Professional Documents
Culture Documents
Monitoring of Deforestation and Fragmentation in Sarawak, Malaysia Between 1990 and 2009 Using Landsat and SPOT Images PDF
Monitoring of Deforestation and Fragmentation in Sarawak, Malaysia Between 1990 and 2009 Using Landsat and SPOT Images PDF
Monitoring of Deforestation and Fragmentation in Sarawak, Malaysia Between 1990 and 2009 Using Landsat and SPOT Images PDF
Regular Article
ISSN 1226-8267
Journal of Forest Science Journal of Forest Science
Vol. 28, No. 3, pp. 152-157, August, 2012
http://dx.doi.org/10.7747/JFS.2012.28.3.152
Abstract
Sarawak is the largest state in Malaysia that covers 37.5% of the total land area. Multitemporal satellite images of
Landsat and SPOT were used to examine deforestation and forest fragmentation in Sarawak between 1990 and 2009.
Supervised classification with maximum likelihood classifier was used to classify the land cover types in Sarawak. The
overall accuracies of all classifications were more than 80%. Our results showed that forests were reduced at 0.62%
annually during the two decades. The peat swamp forest suffered a tremendous loss of almost 50% between 1990
and 2009 especially at coastal divisions due to intensified oil palm plantation development. Fragmentation analysis revealed
the loss of about 65% of the core area of intact forest during the change period. The core area of peat swamp forest
had almost completely disappeared during the two decades.
Key Words: land cover change, fragmentation, remote sensing, GIS, Borneo
Received: June 14, 2012. Revised: August 8, 2012. Accepted: August 9, 2012.
School of International Tropical Forestry, Universiti Malaysia Sabah, UMS Road, Kota Kinabalu, Sabah 88400, Malaysia
Tel: 6088320118, Fax: 6088320876, E-mail: kamlisa@ums.edu.my
Kamlun et al.
linked to fragmentation analysis are invaluable techniques titude, soil conditions and water regime. While there are
of assessing the degree of threats to a forest ecosystem still large tracts of forests remain, the inland areas have been
(Sader et al. 2001). Landscape metrics have been used to modified by shifting cultivation activities. Recent agricul-
monitor the pattern of forest changes that is not readily visi- tural development especially oil palm plantation dominates
ble to human eye or easily detectable by human analysis the coastal plains of Sarawak.
(Frohn 1998). Riitters et al. (2000) developed a forest frag-
Data acquisition and pre-processing
mentation model that classifies forest pixels into six catego-
ries based on forest density and connectivity. This model Multi-temporal Landsat satellite images between 1990
has been useful to multitemporal monitoring of fragmenta- and 2009 were the main data for monitoring deforestation
tion of peat swamp forest in Sabah, Malaysia (Phua et al. in Sarawak. A total of nine scenes were needed to cover the
2008). whole area of Sarawak. Nine images of Landsat5-TM ac-
Sarawak is the largest state (12.3 million ha) in Malaysia quired in 1990 were downloaded from USGS website.
that occupies 37.5% of the total land. One of the driven Multiple Landsat7-ETM+ images in 2009 were down-
forces of land cover change is the dependency on state econ- loaded for each of the nine scene areas for correcting the
omy of forestry that contributes to 52% of state revenue missing line errors. The Landsat7-ETM images contained
(Hansen 2005). Land cover change in Sarawak has been 5% to 40% cloud covers. For each scene area, the best-qual-
monitored for recent decades with Landsat data, which ity image with least cloud cover was placed on top while the
were affected by clouds and shadows. Expansion of agricul- remaining images were used to fill the missing lines. A mi-
tural land especially for oil palm plantation was the main nority filter (3x3) followed by a median filter (5x5) was
cause of deforestation in recent years (Tsuyuki et al. 2011). used to eliminate the linear residues resulted from the miss-
In this paper, we provide an update on land cover changes ing line correction. In addition, three SPOT 4-HRVIR
in the recent two decades and monitor fragmentation of the images acquired in 2010 were used to fill in areas that were
forest types that suffer highest area loss. We used satellite heavily affected by clouds.
images of Landsat and SPOT for monitoring deforestation
Deforestation rates and pattern
in Sarawak at a landscape scale between 1990 and 2009.
