JFS

(1)

Vol. 29, No. 1, pp. 49-57, February, 2013 http://dx.doi.org/10.7747/JFS.2013.29.1.49

Classification of Degraded Peat Swamp Forest for Restoration Planning at Landscape Level Using Remote Sensing Technique

Khali Aziz Hamzah*, Azahan Shah Idris and Ismail Parlan Forest Research Institute Malaysia, Kepong Selangor 520109, Malaysia

Abstract

Malaysia possesses about 1.56 million ha of Peat Swamp Forest (PSF). The PSF safeguard enormous biological diversity, while providing crucial benefits for the sustainable development of human communities. Numbers of threatened plant species are associated with the PSF, including the commercially important Gonystylus bancanus timber. To prevent significant losses of biodiversity, it is important to manage the PSF for both biological conservation and sustainable use. Equally important is to restore all degraded PSF in an attempt to ensure the PSF ecosystem is suitable for the vegetation to grow and rehabilitate back to the normal condition. Prior to plan any forest restoration program, there is a need to properly map the degraded PSF in order to estimate the forest conditions and determine the vegetations status. Most of the time this need to be done at a landscape level and requires a technology that can provide accurate, timely and reliable information for the planner to make decision. This paper describes a study using geospatial technology in combination with ground survey to classify the degraded PSF in South East Pahang Peat Swamp Forest (SEPPSF), Malaysia, into different degree of vegetation classes. With map accuracy of about 83%, the technique proved to be useful in delineating the different degree of PSF degradation from which the information can be used to properly plan forest restoration program in the area. The final output which is in the form of map can be used in developing a Restoration Master Plan for the degraded PSF areas.

Key Words: peat swamp forest, remote sensing, forest restoration, Landsat TM, classification

Received: August 14, 2012. Revised: December 13, 2012. Accepted: December 17, 2012.

Corresponding author: Khali Aziz Hamzah

Forest Research Institute Malaysia, Kepong Selangor 52109, Malaysia Tel: 603-62787201, Fax: 603-62729852, E-mail: [email protected]

Introduction

Malaysia possesses about 1.56 million ha of Peat Swamp Forest (PSF). PSF is waterlogged forest growing on a layer of dead leaves and plant material up to 20 meters thick. The PSF ecosystem is characterized by waterlogging with low nutrients and dissolved oxygen levels in acidic water regimes. This harsh waterlogged environment has led to the evolution of many species of flora uniquely adapted to

these conditions. The PSF safeguard enormous biological diversity, while providing crucial benefits for the sustainable development of human communities. The PSF areas contain some unique and endangered flora and fauna.

Numbers of threatened plant species are associated with the PSF, including the commercially important Ramin (Gony- stylus bancanus). Viable populations of many globally threat- ened animal species also occur in the PSF, such as the Sumatran Rhinoceros, Storm’s Stork, Wrinkled Hornbill

(2)

and False Gharial. To prevent significant losses of biodiversity, it is important to manage the PSF for both biological conservation and sustainable use (Shamsudin 1996).

Equally important is to restore all degraded PSF in an attempt to ensure the PSF ecosystem is suitable for the vegetation to grow and rehabilitate back to the normal condition.

As recommended by the UNDP (2007) there is a need to map the degraded PSF at landscape level preferably using remote sensing for rehabilitation planning. As suc this study was undertaken.

The use remote sensing technique in the field of forestry has been well explored worldwide and in tropical countries among others, studies by Langner et al. (2007), Phua et al.

(2007) and Yoshino et al. (2010) indicates that it can be used for various forestry applications. Seca Gandaseca et al.

(2009) successfully used remote sensing data to assess the impact of logging on PSF in Sarawak, Malaysia where as Khali Aziz et al. (2011) demonstrates that hypersepctral remote sensing data can be used to map distribution of im- portant CITES-listed timber species (Gonystylus bancanus) in the PSF. Various studies also shows that for peat swamp forest remote sensing can provide useful information for hydrological and forest fire modeling, retrieval of bio- physical parameters, and the management of natural resources (Wijaya et al. 2010; Razali and Ainuddin 2012).

