Estimating aboveground biomass in forest and oil palm plantation in Sabah, Malaysian Borneo using ALOS PALSAR data
Highlights
► Tree height measurements are essential for reliable aboveground biomass (AGB) estimates. ► Wet conditions and high forest AGB values limit the applicability of L-band SAR data. ► Given dry and degraded forest conditions, ALOS-PALSAR predicted AGB values (R2 = 0.62). ► ALOS-PALSAR can distinguish between forest and oil palm areas with 97% accuracy.
Introduction
Conversion of secondary forest to oil palm plantation is a huge concern for carbon emissions and biodiversity conservation in Southeast Asia. The forests in this region are a large carbon reserve, having among the highest carbon densities of all undisturbed tropical forests (Slik et al., 2010); however, aggressive timber extraction over the past several decades has severely decreased this carbon store (Houghton, 2005). Borneo has suffered among the highest levels of logging in Southeast Asia with extraction rates of greater than 100 m3 ha−1 (Collins et al., 1991, Sundberg, 1983), with 80% of its lowlands already degraded by selective logging (Curran and Trigg, 2006). Pressure on these forest reserves will continue to increase as volumes of readily harvestable timber dwindle, sometimes resulting in the conversion of these areas to oil palm plantations. As very little if any unprotected “primary forest” in this region remains and only previously logged areas can be cited for conversion, the focus of this study has been to compare AGB of logged and degraded forest with AGB of oil palm plantation.
Past efforts at estimating the degree of forest loss due to expansion of oil palm plantation, such as that by Koh and Wilcove (2008), have used data from the United Nations Food and Agriculture’s (FAO) Forest Resource Assessment (FRA). These data allowed them to estimate the amount of planted oil palm area replacing natural forest in Malaysia and Indonesia. However, this analysis was not able to differentiate the aboveground biomass (AGB) stored in this forest before clearance and therefore could not estimate the carbon emissions due to this land conversion. As a result of studies like these, concerted efforts are underway to improve the mapping of AGB using remote sensing.
This study focuses on Sabah, Malaysian Borneo, which covers 73,731 km2 or 10% of Borneo’s total area. It is also the Malaysian state with the largest area of planted oil palm, covering approximately 17% of the state’s total land area (MPOB, 2008). Under the Malaysian National Forest Policy 45% of Sabah’s land area is designated as Permanent Forest Reserve, which are unlikely to be converted to oil palm; although they will continue to be logged. Berry et al. (2008) predicted that by the end of 2010 all of the remaining natural forest outside of protected areas will have been logged at least once. Better AGB estimates could be pivotal for prioritizing areas needing enhanced protection and/or identifying areas being degraded unsustainably.
Estimating the carbon implications of this forest degradation and large-scale conversion is still relatively uncertain considering the errors in regional carbon stock estimates. While carbon is stored in both vegetation and soil, 89% of carbon losses are due to loss of living biomass (Houghton, 2005); therefore, efforts have been focused on estimating AGB of vegetation at the landscape scale (Saatchi et al., 2007b). Mapping AGB using remote sensing has been a significant challenge to researchers, but is extremely important for future implementation of carbon credit verification in the Land-use Change and Forestry (LUCF) sector (GOFC-GOLD, 2009).
Mapping AGB in tropical regions can be especially challenging due to the complex canopy structure as well as predominant cloud cover. Passive optical data can only sense the canopy in two dimensions making it unable to sense the sub-canopy structure, including canopy height (Almeida-Filho et al., 2007, Anaya et al., 2009, Olander et al., 2008). Therefore, passive optical data has been considered to have limited use for estimating AGB in comparison to synthetic aperture radar (SAR) and light detection and ranging (LiDAR) data, which are sensitive to the forest structure (Drake et al., 2003, Gibbs et al., 2007, Le Toan et al., 2004, Patenaude et al., 2005). SAR data has been effective in directly estimating forest AGB in African (Mitchard et al., 2009), Latin American (Saatchi et al., in press) and more recently in Southeast Asian forests, namely in peatland areas of Kalimantan (Englhart et al., 2011). Also, efforts have attempted to relate height variables (e.g. canopy height and lorey’s height) to AGB estimates to be applicable for techniques able to estimate forest height directly from remote sensing data (Köhler and Huth, 2010, Saatchi et al., 2011).
