tile folder included a projected TIF file. Moreover, there is a picture folder to display images in
Matlab GUI, and folders of the whole of Indonesia for clipping function.
Matlab GUI in this study in the study area of West Java. West Java using MODIS data h28v09 tile with grid size
1600x1600 with the coordinates of the point one and point two 2401 2401.4000, 4000. Data were processed ie as
many as 353 EVI band data. Tif file size 353 data read is 1,778,704 kb with long time 4939.63 seconds so the time
for each reading of the file that is 13.99 files / sec. This study compare as much data as 23 data required a long time
so that the second reading of 64.43 for a single file is read ie 2.80 file / sec. differences in the length of time each of
the 353 files per second between the data and the data 23 that occur in because the file size is increasingly enlarged.
Directory .tif file results are in (processed \ province \ raw \ olahan_grid_Jawa_Barat_2000-2015_EVI.tif)
This research was done in GUI Matlab Test performance for reading data. The performance test used to estimate
the data readout time for the entire Indonesia. Time graph of the number of pixels can be seen in chart 1, while the
graph of the amount of time the amount of data with the same pixel size can be seen in chart 2. The estimated time
for each province with stacking the data 46 is 5.6 hours for the entire Indonesia. The relation in data and time graph
is linear, but the relation between pixel size and time is exponential. The more number of data, the longer time to
 Muhammad Tajul Arifin et al. / Procedia Environmental Sciences 33 (2016) 3 – 13 9

process linearly. But when pixel size increased, the processing time increased exponentially. Estimated time to
process the data for a whole Indonesia map using the equations above is 6419.58 seconds for 23 data stacking. Fig. 5
indicates the data reading of West Java.

800x800pixel 1600x1600pixel 2400x2400pixel 3200x3200pixel 4000x4000pixel

y=14.73xͲ 102.17
1800
R²=0.9506
1600
y=14.096xͲ 295.69
1400
R²=0.9691
1200 y=11.59xͲ 276.05
Time (Second)

1000 R²=0.9715
800
y=5.0805xͲ 115.46
600 R²=0.972
400 y=2.7712xͲ 26.448
200 R²=0.4916
0
0 20 40 60 80 100 120 140
AmountofDataStacking

Fig. 5. Comparison between the amount of data and required processing time

1tahun(23data) 2tahun(46data) 3tahun(69data) 4tahun(92data)

4000
3500
3000
y=193.14e0.0005x
Time(Second)

2500 R²=0.9196
y=68.973e0.0007x
2000 R²=0.987

1500
y=37.069e0.0007x
1000 R²=0.9822

500
y=19.437e0.0006x
0 R²=0.9291
0 1000 2000 3000 4000 5000 6000
AmountofPixel

Fig. 6. Comparison between the amount of pixel and required processing time
10 Muhammad Tajul Arifin et al. / Procedia Environmental Sciences 33 (2016) 3 – 13

Fig. 7. Example read data Jawa Barat Province

800x800pixel 1600x1600pixel 2400x2400pixel 3200x3200pixel

2500

2000 y=39.825xͲ 886.96

R²=0.9877
Time(Second)

1500
y=24.666xͲ 603.25
R²=0.9349
1000

y=4.1363xͲ 88.133
500 R²=0.9984
y=0.3754xͲ 7.0767
R²=0.999
0
0 10 20 30 40 50 60 70 80
AmoundofData

Fig. 8. Comparison between the amount of data and processing time

 Muhammad Tajul Arifin et al. / Procedia Environmental Sciences 33 (2016) 3 – 13 11

1tahun(23data) 2tahun(46data) 3tahun(69data)

3500
3000
y=6.0173e0.002x
2500
Time(Second)

R²=0.9308
2000
y=3.6143e0.0018x
1500 R²=0.9542
1000
y=0.4519e0.0018x
500
R²=0.9536
0
0 500 1000 1500 2000 2500 3000 3500
Pixel

Fig. 9. Comparison between the amount of data and processing time

The projection function on the GUI is to change the projection of the file to be the WGS84 projection as referred
to the basis map of BIG. For a file size of proxies 1770817 kb required 25931.05 second to perform the projection
process in every once looping of 353 data. As given in Fig. 3 and 4, the relation between the amount of data and
processing time is linear, but the relation between pixel size and time is exponential. This result shows that the more
data will affect to the processing time linearly. On the other side, if the pixel size increased, the processing time
increased exponentially. The estimated time to process the whole data of Indonesia required approximately 3169.16
seconds for stacking process of 23 data.
According to [8] wavelet transform DWT Coiflet (order = 4) gives excellent results in determining the
phenological stage and rice growing period. In addition, the WT allows the determination of the characteristics of
the area of rice phenology. According [9], basically, the idea of the wavelet transform (WT) is the decomposition
(process of changing into a simpler form) signal at a different time scale spatial or time a set of basic functions. The
wavelet transform is a filter method to determine the temporal changes in vegetation. According to [10] NDVI
temporal has a different frequency each time, so that the seasonal variations, long-term and short-term fluctuations
can be identified.

