2012 IEEE Colloquium on Humanities, Science & Engineering Research (CHUSER 2012), December 3-4, 2012, Kota Kinabalu,

Sabah, Malaysia

Estimation of above ground biomass of oil palm trees

by PALSAR
Veena Shashikanta,b, Abdul Rashid Mohamed Shariffa,b, Laili Nordinc , Biswajeet Pradhana,d

a
Geospatial Information Science Research Centre, Universiti Putra Malaysia,
43400 Selangor, Malaysia; Tel:+6(01)-4929-2709; Email: vnajogi@gmail.com
b
Department of Biological and Agricultural Engineering, Faculty of Engineering,
Universiti Putra Malaysia,43400 Selangor, Malaysia; Tel:+6(03)-8946-6414;
Email: rashid@eng.upm.edu.my
c
Faculty of Continuous Studies, Geomatika College International,Wilayah Persekutuan,54200
Malaysia; Tel:+6(01)-3394-9059; Email: drnordin@geomatika.edu.my
d
Department of Civil Engineering, Faculty of Engineering,
Universiti Putra Malaysia,43400 Selangor, Malaysia; Tel: + 6(01)-4896-8466;
Email: biswajeet24@gmail.com

Abstract—The main objective of this paper is to show the One of the main issue to solve in SAR imaging is the
potential use of L-Band SAR polarimetry. We used SAR images multiplicative speckle noise. The speckle noise is considered
to estimate the above ground biomass (AGB) of an oil palm undesireable since it interrupts with the radiometric resolution
plantation. The approach used in this study was to analyze the and affects the interpretation of SAR images. It is important to
relationship between the radar backscatter and AGB data by a
apply filters that will preserve the structural and texture
specific allometric equation in variation of age. The four
polarization of PALSAR was adapted in this study and converted information.
to sigma nought for respective ages. Laplacian filter with window
MATERIALS AND METHOD
size 3x3, 5x5 and 7x7 was used on the PALSAR polarimetry
images. Laplacian filter was used to show the best improvement
to the relation of radar backscattering and AGB of oil palm trees.
A. Field Biomass Estimation Data
As a conclusion, improved radar backscattering of oil palm trees
can be useful to estimate AGB of oil palm trees. Fieldwork was conducted at southern part of the state Perak
in Malaysia. The oil palm plantation was chosen to conduct
Keywords-Above Ground Biomass; PALSAR; Oil Palm fieldwork because of the availably of the remotely sense data
Plantation coverage. Fieldwork was done at the proposed site area
according to the palm ages from 6 to 24 years old. GPS
INTRODUCTION devices were used during the fieldwork to determine ground
Plant biomass represents a sink for atmospheric coordinates. Plots were designed to inventory fields of ages
carbon dioxide,which is one of the most important green 6,8,10,12,14,18 and 24 years. A total of 427 palms were
house gases and assumed to contribute more than half of the recorded and the average height for selected age group was
global warming [1]. Densheng(1994) mentioned that determined (Fig.1).
computer-prosessed remotely sensed data are economical to
estimation biomass in tropical regions over large areas. In
contrast with the conventional method [2], these approaches
are often time consuming, labor intensive, and difficult to
implement, especially in remote areas; also, they cannot
provide the spatial distribution of biomass in large areas [3].
With the advancement of remote sensing techinques, many
approaches have shown to estimate the forest biomass using
both optical [4] and radar sensors [5]. In relation to that, radar
backscattering coefficient of SAR images shows correlation
with forest biomass [6,7]. The results of another study showed
that the estimation oil palm biomass from radar backscattering
coefficient is encouraging [8]. Figure 1. shows the oil palm tree of age 6 years in the
United Plantation, Perak.

