Introduction
Global production of agricultural commodities such as palm oil has risen
steadily in recent decades in response to market demand (USDA, 2009, 2010,
2016). Southeast Asia's palm oil sector has grown through expansion of oil
palm plantations in Malaysia, Indonesia, and more recently, Papua New Guinea
(Gunarso et al., 2013; Carlson et al., 2013; Miettinen et al., 2016a; Vijay
et al., 2016). By 2014, Indonesia and Malaysia accounted for nearly 69 %
of harvested oil palm area worldwide (FAO, 2016).
In the past decade, Indonesia had the highest rate of forest loss of any
country in Southeast Asia (Hansen et al., 2013;
Margono et al., 2014; Kim et al., 2015), spurred
by rapid forest conversion for oil palm and other industrial plantations
(Carlson et al., 2012; Gunarso et al., 2013; Abood et al., 2015). Between
1990–2010, more than one-third of new oil palm plantations replaced forested
landscapes in Southeast Asia (Gunarso et al., 2013; Gaveau et al., 2016),
with rates as high as 90 % in regional hotspots such as coastal West
Kalimantan, Indonesia (Carlson et al., 2013). Conversion of primary and
logged forests for oil palm, including vast areas with deep organic peat
soils, contributed to significant greenhouse gas (GHG) emissions from fire,
decomposition, and peat oxidation (Page et al., 2002;
van der Werf et al., 2008;
Hooijer et al., 2012; Ramdani and Hino, 2013;
Field et al., 2016; Huijnen et al.,
2016). Environmental concerns with palm oil production extend beyond GHG
emissions, however, as forest loss threatens biodiversity (Pimm et al., 2014;
Vijay et al., 2016) and particulate emissions from fires are a major public
health concern in Indonesia and downwind population centers such as Singapore
(Murdiyarso et al., 2004; Gaveau et al., 2014; Kunii et al., 2002; Reddington
et al., 2014; Marlier et al., 2015; Chisholm et al., 2016; Johnston et al.,
2015).
Palm oil is the fastest growing certified agriculture commodity, and
Indonesia accounted for > 50 % of certified production areas
in 2016 (Potts et al., 2014; RSPO, 2016). The push for certification within
the palm oil industry reflects a growing consumer awareness of GHG emissions
from palm oil expansion and peat oxidation and an overall rise in producer
and consumer interest in “sustainable” and deforestation-free products
(UNCS, 2014; Butler, 2015; McCarthy et al., 2016). The Roundtable on
Sustainable Palm Oil (RSPO) certification is the most widely adopted
certification standard. By 2016, the RSPO had certified 2.83 Mha of oil palm
that produced 10.8 million tons of palm oil, or approximately 17 % of
global palm oil production, with > 90 % of certified areas in
Southeast Asia (RSPO, 2016). Specific principles and criteria of RSPO
certification promote sustainable palm oil production and processing (Garrett
et al., 2016; RSPO, 2004, 2015b). Among other provisions, RSPO certification
prohibits conversion of primary forests and forests with a high conservation
value (HCV) and bans fire use for land clearing in compliance with the
Indonesian moratorium on fire (RSPO, 2007; RSPO, 2013; Edwards and Heiduk,
2015). Companies interested in certification first become members of the RSPO
and are subjected to the principles and criteria (P&C), including the
prohibition on new plantings through deforestation of primary or HCV forests
after November 2005 without compensation (Criterion 7.3, RSPO, 2013).
However, the RSPO does not independently monitor deforestation within member
plantations. RSPO member companies are notified of satellite fire detections
on their plantations, and a description of fire incidents is included in
monthly reports to RSPO. Recent fire activity has been assessed for a subset
of palm oil concessions in Indonesia (Cattau et al., 2016a), but the use of
fire for forest conversion on oil palm plantations has not previously been
quantified.
