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

net/publication/286285443

Mangrove mapping analysis on: Optical and Synthetic Aperture Radar data
using: ALOS/PLASAR and ALOS/AVNIR-2

Article · January 2012

CITATIONS READS

3 580

4 authors, including:

Chathura Wickramasinghe
RMIT University
7 PUBLICATIONS 65 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Identification of sensitive areas in South eastern river basin Sri Lanka View project

All content following this page was uploaded by Chathura Wickramasinghe on 11 November 2016.

The user has requested enhancement of the downloaded file.

 Mangrove mapping analysis on: Optical and Synthetic Aperture Radar
data using

ALOS/PLASAR and ALOS/ AVNIR-2

Chathura Hasanka Wickramasinghea, Dr. Lal Samarakona ,Mr. Gary Godferyb, Mr.
Melvin Kinsuatc
a
Geoinformatic center ,Asian Institute of Technology, Po.Box 04 Klong Lunag
Pathumthani,Thailand
b
Department of Agriculture , Sabah Jalan Tasik, 88999 Luyang, SABAH
c
Department of Fisheries, Sabah, Fisheries research center, Likas, Jalan Perikanan,
P.O box 89400,SABAH

KEY WORDS: Remote Sensing, classification, Synthetic Aperture Radar, ALOS ,

decomposition

Abstract: The Klias Peninsula also known as the last largest remaining wetland area
of Sabah is located within the district of Beaufort, Kuala Penyu and Membakut. It covers
approximately 130,000 ha with dense mangrove, nipah forest and peat forest. Grassland,
scrub and small scale mixed horticulture also observed throughout the peninsula. It is
important to track the land use change in the area to monitor the mangrove forest growth.
Mangrove forests have been affected by illegal shrimp farming with in the area. This study is
to analyze mangrove mapping using Optical and Synthetic Aperture Radar (SAR) data. Both
full Polarimetric SAR and Optical data for the year 2010, from ALOS/PALSAR and
ALOS/AVNIR-2 sensors were used to produce the year 2010 land use for the study area.
The importance of microwave images was felt as finding cloud free optical images of the
study area was a difficult task. Combination of field data and Unsupervised classification
was used to classify the images. ALOS/PALSAR PLR 1.1 product level data was used to
apply Wishart classification algorithm for the SAR data. Optical image classification was
carried out using maximum likelihood algorithm. Initially it failed to distinguish forest and
mangrove area accurately. Thus to improve the accuracy of the classification Principal
Component Analysis was carried out, this resulted in accurately distinguishing the forest and
the mangrove. Comparison of the results showed SAR data is more suited for mangrove
mapping compared to Optical data from ALSO/AVNIR-2. The lack of Short Wave Infra-Red
band in ALSO/AVNIR-2, made it difficult to apply ALOS/AVNIR-2 data for mangrove
mapping compared to sensors with SWIR band. Biomass based mangrove area extraction
using SAR data also showed very promising method for mangrove mapping using SAR data.
Intorduction

Mangrove forest growing in saline coastal area in the tropics and subtropics region
plays a major part in the costal ecosystem and sea coast conservation. There are more than
100 Mangrove species each specially adapted saline, tide and wave condition of the area.
Specialized root system enables them to grow in this saline environment where other plant
species have failed. Mangrove forest also contributes in reducing the global carbon foot
print. Been able to thrive in salt water it proved solutions to un-usable land area due to
salification with the increasing sea-level. Mangrove forest has its own unique ecosystem and
is the breeding ground for verity of sea species.

The demand for seafood has led to the destruction mangrove forest, large mangrove
areas have been cut down for shrimp farming, and other aquaculture activities. The land is
abandoned after few years due to the build-up of toxicity in the area, which make is even
unsuitable for mangrove. Thus it’s very important to map and monitor mangrove forest to
protect its ecosystem.

Remote Sensing Optical data has been widely used for land cover classification and
have proven to be very efficient. Landsat TM and Landsat ETM are most widely used
sensors; ALOS/AVNIR-2 has also become widely used due to its 10m spatial resolution,
compared to 30m spatial resolution in Landsat TM. Maximum likelihood and K-Mean are
most commonly used Supervised and Unsupervised image classification algorithms used in
optical RS and have been proven to be effective in land cover mapping applications.

