Deforestation risks posed by oil palm expansion in the Peruvian Amazon

Further expansion of agriculture in the tropics is likely to accelerate the loss of biodiversity. One crop of concern to conservation is African oil palm (Elaeis guineensis). We examined recent deforestation associated with oil palm in the Peruvian Amazon within the context of the region’s other crops. We found more area under oil palm cultivation (845 km2) than did previous studies. While this comprises less than 4% of the cropland in the region, it accounted for 11% of the deforestation from agricultural expansion from 2007–2013. Patches of oil palm agriculture were larger and more spatially clustered than for other crops, potentially increasing their impact on local habitat fragmentation. Modeling deforestation risk for oil palm expansion using climatic and edaphic factors showed that sites at lower elevations, with higher precipitation, and lower slopes than those typically used for intensive agriculture are at long-term risk of deforestation from oil palm agriculture. Within areas at long-term risks, based on CART models, areas near urban centers, roads, and previously deforested areas are at greatest short-term risk of deforestation. Existing protected areas and officially recognized indigenous territories cover large areas at long-term risk of deforestation for oil palm (>40%). Less than 7% of these areas are under strict (IUCN I-IV) protection. Based on these findings, we suggest targeted monitoring for oil palm deforestation as well as strengthening and expanding protected areas to conserve specific habitats.


Introduction
The demand for commodity crops has driven the global expansion of agriculture, often at the expense of natural habitats and biodiversity. Most of the world's biodiversity hotspots are in tropical regions [1,2]. In these regions, agricultural expansion may accelerate deforestation and threaten biodiversity and associated ecosystem services.
The African oil palm (Elaeis guineensis) is a commodity crop of singular concern to conservation, as it grows almost exclusively in areas that were once tropical moist forests. Oil palm production has doubled in the recent past [3], with demand projected to drive further expansions [4,5]. If oil palm expands into all biophysically suitable areas, it may affect more than half of all threatened birds and mammals globally [6].
Palm oil, derived from the oil palm fruit and seeds, is the most widely consumed vegetable oil in the world. Processed foods, cosmetics, biofuels, household cleaners and industrial lubricants all use it. Oil palm is expanding rapidly in South America, driven by economic incentives, international investments, and large areas suitable for its production [7]. Although estimates of deforestation for oil palm vary throughout South America, prior studies in Peru suggest that at least half of current oil palm is on land deforested in the past two decades [8][9][10].
We consider the recent deforestation impacts of oil palm within the context of general agricultural development and protected areas in the Peruvian Amazon. We examined the extent and spatial pattern of oil palm plantation development and its distribution relative to the physical environment, human population, and accessibility. From these analyses, we projected long-term and short-term risks of deforestation from oil palm development. We then explored the distribution of such risks relative to administrative regions, ecoregions, recognized indigenous territories, and currently and formerly protected areas.
We asked the following questions: (1) what is the current extent of oil palm compared to other crops in the Peruvian Amazon, how has it changed over time, and how much of it came from recent deforestation?
(2) Do oil palm plantations differ in size, spatial distribution, or biophysical characteristics from sites occupied by other crops? (3) Do human population and the ease of forest access predict the location of oil palm plantations within bounds set by site characteristics? (4) How well do the currently protected areas and recognized indigenous territories conserve forests predicted to be at risk of deforestation for oil palm? (5) How could the expansion of oil palm agriculture impact specific habitats and biological communities in the Peruvian Amazon?
Despite the inaccessibility of much of the Amazonian region, human activities have fragmented the forests near population centers and provided markets for agricultural products. Agricultural expansion in this region concentrates along roads and rivers that provide access to these markets [11]. Common subsistence crops in the Amazon basin include maize, rice, cassava, beans, and fruit trees, while commercial crops include soybean, coffee, cacao, and beef [12]. Thus, oil palm is part of a diverse and rapidly changing agricultural sector in the Amazon basin and this trend mirrors that of South America as a whole, where agriculture is a primary driver of deforestation [13,14]. In 2005, pasture and commercial cropland accounted for 96% of agricultural land in South America, while tree crops such as oil palm accounted for only 0.5% [15]. However, in the past decade, regional studies show marked expansion of tree crops including cacao [16] and coffee [17].
Away from population centers and the roads and rivers connecting them, much of the Peruvian Amazon is still intact. It is one of the most biodiverse areas in the world [18,19] and contains large areas of contiguous primary forest [20]. Extensive areas of forest are legally protected from development in some manner, with protected areas as the primary tool for conservation [21]. Compared to equivalent areas without protection, protected areas and indigenous territories reduced deforestation in the Peruvian Amazon [22][23][24] and studies in the Brazilian Amazon showed that both are barriers for land clearing associated with 'slash and burn' agriculture [25,26].
The placement of protected areas, however, does not always align with broader goals of conservation, but rather with residual areas of human land uses [27,28]. In this analysis, we considered not just the context of agricultural areas and existing human population centers, but also the coverage of currently protected areas and officially recognized indigenous territory. Our focus was their possible role in limiting the future expansion of oil palm agriculture should global demand drive its continued expansion.

