Rise and fall of forest loss and industrial plantations in Borneo (2000–2017)

The links between plantation expansion and deforestation in Borneo are debated. We used satellite imagery to map annual loss of old‐growth forests, expansion of industrial plantations (oil palm and pulpwood), and their overlap in Borneo from 2001 to 2017. In 17 years, forest area declined by 14% (6.04 Mha), including 3.06 Mha of forest ultimately converted into industrial plantations. Plantations expanded by 170% (6.20 Mha: 88% oil palm; 12% pulpwood). Most forests converted to plantations were cleared and planted in the same year (92%; 2.83 Mha). Annual forest loss generally increased before peaking in 2016 (0.61 Mha) and declining sharply in 2017 (0.25 Mha). After peaks in 2009 and 2012, plantation expansion and associated forest conversion have been declining in Indonesia and Malaysia. Annual plantation expansion is positively correlated with annual forest loss in both countries. The correlation vanishes when we consider plantation expansion versus forests that are cleared but not converted to plantations. The price of crude palm oil is positively correlated with plantation expansion in the following year in Indonesian (not Malaysian) Borneo. Low palm oil prices, wet conditions, and improved fire prevention all likely contributed to reduced 2017 deforestation. Oversight of company conduct requires transparent concession ownership.


INTRODUCTION
Industrial plantations of oil palm (Elaeis guineensis Jacq.) and pulpwood (Acacia mangium Willd. and other tree species) have replaced large areas of old-growth forests in Indonesia and Malaysia, where 87% of global palm oil production originates (FAO 2015). This region, rich in threatened forest species, accounts for some of the world's most rapid forest loss Margono, Potapov, Turubanova, Stolle, & Hansen, 2014).
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. The links between plantation expansion and deforestation remain debated (Meijaard et al., 2018). Although some plantations replace old growth forests, a substantial fraction make use of land cleared of forests many years previouslyindicating that not all plantation developments cause deforestation . Assessing such outcomes have been challenging and controversial. Various industry and government representatives dispute that plantations cause deforestation and highlight that plantations are a reasonable use of already deforested land . At the same time, many consumers and activists want products that are free from forest loss (Sheil et al., 2009).
Consumer concern has led to various initiatives aimed at distinguishing products that avoid deforestation. A producerled certification scheme, the Roundtable on Sustainable Palm Oil (RSPO), was established in 2004, requiring oil palm plantation companies to recognize and conserve "high conservation values areas," including species-rich forests within the boundary of their certified concessions (Schouten & Glasbergen, 2011). There have also been national initiatives such as the Indonesian Standard for Sustainable Palm Oil (ISPO) and the Malaysian Sustainable Palm Oil standard (MSPO) (Schouten & Bitzer, 2015). Also In 2011, the Indonesian government instituted a nationwide moratorium on new concessions for oil palm and pulpwood plantations on peatlands and primary forests (e.g., excluding natural forests that have been degraded by selective timber harvesting, and reclassified a "secondary/degraded" forest by the Indonesian government; Busch et al., 2015).
One recent trend has been the widespread adoption of No Deforestation commitments across the private sector. Many of the world's largest traders and producers of palm oil and pulpwood have pledged to eliminate deforestation from their supply chain (Lambin et al., 2018). The coverage of these No Deforestation commitments, some immediate and some with 2020 deadlines, is striking. Already, over 74% of internationally traded palm oil is under companies that have committed to No Deforestation whereas only 20% is certified by the RSPO Steinweg, Drennen, & Rijk, 2017). Seen from this perspective, these commitments offer a promising means to curtail forest losses.
Our objective is to understand recent trends in Borneo's industrial plantation expansion and forest loss and their relationship. For the first time, we present an annual time-series of plantation expansion, forest loss, and their overlap from 2001 to 2017. These 17 years of data derive from visual interpretation of LANDSAT imagery (see Methods) and previously published sources Hansen et al., 2013). We separate results for Indonesian Borneo (Kalimantan) and Malaysian Borneo (Sabah and Sarawak) to allow for different contexts. We exclude Brunei Darussalam where industrial plantations remain negligible . We have shown previously how the greater extent of scrublands in Indonesian versus Malaysian Borneo has facilitated the expansion of industrial plantations without conversion of old growth forest . Furthermore, the oversight and regulatory context in each country differ, with a less centralized and overlapping jurisdiction among agencies in Indonesia versus Malaysia (Cramb & McCarthy, 2016;McCarthy & Cramb, 2009). In addition, the government in Indonesia has tolerated NGO activity and criticism to a greater degree than in Malaysia suggesting that NGO oversight may also play a role.

