Land cover distribution in the peatlands of Peninsular Malaysia, Sumatra and Borneo in 2015 with changes since 1990

Published Date
Global Ecology and Conservation
April 2016, Vol.6:67–78, doi:10.1016/j.gecco.2016.02.004
Open Access, Creative Commons license

Original research article

Author 

• Jukka Miettinen ^,
• Chenghua Shi 
• Soo Chin Liew 

• Centre for Remote Imaging, Sensing and Processing (CRISP), National University of Singapore (NUS), 10 Lower Kent Ridge Road, Blk S17, Level 2, Singapore 119076, Singapore

Received 18 December 2015. Revised 15 February 2016. Accepted 15 February 2016. Available online 5 March 2016. 

Abstract

Insular Southeast Asian peatlands have experienced rapid land cover changes over the past decades inducing a variety of environmental effects ranging from regional consequences on peatland ecology, biodiversity and hydrology to globally significant carbon emissions. In this paper we present the land cover and industrial plantation distribution in the peatlands of Peninsular Malaysia, Sumatra and Borneo in 2015 and analyse their changes since 1990. We create the 2015 maps by visual interpretation of 30 m resolution Landsat data and combine them with fully comparable and completed land cover maps of 1990 and 2007 (Miettinen and Liew, 2010). Our results reveal continued peatland deforestation and conversion into managed land cover types. In 2015, 29% (4.6 Mha) of the peatlands in the study area remain covered by peat swamp forest (vs. 41% or 6.4 Mha in 2007 and 76% or 11.9 Mha in 1990). Managed land cover types (industrial plantations and small-holder dominated areas) cover 50% (7.8 Mha) of all peatlands (vs. 33% 5.2 Mha in 2007 and 11% 1.7 Mha in 1990). Industrial plantations have nearly doubled their extent since 2007 (2.3 Mha; 15%) and cover 4.3 Mha (27%) of peatlands in 2015. The majority of these are oil palm plantations (73%; 3.1 Mha) while nearly all of the rest (26%; 1.1 Mha) are pulp wood plantations. We hope that the maps presented in this paper will enable improved evaluation of the magnitude of various regional to global level environmental effects of peatland conversion and that they will help decision makers to define sustainable peatland management policies for insular Southeast Asian peatlands.

Keywords

Peat swamp forest

Industrial plantation

Kalimantan

Oil palm

Pulp wood

Deforestation

---

1 Introduction

Insular Southeast Asia has faced rapid environmental changes over the past few decades and it is currently one of the global hotspot areas of deforestation, forest degradation, tropical peat fires and plantation development (Achard et al., 2002, Corlett, 2009, van der Werf et al., 2010, Miettinen et al., 2012a, Margono et al., 2014, Miettinen et al., 2014 and Stibig et al., 2014). The intensity and rapidity of these changes, as well as the associated environmental problems, are perhaps best seen in the peatlands of the region (Miettinen et al., 2012b). Due to the difficult working conditions for heavy machinery, low agricultural potential and sufficient availability of land on mineral soils, the 25 Mha of peatlands in Southeast Asia (equal to 56% of all tropical peatland; Page et al., 2011) were left largely undeveloped until the 1980’s. In their natural state, peat swamp forests form a carbon sink which has resulted in an immense carbon deposit (∼69 Gt) in the peatlands of the region (Page et al., 2011). In addition, peatlands support specialized flora and fauna, partially endemic to the region, play an essential role in hydrology by regulating the water flow and have significant societal values for local people (Giesen, 2004, Rieley and Page, 2005 and Corlett, 2009).