Methods
Our analyses proceeded in five parts. (1) We mapped the extent of deforestation for oil palm and other crops in the study area. (2) Using climatic and edaphic factors, we modeled the risk of future deforestation from oil palm expansion ('long-term risk'). (3) We refined the forests at long-term risk using population and accessibility factors to identify those most immediately threatened ('short-term risk'). (4) We examined the distribution of protected areas and indigenous territories, as well as their overlap with atrisk forests. (5) We then evaluated these areas by ecoregion, representing risk to the region's major habitats.

Mapping deforestation
Our focus was the tropical moist forest biome of Peru, an area that occupies 53.9% of the country. We defined our study area as the monitoring area for the forest change dataset [29] generated by Peruvian Government agencies, primarily the Servicio Nacional Forestal y de Fauna Silvestre (SERFOR), the Ministry of the Environment (MINAM), and the Ministry of Agriculture and Irrigation (MINAGRI). Using this dataset, we assessed yearly forest and non-forest areas from 2000-2015 at 30 m spatial resolution. A known problem in the dataset is that river meanders and oxbow lakes contribute substantial false positives to deforestation detection. To account for such errors, we further refined these data by removing all pixels identified as surface water at any time between 2000-2015 in the JRC Global Surface Water Occurrence dataset [30].
Within areas classified as non-forest in 2015, we identified oil palm and other cropland areas. To visually identify patterns of oil palm monoculture, we used high-resolution (0.5 m spatial resolution) Worldview-02 and Worldview-03 imagery for 2014-2016 obtained from DigitalGlobe (NextView License). Where possible we corroborated our identification of oil palm plantations with news and scholarly articles, government and company records, and geotagged photos.
We defined cropland as areas with active or fallow cultivation of annual or plantation crops other than oil palm. To identify cropland, we used an unsupervised classification implemented in Google Earth Engine, based on Landsat 8 2015 data (SI: Addl. Methods). Pastures were excluded from our analysis because they face different biophysical constraints than areas of crop production, including oil palm cultivation. Many of the region's agricultural areas are mixed-use farms that do not conform to the large-scale monoculture typical of industrial agricultural practices, increasing the likelihood of underestimating cropland extent [31].
Using the binary forest cover layers for 2000-2015, we conducted a change analysis within current oil palm and other cropland to determine the percent forested in any given year. We then determined the size of contiguous oil palm or other crop patches in 2015, and repeated this analysis again for 2000 to determine the size of non-forest patches within the 2015 sample areas.
Short distances may separate small patches of deforestation, so a cluster of such patches may be appropriately viewed as a single patch in terms of habitat fragmentation. To address this possibility, we also evaluated the spatial clustering of oil palm and other crop deforestation patches at different aggregation distances, from Euclidean distances of 0-600 m at 60 m intervals (figure 1). We examined how mean patch size increased with aggregation distance for oil palm and other crops to indicate a relative degree of clustering in both types of agricultural deforestation.