RESULTS
Borneo's old-growth forest were reduced by 14% (6.04 Million hectares) between 2000 and 2017 (3.74 Mha and 2.29 Mha lost in Indonesian and Malaysian Borneo, respectively). Annual forest loss trended upwards over most of the period reaching a maximum in 2016 (0.61 Mha lost) before declining to its lowest level since 2004 in 2017 (0.25 Mha lost; Figure 1a). This 2016 maximum is evident in Indonesian Borneo ( Figure 1b Unsurprisingly, the variation in annual expansion of industrial plantations is positively correlated with the variation in forest loss, although the correlation is noisier in Indonesian Borneo (Figure 2a, b). The correlation disappears if we subtract forest conversion to plantations from forest loss, to test whether variation in plantation expansion relates to variation in forest areas cleared but not planted as industrial plantations (Figure 2c, d). We explored whether there were any correlations among these annual data with a time difference of 1 or 2 years before or after the plantations are established, but found no clear relationships (results not shown). Forest converted to plantations as a proportion of total plantation expansion in each year has varied, though not declined (it has increased slightly in Indonesian Borneo), throughout the study period, ranging from 16% to 43% and 52% to 80% in Indonesian Borneo, and in Malaysian Borneo, respectively. Furthermore, the variation in the annual conversion of nonforest areas to industrial plantations remains tightly correlated with the variation in the annual conversion of forest to industrial plantations . The remaining 8% (0.23 Mha) forests converted were cleared in 1 year, then converted to plantation more slowly (within 2, 3,… .17 years). This slower conversion counts as "Nonforest to plantations" in d, e, f (white bars), and constitutes 7% of the total nonforest area converted to plantations (white bars). Ninety three percent (3.14 Mha) of the white bars were already deforested at least since year 2000 indicating that these rates depend on shared causes throughout the period (Figure 2e, f).
We noticed that overall plantation expansion and the price of crude palm oil (CPO) the previous year had similar trends. A formal test showed a significant and positive correlation for Indonesian but not Malaysian Borneo (Figure 3).

DISCUSSION
We have quantified forest loss, industrial plantation expansion and their overlap each year from 2001 to 2017 in Borneo. We found a peak of forest loss in 2016 in Indonesian Borneo, a sharp drop in 2017 and a steady decline in industrial plantation expansion and in associated forest conversion to industrial plantations since 2012 in both Indonesian and Malaysian Borneo. Expansion of industrial plantations has directly contributed to forest loss throughout the study period as seen in the areas of forest cleared and converted within the same year. It is this shared area (forest converted to industrial plantations) included in both annual industrial plantation expansion and annual forest loss that explains why variation in these two measures are positively related (Figure 2a,  b). This correlation is particularly marked in Malaysian Borneo (Figure 2b), where 58% of total deforestation since 2000 resulted in plantations within 1 year (Figure 1c, f). As previously shown, industrial oil palm developments are the main driver of deforestation in the Malaysian states of Sabah and Sarawak . The correlation is less clear in Indonesian Borneo (Figure 2a), where 38% of total deforestation since 2000 involved conversion to plantations within 1 year (Figure 1b, e). More forest was lost in 2015 and 2016 in Indonesia than a linear relationship with industrial plantation developments implies. We know that forests in Indonesian Borneo suffered droughts and fires in these years (Huijnen et al., 2016;Sloan, Locatelli, Wooster, & Gaveau, 2017).
Forest conversion to plantations is not the sole driver of deforestation. Our provisional assessments suggest that >40% of total forest loss recorded in 2015 and in 2016 was caused by Indonesian forest fires (unpublished results). Fires impacted large areas of Central Kalimantan and its remaining forests in 2015 (August to October). Some of these losses have been recorded only the following year (2016)   If we compare variation in industrial plantation expansion versus just those forest losses that were not converted to plantations we don't find a clear correlation (Figure 2c, d). This result was surprising as we know that new plantations require new infrastructure (e.g., roads), cause in-migration of labor, and often encourage small scale plantations through outgrower schemes-each of which may lead to forest conversion (Meijaard et al., 2018). We underline that our results are not evidence against such links. Rather, the impacts of plantation developments on forests occur over substantially longer time scales rather than the discrete year-by-year periods of our analyses.