Improving estimates of fire-driven deforestation is critical
for assessing environmental
compliance by oil palm plantations, reducing uncertainties in deforestation
carbon emissions (Le Quéré et al., 2015;
Houghton et al., 2012; van
der Werf et al., 2009b), and characterizing ignition sources that may give rise
to uncontrolled burning during drought periods (Carlson et al., 2012;
Cattau et al., 2016b). The timing of GHG emissions from forest
conversion to oil palm depends on the degree of fire use for deforestation
(DeFries et al., 2008; Houghton et al., 2012), including the proportion of
clearing activity through fire and the combustion completeness of initial or
repeated burning (van der Werf et al., 2009a). Fires are common in industrial
plantations and smallholder properties (Stolle et al., 2003; Austin et al.,
2015; Marlier et al., 2015; Miettinen et al., 2016b; Cattau et al., 2016a),
yet the link between fire activity and forest conversion is unclear. Many
estimates of carbon emissions from tropical forest conversion report
committed fluxes without separating fire and decomposition losses (Koh et
al., 2011; Carlson et al., 2012;
Austin et al., 2015). Previous studies with bookkeeping or
biogeochemical models suggest that fire accounts for 30 % (Houghton and
Hackler, 1999) to 50 % (van der Werf et al., 2009a) of carbon emissions
from aboveground biomass during forest conversion in Southeast Asia – a
broad range that applies to all forest conversion, not strictly to oil palm
expansion. Fires are not restricted to forested areas; El Niño conditions
suppress precipitation over large parts of Southeast Asia, leading to
widespread fire activity during drought periods, particularly in carbon-rich
peatlands (Page et al., 2002;
van der Werf et al., 2008; Field et al.,
2009, 2016). Understanding the contribution from fire-driven deforestation to
total fire activity is therefore a critical part of mitigating fire risk
during drought years (e.g., Chen et al., 2016).
Here, we combined time series of satellite data on forest loss and active
fire detections with locations of oil palm plantations to assess fire-driven
forest and peatland conversion on and around oil palm plantations. The
combination of land management, forest loss, and active fire data provided an
opportunity to evaluate the relative contributions from different fire types
to spatial and temporal variability in satellite fire detections. Our study
addressed three primary questions regarding oil palm expansion (1) What
fraction of forest and peat forest conversion for oil palm involves
fire? (2) Does certification alter fire use for forest conversion or the
frequency of management or accidental fires in plantation areas? (3) During El Niño years, do certified plantations have fewer
satellite fire detections compared to noncertified plantations and
surrounding lands? Characterizing fire-driven deforestation is critical to
evaluating the influence of RSPO certification on fire activity and to
improving
estimates of GHG emissions from oil palm expansion.
Extent of RSPO certified (red) and noncertified (blue)
oil palm plantations in Indonesia. Regional subsets
highlight plantation boundaries on peatlands (green) in the lowlands of
Sumatra (a), Kalimantan (b), and Papua (c).
![]()
Material and methods
Oil palm plantations
The government of Indonesia allocates land for oil palm production to
companies for a limited period of time. We separated leases for oil palm
production into two categories, certified and noncertified plantations.
Certified plantations are properties certified by the RSPO; noncertified
plantations are properties allocated by the Indonesian government to
companies but that are not certified, even if they are held by RSPO members.
Comparisons between certified (ever) and noncertified (never) plantations
considered forest loss and fire activity over three timescales: (1) following the benchmark date for compliance with RSPO criterion
7.3 (November 2005), (2) following the first issuance of RSPO certificates to
Indonesian producers in 2009, and (3) following the date of certification for
individual plantations. Boundaries of certified plantations were compiled
from several sources, including boundary polygons provided by the RSPO,
digitized boundaries from RSPO audit reports, and spatial data on plantation
boundaries from RSPO member companies provided in annual communication of
progress (ACOP) reports (RSPO, 2015a). Boundaries of noncertified
plantations were obtained from a database of oil palm plantations published
by Greenpeace (Greenpeace, 2016) and supplemented with noncertified
plantations held by RSPO members, as indicated in ACOP reports (RSPO, 2015a)
or by Sawit Watch (2013). In total, we analyzed 154 certified and 1536
noncertified plantation boundaries for Indonesia (Fig. 1). Data on the
location and certification date of certified plantations were also available
for Malaysia (n = 119) and Papua New Guinea (n = 10), but
boundaries of noncertified plantations were not available.