Long-term and short-term risk modeling
To estimate the areas at long-term risk of deforestation for oil palm, we calculated the biophysical envelope for oil palm growth. By this, we mean the current distributions of climatic and edaphic variables for current oil palm plantations in the region. We examined physical and climatic variables likely to influence the planting of crops: mean annual precipitation, minimum monthly temperature, elevation, slope, and soil depth (SI: Addl. Methods). Using these distributions, we identified other forested areas likely suitable for oil palm.
Within forest areas at long-term risk of deforestation, we used a classification and regression tree model (CART) to determine areas of short-term, i.e. more immediate, risk of deforestation (SI: Addl. Methods).
Here, explanatory variables describe population centers and accessibility for development: large urban areas, villages, distance to roads and navigable rivers.

Evaluation of protected areas
Within areas determined at long-term risk of deforestation for oil palm, we assessed the extent of protected areas and officially recognized indigenous territories (henceforth just 'indigenous territories') by administrative region (department). Data on protected areas, including information on IUCN categories, were from MINAM and data on indigenous territories from Rede Amazônica de Informação Socioambiental Georreferenciada (RAISG) [32]. We also assessed National Forest areas degazetted in 1996 and 2000 based on data from the Protected Area Downgrading, Downsizing and Degazettement (PADDD) dataset, as well as the extent of oil palm and oil palm risk classes found within these areas [33].
We evaluated long-term and short-term risk by ecoregion as well as protected areas and indigenous territories coverage for each ecoregion. A lack of finescale data on species abundance and ranges in this region makes it difficult to assess the impact of deforestation on biodiversity. When such data are lacking, ecoregions [34] serve as useful proxies for the habitat and community types required to sustain biodiversity [35,36].

Mapping deforestation
We identified 845 km 2 of oil palm plantation area and 21 997 km 2 cultivated for other crops in the Peruvian Amazon (figure 2). In 2000, deforested areas within current oil palm and cropland areas showed a similar distribution of sizes with a median patch size of 0.81 km 2 and 0.99 km 2 respectively (figure 3). By 2015, we observed a highly divergent pattern for the median size of oil palm deforestation (4.50 km 2 ), more than six times the size of cropland deforestation (0.72 km 2 ). The statistical distributions in patch size of both the oil palm and cropland patches are long-tailed. The top 5% of patches deforested for both oil palm and cropland cultivation account for 82% and 76% of the area deforested for each land use, respectively. The distribution of patch sizes for oil palm is also bimodal, with many small patches in addition to the large ones.
The pace of forest transformed to oil palm or cropland, as a percent of that cover type in 2015, varied. In the 2015 oil palm area, 47% was forested in 2000   the presence of oil palm seedlings within the first year of planting. Some of the deforested areas may indeed be oil palm, but it is too soon to confirm using remote sensing. In the area planted in other crops in 2015, 21% was forest in 2000, with an average of 1.4% of this area converted from forest each year between 2000-2015.
The exact contribution of oil palm to overall agricultural deforestation fluctuated over the study period, reaching 20% in 2013, assuming current oil palm areas deforested since 2000 were converted directly to oil palm from forest ( figure 4(B)). From 2001-2006, oil palm accounted for an average of 5% of annual agricultural deforestation, in the second part of the study period, from 2007-2013, oil palm accounted for an average of 11% of annual agricultural deforestation. The same detection difficulties mentioned above affected the final two years evaluated for this metric. Over the entire study period, deforestation within the largest oil palm patches accounted for 88% of total oil palm deforestation.
We considered how patches of oil palm and other cropland deforestation would aggregate at distances up to 600 m. Using a linear regression of aggregation distance (d A ) to mean patch area after aggregation (figure 4(C)), we found that the rate of increase in mean patch area was five times greater for oil palm patches (0.50 d A +32.0, R 2 =0.99) than for cropland patches (0.10 d A −2.22, R 2 =0.97), indicating a much greater degree of clustering for oil palm.

Long-term risk modeling
The distributions of oil palm and other crop patches were different with respect to minimum monthly temperature, mean annual precipitation, elevation, slope, and soil depth (Kolmogorov-Smirnov test, p<0.001; figure S1 is available online at stacks.iop. org/ERL/13/114010/mmedia). Although the ranges of oil palm and other crops for each environmental factor examined overlap, oil palm tended to occupy a narrower biophysical niche. We determined threshold values that included all the variability in oil palm patch values for each explanatory variable: 1678 mm mean annual precipitation, 17.9°C for minimum monthly temperature, 705 m for elevation, 30°for slope, and 1 m for soil depth. Forest areas meeting all these conditions are considered at long-term risk of deforestation for oil palm plantations ( figure 5(A), table S1). Most of these areas were in Loreto (75%) and Ucayali (14%). None of the other regions contained more than 5% of the area at long-term risk.