With the improvement of full polarimetric SAR (Synthetic Aperture Radar )satellite
data, use of SAR data for land cover classification has grown. R.Cloude & E.Pottier;1997 has
introduced new SAR decomposition for supervised as well as unsupervised algorithms to
accurately classify the SAR data.

Tree biomass calculation provides various benefits, including habitats, food, timber,
forest growth and in reducing the carbon footprint (Nagelkerken et al., 2008). Biomass
estimation of mangrove forest using filed work is a difficult due to the muddy soil condition
and weight of the trees. To overcome this problem researchers have developed various kinds
of allometric algorithms to calculate the Dry Weight of mangrove tree species. (C. Proisy ,
2000 ) has shown that it’s possible to use SAR data to calculate the mangrove biomass with
only few sample point.

The study was undertaken to study mangrove forest mapping using ALOS/PALSAR
and ALOS/AVNIR-2 data the results were compared to see how effective the technology is
for mangrove mapping. The SAR data proved to be more effective to distinguish mangrove
compared to four band AVNIR data. The lack of short wave infrared band proved
ALOS/AVNIR-2 is ineffective for mapping mangrove. Approximate biomass estimation was
done using SAR backscatter coefficient. To map the mangrove area using back scatter
coefficient.
Study area

The Klias Peninsula, Malaysia (5º21’57.77”N 115º 33’53.38”E) also referred to as

the last largest remaining wetland area of Sabah is located within the district of Beaufort,
Kuala Penyu and Membakut which is about 80 km southeast of Kota Kinabalu. It is a large
area approxiately 130,000 ha covered with dense areas of mangrove, Nipah forest and Peat
forest. Grassland, scrub and small scale mixed horticulture also observed throughout the
peninsula (Scott, 1989). There are six Forest Reserves within the Peninsula which include the
Binsuluk, Klias, padas Damit, Kampung Hindian and Menumbok which covered and area of
31,053 ha (Vas, Justin. 1998; Maryati Mohamed et al., 1999). These areas are classified as
class one forest reserves and not allowed to log within the areas. The Peninsula are connected
with several extensive river channel which include the Padas Damit, Klias river flows from
southern part of Binsuluk Forest Reserve towards the Brunei Bay and the Api-Api and Bukau
river flows from Southern part of Klias Forest Reserve towards Weston.

Land cover classification for Optical data

Images from Landsat TM 30m spatial resolution and ALOS/AVNIR-2 with 10m
spatial resolution were used to for the classification. For the Landsat classification six bands
from the Landsat TM was selected band 6 was omitted because of the 120m resolution.
Initial unsupervised classification was done on the 1995 Landsat image to get a general idea
of the land cover in 1995. Created 25 classes in the unsupervised classification were merged
to create 8 classes. By combining results from the unsupervised classification and the land
use map of 1999 the image was re-classified using training data base on the maximum
likelihood algorithm into 8 land use class. Coefficient matrix was generated to check the
accuracy of the classification. The overall accuracy was 95.7794% and Kappa coefficient of
0.9460.

ALOS/AVINR-2 was fist classified using ground truth data collected from field work
using maximum likelihood algorithm but algorithm failed to distinguish between forest and
mangrove. Thus Principle component analysis was performed on the AVINR-2 data set and
the first 4 components were stacked. Unsupervised K-mean algorithm was applied with
initially 30 classes and merged in to the final 8 class to create the 2010 land cover map Figure
2(a). The results were accurate than using the raw optical bands, but was not accurate enough
to use the technique for mangrove mapping. In the post processing stage the classified image
were cleaned and generalized using Sieve and Clump applied to remove isolated pixels and to
add spatial coherency to the classes, for both 1995 and 2010 images.
 Figure 1 1995 Landsat TM image classfication (a) Extracted mangrove area (b)

