The Study of Discrimination of Remotely Sensed Data for Designing

the Separation Technique between Cassava and Sugarcane Farmland

Anuphao Aobpaet

and Soravis Supavetch

Department of Civil Engineering, Kasetsart University, Bangkok, Thailand

Keywords: Cassava, Sugarcane, Remote Sensing for Agriculture.

Abstract: Cassava and sugarcane are the most important agricultural crops in Thailand. The cultivations of those are

similarly in crop season, natural resources, and climate. For decades, the farmers usually switch their plant

depending on unit price and government subsidy. The use of remote sensing data for monitoring change in

farmland has encountered a problem on the similarity of vegetation index and the seasonal variation. In this

work, we investigate the significant differences between cassava and sugarcane plantation by using satellite

data from two sensors systems (Optical and SAR sensor) from Sentinel-1 and Sentinel-2 satellites. The result

of the sampling fields of cassava shows the fluctuation of the growth and the mean of SAVI is slightly lower

than sugarcane at the same age. SAVI values over the cassava farmland seem to approach the homogeneity

of sugarcane when the age of more than 11 months. Thus, the difference between cassava and sugarcane

farmland using this method should be investigated on the growth stage of the age between 4-9 months. For

SAR polarization, the VV, VH of SAR backscatters have little difference in cassava and sugarcane. When

compare the backscatters value of VV and VH from cassava and sugarcane, the sigma0 values in dB show

that VV backscatters have a higher signal return. The variation of VH polarization of cassava and sugarcane

seem difficult to identify due to the diversity of signal targets. Therefore, by using SAR data, the detection of

the difference between cassava and sugarcane should be considered after working on time series techniques

for crop seasoning to remove unwanted objects until only cassava and sugarcane remain. From the results, we

also found that the parcel-based method is a better processing approach to separate cassava from sugarcane

compared to pixel-based, and it requires descriptive statistics to distinguish between cassava and sugarcane

at each age. This method requires the information of two agricultural plantations boundaries. The possible

handling process when harvesting and preparation of the plantation are by observing time-related over an area

to determine the boundary of the farmland. Therefore, the discrimination of remotely sensed data for designing

the decomposition technique between cassava and sugarcane farmland is necessary because of the specificity

of cultivation in Thailand.

1 INTRODUCTION

The Ministry of Agriculture and Cooperatives also

announce the policy to reform Thailand's agricultural

sector mainly focuses on seven areas. The first area is

agricultural zoning to a suitable geographical and

climatic condition. The second area is the

establishment of learning centres for increasing the

efficiency of the agricultural product. The third area

is a grouping of farmers and farmland for better

efficiency. The fourth area is seeking the encourage

farmers producing in response to the market demand.

The fifth area set up the banks for agricultural

https://orcid.org/0000-0001-7638-853X

products through the grouping of farmers. The sixth

area is seeking to promote working as a team through

the "single command" system in order to translate the

reform plan into action. The seventh area is the

reduction of agricultural production costs. In the case

of cassava and sugarcane, the factor of success policy

is based on the monitoring of cultivation behaviour of

the farmers. In recent years, Earth Observation

Satellite (EOS) is a powerful tool for constantly

assessing the status of agricultural production on a

wide range of spatial and temporal scales. It provides

time series data and can rapidly reveal where change

has happened in a consistent and repeatable manner.

Aobpaet, A. and Supavetch, S.

The Study of Discrimination of Remotely Sensed Data for Designing the Separation Technique between Cassava and Sugarcane Farmland.

DOI: 10.5220/0007748102670273

In Proceedings of the 5th International Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2019), pages 267-273

ISBN: 978-989-758-371-1

267

Information from Earth Observation is therefore very

well suited to timely information on cropland

distribution and status at various crop stages.

In a global scale, characteristics of satellite remote

sensing data and processing implementations are

available for the large areas applications such as

forestry, agriculture, fishery, coastal and marine,

environment, disaster mitigation, spatial planning,

and so on. Nevertheless, in a small scale (high

resolution), many factors are relevant due to many

components of the ecological network which are

involved i.e., small pond, farm dike, mixed cropping,

species, soil types, moisture, etc.