Short-term risk modeling
Based on cross-validated error, we selected a tree with five terminal nodes for the CART model representing the short-term risk of deforestation for oil palm in unprotected forests within the long-term risk areas (figure S2). The rarity of oil palm deforestation in the area sampled for this model (0.1%) impacts the precision of the model, 92.6%. Two scenarios emerged for short-term risk of deforestation for oil palm within the long-term vulnerable forest areas: (1) forest that occur close to roads (<23 km) and urban areas (<104 km) and (2) forests that are further from urban areas, but even closer to roads (<12 km) and near previously deforested areas (>1.3% of the area within 300 m radius deforested). The greatest proportions of forest at short-term risk are in the Loreto (31%), Ucayali (15%), Huánuco (6%), and Madre de Dios (5%) regions ( figure 5(A), table S1).

Evaluation of protected and indigenous areas
Protected Areas cover 19 2988 km 2 (24.6%) of the Peruvian Amazon ( figure 5(A)). Officially recognized indigenous territories cover another 12 4626 km 2 . No oil palm is within current protected areas and <0.1% is within indigenous territories. We determined that 107 102 km 2 (23.6%) of forests at long-term risk are protected. Of these, 30 636 km 2 are within IUCN I-IV protected areas (6.7%). Another 75 589 km 2 (16.7%) of forest at long-term risk was within indigenous territories. We determined that 2708 km 2 of forest would be at short-term risk but for its protected areas status. Another 16% of forest at short-term risk is located within indigenous territories.
Peru removed the National Forest designation from its protected areas system in 2000 (SI: appendix B). These areas formerly covered 34 594 km 2 of the study area. In areas formerly protected by the Humboldt and Iparía National forests, we found 96 km 2 of oil palm, 11.3% of total estimated Peruvian oil palm. Some 81 of 100 random samples of oil palm in the former Humboldt and Iparía National forests were established after 2005. Later plantation establishment occurred within the interior of the formerly protected areas (SI: appendix B). We found 16 810 km 2 (3.7%) of forests at long-term risk and 2381 km 2 (12.6%) of forests at short-term risk of deforestation for oil palm are within degazetted National Forest areas. Some 19.2% (6654 km 2 ) of these former National Forests became officially recognized indigenous territory.

Status of ecoregions
Some 99.9% of oil palm deforestation is within two ecoregions, Ucayali Moist Forest (80.9%) and Iquitos Várzea (19.0%) (table S2, figure 5(B)). This reflects the two types of ecoregions most at risk of long-term development for oil palm: moist forests (Napo In each of the six moist forest and várzea ecoregions, 38%-43% of the long-term risk area is under some form of protection or within indigenous territories. However, these differ among those that have primarily protected areas (Iquitos Várzea, Solimões-Japurá and Southwest Amazon Moist Forests), indigenous territories (Ucayali Moist Forest and Purus Várzea) or a nearly equal amount of both (Napo Moist Forest). Less than 7% of the area at long-term risk is under strict (IUCN I-IV) protection.