We used maps of planted oil palm to identify established plantations within
certified and noncertified plantations in Indonesia, Malaysia, and Papua New
Guinea. Data on the extent of planted oil palm were compiled from three
sources: Gunarso et al. (2013) for 2000, 2005, and 2010; Carlson et
al. (2013) for 2000, 2005, and 2010; and Transparent World (TW, 2015) for
2014. Maps of planted palm were generated from 30 m Landsat imagery and
validated using higher-resolution satellite imagery (Carlson et al., 2013;
TW, 2015; Petersen et al., 2016). When multiple estimates were available for
the same epoch, we used the combined area from all sources as a more
conservative estimate of the extent of planted oil palm. Only noncertified
plantations that showed evidence of planted oil palm up to 2014 were included in this study.
We evaluated forest loss and fire activity for a single set of 5 km buffers
surrounding both certified and noncertified plantations in Indonesia. Recent
fire emergencies have intensified the debate over the source of fire
ignitions during El Niño and other drought events (e.g., Austin et al.,
2015). The 5 km buffer was chosen to capture the potential influence of
forest loss and fire activity on the surrounding landscape, including the
direct fire spread from adjacent lands into palm oil plantations, wind-blown
embers from nearby fires, and the most acute impacts of smoke on both human
health and ecosystems. Oil palm plantations in Southeast Asia are frequently
adjacent to other oil palm plantations (Fig. 1), making it difficult to
attribute buffer activities to only certified or noncertified neighbors. We
therefore analyzed a single set of buffer areas to evaluate forest loss and
fire activity surrounding large oil palm plantations. Nearly 12 % of the
area within the 5 km buffer was mapped as planted oil palm in 2010 (Gunarso
et al., 2013; Carlson et al., 2013). As for plantations, areas of planted oil
palm were excluded from estimates of forest loss in buffer areas. However,
total fire activity in the buffer region may reflect differences in oil palm
management, in addition to differences in land use and land cover, based on
the abundance of planted palm outside of the large certified and
noncertified plantations.
Forest definition, cover, and loss
Estimates of forested area and forest loss fundamentally depend on the
definition of forest cover (Sexton et al., 2016). Countries may use canopy
cover thresholds between 10 and 30 % for reporting under the United
Nations Framework Convention on Climate Change (UNFCCC) REDD+ framework
(UNFCCC, 2002). The Indonesian government uses two definitions of forest for
reporting purposes (BP-REDD+, 2015). For the United Nations Food and
Agriculture Organization (FAO) Forest Resources Assessments (FRAs), forest is
defined as canopy cover > 10 % for Global Forest Resource
Assessment (FAO, 2010), whereas the Minister of Forestry Decree uses a higher
threshold of canopy cover > 30 % for forest reference
emissions level (FREL)
reporting (MoF, 2008), at the high end of the REDD+ range. Therefore, we
used the > 30 % canopy cover threshold in this study to be
consistent with the forest definition used by the Indonesian government. The
higher canopy cover threshold also reduced ambiguity associated with
discriminating tropical forests from other land cover types in remote sensing
data for regions with persistent cloud cover, such as Southeast Asia. Forest
and non-forest areas were separated using Landsat-based estimates of
fractional tree cover in 2000 (Hansen et al., 2013). Estimates of annual
forest loss between 2002 and 2014 (Hansen et al., 2013) were used to identify
the timing of forest conversion on and around plantations.