Based on the land cover classifications, patterns of forest Post-classification comparison approach was used to
fragmentation were monitored using the forest fragmenta- identify the land cover change in Sarawak between 1990
tion model of Riitters et al. (2000). and 2009. Most of the land cover change studies adopt ei-
ther supervised or unsupervised classification in determin-
Materials and Methods ing the spectrally distinctive land cover classes (Smits et al.
1999). In this study, all the preprocessed satellite images
Study area
were classified using supervised classification with max-
Sarawak is the largest state of Malaysia that is situated in imum likelihood algorithm. Ten land cover classes were de-
o o
Borneo Island. It is located between 0 50’ and 5 00’ latitude fined; intact forest, degraded forest, peat swamp forest
o o
and 109 35’ and 115 40’ longitude. The total area of (PSF), mangrove, rubber, oil palm, grassland, bareland,
2
Sarawak is 123,300 km and about 720 km from Peninsular water and cloud were classified. 5x5 majority filtering was
Malaysia. Sarawak is categorized as humid tropical region applied on the land cover classification maps to reduce the
with a wet season from October to February. The topography salt and pepper effect. Digital Elevation Model (DEM) of
is flat and undulating. Hilly area is found in the southern the Shuttle Radar Topography Mission (SRTM) data was
part, near the border between Sarawak and Kalimantan. used to correct forest areas above an elevation of 60m that
There are five main natural forest types in Sarawak were misclassified as PSF. The areas covered by clouds
namely; hill mixed Dipterocarp forest, peat swamp forest, were masked out from both images.
mangrove forest, kerangas forest (heath forest) and mon- For land cover classification of 1990, reference data were
tane forest. Distributions of these forest types depend on al- generated from historical land use maps with scales of
1:50,000 and 1:25,000 from the Department of Survey and iv. edge, if density>60% and density<connectivity
Mapping, Malaysia. The maps covered areas of Nanga v. transitional, if density is between 40% and 60%
Tamin, Tian, Sibu, Sarikei, Serantok and Kuching. We vi. patch, if density<40%
generated 660 points using user defined point for assessing Forest fragmentation classes can be differentiated using
the accuracy of the land cover classification of 1990. For different threshold values. The interior classis defined with
land cover classification of 2009, 450 points were collected density above 90% and core pixels with density of 100%.
for accuracy assessment. Of these reference points, 105 Any remaining pixels, although rare, is labelled as un-
points were collected in the field works in December 2009 determined (Fig. 1).
and July 2010. The remaining ground truth points were In this study we examined the fragmentation for intact
obtained from high-resolution quick-looks images in forest, peat swamp forest and mangrove forest. These forest
Google Earth. types were extracted from the land cover classifications of
Annual deforestation rate can be calculated as follows 1990 and 2009 for forest fragmentation analysis using the
(Puyravaud 2003), Riitters model in SAGA-GIS. The connectivity values
were calculated with an immediate neighborhood, 3x3
P
moving windows.
Where A1 and A2 are the forest cover at time t1 and t2 re-
spectively (the unit indicates per year or percentage per Results and Discussion
year).