In Thailand, the National Statistical Office (2013)

reported that 87.8% of 5.9 million farmers owned

land area lower than 6 ha. However, this small farm

size alone is not only a limiting factor for determining

the productivity (yield) and the homogeneity of the

plant by using spectral reflectance determination.

Moreover, Thai government pay attention to promote

“Philosophy of Sufficiency Economy and New

Theory” in agricultural sector following his Majesty

King Rama 9 suggestion that the farmers are

suggested to divided his land into four parts with a

ratio of 30:30:30:10. The first 30% is designated for

a pond to store rainwater during rainy season. The

second 30% is set aside for rice cultivation during the

rainy season for family’s daily consumption

throughout the year. The third 30% is used for

growing fruit and perennial trees, vegetables, field

crops and herbs for daily consumption. The last 10%

is set aside for accommodation, animal husbandry,

roads, and other structures (The Chaipattana

Foundation, 2017). Thus, these activities made more

complexity on a small farm in term of remotely

sensed data. For an agricultural country such as

Thailand, the implementation of remote sensing

technologies for monitoring and controlling the

government’s policy in agriculture using a field-

based approach is the importance for those

smallholder farmers.

In this study, the cassava and sugarcane are

selected as a feasible crop. The influence of the

successive policy depends on the monitoring of the

cultivation behaviour of the plants following farmers’

decision which sometimes varied and difficult to

identify. The crop growing process of cassava and

sugarcane are likewise in crop season, natural

resources, and climate (Thai Meteorological

Department, 2015), as already mentioned above. The

farmers have their choice to select the plants base on

the unit price and government subsidy made the

difficulty of policy control. The potential on using

remote sensing for farmland monitoring is the

challenge technique due to the fact about the

correspondence of vegetation index term, the

seasonal effect on a reflectance make the signature of

a phenotype cannot be determined. This study used

permitted available data from Sentinel-1 and 2 for

data processing and uses the parcel-based descriptive

statistics in analysing and interpreting the results. In

this stage of the study, we only focus on the technique

for the separation of cassava and sugarcane which

will make possible for crop forecasting in Thailand.

2 OBJECTIVE

As by the location of the country, there is a little

difference in environments as well as the planting

season, so we are unable to classify the agricultural

farmland of both plants by means of single temporal

classification. Therefore, the main objective of this

study is to investigate the significant differences

between cassava and sugarcane plantation by using

satellite data from two sensors systems which are

Optical and SAR sensor from Sentinel-1 and

Sentinel-2 satellites. Both satellites that we applied

are available at no cost and it is possible to develop

the automatic system for future operation.

3 DATA PREPARATION

The sampling areas of cassava and sugarcane are

selected in Nakhon Ratchasima province which is the

largest cassava plantations in Thailand (Figure 1).

This province economy has traditionally been heavily

dependent on agriculture which is the production of

rice, tapioca, and sugar.

Figure 1: Nakhon Ratchasima province, the largest cassava

plantations.

GISTAM 2019 - 5th International Conference on Geographical Information Systems Theory, Applications and Management

268

The Sentinel-1 and Sentinel-2 satellites data are

able to access via Copernicus Open Access Hub

(https://scihub.copernicus.eu/) for this investigation

(European Space Agency). Satellite data processing

from Sentinel-1 uses the Orfeo Toolbox software as a

processing tool (Inglada et al., 2009) (Christophe et

al., 2009) (Teodoro et al., 2016). The data

downloaded from the Copernicus Open Access HUB

is in Level-1 Ground Range Detected (GRD) and

Level-1 Single Look Complex (SLC). The processing

step started with the adjustment of the satellite's orbit

value with the most accurate orbit data (precise orbit),

and then apply radiometric correction by calibrating

to adjust the pixel value to backscatter and convert

them to decibel (dB). The data was then filtered using

speckle filter for noise reduction with speckle

filtering which offers many options for filtering as

follows: Boxcar (mean), Median, Frost, Lee (Lee et

al., 1999), Refined Lee, Gamma-MAP, Lee Sigma,

and IDAN, etc.