Landsat TM data also had been used successfully to detect timber extraction roads in peat swamp forest (Khali et al.

1995) of Malaysia.

The main objective of this study is to classify the degraded tropical PSF into different degree of vegetation classes based on remote sensing data. The main expected outcome from the study will be a map showing the different classes of vegetations conditions in the area. The use of satellite imagery with the capability of extracting information at a landscape level was chosen and this is in line with the ITTO/IUCN (2005) forest restoration concept in the tropics as detailed out in the Guidelines for the Restoration, Rehabilitation and Management of Degraded and Secondary Tropical Forests and the IUCN (2002). The ITTO/IUCN (2005) defines Forest Landscape Restoration (FLR) as a process that aims to regain ecological integrity and enhance human well-being in deforested or degraded forest landscapes. The final output which is in the form of map can be used in developing a Restoration Master Plan for the degraded PSF areas.

Materials and Methods

Styudy area

The study area covering an area of 8,944 ha is located in the state of Pahang, Malaysia, better known as the South East Pahang Peat Swamp Forest (SEPPSF) (Fig. 1). The area is adjacent to a large block of pristine Pekan Perman- ent Reserved Forest (PRF). This study area was just been gazetted as forest reserve in 2008 due to a strong proposal by the UNDP/GEF PSF Project that was conducted in the SEPPSF in 2004-2007 (UNDP/GEF 2007). Prior to this gazettment, the area is a stateland forest and mostly have been disturbed with different degree of degradations due to various human activities (logging and encroach- ment.) as well as other uncontrolled phenomenon (fires, flood, etc.). Fig. 2 shows the general land cover conditions and the extent of the study area as seen in the Landsat TM image.

Rainfall

Rainfall intensity and distribution across the PSF is important as the peat swamp’s water source is principally from rain. Rainfall information of the study area was collected from tree selected stations located within the vicinity of the study area over a 7 year span, from 1995-2001 and presented in Fig. 3. The project area experiences a relatively drier period lasting 8 months from February to September, followed by four months of heavy rain between October and January, the peaks being in December and January.

December records the highest mean monthly rainfall at the three selected stations with an average of 483 mm and the driest July with 106 mm. Information on rainfall is very important for this research since it will influence the field survey planning as it is not suitable to enter the PSF area during heavy rain.

Vegetation

Studies undertaken indicated that the virgin PSF is relatively rich in plant species though less than the dry inland forest and has a significant number of endemics species.

Based on the multidisciplinary assessment (MDA) report (UNDP 2003; Nizam et al. 2009), the PSF can be charac- terized by the abundance of Shorea platycarpa, Gonystylus bancannus, Kompassia malaccensis, Durio carinatus and

(3)

Fig. 1. Map showing the location of the study area in the South East Pahang Peat Swamp Forest, M alaysia.

Fig. 2. Landsat TM image of the study area (Bands 321).

Fig. 3. Mean monthly rainfall distribution at three selected stations in the vicinity of the study area (Source: MMS 2001).

Calophyllum spp. However, in the study area which is basi- cally a degraded PSF due to past status as stateland forest, the vegetations are dominated by pioneer species such Rhodomyrtus tomentosa, Rhodomina cineria, Fragrea fragrans, Ploiarium alternifolium, Nephentes spp., Vitex spp., Macaranga

gigantia, Dillenia spp., and with few important timber spe- cies such as Callophylum ferrugineum and Madhuca montleyana.

Information on vegetation composition is very important as only suitable tree species should be planted in the degraded areas during PSF restoration activities.

Data analysis

Satellite data of Landsat 5 TM acquired on 13/9/2004 with less than 30% cloud cover was used for the analysis.

Other ancillary data includes topographic map L7030 ser-

(4)

Fig. 4. Digital image processing methodology used to derive the vegetation map.

ies (scale 1:50,000) and geospatial-based data, which include district and forest reserves boundaries were used for the analysis.