However, since the 1980’s the peatlands of insular Southeast Asia have been increasingly utilized (Silvius and Diemont, 2007), inducing significant ecological, hydrological and atmospheric effects. Extensive logging activities over the 1990’s made peat swamp forests highly susceptible to fire (Siegert et al., 2001) and lead to catastrophic fire damage during the 1997–1998 El Niño season resulting in massive carbon emissions (Page et al., 2002). Due to very slow natural regeneration of burnt peat swamp forests, often hindered by dense ferns and recurrent fire activity (Langner and Siegert, 2009, Page et al., 2009 and Blackham et al., 2014) the majority of the 1997–1998 burnt areas remained as degraded peatlands. Drainage and conversion of peatland areas to plantations and agriculture gained momentum over the first decade of the new millennium, leading to remarkable expansion of fire prone peatland areas with lowered water table levels. Aerobic conditions in the upper peat profile, often combined with change in vegetation cover and use of fertilizers, result in increased carbon emissions from peat oxidation (Hooijer et al., 2012, Hooijer et al., 2014, Jauhiainen et al., 2012, Jauhiainen et al., 2014, Hirano et al., 2014 and Sakata et al., 2015) and make the top layers of peat vulnerable to fires (van der Werf et al., 2008 and Gaveau et al., 2014). Carbon emissions associated with peatland drainage and cultivation (Couwenberg et al., 2010, Hooijer et al., 2010 and Miettinen et al., 2012a) as well as with recurrent peat fires (Page et al., 2002, van der Werf et al., 2008, van der Werf et al., 2010 and Gaveau et al., 2014) make Indonesia one of the top emitters of greenhouse gases in the world and directly affect global climate change.

Peatland deforestation, drainage and conversion to agriculture drastically changes peatland ecosystems and may jeopardize the existence of plant and animal species endemic to Southeast Asian peatlands (see e.g. Giam et al., 2012). The region is one of the biodiversity hotspots in the world but is currently experiencing high levels of extinctions (Myers, 1988 and Wilcove et al., 2013). Peatlands serve increasingly as refuge for endangered animal species (e.g. orangutan, Sumatran tiger and Sumatran rhino) which are losing their habitats in mineral soils (Giesen, 2004, Morrogh-Bernard et al., 2003 and Meijaard et al., 2012). Furthermore, peatland drainage causes fluvial runoff of carbon from the peat domes, easily leads to flooding in nearby areas and may have feedback effects on local and regional climate patterns due to changes in evapotranspiration (Rieley and Page, 2005 and Evans et al., 2014).

By 2007, forest cover in the peatlands of Peninsular Malaysia, Sumatra and Borneo had decreased to 42% (Miettinen and Liew, 2010) and deforestation rates remained high (Miettinen et al., 2012b). Over a quarter of peatlands had been converted to managed land cover types (11% small-holder areas and 18% industrial plantations), with lowered water table levels, and further 23% of the peatland areas were covered by highly fire prone degraded fern, shrub and secondary regrowth (Miettinen and Liew, 2010). Deforestation and conversion to managed land cover types is expected to have continued since 2007 but the current land cover distribution in the peatlands of Southeast Asia is unknown.

Meanwhile, peatland deforestation and conversion taking place in insular Southeast Asia, and particularly the role of industrial plantation development in it, has become one of the most discussed topics in natural resource management and conservation (see e.g. Jewitt et al., 2014; Law et al., 2014 and Austin et al., 2015). Indonesia and Malaysia are constantly under international pressure to implement sustainable peatland management policies protecting the remaining peat swamp forests and improving management practices and rehabilitation efforts in deforested peatlands. Annual peatland fires with repeated transboundary haze episodes (see e.g. Gaveau et al., 2014) cause significant health problems and economic losses throughout the region, while the ecological, biodiversity and carbon emission effects of peatland conversion highlighted above have consequences in varying levels from local to global scale. Due to the broad and far reaching consequences of peatland management, current peatland policy discussion in Southeast Asia involves governmental, non-governmental and business stakeholders from all over the world with a common aim to find solutions to the pressing peatland management challenges in the region.