978-1-4673-4617-7/12/$31.00 ©2012 IEEE 838

 2012 IEEE Colloquium on Humanities, Science & Engineering Research (CHUSER 2012), December 3-4, 2012, Kota Kinabalu,
Sabah, Malaysia

In all age group, an allometric equation was used to calculate ground biomass and the other is the filters used on the radar
the AGB of the palm trees direct from the diameter at breast backscattering of the PALSAR image. The coefficient of

height (dbh) and tree height. Global and regional level data is determination (R2) is indication of the regression model. The
also valuable for forest AGB estimation and carbon R2 value will show the percentage of variation as the regression
accounting. International land-use and land cover datasets models variables.
exist, largely from remote sensing imagery, although image
resolution and the accuracy of ground-referenced data are
generally limited. Sources of data include international experts,
international organizations publishing statistics, such as the
United Nations and OECD, and international scientific
journals. In particular, the FAO Forest Resources Assessment
(FAO, 2006), the IPCC Agriculture, Forestry and Other Land
Use (AFOLU) inventory guidance volume (IPCC, 2006), and
FAO’s primer for estimating biomass (Brown, 1997) all
provide parameter information that can be used estimation of
AGB. It is important to ensure that data sets are harmonized so
the use of secondary data and accuracy of accounting are
acknowledged. According to IPCC 2003, locally-specific
equations exist to convert direct measurements of tree height
and diameter to volume; regression equations to directly
estimate biomass from tree diameter are preferable. Therefore,
the following equation was adapted from a Malaysian case
study [10]
` W =725 + 197H (1)
where H is the palm height (cm) and W the total fresh weight
(kg) . Using equation (1), the AGB was calculated based on the
summation values over all the trees.
B. Remote Sensing Data
In this study, satellite image was obtained for the oil palm
plantation in Hilir Perak, Perak. Two radar scenes were
captured by PALSAR; ALOS was used in this research. The
image was captured in Polarimetric (PLR) mode. The PLR
image covers a swath of 30km at 10m azimuth spatial
resolution. The images were acquired at level 1.5 which is
similar to a georeferenced SAR image in the ascending orbit
direction. The acquisition time for both was about 15:50 local
time (GMT+8) on 3rd April 2011 with off Nadir angle of 21.5˚.
Fig. 3 shows the PALSAR image of the palm trees selected
in this study. At first, PALSAR image was geometrically
corrected and then converted to Sigma Naught for all four
polarizations. The laplacian filter was selected in this study Figure 3. PALSAR Image of Study Site
and applied on the image at five different location of each age.
Variations in window size for the filters were 3x3, 5x5, and
7x7. Average of 5 different location of each age in dB was RESULTS AND DISCUSSION
selected to represent the radar backscattering of the respective
palm age.
AGB was estimated through the fieldwork which
requires palm to palm assessment for height measurements
C. Integration of biomass and remotely sensed data. totaling 427 palms using for this study. Table I shows the AGB
Remotely sensed data contain responses from the of palm from age 6 until 24 years, that was observed during the
vegetation structure, vegetation density and vegetation species fieldwork.
composition. All forest stand and even for that matter various
canopy covers have different reflectance and texture pattern in
various wavelengths. Thus the relationships between biomass
and remotely sensed data are unique. To analyze these
relationships, Pearson’s correlation coefficient is used. If the
coefficient is close to 1, it means there is a strong relationship
between them. In this study, the first variable is the field above

839
 2012 IEEE Colloquium on Humanities, Science & Engineering Research (CHUSER 2012), December 3-4, 2012, Kota Kinabalu,
Sabah, Malaysia

TABLE I. ESTIMATED AVERAGE AGB OF PALM TREES BY AGE

AGB Std Dev
Palm Age No of tress TABLE IV. AVERAGE RADAR BACKSCATTERING OF AGB OF PALM TREES
(kg/tree) (kg/tree)
WITH LAPLACIAN FILTER 5X5
6 40 1472.94 802.82
Palm Polarization (dB)
8 40 1542.33 805.77
Age HH HV VH VV
10 40 1804.12 791.98
6 -0.15 -4.02 -4.10 -1.24
12 40 1926.92 788.04
14 40 2272.54 807.54 8 -0.11 -3.99 -4.06 -1.28