SAR data acquisition entails emission of a microwave of discrete wavelength, 1–150 cm, which interacts with the earth’s surface described by a scattering coefficient, σ0. This coefficient is a dimensionless value, which is mapped as intensity using a logarithmic scale in decibels [dB] (Waring et al., 1995). Each pixel of a SAR image is a combination of several backscattering coefficients, which can lead to either constructive or deconstructive interference creating a speckle effect in an image (Balzter, 2001). Smoothing this speckle through kernel filters or reducing the resolution of SAR backscatter to 50 or 100 m improves the quality of intensity information (Le Toan et al., 2004, Saatchi et al., in press). Finally, more advanced SAR sensors discriminate returning signals by polarization, providing information on the structure of the backscattering surface. With polarimetric SAR, the backscatter signal from the surface is measured in a combination of horizontal (H), transmitted or received parallel to the ground surface, and vertical, (V) transmitted or received perpendicular to the surface, polarizations.
Before the launch of the Advanced Land-Observing Satellite (ALOS), it was not considered possible to generate biomass maps from the radar sensors available (e.g. JERS-1, ERS, etc.). The Japanese Earth Resources Satellite 1 (JERS-1), launched from 1992 to 1998, has been used in conjunction with optical sensors to aid in deforestation monitoring; however the use of only one polarization (HH) severely limited its ability to differentiate between disturbance types. The longer wavelength of the L-band is particularly sensitive to the primary, secondary branches and stems of forests, although it also exhibits saturation to dense forest (Quegan and Le Toan, 2002). ALOS launched in 2006 with the L-band (wavelength: ∼24 cm) synthetic aperture radar sensor, PALSAR, onboard was hailed as a significant contribution to the field of forest monitoring (Rosenqvist, 2003), namely for biomass estimation and/or growing stock volume (Eriksson et al., 2003). Unfortunately, the PALSAR sensor failed on May 12, 2011; therefore, new imagery will not be available for future monitoring efforts.
ALOS-PALSAR acquired L-band data in five different modes; however, this study used fine beam dual polarization (FBD) data HH and HV. Dual and quad band polarizations increase the sensitivity of the signal in order to overcome saturation for biomass values greater than 50 Mg ha−1 (Quegan and Le Toan, 2002). Nevertheless, PALSAR FBD may not be an effective dataset for mapping very biomass-rich forest types due to saturation of the signal (Gibbs et al., 2007, Magnusson et al., 2008) or its inability to distinguish between types of severely degraded forests (Watanbe et al., 2007).
For this study it was considered necessary to explore the range of AGB values by forest disturbance type and oil palm age in order to reliably estimate changes in AGB from forest to oil palm plantation. Also, to map AGB in these two structurally different land-cover types a reliable means of differentiating their respective areas was needed. Therefore, the main aims of this paper are: (1) present mean carbon stock values for different land-cover types across Sabah (2) investigate the potential for ALOS-PALSAR data to differentiate oil palm plantation from forest area and (3) assess generated logarithmic relationships between AGB and SAR backscatter for estimating biomass across Sabah.
Section snippets
Study area
The study sites sampled were located in six forest reserves and three oil palm plantations across much of eastern, lowland Sabah. Annual precipitation ranges from 2000 to 3000 mm due to the influence of two monsoons acting in November–March and a drier one in June–July, creating relative dry seasons in April–May and August–September (Marsh and Greer, 1992). Temperatures are typical for a moist, tropical climate, and in the lowlands rarely go below 20 °C or above 30 °C, with means of 26.7–27.7 °C.
Differentiation of disturbance levels by height estimates
Due to the range of plot sizes available for analysis, plot-level mensuration variables were plotted as histograms to assess their changes in distribution with increasing plot size (see Fig. A2). The evolution of average height values across plot size were illustrative, showing reduced variation with increasing plot size across disturbance levels indicating it was not an effective means of distinguishing between them. Dominant height showed a better distribution across plot sizes (not
Relationship between AGB and forest structural metrics
Average height values did not change significantly across disturbance level, but exhibited a stronger relationship with biomass compared to dominant height. This was surprising considering that biomass did change across disturbance levels. It appears that neither of these height variables would make a reliable estimator of plot biomass levels, however other studies have had more success relating height variables to AGB. Saatchi et al. (2011) have found significant relationships using Lorey’s
Conclusion
This study has accomplished two of its three aims, being unable to map AGB in oil palm plantations. It has also highlighted some of the limitations to monitoring this region with SAR data, due to the high AGB values in unlogged and secondary forest as well as the difficulty in avoiding rain events. Similar studies in savannah forests of Africa have shown much higher degrees of correlation (Mitchard et al., 2009), both due to the relatively lower range of AGB levels and drier state of the
Acknowledgements
The authors would like to acknowledge Dr. Glen Reynolds, Simon Siburat, Philip Ho, Amy Bosi, Unding Jemy, Mike Bernadus and Alex Karolus for support in the field and Dr. Richard Lucas and Matthew Walthram for advice on SAR analysis. Finally, Dr. Heiko Balzter provided valuable feedback on an interim draft of this paper.