In order to provide information on the current status and recent change trends on peatlands to evaluate the effects of the changes and to support formulation and implementation of peatland management policies, we here present land cover and industrial plantation distribution in the peatlands of Peninsular Malaysia, Sumatra and Borneo in 2015. We analyse the 2015 maps together with fully comparable and completed land cover maps of 1990 and 2007 (Miettinen and Liew, 2010) as well as industrial plantation maps of 1990, 2000, 2007 and 2010 (Miettinen et al., 2012a). In our land cover change and industrial plantation expansion analyses we concentrate on previously unpublished changes since 2007 (and since 2010 for plantation extent) building on known peatland land cover change history 1990–2010.

2 Materials and methods

2.1 Study area

The study area covers 15.7 Mha of peatland (Fig. 1) as defined by the peatland maps used in the analysis. The peatland areas for Sumatra and Kalimantan (Indonesian part of Borneo Island), were extracted from the Wetlands International 1:700 000 peatland atlases (Wahyunto et al., 2003 and Wahyunto et al., 2004). For Malaysia, the European Digital Archive of Soil Maps (Selvaradjou et al., 2005) was used to outline peatland areas as described in Miettinen and Liew (2010). For Brunei, we could not find any existing maps of peatland extent. The peatlands in Brunei were manually digitized using the Landsat data described below, Shuttle Radar Topography Mission (SRTM) elevation product (Jarvis et al., 2006) and an image originally published by Anderson and Marsden (1984), providing the extent of peat swamp forests in Brunei (available online at https://sites.google.com/site/peterengbersbrunei/brunei, accessed Oct 2015).

Fig. 1. Peatlands of the study area (dark grey). Accuracy assessment areas outlined in yellow (2015 very high resolution data) and red (2014 very high resolution data). Administrative area abbreviations: NS  =  North Sumatra, WS  =  West Sumatra, Ria  =  Riau, Jam  =  Jambi, Ben  =  Bengkulu, SS  =  South Sumatra, Lam  =  Lampung, WK  =  West Kalimantan, CK  =  Central Kalimantan, SK  =  South Kalimantan, EK  =  East Kalimantan, PM  =  Peninsular Malaysia, Sar  =  Sarawak, Sab  =  Sabah and BR  =  Brunei. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

2.2 Satellite data

A 30 m spatial resolution composite image created with Landsat 7 ETM+ (Enhanced Thematic Mapper) and Landsat 8 OLI (Operational Land Imager) data was used in the 2015 land cover mapping. The composite image was created using the Google Earth Engine JavaScript Application Program Interface (https://developers.google.com/earth-engine/, accessed Sep 2015). The compositing script developed for this study utilized all available Landsat 7 and 8 data acquired between the 1st Jan 2015 and the 31st Aug 2015. Pixels were considered valid if the Top of the Atmosphere (TOA) reflectance in blue band was less than 0.2 and an additional criterion was valid. The additional criterion was dependent on vegetation greenness. (1) If NDVI was less than 0.6, the ratio between red and the 2.1 μm band needed to be less than 1.0. (2) If NDVI was greater than or equal to 0.6, the ratio between red and the 2.1 μm band needed to be less than 2.5. The first criterion masked out thick clouds, taking advantage of the sensitivity of the blue wavelength to clouds and haze. The idea behind the additional criterion is the high correlation between red and the 2.1 μm shortwave infrared (SWIR) band in cloud free conditions presented by Kaufmann et al. (1997). The ratio of red and the 2.1 μm is dependent on land cover, hence the separation into densely vegetated and sparsely vegetated or bare areas. A median value of all valid pixels was used for the composite image. The composite image contained three bands: 2.1 μm SWIR, near infrared (NIR) and red.

The rather limited observation period (eight months) was chosen (1) to limit the data acquisition to year 2015 in order to ensure inclusion of the latest land cover changes and (2) to minimize the inclusion of images with burned areas, thereby essentially creating a pre-fire land cover map for the exceptionally bad fire year of 2015. The quality of the composite was generally very good, with some limitations for the detection of the degradation level in peat swamp forests (see more below). The only area where classification could not be completed with the 2015 composite image due to missing data was the southern part of Riau province in Sumatra. For this area, a compositing period from the 1st Sep 2014 to the 31st Aug 2015 was used instead.