Information extraction from the satellite image was done based on the standard digital image processing procedure as shown in the flowchart (Fig. 4). The preprocessing tasks including image geocoding, enhancement and masking were done prior to final supervised classification. Data quality in the study area is affected by some noise due to haze and clouds covers. This requires a special noise removal processes including haze removal, cloud and water body masking prior to further analysis. The area of interest in the imagery was then subseted before applying Unsupervised and Supervised normalized difference vegetation index (NDVI) classifications.

The NDVI is a non-linear transformation of the visible (red) and near-infrared bands of satellite information.

NDVI is defined as the difference between the visible (red)

and near-infrared (nir) bands, over their sum. The NDVI is an alternative measure of vegetation amount and condition. It is associated with vegetation canopy character- istics such as biomass, leaf area index and percentage of vegetation cover. In this study the NDVI algorithm is used in order to classify the vegetation condition in study area.

NDVI is calculated using the following formula:

[Near-infrared (NIR)-Red]/[Near-infrared(NIR)＋Red]

or for Landsat 5 TM image;

[Band 4-Band 3]/[Band 4＋Band 3]

Unsupervised classification

The unsupervised method used the ISODATA approach to classify and recalculates statistics of the image signature. This method uses minimum spectral distance to assign a cluster for each candidate pixel. The process begins with a specified number of arbitrary cluster means or the means of existing signatures, and then it processes repeti- tively, so that those means will shift to the means of clusters in the data.

Ground truth survey

Ground survey is necessary to verify the vegetation classes derived from the unsupervised classification. The survey will confirm the actual ground conditions and to ass- es the accuracy of the overall image classification. Other ground information in particular on the vegetation types, tree diameter size as well as soil conditions were also collected which can be used to assist the restoration planning.

Due to accessibility problem in the peat swamp forest, only a total of 17 randomly distributed sampling points were vis- ited during the ground survey.

Supervised classification

Supervised classification is an approach where the image interpreter has an experience and first hand knowledge of ground data and also with other references such as aerial photo, topographic map or other classified satellite images.

Supervised classification was performed to the satellite image taking into account the unsupervised classification results as well as information from the field ground truth survey.

(5)

Fig. 5. Landsat NDVI image of the degraded PSF in the study area.

Fig. 6. Vegetation classes map of the study area based on unsupervised classification of Landsat TM image with locations of ground truth points.

Results

NDVI

The NDVI was calculated from Landsat 5 TM information by using the combinations of bands 3 (0.63-0.69 mm) and 4 (0.76-0.90 mm). Healthy vegetation will have a high NDVI value. Bare soil and rock reflect similar levels of near-infrared and red and so will have NDVI values near zero. Clouds and water are the opposite of vegetation in that they reflect more visible energy than infrared energy, and so they yield negative NDVI values. The result of NDVI image of the study area is shown in Fig. 5. The image of the NDVI alone is a meaningful picture of the vegetation pat- terns in the PSF. The high dense vegetations (with high NDVI value) can be distinguished from the low dense vegetation and grass land area. The NDVI value decreases as the tree canopy opens and mixes with shrub and grassland as well as bare land.

Unsupervised classification

Unsupervised classification for the study area was set to five (5) classes with gray scale level from low to high value.

Different colours were assigned for each class to make it easier to visualize for ground truth purposes. The five unsupervised classes of vegetation are shown in Fig. 6.

Ground truth survey

The ground checking in the study area shows that few points that were expected to be forested areas were found to be burnt area and oil palm plantation. Apparently these oil palm plantations belong to an indigenous local community in the Simpai Village. The ground survey also indicates that the Class 1 and Class 2 was consistently an open area with grass and lallang but more dense in Class 2. Class 1 is an open area with patches of grasses and absent of trees, some repeatedly burnt (leaving old stump in some areas).

Melastoma malabatricum, and Scelaria spp. are common and in some area Imperata cylindrical dominated the area. The Class 2 is basically areas dominated with shrubs and small trees (some up to 5 m in height), some of the areas area rather open. Only few of species of vegetations can be found in this area such as Rhodo myratus spp., F. fagrans, and Vitax spp. In some cases ferns and Scelaria spp. dominated the areas.