Figure 2 (a) ALSO/AVNIR-2 2010 image classification (b) Extracted mangrove area showing miss classification
 Combination of unsupervised and supervised classification for 1995 landst5 TM
image had an overall accuracy of 95.7794% but the initial supervised classification of the
2010 ALOS/AVNIR-2 was 67.2342% with forest and mangrove classified as forest. PCA
was applied with K-mean algorithm to increase the accuracy of the result, which gave an
better result but the result still had miss classified mangrove as shown in Figure-2 (b) the area
circled by white circle is an non mangrove are but misclassified as mangrove and the area
marked by an white square should have mangrove but was classified as oil palm and
agriculture. It proved that it’s difficult to use ALSO AVNIR-2 data for mangrove forest
mapping. This is due to the lack of SWIR band in the sensor ( Masayuki, 2010 ). Thus
Landsat and other sensors with SWIR wave observation should be used if mangrove mapping
is to be done using optical data. Masayuki, 2010 also has shown it’s possible to use a
combination of AVNIR -2 data and DEM (Digital Elevation Model ) with decision tree
technique to increase the accuracy of the AVNIR-2 based mangrove mapping. Figure 3 below
shows the significant difference in the spectral signature in the SWIR region compared to the
four bands used by ALOS/AVNIR-2.

Figure 3 Leaf spectral reflectance of 6 species of mangorves and 7 species of non-mangorve

trees, with AVNIR-2 band 4 range , and landsat TM band 5 and band 7 ranges (source : Masayuki,
2010 )
SAR data based mangrove mapping

ALOS/PALSAR Full Polarimetric data was used to analyse feasibility of SAR data
for mangrove mapping as well as overall land cover classification. Two PALSAR scenes
were used to cover the study area. The scenes were classified separately and mosaicked to
create the full map. ALOS/PALSAR product level 1.1 was used, which is single look
complex (SLC) data in slant range, data file contains 32 bit real and 32 bit imagery
components. The data will reduce its resolution from 12m to 30m after multi-looking process.

Coherency matrix [T3] matrix was generated which is needed to identify incoherent
objects (object having distributed scatter) such as natural and man-made object. Lee refined
filter was used to remove the speckle in the image, first run was done using 3X3 kernel, for
the second run 5X5 kernel was used on the already filter image which resulted in an smooth
image. Coherency matrix was geo referenced using sensor parameters contained in the
“LED” file provided by with the data. The H/A/α polarimetric decomposition theorem by
Cloude and E.Pottier, 1997 was used to decompose the [T3] matrix into Entropy, Ansitropy
and Alpha Eigen vector components. Using Wishart supervised classification algorithm the
SAR images was classified in to 8 land cover classes. Due to the limitation of the software it
was not possible to create the error matrix to carryout accuracy assessment of the results.
Thus quantitative analysis was done comparing results from AVNIR-2 and PALSAR data.

The result has shown it’s possible to use the SAR data for land cover mapping. It was
able to distinguish forest and mangrove without difficulty, the forest and mangrove which
were difficult distinguish in ALOS/AVINR -2 data was easily distinguished. SAR results
were compared with the AVNIR-2 results as a qualitative analysis. The area shown in red on
Figure 6(a) is classified as mangrove in SAR data but classified as some other land use in
AVNIR-2. Field data confirms this area to be mangrove. The SAR data has also has
identified the missing mangrove area in the ALSO/AVINIR-2 classification (Marked by
White Square in Figure-2 (b)). In Figure 6 (b) Red spots indicates scrubs area in according to
AVNIR-2 but classified as mangrove in the SAR data. There has been some miss
classification in the SAR data, but the result has to verify through field verification. Figure 6
(c) shows area of agriculture in the AVNIR-2 data but misclassified as mangrove in the SAR
data. There is considerable amount of error in the west costal area, where there is lots of
coconut trees and other agriculture. This can be explained by lack of agricultural training area
for SAR classification in the second SAR image, as the field work was collected only
considering ALOS/AVNIR-2 images.

Figure 5 shows the area extent of each land cover in ALSO/PALSAR and
ALOS/AVNIR-2 images. Most significant different is between forest and agriculture land
use. This is due to the miss classification in the SAR image in the southern west costal area of
the study area. The results show the SAR classification has more mangrove area and less
forest area compared to the ALOS AVINIR-2 this is due to the miss classification in the data.
.