The satellite data used for the 1990 and the 2007 land cover maps are described in detail in Miettinen and Liew (2010). In short, the 1990 map was created using the 28.5 m spatial resolution GeoCover 1990 mosaic of Landsat 5 TM (Thematic Mapper) images acquired between 1987 and 1993 (MDA Federal, 2004). The 2007 map was created using 121 10–20 m resolution images from the SPOT (Satellite Pour l’Observation de la Terre) satellite, with acquisition dates varying from 2005 to 2008. And finally, two additional satellite image datasets were used for the industrial plantation mapping reported in Miettinen et al. (2012a) and referred in this study. The 2000 mapping was based on the 2000 GeoCover product composed of Landsat 7 ETM + images acquired between 1997 and 2003 (MDA Federal, 2004). The 2010 mapping was performed using 74 Landsat 7 ETM + images acquired between the 1st January 2010 and the 11th March 2011.

As described in Miettinen and Liew (2010) the 1990 and 2007 satellite image datasets did not allow full coverage mapping of the peatland areas due to clouds and missing data. The valid data proportions for the 1990 and 2007 maps were 90% and 82% respectively. During the 2015 mapping effort these missing areas were filled using Landsat composites created with the same 2015 compositing approach described above. The 1990 additional composite images were built with Landsat 4 and 5 TM data acquired between the 1st Jul 1989 and the 30th June 1991. The 2007 additional composite images were produced with Landsat 5 TM and Landsat 7 ETM+ images acquired between the 1st Jul 2006 and the 30th Jun 2008.

2.3 Land cover classification

To maximize comparability with earlier results, the classification was performed using exactly the same approach as in Miettinen and Liew (2010). The classification was performed by visual inspection and on-screen digitizing of land cover features. The work was done using varying scales between 1:50 000 and 1:100 000, depending on the complexity of the area. The 2015 classification was performed by the same person who had done the 1990 and 2007 classifications. The interpreter has nearly 15 years of working experience with insular Southeast Asian land cover mapping and extensive experience on visual satellite image interpretation of tropical peatlands with high resolution images.

The classification scheme includes eleven classes (Table 1). The ‘Degraded PSF’—class includes all areas where some sort of disturbance (most typically logging) has been detected at least once during the three rounds of classification. It essentially outlines those forest areas that are considered to be non-intact, as opposed to the intact forest areas where no disturbance has been detected in any of the mapping efforts. We considered to change the name of the class to better describe the reality of the classification, but finally decided to retain the old class names for consistency with previous publications. However, it is important to understand that the level of degradation varies within the class, from recovered logging areas where no signs of human disturbance can be seen anymore to areas which exhibit clear signs of recent selective logging. The 30 m Landsat composite images used for the 2015 mapping did not provide as detailed and sharp picture as the individual 10–20 m SPOT images used for the 2007 classification. Therefore, the 2007 classification of forest degradation level was used as the default for the 2015 forest degradation mapping, and changes to the 2007 classification were made only if clear signs of disturbance were noticed in the forest area.

Table 1. Description of land cover types.

Land cover type	Description
Water	Permanent water bodies. This class also includes fish and crab farming ponds.
Seasonal water	Areas that are inundated part of the year. Typically either extremely degraded areas or flood zones of rivers. This class also includes small-holder mining sites.
Pristine peat swamp forest (PSF)	PSF with no clear signs of human intervention.
Degraded PSF	PSF with clear signs of disturbance (e.g. logging), typically in the form of logging tracks and canals and/or opened canopy.
Tall shrub/secondary forest	Shrub land or secondary forest with average height above 2 m.
Ferns/low shrub	Ferns and grass or shrub land with average height less than 2 m.
Small-holder area	Mosaic of housing, agricultural fields, plantations, gardens, fallow shrubland etc. Note that the name of the class refers to the patchy land scape patterns, typical in small-holder dominated areas, but the actual land tenure of the areas is unknown.
Industrial plantations	Large scale industrial plantations assumed to have been already planted with the plantation species. Mainly oil palm and pulp wood.
Built-up area	Towns, industrial areas etc.
Clearance	Open area with no vegetation, including recently burnt areas.
Mangrove	Areas that were considered to be mangrove forest in the satellite image interpretation although they were located within the peatland maps used in this study.