Areas with big and small tress were mixed up in Classes 3, 4 and 5. The area is dominated with secondary forest species, having rather closed canopy. It is the densest vegetated areas as compared to the previous two classes. Some of the tree species recorded in the area include C. ferrugineum, Kokona spp. Litsea spp., M. montleyana, Pometia pinnata,

(6)

Table 1. Final vegetation classes in the degraded PSF area

Vegetation classes Area extent (ha) Vegetation description Class 1

Class 2

Class 3

Class 4 Class 5

Grass/lallang

Shrubs/small trees

Vegetated/forested

Oil palm Unclassified Total

1,272.98

173.67

6,891.19

419.78 186.38 8,944.00

Open area with patches of grasses and absent of trees, some repeatedly burnt (leaving old stump in some areas). Melastoma malabatricum and Scelaria spp. are common.

Presence of Imperata cylindrical (lallang).

Areas dominated with shrubs and small trees (some up to 5 m in height), some of the areas area rather open. Some common tree species include Rhodo myratus spp, Fagrea fagrans and Vitax spp. In some cases ferns and Scelaria spp.dominated the areas.

The area is dominated with secondary forest species, having rather closed canopy. It is the densest vegetated areas as compared to the previous two classes. Some of the tree species recorded in this class include (figures in bracket is the diameter size measured at breast height (dbh) in cm):

Callophylum ferrugineum-(34 cm), Kokona spp.-(35 cm)

Litsea spp.-(32 cm)

Maduca montleyana-(23.5 cm) Pometia pinnata-(21.5 cm) Artocarpus-(21 cm) Eugenia spp.-(25 cm)

Callophylum ferrugineum-(17 cm) Macaranga gigantia-(15 cm) Ferns

Bamboos More than 6 years old Rivers and cloud shadow Artocarpus spp. and Eugenia spp. Most of them are timber

species with diameter sizes ranges from 15-35 cm dbh (diameter at breast height). With proper management it is anticipated that in the long term this forest area will be able to recover back to the normal condition.

Supervised classification

Data from ground truth was analyzed and final classification was produced for the study area. After incorporat- ing information gathered from the ground truth, the unsupervised 5 vegetation classes were reclassified into three vegetation classes as summerised in Table 1. The classes 3, 4 and 5 were merged to one class known as the Vegetated/

forested type due to the almost similar vegetation conditions as verified in the field surveys. The final vegetation classes after taking into consideration the ground truth data are, Grass/lallang (Class 1), Shrubs/small trees (Class 2), Vegetated/forested (Class 3), Oil palm plantation (Class 4) and Class 5 for the unclassified area. The supervised classi-

fication results of the study area with brief description of the vegetation conditions are presented in Table 1 and Fig. 7 shows the final map of the distribution of the different vegetation classes in the study area after performing edge enhancement filter.

The results show that majority of the areas fall in the Class 3 (Vegetated/forested) which is having dense vegetation conditions. These areas cover about 77% of the PSF in the study area. These areas can be considered as a secondary forest undergoing succession resulting from dis- turbances including logging. The most degraded areas are the Class 1 (Grass/lallang) which is normally an open area with no tree vegetation. This is an area which requires at- tention in the restoration programme in which it is almost impossible to let the area to regenerate naturally.

Accuracy assessment

Accuracy assessment is a process of comparing the classification results to geographical data or ground truth data.

(7)

Fig. 7. Vegetation classes map of the study area after supervised classification of landsat TM image.

Table 3. Accuracy results for the classified image in the study area

Class name Reference totals Classified totals Number correct Producers accuracy (%) User accuracy (%) Grass/lallang

Shrubs/small trees Vegetated/forested

5 4 9

3 4 11

3 3 9

60.00

75.00

100

100 75.00 81.82 Overall classification accuracy=83%.

Table 2. Accuracy results for the classified image in the study area

Classified data Reference data

Total classified data Grass/lallang Shrubs/small trees Vegetated/forested

Grass/lallang Shrubs/small trees Vegetated/forested Total reference data

3 1 1 3

0 3 1 8

0 0 9 6

3 4 11 The producer’s accuracy is refers to the probability that a

certain land-cover of an area on a ground is classified as such, while user’s accuracy refers to the probability that a pixel labelled as a certain land-cover class in the map is true.