Discussion
Oil palm covers a small area, but one expanding rapidly at the expense of forests Between 1990 and 2010, demand for palm oil resulted in an expansion of plantation area from 60 000-160 000 km 2 globally [3]. As suitable land becomes limited in Southeast Asia, producers are targeting expansion in tropical areas, including in South America [37]. The impacts of such expansion are of concern when viewed in conjunction with those of mining, logging, petroleum extraction, infrastructure development and urban expansion in the Peruvian Amazon [38][39][40][41][42][43][44][45][46].
We identified 845 km 2 of oil palm plantation area in the Peruvian Amazon (figure 2). The area is greater than reported in previous studies and nearly twice the FAO estimate of harvested oil palm area in 2015 (431 km 2 ) [3]. Almost half of the oil palm plantations we mapped were forest in 2000, compared to about 21% of other crops ( figure 4(A)). Oil palm has become a major source of agricultural deforestation in the Peruvian Amazon, accounting for 11% of agricultural deforestation from 2007-2013, while occupying less than 4% of the agricultural area. These results support general trends from previous studies suggesting recent deforestation for oil palm [9][10][11], including increases in the rate of deforestation since 2006 ( figure 4).
The spatial pattern of deforestation for oil palm differs from other crops in terms of scale, organization, and physical characteristics The spatial pattern of oil palm agriculture differed from other crops, with larger and more clustered deforestation patches (figures 3 and 4). Most oil palm cultivation comes from a small number of large plantations. However, the bimodal nature of oil palm patch distribution suggests both smallholder and large industrial actors are responsible for plantation establishment. A recent study of deforestation in the Peruvian Amazon noted that small patch deforestation is not always attributable to smallholder or subsistence agriculture but to a diversity of actors [47]. This motivates a need for further field surveys in the region. A greater percentage (50%) of large plantations came from deforestation since 2000 than small ones (30%). This is in broad agreement with earlier studies of this region [9,11]. The dramatic shift we observed in median patch sizes from 2000-2015 suggests a fundamental change in the scale of deforestation within areas currently occupied by oil palm.
While small plantations may each have a limited impact, oil palm plantations were also more spatially clustered than cropland areas ( figure 4(C)). Clustering of oil palm plantations may result from the need to transport oil palm fruits to mills soon after harvest. Rapid processing of the fruit is needed to prevent a rapid rise in free fatty acids that reduces the quality of the crude palm oil produced [48]. Such aggregation possibly leads to larger effective patches of oil palm deforestation, when we consider the effects of fragmentation on the remaining forest between patches. Studies indicate that increasingly fragmented forest areas support fewer species in the long-term and affect the function of forest areas [49][50][51][52][53].
Oil palm occupies sites at lower elevations, higher precipitation, lower slopes, and greater soil depths than many crops in this region (figure S1). Such habitat envelope differentiation suggests that oil palm development leads to deforestation in different areas than for other crops. If oil palm agriculture continues to expand, it could greatly change the distribution of agriculture on this landscape.
Population and accessibility factors predict the location of current oil palm development within the bounds set by climate and soils Large areas of forest are biophysically suitable for oil palm. We do not suggest that the entirety of this area will convert to oil palm. Rather we suggest that deforestation for oil palm could become a substantial part of overall deforestation within this region, given the extent of suitable forest area. Within the area determined suitable for oil palm, approximately 13% are seasonally or permanently flooded wetlands, including lowland peatlands [54]. We have not observed conversion of wetlands for oil palm in Peru, though logging, agriculture, and grazing activities increasingly degrade and convert these systems [55]. Peatland conversion for oil palm is an issue in Southeast Asia [56] and in the flooded savannah ecosystems of the Colombian Llanos [57]. Because of the uncertainty around future oil palm development scenarios in the Peruvian Amazon, we include wetlands in our areas projected to be at risk of development.
We highlight a subset of the biophysically suitable envelope as under short-term threat due to population and accessibility factors. In the short-term, the proximity of urban areas, roads and previous deforestation increase the risk of deforestation from oil palm ( figure 5(A)). Another important factor is the proximity to existing oil palm plantations, as evidenced by the spatial clustering of oil palm plantations ( figure 4(C)). Other studies find similar factors to be important in other regions of the world [58][59][60].
Frequent satellite images are few in areas of active oil palm expansion. Thus, the exact sequence of events in deforestation for new plantations and the development of infrastructure to bring palm oil to market requires further investigation. Our fieldwork near Pucallpa supports the idea that, while major road infrastructure is predictive, the establishment of largescale plantations involves construction of new roads and improvement of existing ones. Our short-term risk analyses suggest that monitoring should continue throughout this region, including areas where there is no oil palm currently. A notable example is the province of Madre de Dios, which shares many of the biophysical, population and accessibility factors with areas that already have oil palm development.