2.4 Plantation classification and species identification

Industrial plantation areas were outlined as part of the 1990, 2007 and 2015 land cover classifications as described above. The same classification approach was used in the additional 2000 and 2010 industrial plantation classifications published in Miettinen et al. (2012a). Together these datasets enable analysis of the expansion of industrial plantations in the peatlands of Peninsular Malaysia, Sumatra and Borneo since 1990 in five time steps: 1990, 2000, 2007, 2010 and 2015.

Whenever possible, the plantation species identification was derived from the 2007 plantation map (Miettinen et al., 2012a), with some corrections of known identifications errors. The species had been originally identified using the 2007 SPOT satellite dataset described above. The species information for plantations established after 2007 was derived from the 2015 Landsat composite. The two main plantation species used in the peatlands of insular Southeast Asia, oil palm and pulp wood, have distinctly different appearance in 10–30 m spatial resolution satellite images, enabling reliable identification in most cases. In addition to the visual appearance (e.g. tone and texture), spatial arrangement of the plantation canals and roads, location, context, the interpreter’s personal knowledge and available land use allocation information was used to support the decision making.

Three plantation species classes were used in the identification: ‘Oil palm’, ‘Pulp’ and ‘Other/unknown’. The ‘Other/unknown’-class was used for plantations which displayed characteristics not typical for either oil palm or pulp, or which were known to have other species. The same plantation species was assigned for each individual plantation in all of the time steps. This decision was assumed to be generally valid in the peatlands of insular Southeast Asia due to the relatively short period of existence for most of the plantations (predominantly less than 15 years) and the fixed infrastructure and plantation design needed for the two main plantation species found in the peatlands of the region.

2.5 Accuracy assessment

Very high resolution satellite images available in Google Earth were used in the accuracy assessment. Most of the images were acquired in 2015. A few images acquired in 2014 were also selected in order to get wider distribution of images over the entire study area (Fig. 1). The accuracy assessment areas covered 4.5 Mha or around 30% of the study area with 4.3 Mha covered by images acquired in 2015.

The sample plots were selected using stratified random sampling approach. Altogether 853 sample plots were used for the land cover accuracy assessment and 267 for the plantation species accuracy assessment. The 853 plots for the land cover accuracy assessment were composed of 800 basic plots (allocated based on the proportional areas of land cover types), and additional 53 plots to assure minimum of 20 plots for each class. The 267 sample plots for the plantation species accuracy assessment were composed of 250 basic plots with additional 17 plots in the ‘Other/unknown’-class to reach the minimum of 20 sample plots.

The classes of ‘PSF’ and ‘Degraded PSF’, as well as the classes of ‘Seasonal water’, ‘Fern/low shrub’ and ‘Clearance’ were combined into two classes: ‘Peat swamp forest’ and ‘Open undeveloped’. In the case of the ‘Peat swamp forest’-class this was done since it was considered impossible to determine the correctness of the classification of the degradation level based on a single sample plot in a very high resolution image. This was partly due to the cumulative nature of the ‘Degraded PSF’-class and partly due to the generalized delineation of logging areas, in most cases without any clear boundary in the forest. For the ‘Open undeveloped’-class, the differences between the three combined classes are by definition somewhat ambiguous and they are easily interchangeable in short periods of time. It was thus not considered meaningful to evaluate the correctness of the classification of the three original classes separately using a single very high resolution image.

The visual interpretation of the very high resolution images was performed using a 300 × 300 m box as a sample plot, except for the ‘Small-holder area’-class where the size of the plot was 500 × 500 m. The dominant land cover or plantation species within these boxes was recorded. The use of sample plots was considered more suitable than sample points (1) for the rather coarse manual classification approach used in the mapping, rarely resulting in polygons less than 10 ha in size and (2) for the mosaic nature of the ‘Small-holder area’-class.