18 27 2377.79 788.43 10 0.07 -3.80 -3.85 -0.80

24 40 3406.17 786.46 12 0.03 -4.00 -4.04 -0.69

14 0.07 -3.91 -3.93 -0.70

The estimated AGB of palm trees was used to validate the 18 0.86 -3.86 -3.94 -0.24
correlation of PALSAR radar backscattering. Table II shows 24 0.55 -3.70 -3.87 -0.55
the average radar backscattering of AGB of Palm Trees. AGB
was determined by taking the average radar backscattering of
TABLE V. AVERAGE RADAR BACKSCATTERING OF AGB OF PALM TREES
the area based on the palm area by age [11]. WITH LAPLACIAN FILTER 7X7

TABLE II. AVERAGE RADAR BACKSCATTERING OF AGB OF PALM TREES Polarization (dB)
Palm
BY AGE
Age HH HV VH VV
Palm Polarization (dB)
Age HH HV VH VV 6 -0.23 -4.46 -4.53 -1.21
6 -0.11 -4.46 -4.47 -1.38
8 -0.08 -4.45 -4.50 -1.23
8 -0.13 -4.45 -4.50 -1.53
10 0.09 -3.86 -3.90 -0.85
10 -0.02 -4.19 -4.31 -0.92
12 0.10 -4.50 -4.50 -0.65
12 -0.04 -4.50 -4.57 -0.78
14 0.02 -3.80 -3.89 -0.70
14 -0.06 -4.38 -4.46 -1.08
18 0.91 -4.31 -4.39 -0.23
18 -0.07 -4.39 -4.46 -1.14
24 0.49 -4.24 -4.28 -0.67
24 -0.06 -4.38 -4.46 -1.09
The results are shown in the graph of Fig.4 that the HV-
Laplacian with the window 5 x 5 has the coefficient of
Overlaying oil palm tree plots of the fieldwork in the
PALSAR image was to allow Laplacian filtering process. correlation, R2, a value of 0.652. Results shown that generally
Table III, IV and V shows the filtering results obtained through all the polarization improved the best in the 5 x 5 window size
three variation window size on all polarization of PALSAR. in showing the radar backscattering values of AGB of palm
The Laplace filter allows detecting large changes in pixel value trees. Laplacian filter was used in other research to preserve
[12]. The Laplace filters often result in "negative" images with the edges of SAR images.
brighter tones at detected features and dark or black tones
elsewhere. These filters will emphasize linear features such as
edges [13].

TABLE III. AVERAGE RADAR BACKSCATTERING OF AGB OF PALM TREES

WITH LAPLACIAN FILTER 3X3

Palm Polarization (dB)

Age HH HV VH VV
6 -0.11 -4.46 -4.08 -1.24
8 -0.11 -4.23 -4.00 -1.34

10 0.06 -3.78 -3.83 -0.80

12 0.01 -3.81 -4.05 -0.70

14 0.08 -3.94 -3.93 -0.70

18 0.84 -3.86 -3.95 -0.23 Figure 4. Polarization HV with Laplacian filtering of window
size 5x5
24 0.57 -3.82 -3.86 -0.51

840
 2012 IEEE Colloquium on Humanities, Science & Engineering Research (CHUSER 2012), December 3-4, 2012, Kota Kinabalu,
Sabah, Malaysia