Land cover classification accuracy assessment
Fragmentation analysis
Accuracies of land cover classifications in 1990 and 2009
Two key parameters in the Riitters model are forest den- were assessed using separate sets of reference points. The
sity (Pf) and forest connectivity (Pff). Both parameters are overall accuracies of land cover classifications of 1990 and
calculated for a focal pixel based on neighborhood analysis. 2009 were 86% and 80% respectively. Land cover classi-
Pf within a moving window is calculated as follows, fication accuracy that is higher than 80% is considered ac-
Pf =
number of forest pixels curate and reliable (Swain and Davis 1978). We also calcu-
number of all non water pixels
lated the kappa accuracy for each of the land cover
Similarly, Pff is calculated from the neighborhood with classifications. The Kappa accuracy was 0.84 and 0.75 for
the following formula: land cover classification in 1990 and 2009 respectively. The
number of pixel pairs with both pixels forest
Pff =
number of pixel pairs with at least one pixel forest
Core
Based on the two parameters, the fragmentation model Interior
100
identifies six categories of fragmentation classes of interior,
undetermined, perforated, edge, transitional and patch. Perforated
80
The calculated Pf and Pff were converted to percentage with Edge
a multiplication factor of 100. Using density and con-
Density [%]
60
nectivity, each pixel can be assigned to one of six forest frag- Transitional
mentation classes. If we express the density and con- 40
nectivity ratios as percentages by multiplying it with factor
Patch
100, the fragmentation classes can be defined as follow, 20
i. interior, if density=100%
No Forest
ii. undetermined, if density>60% and density=con- 0
nectivity 0 20 40 60 80 100
iii. perforated, if density>60% and density>con- Connectivity [%]
lion ha, was the largest land cover type in 1990. Although it
remained as the largest land cover type in 2009, more than
550,000 ha were deforested (Fig. 3). Half a million ha of
peat swamp forest was cleared from 1990 to 2009 (Fig. 5).
The forests of Sarawak have been pressured by large-scale
expansion of oil palm development (Tsuyuki et al. 2011).
Our results showed that the forests (intact forest, degraded
forest, peat swamp forest and mangrove forest) were re-
duced at 0.62% annually during the two decades. The oil
palm was the fastest growing agricultural crop in Sarawak.
Oil palm plantation increased about 70% annually between
1990 and 2008 (Mohd Basri and Mohd Arif 2009). Our
results indicated an annual increment of 77.4% for the oil
palm plantation. The Sarawak state government had tar-
geted in 2007 to develop about 1 million ha of oil palm
Fig. 3. Land cover classification of 2009.
plantation by 2010 (MPOB 2007). Our findings showed
that Sarawak’s oil palm plantation has reached about 1 mil-
missing lines in Landsat7-ETM+ data were likely the lion ha in 2009. These recent oil palm plantation develop-
cause of the lower accuracy in the land cover classification ments took place at the peat swamp forest, mainly located at
of 2009. the coastal divisions of Sibu, Mukah, Bintulu and Miri
(Fig. 4).
Land cover change and deforestation
In 1980, 1.6 million ha of peat swamp forest was re-
Fig. 2. shows that the intact forest, with more than 6 mil- ported (Wong 1991). Our results indicated that only about
during the two decades reflect the emphasis of agricultural Forest Ecology and Management 167: 263-284.
development over sustainable forestry in the largest state of Miettinen J, Shi C, Liew SC. 2011. Deforestation rates in insular
Southeast Asia between 2000 and 2010. Global Change Biology
Malaysia. Similar trend is expected to continue unless more
17: 2261-2270.
forests are gazetted as permanent forest reserves. Mohd Basri W, Mohd Arif S. 2009. Issues related to Production
Cost of Palm Oil in Malaysia. Oil Palm Industry Economic
Acknowledgements Journal 9: 1-12.
Nor Rasidah H. 2010. Analyzing long-term landscape changes in a
borneon forest reserve using. Aerial photographs. Open Geogr J
We thank Sarawak Forestry Department for support and
3: 161-169.
cooperation. This study was supported from the Grant of Ozesmi SL, Bauer ME. 2002. Satellite remote sensing of wetlands.
Asia Pacific Network for Global Change Research Wetlands Ecology and Management 10: 381-402.
(ARCP2010-02CMY-Phua) and Postgraduate Research Phillips VD. 1998. Peatswamp ecology and sustainable develop-
Grant of Universiti Malaysia Sabah (GPS0016). ment in Borneo. Biodiversity and Conservation 7: 651-671.
Phua MH, Conrad O, Kamlisa UK, Fischer M, Böhner J. 2008.