Existing protections are working, but few areas are strictly protected Less than 7% of the area at long-term risk is under strict (IUCN I-IV) protection. Although Peru's forestry laws mandate preservation of vegetation cover and forestry resources, agricultural deforestation has occurred under the 'best land use capacity' exemption to them [61]. This highlights the importance of protected areas and indigenous territories. A new governmental resolution, signed by the Vice-Minister of Culture in 2017, mandates consultation with communities in indigenous territories (Resolution 014-2017-VMI-MC). Recent news items indicate that some indigenous communities are vocal in their concern about development within their territories [62].
It is also important to maintain long-term protections. From 1996-2000, Peru degazetted National Forests from their protected area system. This reduced the total protected areas in the study region by 12.3%. After degazetting, former National Forests opened to commercial logging and agriculture. From 2000-2010, deforestation rates exceeded those in both protected areas and areas that were never protected [63]. Most of the oil palm we identify in the former Humboldt and Iparía National Forests was planted after 2005 (SI: appendix B). Maintaining and enforcing protections preventing agricultural deforestation in National Forests could have reduced the areas at long-term and short-term risk of deforestation for oil palm by 3.7% and 12.6% respectively. While our findings support the conclusion that protected status prevents development of protected lands, it could displace development to other areas and, therefore, may not reduce overall risk of deforestation.
Oil palm expansion could heavily impact certain habitats and biological communities The two ecoregions currently containing oil palm (Ucayali Moist Forest and Iquitos Várzea) also contain much of the area at short-term risk of oil palm deforestation. In the long term, all six ecoregions of low elevation moist forests and várzea have large areas at risk. Little of these are under strict protection (<7%), though indigenous territories and other categories of protection cover ∼40% of each. This suggests that increasing protections of already designated areas could be a major component of conservation in this region. Indigenous territories are critical to the protection of the Purus Várzea, Napo Moist Forests and Solimões-Japurá Moist Forests, which have more than 90% of their area at long-term risk. Eighty percent of at-risk area currently under IUCN I-IV protection lies within the Southwest Amazon Moist Forest. It has considerable area under short-term risk despite containing no oil palm currently.
The role of policy and voluntary commitments in the future of oil palm expansion in Peru Peru's recent National Oil Palm Plan [64] indicates that an area of more than 14 000 km 2 may be suitable for oil palm development, a claim that conflicts with the goal of reducing deforestation within the Amazonian region. In recent years, companies have experienced increasing pressure to purchase responsibly produced palm oil, certified through the Roundtable on Sustainable Palm Oil (RSPO) [65]. RSPO certified palm oil now accounts for about 20% of global oil palm production. A recent study of RSPO certification effectiveness in Indonesia found it to be insufficient in reducing deforestation for oil palm [66].
In a recent instance of RSPO activity in Peru, the Native Community of Santa Clara de Uchunya made a formal complaint regarding the activities of one of its member companies: plantations of Pucallpa. An RSPO investigation ruled the company guilty of land clearing within traditional forest lands, violating the code of conduct for member companies [67]. Although Plantations of Pucallpa withdrew from the RSPO prior to the verdict, the Peruvian government fined it for compliance issues related to the case. Outside of the RSPO framework, a combination of societal pressure and legal action prevented deforestation of 230 km 2 of primary forest by Grupo Romero, a Peruvian producer in 2015 [61].
Voluntary commitments form an important part of deforestation regulation in Peru, but their effectiveness is limited when unaccompanied by policy interventions. Plantations will likely continue to expand in the coming years with a combination of international and national producers active in the region. As evidence of this, after the end of our study period in 2016 and 2017, deforestation occurred to establish a large plantation in Northern San Martin and for smaller plantations in the Ucayali and Huánuco regions.

Conclusions
This study proposes an approach to limit deforestation from oil palm expansion and to consider its implication for conservation in the Peruvian Amazon. Deforestation monitoring programs, like the newly established National System of Monitoring and Control in Peru, can target the unique spatial patterning associated with oil palm development within areas shown to be at risk. The implementation of real-time monitoring using satellite imagery enhances this approach [68]. In habitats disproportionately at risk of oil palm deforestation and with little protected area coverage, it may be advisable to establish new protected areas, increase protection levels in existing ones, and effectively implement consultation with indigenous communities. These strategies, along with regulation on infrastructure expansion such as road construction, could be essential to reduce the impacts of oil palm expansion to biodiversity. resolution imagery from DigitalGlobe, Digital Globe (Nextview), and USAID GeoCenter Remote Sensing Program.