In contrast, the AGB of palm trees for the window size 7 x 7 and understand the oil palm AGB characteristic in the
did not improve the relationship of the radar backscattering. PALSAR images.
An increment of less than 8% was obtained in the HV and VH
polarization. Highest increment of the filtered PALSAR image ACKNOWLEDGMENT
is >60% in the HV polarization for the window size 5 x 5. We appreciate deeply these research results have been
Next highest increment was shown in the VV polarization made possible through much help from many parties. Data that
where the window size reduces to 3 x 3 have been as the Research results have been made available
through the cooperation with JAXA/METI for PALSAR in the
Fig.6 shows correlation to observe the relationship of the radar JAXA’s ALOS RA. We acknowledge the assistance of
backscattering and palm AGB. Research Management Centre of Universiti Putra Malaysia,
WĂůŵ
Serdang, Malaysia for aiding this project financially with
 ' ,, ,s s, ss ‘Matching Grant-Vote No: 9300401’. We also would like to
thank Dr. Xavier Arulandoo and his associates at Research and
WĂůŵ' ϭϬϬ͘ϬϬй Development Department of United Plantations, Perak for
,, ϯϭ͘ϲϵй ϭϬϬ͘ϬϬй providing support in fieldwork and support.
,s ϭϮ͘ϯϱй ϱϰ͘ϯϵй ϭϬϬ͘ϬϬй
REFERENCES
s, Ϭ͘ϰϯй ϯϰ͘ϲϳй ϵϱ͘ϰϲй ϭϬϬ͘ϬϬй [1] Syahrinudin, "The potential of oil palm and forest plantations for carbon
ss ϯϭ͘Ϯϯй ϵϲ͘ϱϮй ϯϬ͘ϳϳй ϭϬ͘ϲϬй ϭϬϬ͘ϬϬй sequestration on degraded land in Indonesia," 2005.
[2] Basuki, T., P. v., A. S., & Y. H. (2009). Allometric equations for
,,Ͳ>ϯyϯ ϳϰ͘ϰϮй Ϯϭ͘ϲϴй ϭϰ͘ϰϱй ϴ͘ϯϱй ϭϵ͘ϳϳй
estimating the above-ground biomass in tropical lowland Dipterocarp
,sͲ>ϯyϯ ϱϳ͘ϴϲй ϴϲ͘ϳϯй ϰϲ͘ϲϳй Ϯϭ͘Ϭϰй ϴϮ͘ϰϱй forests. Forest Ecology and Management, 257(1684–1694), 1684-1694.
s,Ͳ>ϯyϯ ϱϵ͘ϲϮй ϱϰ͘ϴϭй ϴϰ͘ϲϵй ϳϮ͘Ϯϱй ϯϰ͘ϵϬй [3] R. Cerrillo and P. B. Oyonarte, "Estimation of above-ground biomass in
shrubland ecosystems," Invest Agrar: Sist Recur, vol. 15, no. 2, pp. 197-
ssͲ>ϯyϯ ϳϮ͘ϭϴй ϲϭ͘ϲϳй Ϯϯ͘ϯϮй ϴ͘ϯϰй ϲϮ͘ϯϲй 207, 2006
,,Ͳ>ϱyϱ ϳϮ͘ϲϯй Ϯϯ͘ϳϵй ϭϰ͘ϭϵй ϲ͘ϵϰй ϮϮ͘ϭϰй [4] D. Zhen, J. Rademacher, T. Crow, M. Bresee, J. L. Moine and S.-R.
Ryu, "Estimating above ground biomass using Landsat 7 ETM+ data
,sͲ>ϱyϱ ϴϭ͘ϱϵй ϰϳ͘Ϭϴй ϲϰ͘ϲϭй ϱϰ͘ϴϭй ϯϮ͘ϳϱй across a manage landscape in northen Winconsin,USA," Remote Sensing
of Environment, vol. 93, pp. 402-411, 2004.
s,Ͳ>ϱyϱ ϲϳ͘ϭϯй ϲϲ͘Ϭϳй ϳϵ͘ϴϱй ϲϱ͘ϳϵй ϰϵ͘ϰϬй
[5] Z. a. M. Y. Yichun Xie, "Remote Sensing imagery in vegetation
ssͲ>ϱyϱ ϲϵ͘ϲϮй ϲϬ͘ϳϴй Ϯϭ͘ϵϭй ϲ͘ϭϯй ϲϭ͘ϲϰй mapping: a review," Journal of Plant Ecology, vol. 1, no. 1, pp. 9-23,
,,Ͳ>ϳyϳ ϲϱ͘ϵϳй Ϯϳ͘ϳϬй ϭϯ͘ϴϮй ϰ͘ϭϱй Ϯϲ͘ϰϰй Mac 2008.
[6] S. Masanobu, "Coherence dependency of the PALSAR POLinSAR on
,sͲ>ϳyϳ ϮϬ͘ϴϯй ϱϮ͘ϭϯй ϳϵ͘ϭϲй ϳϬ͘ϵϬй ϯϰ͘Ϯϯй forest in Japan and Amazon," IEEE, pp. 2636-2639, 2007.
s,Ͳ>ϳyϳ Ϯϭ͘ϮϮй ϱϴ͘ϳϮй ϴϬ͘ϵϴй ϳϭ͘Ϭϴй ϰϭ͘ϭϭй [7] J. A. a. Josaphat, "SAR and Optical Images for Forest Biomass