In: Multitemporal Fragmentation Analysis of Peat Swamp
References Forest in the Klias Peninsula, Sabah, Malaysia using GIS and
Remote Sensing Techniques (Böhner J, Blaschke T, Montanare-
Abdullah SA, Nakagoshi N. 2006. Forest fragmentation and its lla L, eds). Hamburger BeiträgezurPhysischenGeographie und
correlation to human land use change in the state of Selangor, Landschaftsökologie, Universität Hamburg Institutfür Geo-
peninsular Malaysia. Forest Ecology and Management 241: graphie, Hamburg, pp 81-90.
39-48. Puyravaud JP. 2003. Standardizing the calculation of the annual
DeFries R, Achard F, Brown S, Herold M, Murdiyarso D, rate of deforestation. Forest Ecology and Management 177:
Schlamadinger B, de Souza Jr C. 2007. Earth observations for 593-596.
estimating greenhouse gas emissions from deforestation in de- Riitters K, Wickham J, O'Neill R, Jones B, Smith E. 2000.
veloping countries. Environmental Science and Policy 10: Global-scale patterns of forest fragmentation. Conservation
385-394. Ecology 4: 3.
FAO. 2007. Definitional issues related to reducing emissions from Sader SA, Hepinstall DJH, Coan M, Soza C. 2001. Forest change
deforestation in developing countries. Forest and Climate monitoring of a remote biosphere reserve. Int J Rem Sens 22:
Change Working Paper. Vol. 5. 1937-1950.
Frohn RC. 1998. Remote Sensing for Landscape Ecology: New Skole D, Tucker C. 1993. Tropical deforestation and habitat frag-
Metrics for Monitoring, Modeling, and Assessment of Ecosy- mentation in the Amazon: satellite data from 1978 to 1988.
stems. Lewis Publishers, Florida. Science 260: 1905-1910.
Gasparri NI, Grau HR. 2009. Deforestation and fragmentation of Smits PC, Dellepiane SG, Schowengerdt RA. 1999. Quality as-
Chaco dry forest in NW Argentina (1972-2007). Forest Ecol sessment of image classification algorithm for land-cover map-
Manage 258: 913-921. ping: review and a proposal for a cost- based approach. Int J
Hansen TS. 2005. Spatio-temporal aspect of land use and land cov- Rem Sens 20: 1461-1486.
er change in the niah catchment, Sarawak, Malaysia. Singapore Sohel MSI, Rana MP, Islam AZMZ, Akhter S. 2011. Monitoring
J Trop Geogr 26: 170-190. land-use changes by remote sensing and GIS techniques: case
IPCC. 2007. Climate Change 2007: Synthesis report. Intergovern- study of barind tract, Bangladesh. Journal of Forest Science 27:
mental Panel on Climate Change (IPCC). http://www.ipcc.ch. 73-79.
Accessed 24 Dec 2011. Swain PH, Davis SM. 1978. Remote sensing: the quantitative
Majid-Cooke F. 1999. The challenge of sustainable forests: forest approach. McGraw-Hill, New York.
resource policy in Malaysia, 1970-1995. Tsuyuki S, Goh MH, Teo S, Kamlisa UK, Phua MH. 2011.
Malaysian Palm Oil Board (MPOB). 2007. Sarawak Sasar 1 Juta Monitoring deforestation in Sarawak, Malaysia using multi-
Hektar 2010. temporal landsat data. Kanto For Res J 62: 87-90.
Malaysia Oil Palm Board (MPOB). 2007. Industri Sawit Tingkatkan Turner MG, Gardner RH, O'Neill RV. 2001. Landscape Ecology
Sosioekonomi. Malaysia Palm Oil Board. http://mpob.gov. in Theory and Practice: Pattern and Process. Springer Science,
my/html/05_pi/gfx_bs/bs66_01a.jpg. Accessed 18 Aug 2012 New York.
Manier DJ, Laven RD. 2001. Changes in landscape patterns asso- Wong MH. 1991. The Distribution Characteristics and Agricul-
ciated with the persistence of aspen (Populus tremuloides tural Utilization of Peat in Sarawak. Kuching, Malaysia.
Michx.) on the westernslope of the Rocky Mountains, Colorado. Sarawak Agriculture Department, Kuching.