Active fires
We used the time series of active fire detections from the Moderate
Resolution Imaging Spectroradiometer (MODIS) instruments on NASA's Terra and
Aqua satellites to evaluate the spatial and temporal patterns of daily fire
activity during 2002–2015. The global monthly fire location product
(MCD14ML) identifies the location of actively burning fires and thermal
anomalies at the time of satellite overpass at 1 km nominal spatial
resolution (Giglio et al., 2003). Fire pixel counts from Terra and Aqua MODIS
sensors were combined using a 1 km grid to evaluate monthly and annual fire
activity from 2002 to 2015. We compared fire pixel density (km-2) across
certified plantations, noncertified plantations, and a 5 km buffer region
surrounding both certified and noncertified plantations.
Forest loss within the boundaries of (a) certified
plantations, (b) noncertified plantations, and (c) the 5 km buffer region surrounding
certified and noncertified plantations in Indonesia from
2002 to 2014. (a1–c1) Forest loss (grey) and fire-driven forest loss (orange); (a2–c2) peat
forest
loss (grey) and fire-driven peat forest loss (orange). Solid black lines
indicate residual forest cover as a percentage of management areas. Annual
estimates of total forest loss and fire-driven forest loss are summarized in
Tables B2 and B3, including the proportion of forest losses on plantations
after the receipt of RSPO certification. Orange dots indicate El Niño
years.
![]()
For 2014 and 2015, active fire detections with higher spatial resolution were used
to confirm patterns in 1 km MODIS fire data. These complementary active fire
detections were derived from the Visible Infrared Imaging Radiometer Suite
(VIIRS) I-Band (375 m) on the Suomi National Polar-orbiting Partnership
(S-NPP) (Schroeder et al., 2014) and Landsat-8 Operational Land Imager (OLI)
data at 30 m resolution (Schroeder et al., 2015). The finer spatial
resolution of these fire data capture additional details regarding fire
activity that can be difficult to evaluate at MODIS resolution, including the
precise location of active fire fronts, separation of flaming and smouldering
fires (Elvidge et al., 2015), and detection of small and/or lower-intensity
fires (Schroeder et al., 2015) – an important component of fire activity in
agricultural landscapes (Randerson et al., 2012). In this study, the improved
spatial resolution of VIIRS and OLI fire data aided the attribution of active
fires to specific land management areas.
Fire-driven forest conversion for oil palm expansion
We combined satellite remote sensing data on forest cover (2000; Hansen et
al., 2013), forest cover change (2002–2014; Hansen et al., 2013), and active
fire detections (2001–2014; Giglio et al., 2003) to identify fire-driven
forest conversion on certified and noncertified plantations. Our assessment
excluded forested areas identified as oil palm (Gunarso et al., 2013; Carlson
et al., 2013). Deforestation on oil palm plantations was therefore limited to
Hansen et al. (2013) tree cover loss in forested areas (tree cover
> 30 %) outside of planted palm. Oil palm expansion into peat
forests was assessed using peatland layers created by Wahyunto and
Subagjo (2003, 2004) and Wahyunto et al. (2006) and Wetlands International
(WI, 2016) (Fig. A1). Co-located forest loss and active fire detections were
considered fire-driven deforestation. Given the potential for fire activity
to predate the detection of forest loss (Morton et al., 2008), active fire
data from the year of forest loss and 1 year before were combined to identify
fire activity associated with forest conversion.
Total and fire-driven forest loss for oil palm expansion in
Indonesia from 2002 to 2014 within certified and noncertified
plantations. See Tables B2 and B3 for annual estimates of
total forest loss, fire-driven forest loss, and the proportion of forest
losses on plantations after the receipt of RSPO certification.
Lease area
Planted palm by 2010
Forest lossa
Peat forest loss
Fire-driven lossb
(ha)
(ha)
(ha)
(ha)
(ha)
RSPO certified
1 652 644
1 212 669
267 931
61 750
85 524 (26 %)
Noncertified
11 266 875
3 416 302
2 522 909
742 025
810 346 (25 %)
5 km buffer
25 092 975
2 931 778
3 853 782
937 206
1 073 664 (22 %)
a Forest loss outside of peat areas.
b Combined (peat and non-peat) forest loss related to fire.