Estimation," in Biomass – Detection, Production and Usage, InTech,
ssͲ>ϳyϳ ϲϬ͘ϯϬй ϱϲ͘ϴϱй ϭϯ͘Ϯϵй ͲϮ͘ϯϮй ϱϵ͘ϴϯй 2011, pp. 53-74.
Figure 6. Correlation Palm AGB and Radar Backscattering for [8] N. Laili, "Application of AIRSAR data to Oil Palm Tree
PALSAR. Characterization," Malaysian Centre for Remote Sensing, 1996.
[9] International Protocol of Carbon Credit 2003. Good Practice Guidance
Window size 7 shows a negative correlation with the VH for Land Use, land-use Change And Forestry. Reported at Vienna, 3-7
Nov 2003.
polarization after filtering. For window size 3, the best
[10] H. K. a. J. M. A. Z Z Zin, "Quantification of Oil Palm Biomass and
correlation was with polarization VH and VV where both were Nutrient in Mature plantation," Journal of Oil Palm Research, vol. II,
72.25% and 62.36% respectively after applying the filter. pp. 23-32, 1999.
Whereas, window size 5 showed the best correlation [11] Sandra Englhart, Vanessa Keuck and Florian Siegert (2011). “Tropical
increment for HH with 23.79% and increment of window size forest biomass assessment using multi-frequency radar imagery,” SPIE
Newsroom. DOI: 10.1117/2.1201108.003684
7 for HV with 79.16%. There was 86.73% correlation of HH
[12] Hisanaga, S., Wakimoto, K., & Okamura, K. (2011). Tone Mapping and
polarization with HV after filtering window size 3. Oil Palm Blending Method to Improve SAR Image Visibility. IAENG
AGB correlation was best with the HV polarization, 81.59% International Journal of Computer Science.
after filtering window size 5 and followed by HH with the [13] Fathee, H. N. (2010). Despeckling of SAR Imagery Using Contourlet
same window size. Transform in A Homomorphic Framework. Raf. J. of Comp. & Math’s. ,
Vol. 7, 151-164.
CONCLUSION [14] L. Densheng, M. Paul, B. Eduardo and M. Emilio, "Above-Ground
Biomass Estimation of Successional and Mature Forest Using TM
The AGB of Oil Palm trees was determined in this study. Images in Amozon Basin," 1994.
The availability of PALSAR data has exploited for the [15] FAO,2006 Global forest resources assessment 2005: Progress towards
computation of oil palm AGB and PALSAR backscattering. It sustainable forest management.FAO Forestry paper,147,Rome.
was shown that there was a weak correlation of 31.69% for oil [16] International Protocol of Carbon Credit 2006. Guidelines for National
palm AGB and HH polarization radar backscattering. Thus, Greenhouse Gas Inventories Vol. 4 Agriculture, Forestry and Other
filtering results has shown correlation obtained up to 81.59% Land Use. Kanagawa,Japan.
with a single Laplacian filter with the window size 5. However, [17] Brown, S. (1997). Estimating Biomass and Biomass Change of Tropical
regression only shows 65.20%. Future research will be focus Forests: a Primer. Illinois: A Forest Resources Assessment.
on applying other filters such as High pass and Median to learn

841