Figure 5 ALOS/PALSAR SAR based land use classification

Figure 4 ALOS/PALSAR & ALOS/AVNIR land cover area comparison

 Mangrove area Scrub/Grassland area in (AVNIR 2010) Agriculture area in (AVNIR 2010)
in (AVNIR 2010)
Area classified as Forest in AVNIR
Area classified as Scrub in AVNIR Area classified as Agriculture in
but classified as Mangrove in SAR
but classified as Mangrove in SAR AVNIR but classified as Mangrove in
Other Land Cover SAR
Other Land Cover Other Land Cover

(a)
(b) (c)

Figure 6 Shows the comparison of mangrove cover comparing 2010 PALSAR & AVNIR results
Mangrove mapping using SAR backscatter coefficient

Due to low above ground biomass value in mangrove it’s possible to accurately calculate
the biomass using SAR backscatter coefficient (Proisy,C. et al., 2000 ). Biomass estimations
from allometric method using the sample data from field work was used to calculate the biomass
values of mangrove the study area. Sample points were only spread along the river due to the
difficulty of field data collection inside the mangrove forest. Mangrove in the river bank and the
near the coast line showed low biomass but increase to the level of saturation point reaching the
100tDMha-1 value when moving in land. Using the calculated values the mangrove biomass was
categorized in to 3 class as 40 t DM ha-1 , 40-80 t DM ha-1 And 81-100 t DM ha-1 . Using the
field data backscatter coefficient ranges were estimated for the three biomass classes. This range
was then use to identify the mangrove area from the SAR image. To improve the accuracy a
buffer zone of 700m was created around water bodies as the most suitable area for mangroves.
Table 2 contains the SAR backscatter based calculated biomass and biomass calculated from
allometric methods for field data points(Komiyama, Eong, & Poungparn, 2008). It can be
observed the SAR base method is accurate for biomass estimation. The SAR data was only
calculated as three ranges to get a homogenous area of pixels. Thus this method can be used to
approximately extract the mangrove forest by building the relationship between the biomass and
the backscatter coefficient of the SAR image. The method has to be improved to get an accurate
map.

Table 1 Relationship between calculated Biomass and Backscatter Coef.

DBH Calculated (t Backscatter Biomass

Tree Name
(cm) DM ha-1) Coef. class
R. Apiculata 49 54.30708 -17 41-80
R. Apiculata 51 56.43706 -17 41-80
R. Apiculata 32 36.20225 -19 41-80
R. Apiculata 52 57.50205 -18 41-80
R. Apiculata 49 54.30708 -17 41-80
R. Apiculata 47 52.1771 -18 41-80
R. Apiculata 51 56.43706 -14 81-100
R. Apiculata 65 71.34692 -12 81-100

Table 2 Backscatter coef. range value for biomass classes

Class Back scatter coefficient

range Sigma ∂
Class 1 (40 t DM ha-1 ) < -20
Class 2 (40-80 t DM ha-1 ) -19 ~ -13
Class 3 (81-100 t DM ha-1 ) -13 ~ -12

Figure 7 SAR Mangrove area map using Backscatter coef.

Conclusions and Recommendations

The study has shown that it’s difficult to use AVNIR-2 for mangrove mapping due to the
lack of SWIR band in the sensor. Thus carrying mangrove forest change detection using
AVNIR-2 results is not suitable due to the inaccuracy of the classification results. Good
alternative would be to use the ALOS/PALSAR sensor for mangrove mapping which has shown
good results in the study. Any other optical sensor with SWIR band will also be good for
mangrove forest identification.

It’s important to use the proper decomposition algorithm when classifying SAR data.
Decomposition technique will extract more meaning full information from the raw SAR data,
thus different decomposition technique will have different result in the classification. Accurate
speckle removal technique applied without using too much data will ensure there will be less
misclassified isolated pixels in the image. Application of SAR data is new for land cover
mapping; the result of the classification showed good accuracy from visual interpretation.
Accurate analysis of SAR base classifications should be done involving more intensive field data
collection. Different decomposition techniques should be tested to see which technique is most
suitable for mangrove mapping. Improvements in the spatial resolution of SAR full polarimetric
data and in the algorithms will ensure that the SAR data will be widely used for land cover
mapping in the coming years.