2.6 Analysis of land cover and industrial plantation distribution

With the new 2015 maps and the completed versions of the older maps combined, we now have a time series of comparable full coverage land cover maps for 1990, 2007 and 2015, as well as industrial plantation maps for 1990, 2000, 2007, 2010 and 2015. In this paper we concentrate on reporting the previously unpublished 2015 land cover and industrial plantation distribution, as well as their changes since the latest available maps (i.e. 2007 for land cover and 2010 for industrial plantations). However, we also refer to the older datasets to provide longer historical perspective. It is important to understand that the numbers presented in this paper may vary from the earlier publications. This is mainly due to the fact that both of the previous analyses (Miettinen and Liew, 2010 and Miettinen et al., 2012a) have been based on incomplete samples, whereas in this study all the datasets have been completed to fully cover the peatlands of the study area.

For all the change analyses, the three classes of ‘Seasonal water’, ‘Ferns/low shrub’ and ‘Clearance’ are combined into one class of ‘Open undeveloped’ (exactly as in the accuracy assessment). The original three classes are easily interchangeable (e.g. due to changes back and forth to the ‘Clearance’-class as a result of recurrent fire activity) without causing a significant change in the land cover/use from the perspective of this study. Therefore, it was not considered meaningful to analyse the changes between these classes. Please see more discussion on the problems related to the classification and analysis of these classes in the discussion section. Note, however, that the ‘Tall shrub/secondary forest’-class was retained as a separate class also in the change analysis. Although the border between this class and the ‘Open undeveloped-class (i.e. 2 m average height) is difficult to define in Landsat type data, and there are surely some misclassifications, the great majority of the ‘Tall shrub/secondary forest’-class is considered to have canopy height over 5 m, in some cases even resembling primary forest in the Landsat images. Thereby, changes between these two classes from one time step to another may provide valuable information e.g. on the direction of change in the area (i.e. either natural regeneration or further degradation).

3 Results

3.1 2015 land cover distribution with changes since 1990

The 2015 land cover distribution in the peatlands of Peninsular Malaysia, Sumatra and Borneo (Fig. 2; Table 2) reveals that peat swamp forests cover 29% (4.6 Mha) of the study area, while 50% (7.8 Mha) of the peatlands are covered by managed land cover types (22.4% small-holders and 27.4% industrial plantations). Furthermore, the great majority of the remaining forest areas have been selectively logged, with only 6% of peatlands showing no signs of human influence since 1990. Currently one fifth (20%; 3.2 Mha) of the peatlands in the study area are covered by open undeveloped areas and secondary regrowth (i.e. the classes of ‘Seasonal water’, ‘Fern/low shrub’, ‘Clearance’ and ‘Tall shrub/secondary forest’).

Fig. 2. Land cover 2015 in the major peat domes of the study area. Administrative areas referred in the text are identified as: PM  =  Peninsular Malaysia, Sar  =  Sarawak, BR  =  Brunei, SS  =  South Sumatra, WK  =  West Kalimantan and CK  =  Central Kalimantan.

Table 2. Land cover distribution in the peatlands of Peninsular Malaysia, Sumatra and Borneo in 2015 (in 1000 ha and %).

	Water	Seasonal water	Pristine PSF	Degraded PFS	Tall shrub/secondary forest	Ferns/low shrub	Small-holder area	Industrial plantation	Urban	Clearance	Mangrove	Total
Peninsular Malaysia	9.84	3.06	41.15	152.17	48.91	34.02	267.13	279.35	24.68	30.01	1.38	891.70
	1.1	0.3	4.6	17.1	5.5	3.8	30.0	31.3	2.8	3.4	0.2
Sarawak	1.73	1.31	7.60

For further details log on website :
http://www.sciencedirect.com/science/article/pii/S2351989415300470