The assessment can be used as a basis in determining the overall accuracy of the maps produced. Tables 2 and 3 show the error matrix and the accuracy results respectively of the map produced. The overall accuracy is 83% which can be accepted for Landsat TM satellite-based forest cover mapping.

Discussion

Forest restoration planning

In Peninsular Malaysia, as part of the forest management and silvicultural systems, there is a requirement to en- rich poorly-stocked degraded forest. One of the approaches is by undertaking forest restoration in the degraded areas.

Map produced using remote sensing data in this study was used to plan a forest restoration plan in the study area. The restoration planning was also in line with the ITTO/IUCN (2005) FLR in the sense that a satellite Landsat TM 5 image with 30 meter spatial resolution was used to map the vegetation conditions in the focus area. The use of the satellite data enable the assessment of the overall condition of the ecosystem in the study be made at a landscape level in line with the ITTO/IUCN (2005) FLR approach. Another advantage of the satellite data is on the possibility of captur- ing information even in the area where accessibility is a con- straint such as the PSF under study.

From the map, the study area can be broadly categorized into three different vegetation classes namely the Class 1- the Grass/lallang dominated area, Class 2-the Shrub/small trees dominated area and the Class 3-Vegetated/forested area. The restoration of the three different vegetation classes in the area requires different approaches depending on various factors including the existing vegetation conditions and the accessibility to the areas. Furthermore the restoration should be carried out in stages and priorities depending on the availability of planting materials as well as budget

(8)

allocation. For the Class 1 and Class 2 areas which is rather open and absent of trees, restoration in this area ideally should be carried out based on an open planting technique.

In the absent of big trees, the area generally receives direct and full sunlight, hence only certain plants species that can withstand sunlight are suitable to be planted in this area such as Anisoptera marginata and Shorea platycarpa. The open planting can be done in a row base on a planting distance of 5x5 m (5 m in line and 5 m between lines) in which a total of about 400 seedlings are required per hectare.

Ismail et al. (2006) describes in detail the different planting techniques in the PSF in Malaysia.

The area under the Class 3 is basically a secondary forest with different degree of succession stage. The tree canopy are rather closed and in some cases there are still big trees with common tree species including Calophyllum spp., Koompassia malaccensis, Polyalthia glauca, Syzgium napiforme, Ficus spp., Cratoxylum formosum and also the presence of other lower ground species including palm, rattan and bamboo. In some areas, a Macaranga spp. dominated stands can also be found in this vegetation class. Generally the areas have been logged previously and hence it is dominated with secondary species and poor in term of quality timber species. Hence, they still need to be replanted with suitable PSF species including Anisoptera marginata, Shorea platycarpa, Dyrea costulata, Gonystylus bancanus and C. ferugi- neum. Due to the presence of trees, the most suitable plant- ing technique for this area is to be based on a line planting method. Based on several studies, line planting is the most successful method in restoring the degraded PSF (Ismail et al. 2006). The line planting method is also known as

‘under-canopy planting’. It is important to note that a total line clearance is needed in order to allow for sufficient light penetration to the planted seedlings. Another important ele- ment in this technique is the need to do continuous weeding along the line to avoid competition with weeds.

Conclusion

Using satellite data, the PSF area was finally stratified into three different vegetation classes. The final vegetation map produced can further be used to plan the restoration program in the area in an effort to improve vegetation growth in the degraded PSF area. The PSF plays im-

portant roles in economic development and maintaining sound environment. Therefore it is important to manage the PSF on sustainable basis. Several studies proved that even high degraded PSF could be restored besides the more moderate degraded areas. Through appropriate restoration programs, productivity of degraded PSF could be improved and this will sustain the production of high quality timber species and maintaining sound environment.