F I G U R E 3 Scatterplot and associated
correlation lines and Pearson's correlation coefficient (r 2 ) between the annual expansion of industrial plantations and the average yearly crude palm oil (CPO) price the year before. The shaded areas show the 95% confidence interval for predictions from the linear model. Average yearly price was calculated from monthly prices in USD using IMF data (CPO prices 2018) We were also surprised that the area of forest converted to plantation annually remains tightly correlated with the annual plantation expansion in nonforest areas in both parts of Borneo throughout the study period (Figure 2e, f). These correlations suggested year-by year variation in the factors that influence investments in new plantations (e.g., business forecasts, land-availability). Exploring these patterns, we noted that the peak rates in plantation expansion seen in 2009 and 2012 followed peaks in the price of CPO in 2008 and 2011, leading us to assess and identify the strong correlation between CPO prices and plantation expansion in the subsequent year in Indonesian Borneo (Figure 3). The pattern in Malaysian territory remains less clear, though we speculate that Malaysian plantation developments are less responsive to short-term prices with expansion determined by more centralized long-term planning (see, e.g., Cramb & McCarthy, 2016;McCarthy & Cramb, 2009). Whatever the causes, Figure 2(e, f) and Figure 3 jointly indicate a stronger and more rapid short-term coupling of the variation in these investment decisions than we had anticipated.
Why did forest loss slow so markedly in 2017? One plausible factor is the implementation of the national peat conversion moratorium, that aims to protect peatland forests from conversion and fires (Norwegian Government, 2016). Additionally, 2017 a non-El Niño year, brought wetter conditions with fewer fires compared to (BMKG, 2017. Reduced use of fire in land management has likely followed intensive awareness campaigns and enhanced enforcement (Medrilzam et al., 2017). Our data show that industrial plantation expansion and associated forest conversion had slowed year on year since 2012, and in 2017 had dropped to their lowest level since 2003 (Figure 1). So why has this expansion slowed?
Various factors have likely influenced the spread of industrial plantations and whether they replace forest. The 2011 moratorium on new plantations in Indonesia likely reduced investments in developing new areas in Indonesian Borneo (Busch et al., 2015). We also note that the improved legal context for community based land claims in Indonesia have likely made it harder for companies to access land (Astuti & McGregor, 2017). In addition, lower CPO prices appear to have discouraged investors from pushing expansion in Indonesian Borneo, and it is likely that it would impact Malaysian plantation expansion too given time. Many of the big plantation companies are expanding their concessions elsewhere, for example, in Papua, in Central and Western Africa and the Americas (Austin et al., 2017;Furumo & Aide, 2017). Such trends are likely drawing on the large but finite sources of investment that were previously focused on Borneo (Potter, 2015).
Much remains uncertain. The overall impact of past initiatives to regulate expansion of plantations into forests are unknown. For example, although comparative studies indicate that RSPO obligations have had little impact in certified concessions Meijaard, Morgans, Husnayaen, & Ancrenaz, 2017)-these concessions tend to be older with little forest and we don't know how these company obligations have influenced the development of new concessions. Similarly, though No Deforestation commitments have had little obvious effect so far-the proportion of plantation expansion that involves direct conversion of forest has not noticeably declined-they may influence longer term investment choices. We also suspect that land and labor are becoming harder to source and sustain in Borneo. Furthermore, attention from NGOs and journalists, pressure from consumers and consumer nations, and (at least in Indonesian Borneo) increased government oversight, may all have constrained expansion. We also underline that we have not been able to quantify the expansion of small-scale smallholder plantations, which from our own informal observations and discussions, we suspect are playing an increasing role in recent years (Meijaard et al., 2018).
Regulations and commitments are necessary but insufficient to halt forest loss. Companies alone cannot prevent all significant losses such as that due to fires and smallholder expansion that arise inside, let alone outside, their concessions Cattau, Marlier, & DeFries, 2016;. Good policies and strong enforcement remain crucial.
Some policies to address forest conversion have potential for perverse outcomes. For example, we feared that delayed No Deforestation commitments might encourage some plantation companies to accelerate forest clearance while they can to avoid possessing stranded assets (forested lands that they cannot use). Thus, we anticipated that some companies might rush to clear such lands ahead of their self-asserted deadlines. Reassuringly, when we examined our data concerning forest losses by concessions by year over the period we see no evidence that the rates of clearance in the concessions with the most rapid losses have been increasing (unpublished results). Another concern is that unscrupulous plantation owners may seek to actively conceal their ownership so as to avoid accountability (see, e.g., Greenpeace, 2018;Jacobson, 2018). Indeed while current technology is sufficient to explore forest loss and link this to local concessions and their commitments, our concerns over unverified ownership have prevented us from making such an assessment (see Methods for a fuller discussion of the technical caveats). Companies who have nothing to hide have no reason to accept this murky situation and-despite the often complex and sometimes obstructive legal context (various personal communication to authors)should look for ways to publish and share the locations, boundaries and ownership details of all their suppliers. Without such information company conduct, including adherence to any No Deforestation commitments, cannot be overseen.
For those concerned for the future of Borneo's forests the bigger picture remains mixed. On the positive side No Deforestation commitments hold promise, and our results show that plantation expansion and deforestation have both slowed. On the negative side, transparency remains incomplete. Fires and industrial plantations continue to cause deforestation. We see no sign that plantation developments are seeking to avoid forest conversions. The links between the price of CPO and plantation expansion, and between plantation expansion and forest loss, indicate that markets play a major role in driving recent trends. Although our observations justify cautious optimism, much work remains to be done to ensure a future for Borneo's forests.