References (65)
- et al.
Aboveground biomass assessment in Colombia: a remote sensing approach
Forest Ecology and Management
(2009) - et al.
Aboveground biomass retrieval in tropical forests – the potential of combined X- and L-band SAR data use
Remote Sensing of Environment
(2011) Wood density for estimating forest biomass in Brazilian Amazonia
Forest Ecology and Management
(1997)- et al.
Reducing uncertainty in the use of allometric biomass equations for predicting above-ground tree biomass in mixed secondary forest
Forest Ecology and Management
(2001) - et al.
Synthesis of remote sensing approaches for forest carbon estimation: reporting to the Kyoto Protocol
Environmental Science and Policy
(2005) - Adler, R., Thomas, Wilheit, J., Kummerow, C., Ferraro, R. 2007. AMSR-E/Aqua L2B Global Swath Rain Rate/Type GSFC...
A study of land cover classification using polarimetric SAR parameters
International Journal of Remote Sensing
(2007)- et al.
Detecting deforestation with multitemporal L-band SAR imagery: a case study in western Brazilian Amazo?nia
International Journal of Remote Sensing
(2007) Forest mapping and monitoring with interferometric synthetic aperture radar (InSAR)
Progress in Physical Geography
(2001)- Banin, L. 2010. Cross-continental comparisons of tropical forest structure and function. In: School of Geography,...
Impacts of selective logging on tree diversity across a rainforest landscape: the importance of spatial scale
Landscape Ecology
Tree allometry and improved estimation of carbon stocks and balance in tropical forests
Oecologia
Error propagation and scaling for tropical forest biomass estimates
Philosophical Transactions of the Royal Society of London Series B-Biological Sciences
Regional and phylogenetic variation of wood density across 2456 neotropical tree species
Ecological Applications
The Conservation Atlas of Tropical Forests: Asia and Pacific
The Oil Palm
Sustainability science from space: quantifying forest disturbance and land-use dynamics in the Amazon
Proceedings of the National Academy of Sciences of the United States of America
Above-ground biomass estimation in closed canopy Neotropical forests using lidar remote sensing: factors affecting the generality of relationships
Global Ecology and Biogeography
Wildlife-friendly oil palm plantations fail to protect biodiversity effectively
Conservation Letters
The potential of ALOS single polarization InSAR for estimation of growing stock volume in boreal forest
International Geoscience and Remote Sensing Symposium (IGARSS)
Biophysical forest type characterization in the Colombian Amazon by airborne polarimetric SAR
IEEE Transactions on Geoscience and Remote Sensing
Aboveground forest biomass and the global carbon balance
Global Change Biology
Statistical fusion of lidar, InSAR, and optical remote sensing data for forest stand height characterization: a regional-scale method based on LVIS, SRTM, Landsat ETM+, and ancillary data sets
Journal of Geophysical Research
A microwave scattering model for the remote sensing of oil palm plantations
PIERS Online
Remotely sensed evidence of tropical peatland conversion to oil palm
Proceedings of the National Academy of Sciences
Is oil palm agriculture really destroying tropical biodiversity?
Conservation Letters
Cited by (164)
Sustainable management of Eucalyptus pellita plantations: A review
2023, Forest Ecology and ManagementDeforestation detection using a spatio-temporal deep learning approach with synthetic aperture radar and multispectral images
2023, ISPRS Journal of Photogrammetry and Remote SensingExpert systems in oil palm precision agriculture: A decade systematic review
2022, Journal of King Saud University - Computer and Information SciencesSpatiotemporal assessment of land use/land cover change and associated carbon emissions and uptake in the Mekong River Basin
2021, Remote Sensing of EnvironmentCitation Excerpt :Additionally, post-disturbance recovery dynamics are usually ignored when only deforestation and forest degradation are being tracked, which is particularly important as tree plantations, such as rubber, cashew, timber, and oil palm, are one of the main drivers of deforestation in Southeast Asia (Hurni and Fox, 2018). Growth rates of plantations can range from 4 to 8 Mg ha−1 yr−1 depending on the species (Avtar et al., 2014; Lewis et al., 2020; Morel et al., 2011). Instead of using regional aggregates, Tang et al. (2020b) developed a spatiotemporal carbon bookkeeping model based on the carbon bookkeeping approach introduced by Houghton et al. (2000) to account for the spatial variation in AGB, regrowth rates, and post-disturbance land use dynamics.
Mapping thins to identify active forest management in southern pine plantations using Landsat time series stacks
2021, Remote Sensing of EnvironmentA multi-scaled analysis of forest structure using individual-based modeling in a costa rican rainforest
2020, Ecological Modelling