Forest loss patch size distribution in Indonesia within the
boundaries of (a) RSPO certified plantations, (b) noncertified
plantations, and (c) the 5 km buffer region.
Patch sizes were assessed at the plantation level and summarized annually
based on the proportion of total forest losses in each size category during
2002–2014.
![]()
Results
Certification and fire-driven deforestation
In Indonesia, forest loss on and around oil palm plantations reduced
remaining forest cover by 30–36 % between 2002 and 2014 (Fig. 2). Gross
forest loss on plantations but outside of planted palm areas totalled
3.59 Mha (Table 1). Average annual rates of forest loss were similar in
certified (1.25 % yr-1) and noncertified plantations
(1.72 % yr-1) over this period. However, trends in annual forest
loss differed between certified plantations and both noncertified and buffer
areas. Following the cutoff date for new deforestation (November 2005), rates
of forest clearing actually increased on plantations that would later be
certified (2006–2008, 38 636 ha yr-1) before declining sharply after
the first certificates were issued (2009–2015, 10 943 ha yr-1;
Table B2). Between 2009 and 2012, the majority of forest loss occurred on
plantations that had not yet received RSPO certification (Table B2), whereas
plantations with certificates accounted for most forest losses identified in
2013–2014. Patterns of peat forest loss on certified plantations were
similar to lowland forest loss outside of peat (Fig. 2, Table B3), with a
peak in peat forest loss around the 2006 El Niño event followed by a
steady decline after 2009. In contrast, rates of forest and peat forest loss
on noncertified plantations increased over time, with peak clearing in 2009
and 2012 (Fig. 2, Tables B2, B3). Temporal patterns of forest loss for buffer
areas within 5 km of plantations (both certified and noncertified) were
similar to noncertified plantations. Given the larger extent of noncertified
plantations, mean annual forest losses differed nearly 10-fold between certified and
noncertified plantations (20 610 and 194 070 ha yr-1, respectively).
Total forest loss (grey) and fire-driven forest loss
(orange) on certified plantations in (a) Indonesia, (b) Malaysia,
and (c) Papua New Guinea. Forest loss was estimated outside of
planted palm (Carlson et al., 2013; Gunarso et al., 2013). Black lines
indicate remaining forest area as a fraction of the total lease area of
certified plantations in each country. Annual estimates of
total forest loss and fire-driven forest loss for certified plantations are
summarized in Tables B2–B5, including the proportion of forest losses on
plantations after the receipt of RSPO certification. Orange dots indicate El
Niño years.
![]()
Although the use of fire for forest conversion is prohibited in Indonesia,
satellite data suggest that one-quarter of forest clearing in both certified
and noncertified plantations involved fire (Table 1). For certified
plantations in Indonesia, the proportion of fire-driven deforestation in both
lowland and peat forests declined sharply following the cutoff date for new
deforestation (November 2005) but before the first RSPO certificates were
issued (Fig. 2), from 37 % during 2002–2005 to 19 % in 2006–2008,
and only 8 % of all forest loss was identified as fire-driven
deforestation after the first certificates were issued (2009–2014;
Tables B2, B3). More than one-third of Indonesian plantations that would later be
certified had fire-driven deforestation between 2002 and 2007 (Table B4). As
total deforestation declined from 2008 to 2014, fire-driven deforestation
activity was still distributed across 34–50 plantations (22–32 %,
Table B4). The proportion of fire-driven deforestation in noncertified
plantations also declined over time, but not as rapidly as in certified
plantations, especially in peat areas. Fire-driven forest losses accounted
for 14 % of total forest loss in noncertified and buffer areas from
2009–2014 (Table B2), but 34 % of peat forest loss was identified as
fire-driven deforestation (Table B3). Notably, the proportion of fire-driven
deforestation in El Niño years (2002, 2006, 2009) was similar to or lower
than non-El Niño years (e.g., 2003, 2007, 2010) for all three management
types – certified plantations, noncertified plantations, and buffer areas.