Definition of forest
Our definition of old-growth forests (abbreviated to "forests") excludes young forest regrowth, scrublands, tree plantations, forests that have been cleared by wildfires, agricultural land, and nonvegetated areas. These old-growth forests are closedcanopy (>90% cover) and high carbon stock (Above Ground carbon: 150−310 Mg C/Ha) evergreen dipterocarps growing on either mineral or peat soils. On peat domes, forests include low carbon stock pole forests. In highland regions, forests include "kerangas" and in coastal areas, mangroves. Our definition of "forests" includes intact and selectively harvested forests. See Supplementary Methods for more detailed information.

Annual loss of natural forest
Annual forest loss represents the area of old-growth forest (including selectively logged forest) that has been cleared (or burned) year since 2001 until 2017. It is caused by conversion to industrial and smallholder plantations, forest fires, infrastructure developments, open-pit mining, and reservoirs for hydropower dams. We extracted annual forest loss from the most recent annual tree loss map created by Hansen et al. (2013), revealing areas of tree cover (>30%; tree height >5 m) cleared every year from 2001 to 2017. See Supplementary Methods for more detailed information.

Annual expansion of industrial plantations
Annual expansion of industrial plantations represents the area of land (whether forest on nonforest) that has been converted to a large-and medium-scale industrial plantation of oil palm or pulpwood in a given year since 2001 until 2017. To map plantations, we scanned seventeen annual cloud-free LANDSAT image mosaics to detect plantations. We declared an area "developed (or under development)" by companies, the moment we observed large rectangular elements, long linear boundaries, and distinctive grid-or contour-planting patterns appear on our sequence of images. See Supplementary Methods for more detailed information.

Annual conversion of forest and nonforest to industrial plantations
We determined the area of forest (and nonforest) converted every year into industrial plantations across Borneo from 2001 until 2017. This was determined by measuring the overlap between the annual plantation map and the annual forest loss map. We declared an area of forest converted to plantations, if it became a plantation in the same year that it lost forest cover. This rapid within-1-year conversion captures 92% of all forest conversion to industrial plantations since 2000. Ninety three percent of the nonforest areas converted to plantations since 2000, were already nonforested prior to year 2000. Seven percent of the nonforest areas converted, were forests cleared in 1 year (in 2001, 2002,…2015), then converted to plantation more slowly (within 2, 3,….17 years). See Supplementary Methods for more detailed information.