Subsequently, geometric correction is performed

using the Range Doppler Terrain Correction method

because the terrain pattern along the plane of the

scene and the inclination of the sensor on the satellite

will cause distances in the radar image distorted.

Since, the image data is not always in the middle

position (nadir) of the image sensor; there will be

distortion. Thus, the terrain is corrected to

compensate for these distortions in order to show the

geometric patterns of the image as close to the most

accurate. However, we skip the step for applying

SRTM DEM in height data correction because of the

plane study areas. The image has been corrected

(geocoded) and then subset to the study area and

classify the area features using the statistical values.

The polarization was used to distinguish and display

by mixing polarized colors such as Dual pol multiple

sigma0, Dual pol ratio sigma0, or Dual pol difference

sigma0 depending on appropriateness. In our

investigation, the program was developed to an

automatic SAR data processing using Orfeo Toolbox

together with the python language program for

creating Dual Polarization data in VV and VH that

will be used to study the characteristics of the

backscatter signal from cassava and sugarcane

plantation.

While, the reflectance obtained from the L1C

product of the data from Sentinel-2 is a reflection that

has been mixed with the diffusion effect in the

atmosphere. This information is called Top of

Atmosphere (TOA). Therefore, it is necessary to

adjust and remove the reflection by atmosphere out of

the data set before being used in order to be a

reflection value at the canopy layer (Bottom of

Atmosphere, BOA). The effect of modifying the

atmosphere to bring the reflection in the atmosphere

makes it possible to know the particles in the

atmosphere which makes the additional data layer

called Aerosol Optical Thickness (AOT). When

comparing the reflection values to the results from

adjustment, it is found that the values that have been

removed, such as clouds, shadows of clouds, etc., are

correlated in each band in the same proportion. The

data of the Soil-adjusted vegetation index (SAVI) (Qi

et al., 1994) is an index data that is closer to the crop

and plant health than NDVI (Normalized Difference

Vegetation Index) (Senay et al., 2000) because it has

been modified to reflect the influence of the soil.

Therefore, the SAVI data is an appropriate index to

be applied to the crops such as cassava and sugarcane.

When observed from satellite data, the effect of soil

reflection will be mixed with the sensor's detection

value. Reducing the impact of this soil will result in

more information that reflects evidences about plants.

The calculation of SAVI (Huete, 1988) is calculated

as by equation below using L = 0.48 according to the

recommended values from European Space Agency

(ESA).

SAVI = (1+L)*(NIR-R) / (NIR+R+L)

(1)

When verified to the cassava plantation area and

define the display with false color combinations, we

will see the data as shown in Figure 2. These time

series data shows the crop stage measurement and the

growing season, respectively.

Figure 2: SAVI information displayed with false color

combinations showing the amount of plant cover (green)

and soil without plants covered (yellow to red colors shade).

4 HYPOTHESIS

From the inspection area, the growth of cassava and

sugarcane taxonomy can be recognized as soil

The Study of Discrimination of Remotely Sensed Data for Designing the Separation Technique between Cassava and Sugarcane Farmland

269

preparation, planting time, and yield period. In

Thailand, there are many species of cassava crop

cultivation. Each type has a leaf size, height varies,

and some types are 2 meters high, some 4 meters tall,

while the sugarcane has a height between 2-5 meters

as well. When both plants are fully grown and yield

8-12 months, the top cover is similar, not significantly

different.

Figure 3: The cassava farmland.

Figure 4: The sugarcane farmland which follows the

“Philosophy of Sufficiency Economy and New Theory”.

In this study, we measured the inspection area

using UAV (Unmanned Aerial Vehicle) to capture a

vertical shot an overlaid with SAVI (Soil-adjusted

vegetation index) and SAR (VV-VH) data. Some

important features that can be hypothesized to find

differences in satellite data between cassava and

sugarcane plants are the growth of cassava (Grown at

the same time in the farmland). It has a variable

appearance according to the environment of the

farmland, rather than the growth of sugarcane (Figure

3 and Figure 4 respectively).