Density of MODIS active fire detections within
certified
plantations, noncertified plantations,
and the 5 km buffer region around plantations in Indonesia from 2002 to
2014. (a) Time series of all MODIS active fire
detections; (b) time series of MODIS active fire detections
associated with fire-driven deforestation. El Niño years are underlined
in red. Table B6 provides annual estimates of fire counts on certified and
noncertified plantations, including the proportion of fire detections on
plantations after the receipt of RSPO certification.
![]()
However, certification did not halt forest conversion altogether. In
Indonesia, forest loss continued on certified plantations following the
start of RSPO certification efforts, including fires for forest conversion,
leading to an additional 8 % loss of remaining forest cover between
2009 and 2014 (Fig. 2; Tables B2, B3, B4). Lower rates of forest loss on
certified plantations are consistent with RSPO restrictions on clearing HCV
forest areas and other lands deemed unsuitable for palm oil production.
Declining rates of forest loss after 2009 may also reflect limited remaining
forest cover on certified plantations by 2014 (15 %, Fig. 2), leading to
smaller clearing sizes that are more difficult to assess with remote sensing
data on forest loss and fire activity (Fig. 3). In contrast, the contribution
from larger clearing sizes increased over time on noncertified plantations
and remained stable for buffer areas.
Patterns of fire-driven forest loss on certified plantations differed across
Indonesia, Malaysia, and Papua New Guinea (Fig. 4). Overall forest loss rates
were higher in Indonesia than Malaysia and Papua New Guinea (Table B1).
However, large forest clearing events were more common on certified
plantations in Malaysia and Papua New Guinea, with more than two-thirds of
forest loss in patches > 10 ha (Fig. A2). Annual forest loss
rates in Malaysia remained high following certification, with little change
from pre-certification patterns (Fig. 4; Tables B5, B6). In Malaysia, oil
palm expansion on certified plantations rarely involved fire, and only
5 % of total forest loss was identified as fire-driven deforestation
(Tables B4, B5, B6). Fire detections associated with forest loss declined in
all three countries following the start of certification in 2008–2009. In
Malaysia and Papua New Guinea, plantations with RSPO certificates had little
fire-driven deforestation and few total fire detections for land management
(Tables B4–B7).
High-resolution active fire detections confirm lower fire
activity on certified plantations during the 2015 El
Niño event. Map panels show active fire detections on 30 September 2015 for peat
fires in southern Sumatra from (a) Terra (blue) and Aqua
(yellow) MODIS (1 km), (b) S-NPP/VIIRS (375 m), and (c) Landsat-8/OLI
(30 m). Background images in (a–c) are a false-color composite of Landsat
8/OLI bands 7-5-3 from the same date (path/row: 124/62). Adjacent panels show
total annual fire detections in 2014 and 2015 for certified
plantations from (d) MODIS, (e) VIIRS, and (f) Landsat 8/OLI.
![]()
Certification decoupled fire detections from variability in fire
risk driven by El Niño–Southern Oscillation in Indonesia. Interannual variability in Indonesian fire activity is
largely governed by the timing and magnitude of El Niño events (Fig. A3;
Chen et al., 2016). Prior to 2009, interannual variability in fire detections
was similar for certified plantations, noncertified plantations, and buffer
areas in Indonesia (Fig. 5). Mean fire rates across land management classes
were also consistent during El Niño events in 2002, 2004, and 2006, with
important contributions from fire-driven deforestation to total fire
detections in these years. Following certification, fire activity declined in
certified plantations in all years, with 75 and 66 % fewer fires
per square kilometer than noncertified plantations during the 2009 and 2015 El Niño
events, respectively (Fig. 5, Table B6). Monthly fire counts confirmed the
reduction in fire activity on certified plantations during peak burning
months (August, September, October) of the 2009 and 2015 El Niño events
(Fig. A4). Evidence for reduced fire activity on certified plantations
highlights the potential for management of fire risk within oil palm
plantations, even during strong El Niño drought conditions (Fig. A5).