Use of SAR backscatter coefficient for mangrove biomass calculation and mapping is a
very easy procedure, for biomass estimation and approximate the mangrove forest cover. Thus
this method can easily be used for monitoring of mangrove forest growth and productivity.
Allometric method used for mangrove dry weight calculation is a simple less field intensive
method which can be adopted for nearly all kinds of mangrove forest. The drawback of this
technique is that it can only be adopted in the pioneer and mature stages forests with biomass
less than 100 t DM ha-1 and the area should not contain other vegetation cover with similar
biomass density in the area.
 References

Baker, C., Lawrence, R.L., Montagne, C. and Patten, D. 2007. Change Detection of
Wetland ecosystem using Landsat Imagery and Change Vector Analysis. Wetland. 27: 610-
619.

Ferguson, R.L. and Korfmacher, K.1997. Remote sensing and GIS analysis of seagrass
meadows in North Carolina, USA. Aquatic Botany. 58: 241–258.

Farid Dahdouh-Guebas. 2002. The use of remote sensing and GIS in the sustainable
management of tropical coastal ecosystems. Environment, development and sustainability.4:
93–112.

Holden, H. and Ledrew, E. 1999. Hyperspectral identification of coral reef features’,

International Journal of Remote Sensing. 20(13): 2545–2563.

Hongyu Liu, Shikui Zhang, Zhaofu Li, Xianguo Lu, and Qing Yang.2004. Impacts on
Wetlands of Large-scale Land-use Changes by Agricultural Development: The Small
Sanjiang Plain, China A Journal of the Human Environment. 33(6):306-310.

Komiyama, A. Ong, J.E and Poungparn, S. 2008. Allometry, biomass, and productivity
of mangrove forests: A review . Aquatic Botany. 89:128–137.

Maryati Mohamed, Mohd. Noh Dalimin, and Danny Chew. "Nature Tourism in Binsulok,
Sabah." In Klias-Binsulok Scientific Expedition 1999, edited by Maryati Mohamed, Mashitah
Yusoff, and Sining Unchi, pp. 87-98. Kota Kinabalu: Universiti Malaysia Sabah, 2000.

Phua, M H, Conrad. O, Kamlun, K. U., Michael Fischer, Jürgen Böhner. 2008.

Multitemporal fragmentation analysis of peat swamp forest in the Klias Peninsula, Sabah,
Malaysia using GIS and remote sensing techniques. Hamburg Beitr zur Phys Geograph und
Landschaft.19: 81–90.

Mahfud M Zuhair, Yousif Ali Hussin and Michael Weir. 2003. Monitoring mangrove
forests using remote sensing and GIS. Forest Science.31: 251-258

Pasqualini,V., Pergent-Martini, C., Clabaut, P., Marteel, H. and Pergent, G. 2001.

Integration of aerial remote sensing, photogrammetry, and GIS technologies in seagrass
mapping. Photogrammetric Engineering and Remote Sensing.67(1): 99–105.
 Proisy, C., Mougin, E., Fromard, F. and Karam, M.A. 2000. Interpretation of Polarimetric
Radar Signatures of Mangrove Forests. Remote Sensing Environment. 71:56–66.

Sabah Forestry Department (SFD), Malaysia. 2009. Mangrove forest management &
restoration. Accessed: http://www.forest.sabah.gov.my/download/2008/23_Mangrove.pdf

Scott, D. A. (ed.) 1989. A Directory of Asian Wetlands. IUCN, Gland, Switzerland and
Cambridge, UK. Xiv + 1181 pp, 33 maps

Smits, P.C., Dellepiane, S.G., and Schowengerdt, R.A. 1999. Quality assessment of
image classification algorithms for land-cover mapping: a review and proposal for a cost
based approach. International Journal of Remote Sensing. 20: 1461-1486.

Upanoi, T. and N.K. Tripathi. 2003. A Satellite Based Monitoring of Changes in

Mangroves in Krabi, Thailand. Map Asia 2003: Natural resource management. Accessed:
http://www.gisdevelopment.net/application/nrm/coastal/mnm/ma03213.htm, April 10, 2009

Vas,Justin. 1998. Sabah biodiversity conservation project-Identification of Potential

Protected Areas, Klias Peninsula- An assessment of Tourism Potential. Ministry of Culture,
Environment and Tourism Sabah and Danish Co-operation for Environment and
Development (DANCED) report. pp: 1-32.

View publication stats