For cassava aged 9-11 months in the sample

farmland in Figure 3, the different heights can be seen

with some little leaves, some plant looks good growth

mixed together while the sugarcane has a relatively

stable growth. Therefore, assuming that cassava at the

same age as sugarcane as shown in figure 4 has a

variation of the LAI (Leaf Area Index) and results in

the plant index or the different altitude of the variable

canopy layer are important to be able to distinguish

between 2 types plantation.

5 METHODOLOGY

The method on building the phenotype from such

time-series of Normalized Difference Vegetation

Index as Savitzky-Golay, Gaussian, Logistic function

and so on, are using the data in a term of pixel-based

time-series for describing the plant's growth. Some

factor such as unstable of a growth rate of the cassava

over the farmland cannot be detected due to the pixel-

based time-series does not demonstrate the spatial

autocorrelation.

Cassava SAVI

Sugarcane SAVI

Cassava SAR VV

Sugarcane SAR VV

Cassava SAR VH

Sugarcane SAR VH

Figure 5: The 8 samples areas of cassava and sugarcane

with background of SAVI and SAR VV/VH.

Thus, in this research, the parcel-based with the

descriptive statistics is selected for use as the detector

of the homogeneity different between the growth of

cassava and sugarcane over the farm parcel. Soil

20 m

GISTAM 2019 - 5th International Conference on Geographical Information Systems Theory, Applications and Management

270

Adjusted Vegetation Index (SAVI), is used instead of

NDVI for reducing a soil effect in the reflectance,

especially for the field crop in which LAI is a low

value over the growth period. The descriptive

statistics such as min, max, mean, standard deviation,

and percentile are used in the calculation for

demonstrating the homogeneity of the plant's growth

over the farmland. The different of 25 and 75

percentile of the cassava should be more than the

sugarcane following the research hypothesis. The 8

samples of cassava and sugarcane are examined

SAVI and SAR backscatter shown in figure 5.

Cassava

Sugarcane

Figure 6: The images of cassava and sugarcane in difference

growing stage.

Due to the field visit on the harvesting season,

cassava and sugarcane are mostly in the maturity

stage as shown in Figure 6. The density of leaves and

stems of the sugarcane seem to be more than the

cassava in every farmland. The height of cassava is

varied and unequally growth except for the age more

than 12 months due to cassava stop growing on root

after 8 month then leaf and stem are growers at 8-12

months.

6 RESULT AND DISCUSSION

The boxplot of the sampling fields of the cassava in

Figure 7 shows the fluctuation of the growth and the

mean SAVI quite lower than the sugarcane in which

the same age. SAVI values over the farmland seem to

approach homogeneity the same as in sugarcane when

the age of the cassava for more than 11 months. Thus

the detection on the difference of the cassava and

sugarcane farmland using this method should be

investigated on the growth stage that age between 4-

9 months. In our assumption, the physical

characteristics of cassava and sugarcane from start

planting on the farmland is supposed to show some

difference of the backscatter signal when returning

back after reflected the objects. In the investigation,

SAR sensor in VV and VH polarization on board

Sentinel-1 satellite were generated and analysed

using Orfeo and then interpreted with parcel-based

descriptive statistics for cassava and sugarcane crop

separation.

Figure 7: The boxplot of cassava and sugarcane parcel-

based from SAVI.

For SAR polarization in Figure 8, the VV, VH of

SAR backscatters have little difference in cassava and

The Study of Discrimination of Remotely Sensed Data for Designing the Separation Technique between Cassava and Sugarcane Farmland

271

sugarcane. The VV backscatters of sugarcane seem to

have more outlier value due to the better backscatter

compare to VH polarization. When compare the

backscatters of VV and VH from cassava and

sugarcane, the sigma0 value in dB show that VV

backscatters have a higher signal return. The variation

of VH polarization of cassava and sugarcane appear

difficult to identify due to the diversity of the signal

target. Therefore, by using SAR data, the detection on

the difference of the cassava and sugarcane should be

considered after working on time series technique for

the crop season to filter or remove unwanted objects

until only cassava and sugarcane remain. In the next

step of our study, it ought to focus on the Polarimetric

SAR Classification. This technique can provide more

information and a variety of method to calculate on

the structure of farmland which is a function spatial

variation in canopy structure and density.