Attribution of fire activity is a critical component of satellite-based
monitoring for environmental compliance. Higher-resolution active fire data
from VIIRS (375 m) and Landsat 8/OLI (30 m) confirmed the relative decline
in fire activity on certified plantations compared to noncertified
plantations and buffer areas in both 2014 and 2015 (Fig. 6). The VIIRS 375 m
fire data provided a more complete characterization of the fire perimeter
than MODIS on a daily basis. Although less frequent, Landsat 8 coverage every
16 days captured the precise location of active fire fronts, small fires, and
persistent smouldering in peat areas that may last for many days (Figs. 6
and A6). High-resolution fire data offer improved understanding of fire use
for deforestation and agricultural management, with detections that can be
more definitively attributed to specific actors in support of monitoring,
reporting, and verification.
Discussion
Following the issuance of the first RSPO certificates in 2009, certified oil
palm plantations in Indonesia had lower fire-driven deforestation and total
fire activity than noncertified plantations, including during El Niño
events. These reductions point to the potential for RSPO to contribute to
REDD+ and to decrease fire ignitions during drought conditions. Our
findings of lower fire activity on certified plantations for both
deforestation and land management during the 2009 and 2015 El Niño events
contradicts earlier work by Cattau et al. (2016a) showing higher fire
activity in a small subset of certified plantations, possibly due to a larger
sample size of certified plantations in our study (N= 154 compared to 28,
with only four plantations on peat). However, certification did not halt
forest losses or fire activity altogether, and rates of forest loss actually
increased following the cutoff date for new deforestation but before the
first RSPO certificates were issued. In addition, certified plantations
currently account for a small fraction of total oil palm leases (e.g.,
13 % in Indonesia); noncertified plantations maintained higher rates of
fire-driven deforestation and fire activity in recent years, including the
2015 El Niño. The opportunity therefore exists to enhance the
environmental benefits of RSPO certification through expansion of certified
plantations and strengthening of certification standards, including the use
of satellite monitoring of fire activity and forest loss.
Our study confirmed the pervasive use of fire for forest conversion to oil
palm in Indonesia, with one-quarter of forest loss identified as fire-driven
deforestation. Fire-driven deforestation was less common on certified
plantations in Malaysia and Papua New Guinea, and fire use for forest
conversion declined to near zero after the start of certification in
2008–2009 in these countries. The long-term records of Landsat forest loss
and MODIS fire detections provide robust evidence of changing fire use for
land management on certified oil palm plantations.
Several factors may account for the reduction in fire activity on certified
plantations following certification. First, certification may reduce
fire-driven deforestation by directly influencing land management practices.
Collectively, all certified plantations in Indonesia, Malaysia, and Papua New
Guinea showed declines in fire-driven forest losses after 2009. Second, the
observed decline in fire activity may indicate an end of the expansion
process rather than a change in fire-driven deforestation. Remaining forest
covered only 8–15 % on certified plantations in Malaysia and Indonesia;
remaining forest areas may not be suitable for oil palm or accessible based
on RSPO restrictions. In a future study, it may be possible to control for
differences in remaining forest cover, plantation age, or company management
practices using a matched study design. For Indonesia, a reduction in overall
fire activity may be less important for GHG emissions than a reduction in
peat fires (e.g., van der Werf et al., 2008; Cattau et al., 2016a; Field et
al., 2016). Regardless, the potential exists for RSPO to promote fire-free
management of plantations to protect high-value tree crops and remaining
carbon stocks in forests and peatlands. Large labor forces needed for oil
palm production (Lambin et al., 2013) may aid regional fire suppression
efforts, allowing established plantations to maintain lower fire activity on
and around plantations during El Niño years.