Before Denoising After Denoising

1

0.8

0.6
EVI

0.4

0.2

0
2/18/2000 2/18/2001 2/18/2002 2/18/2003 2/18/2004 2/18/2005 2/18/2006 2/18/2007 2/18/2008 2/18/2009 2/18/2010 2/18/2011 2/18/2012 2/18/2013 2/18/2014 2/18/2015

Fig. 10. Before and after filter wavelet transform

12 Muhammad Tajul Arifin et al. / Procedia Environmental Sciences 33 (2016) 3 – 13

In this study, the coiflet order 1 level app 1 was applied in the wavelet transform process. Regarding to the
previous study, wavelet coiflet 1 and level app 1 can be used for reduce noise with a thick cloud on MODIS time
series of data. Moreover, this level 1 wavelet transform app can be applied to filter the temporal wavelet transform
without significant information loss of the rice cropping pattern.
In the study area, the paddy field can be classified into 32 classes totally, which indicates a different temporal
pattern of EVI. The processing time of this classification process of k-mean required about 1.5 hours.

Fig. 11. Result of k-mean clustering

4. Conclusion

The Graphical user interface (GUI) Matlab developed in this study can be used to support the application of
MODIS image processing to analyze the dynamics change in paddy fields. The system consists of data reading,
stacking, projection, wavelet transform and k-mean classification. Data can be processed in regard to the grid size or
province. The data reading, projection and clipping took a long time, but the memory used was small. Meanwhile,
the wavelet transformation and k-mean classification requires a large enough memory, but required time is quite
short. Over all the time for data processing is shorter than the manual data processing.

Acknowledgments

The Author would like to thank Mr. Million Muhammad and Mr. Rizky Aulia for giving support and constructive
comments to enhance this study.

References

1. Ministry of agriculture indonesian, Production, Quality and Food Safety Rice, 2015. [online]. Available:
http://pustaka.litbang.pertanian.go.id/berita.php?newsID=br20150812#.Vr0NhPnCYSk. [Accessed: 10- Oct- 2015]
2. Liu Y, Wang X, Guo M, Tani H, Matsuoka N, Matsumura S. Spatial and Temporal Relationships among NDVI, Climate Factors, and Land
Cover Changes in Northeast Asia from 1982 to 2009. GIScience Remote Sens. 2011; 48: 3, 371–393.
 Muhammad Tajul Arifin et al. / Procedia Environmental Sciences 33 (2016) 3 – 13 13

3. Liyantono T, Kato H, Kuroda, Yoshida K. GIS analysis of conjunctive water resource use in Nganjuk district, east Java, Indonesia. Paddy
Water Environ. 2013; 11, 193–205.
4. Xiao X, Boles S, Liu J, Zhuang D, Frolking S, Li C, Salas W, Moore B. Mapping paddy rice agriculture in southern China using multi-
temporal MODIS images. Remote Sens. Environ. 2005; 95: 4, 480–492.
5. Setiawan Y, Yoshino K, Philpot WD. Characterizing temporal vegetation dynamics of land use in regional scale of Java Island, Indonesia.
J. Land Use Sci. 2013; 8, 1–30.
6. Setiawan Y, Rustiadi E, Yoshino K, Liyantono, Effendi H. Assessing the Seasonal Dynamics of the Java’s Paddy Field Using MODIS
Satellite Images, ISPRS Int. J. Geo-Information 2014; 3: 1, 110–129
7. O’Brien JA. Management Information Systems: A Managerial End User Perspective. Irwin, Homewood, IL. 1993
8. Sakamoto T, Yokozawa M, Toritani H, Shibayama M, Ishitsuka N, Ohno H. A crop phenology detection method using time-series MODIS
data. Remote Sens. Environ.2005; 96, 366–374.
9. Daubechies I. Orthonormal Bases of Compactly Supported Wavelets II. Variations on a Theme. SIAM J. Math. Anal. 1993; 24: 2, 499–519.
10. Martínez B, Gilabert M. Vegetation dynamics from NDVI time series analysis using the wavelet transform. Remote Sens. Environ. 2009;
113: 9, 1823–1842.