Figure 8: The boxplot of cassava and sugarcane parcel-

based from SAR VV and VH polarization.

7 RECOMMENDATION

From the results, we found that the parcel-based

method is a better processing tool to separate cassava

from the sugarcane compared to pixel-based, and it

requires descriptive statistics to distinguish between

cassava and sugarcane at each age. This method

requires information on the boundaries of the two

agricultural plantations. The possible handling

process is when the harvesting and preparation of the

plantation by observing time-related over such area to

allow the boundary of the farmland to be determined.

With the assumption that each farmland belongs to a

small farmer, usually harvesting in different periods

of time due to the small amount of labor in the harvest

season compared to the output quantity causing the

need to be in the waiting list for harvesting queue.

Therefore, determining the boundary of the farmland

by using the time series of the cutting tracking data

set will help determine the farmland boundary.

ACKNOWLEDGEMENTS

This work was prepared in the frame of the cassava

crop monitoring project which received research

grants from Kasetsart University Research and

Development Institute, KURDI). The authors

acknowledge the European Space Agency (ESA) for

providing the Sentinel-1 dual polarization images

available online contributed to scientific community.

This work contains modified Copernicus Sentinel

data.

REFERENCES

Christophe, E., Inglada, J. and Giros, A., 2008. Orfeo tool-

GISTAM 2019 - 5th International Conference on Geographical Information Systems Theory, Applications and Management

272

box: a complete solution for mapping from high

resolution satellite images. International Archives of

the Photogrammetry, Remote Sensing and Spatial

Information Sciences, 37(PART B4), pp.1263-1268.

European Space Agency, (n.d.). Open Access Hub.

(European Union) Retrieved 1230 2018, from

https://scihub.copernicus.eu/userguide /ODataAPI.

European Space Agency, (n.d.). ESA Sentinels Help

Monitor Grasslands for Agricultural Subsidy Checks in

Europe. Retrieved 12 25, 2018, from http://www.dhi-

gras.com/news/2017/4/7 /successful-demonstration-of-

using-sentinel-1-and-2.

Huete, A. R., 1988. A soil-adjusted vegetation index

(SAVI). Remote Sensing of Environment, 25(3), 295-

309.

Inglada, J. and Christophe, E., 2009. The Orfeo Toolbox

remote sensing image processing software. In

Geoscience and Remote Sensing Symposium, 2009

IEEE International, IGARSS 2009 (Vol. 4, pp. IV-

733), IEEE.

Lee, J.S., Grunes, M.R. and De Grandi, G., 1999.

Polarimetric SAR speckle filtering and its implication

for classification. IEEE Transactions on Geoscience

and remote sensing, 37(5), pp.2363-2373.

National Statistical Office, 2013. Preliminary Report 2013

Agricultural Census.

Qi, J., Chehbouni, A., Huete, A.R., Kerr, Y.H. and

Sorooshian, S., 1994. A modified soil adjusted

vegetation index. Remote sensing of environment,

48(2), pp.119-126.

Senay, G.B. and Elliott, R.L., 2000. Combining AVHRR-

NDVI and landuse data to describe temporal and spatial

dynamics of vegetation. Forest Ecology and

management, 128(1-2), pp.83-91.

Teodoro, A.C. and Araujo, R., 2016. Comparison of

performance of object-based image analysis techniques

available in open source software (Spring and Orfeo

Toolbox/Monteverdi) considering very high spatial

resolution data. Journal of Applied Remote Sensing,

10(1), p.016011.

Thai Meteorological Department, 2015. The Climate of

Thailand.

The Chaipattana Foundation, 2017. Philosophy of

Sufficiency Economy.

The Study of Discrimination of Remotely Sensed Data for Designing the Separation Technique between Cassava and Sugarcane Farmland

273