The main output of this research is a map showing the distribution and extent of degraded PSF which can be used by the Pahang Forestry Department to plan forest restoration programme in the area. The ability of satellite image in classifying the different vegetation conditions in the degraded PSF will assist planner to have better under- standing on the area and prepare appropriate restoration or reforestation strategies. Since restoration activity in PSF is relatively expensive and more difficult due to accessibility problem, the information provided through the use of satellite image is very important and crucial in ensuring the restoration program can be successfully implemented. This is of special value in which remote sensing application and information able to support restoration planning and strategies in the PSF ecosystem. The method developed in this study can be applied to other PSF in the country when mapping of degraded PSF for restoration planning is desired.

References

Ismail P, Khali Aziz H, Zainudin MA. 2006. Rehabilitation trials in peat swamp forests of Peninsular Malaysia. Malaysian Fore- ster 69: 1-14.

ITTO/IUCN. 2005. Restoring forest landscapes. An introduction to the art and science of forest landscape restoration. ITTO Technical Series No 23 International Tropical Timber Organi- zation.

IUCN. 2002. Guidelines for the restoration, rehabilitation and management of degraded and secondary tropical forests.

Khali Aziz H, Azman H, Ismail H, Chong PF. 1995. Detection of timber extraction roads in peat swamp forest using Landsat TM imagery. Paper presented during the Asian Conferecne on Remote Sensing, Thailand, 20-25 November 1995.

Khali Aziz H, Mohd Azahari F, Tan SA, Hamdan O. 2011.

Generation of Spatial Distribution Maps of Gonystylus bancanus (ramin) using Hyperspectral Technology. In: Proceedings of the Regional Workshop on the Sharing of Findings from the Activities Implementation in Indonesia and Malaysia under the

(9)

ITTO-CITES Project on Ensuring International Trade in CITES-Listed Timber Species are Consistent with their Sustainable Management and Conservation (Khali Aziz H, Ismail H, Ismail P, Mohd Azahari, F, Abd Razak O, eds).

Forest Research Institute Malaysia and Ministry of Natural Resources and Environment Malaysia, pp 27-33.

Langner A, Miettinen J, Siegert F. 2007. Land cover change 2002-2005 in Borneo and the role of fire derived from MODIS imagery. Global Change Biology 13: 2329-2340.

Malaysian Meteorological Service (MMS). 2001. Monthly abstract of meteorological observations. Petaling Jaya, Selangor, Malaysia.

Nizam MS, Ismail P, Latiff A, Shamsudin I, Faridah-Hanum I.

2009. Diversity of tree communities and its relationships with soil properties in a peat swamp forest in Pahang, Peninsular Malaysia. Ecology, Environment and Conservation 15: 307- 318.

Phua MH, Tsuyuki S, Lee JS, Sasakawa H. 2007. Detection of burned peat swamp forest in a heterogeneous tropical landscape:

A case study of the Klias Peninsula, Sabah, Malaysia. Land- scape and Urban Planning 82: 103-116.

Razali SM, Ainuddin AN. 2012. A method of mapping forest fuel

ypes in peat swamp forest. African Journal of Agricultural Research 7: 901-1909.

Seca Gandaseca, John Sabang, Osumanu Haruna Ahmed, Nik Muhamad Ab. Majid. 2009. Vegetation Assessment of Peat Swamp Forest Using Remote Sensing. American Journal of Agricultural and Biological Sciences 4: 167-172.

UNDP/GEF. 2003. South East Pahang Peat Swamp Forest Multi Disciplinary Assessment (MDA). Volume II: Site Assessment Report. Unpublished Report.

UNDP/GEF. 2007. Integrated Management Plan of the South- East Pahang Peat Swamp Forest. PSF Technical Series No. 9.

Kepong: FRIM-UNDP/GEF Peat Swamp Forest Project in collaboration with the Pahang Forestry Department.

Wijaya A, Liesenberg V, Gloaguen R. 2010. Retrieval of Forest Attributes in Complex Successional Forest of Central Indonesia:

Modeling and Estimation of Bitemporal Data. Forest Ecology and Management 259: 2315-2326.

Yoshino K, Ishida T, Nagona T, Setiawan Y. 2010. Land cover pat- tern analysis of tropical peat swamp lands in Southeast Asia.

International Archives of the Photogrammetry, Remote Sensing and Spatial Information Science, Volume XXXVIII, Part 8, Kyoto Japan 2010: 941-946.