The proportion of fire-driven deforestation on oil palm plantations in
Indonesia (∼ 25 %) was similar to the estimate of combustion losses
in bookkeeping models (30–40 %; Houghton and Hackler, 1999), but fire
use was much lower in Malaysia and Papua New Guinea. However, our study only
confirms the coincident timing and locations of fires and forest losses, not
the combustion completeness of fires for forest conversion. Removal of forest
vegetation is critical for
establishing an oil palm plantation, but combustion completeness may be lower
for these fires, given higher fuel moisture and less need for complete
combustion of aboveground biomass than for expansion of row crop agriculture
(Morton et al., 2008). Fuel moisture also has a substantial influence on
trace gas emissions from fire, including smouldering fires in peatlands
(Miettinen et al., 2012; Page and Hooijer, 2016). By combining active fire
detections with satellite observations of trace gas emissions, it may be
possible to characterize regional GHG emissions directly associated with
fires on oil palm plantations. Complementary data on trace gas emissions may
also compensate for missing satellite fire detections. The estimated fraction
of fire-driven deforestation for different land management categories in this
study is likely conservative because satellite platforms do not detect all
fires. Satellite sensors may not sample at the peak of diurnal fire activity
(Giglio et al., 2000), and cloud obscuration (Giglio et al., 2003) and
orbital coverage (Schroeder et al., 2005) reduce the probability of fire
detections, particularly for low-latitude regions with persistent cloud cover
such as Southeast Asia. New satellite products partially overcome these
limitations through improvements in orbital coverage and spatial resolution
(Schroeder et al., 2014), especially for detection of small and low-intensity
fires in deforestation or peatland areas (Schroeder et al., 2015; Elvidge et
al., 2015).
Aligning certification criteria with existing satellite monitoring
capabilities could improve the transparency, accountability, and impact of
RSPO and other certification efforts. RSPO certification prohibits specific
categories of forest clearing that cannot be readily distinguished using
satellite data. For example, total forest loss can be identified using freely
available satellite data products, but HCV or primary forest types cannot be
confirmed with Landsat or MODIS data. Changing RSPO criteria to more closely
match existing satellite data products on forest cover and forest loss would
enable more rigorous monitoring of environmental compliance. Our study
identified forest loss on plantations seeking certification after the cutoff
date set by RSPO, but observations of forest loss do not confirm
noncompliance with the P&C of certification because HCV and primary forest
areas cannot be confirmed using satellite data. Alternatively, public
databases of set-aside areas on certified plantations (e.g., stream buffers,
areas deemed unsuitable for production, or HCV) could improve transparency
and support monitoring efforts without the need to derive forest conditions
directly from satellite data. New, higher-resolution active fire data also
complement the time series of MODIS active fire observations. Landsat and
VIIRS active fire data offer sufficient spatial detail to unambiguously
attribute fire activity to specific land owners – an important step forward
in satellite monitoring by governments, nongovernmental organizations, or
certification bodies such as RSPO. Fire suppression is important to safeguard
carbon stocks in peatlands, and Landsat resolution is particularly beneficial
for early detection of new wildfires and for
identifying small, smouldering
fires in peat areas (Schroeder et al., 2015; Elvidge et al., 2015) that may
persist for weeks and exacerbate GHG emissions and regional air quality
during drought events.
By 2020, Indonesia has pledged to double its palm oil production (Maulia,
2010), and expanding production threatens remaining rainforest and peatland
areas. Certification offers a path for low-carbon development of additional
oil palm production, provided that certification standards are consistent
with capabilities for routine satellite monitoring. RSPO certification has
reduced but not eliminated forest loss and fire use on certified plantations.
To realize the full potential of certification, requirements for RSPO
certification must be updated to align environmental goals with objective
measures of compliance, including industry commitments to a goal of zero
deforestation. Such transparency would also provide more direct insight into
the key mechanisms through which agricultural intensification and expansion
contribute to feedbacks in the Earth system.