Rubber plantation size and global rubber price are linked to forest loss and degradation in Jambi, Sumatra
印度尼西亚苏门答腊占碑省的森林丧失和退化与橡胶种植园的面积和全球天然橡胶价格有关
Luas perkebunan karet dan harga karet dunia menjadi penyebab degradasi hutan alam di Jambi, Sumatra
Editor-in-Chief & Handling Editor: Ahimsa Campos-Arceiz
Abstract
enNatural rubber cultivation is one of the main drivers of tropical deforestation and biodiversity loss. This study examines regulatory and socio-economic conditions that increase the susceptibility of rubber plantations to deforestation and degradation, aiming to support zero-deforestation pledges and sustainability commitments made by the natural rubber industry. By combining bottom-up socio-economic survey data from rubber smallholder farmers in Indonesia with top-down spatial datasets on forest loss and degradation, this study identifies factors associated with deforestation, tree cover loss, and degradation of high-risk plantations. In Jambi Province, Indonesia, from 1991 to 2018, the overall tree cover loss in areas adjacent to rubber plantations was positively correlated to plantation size, remoteness (travel time to cities), and distance to the nearest protected areas, indicating that larger, remotely located plantations likely expanded more into forests between 2000 and 2018. Similarly, tropical forest degradation was positively associated with plantation size, travel time to cities, and distance to protected areas. A higher rubber price in the preceding year correlated with increased annual deforestation and forest degradation, whereas lower prices had the opposite effect. These results suggest that monitoring price changes and identifying plantations that are near non-protected forest frontiers could enable early detection and potential mitigation of deforestation threats.
摘要
zh生产天然橡胶的橡胶人工林种植是热带森林砍伐和生物多样性丧失的主要因素之一。本研究探讨了与能够导致森林砍伐和退化的高风险种植园相关的政策及社会经济因素，旨在支持天然橡胶行业做出的零砍伐承诺和可持续发展承诺。本研究通过将自下而上的研究区橡胶小农户社会经济调查数据与自上而下的森林丧失和退化空间数据相结合，识别了导致森林砍伐、林木丧失和森林退化的高风险种植园的政策及社会经济因素。研究结果显示在印度尼西亚占碑省，自1991年到2018年间，橡胶种植园邻近的林木覆盖率丧失与橡胶种植园面积大小、偏远程度（到城市的交通时间）以及到最近邻的保护区的距离呈正相关关系，这表明在2000—2018年期间，面积越大、越偏远的种植园越可能向森林扩张了更多。研究还发现，热带森林退化与种植园面积大小、到城市的交通时间以及到保护区的距离也同样呈正相关关系。上一年较高的天然橡胶价格与增加的年均森林砍伐和森林退化显著正相关，而价格下跌则与低的年均森林砍伐和退化相关。以上研究结果表明，通过监测天然橡胶价格变化以及识别那些处于非保护的森林附近的种植园，可以及早发现和缓解毁林的潜在威胁。【审阅:翟德利】
Abstrak
idPerkebunan karet alam merupakan salah satu faktor utama deforestasi dan hilangnya keanekaragaman hayati di hutan tropis. Penelitian ini bertujuan untuk mengkaji peran dari regulasi pemerintah dan kondisi sosial-ekonomi masyarakat yang meningkatkan kerentanan deforestasi dan degradasi hutan yang disebabkan oleh perkebunan karet dengan tujuan mendukung komitmen nol-deforestasi dan keberlanjutan yang digagas industri karet alam.
Dengan menggabungkan data survei sosio-ekonomi “bottom-up” yang didapat dari petani karet kecil dengan data spasial “top-down” perubahan tutupan lahan, studi ini mengidentidentifikasi faktor-faktor terkait dengan deforestasi, dinamika tutupan hutan, dan degradasi lahan perkebunan berisiko tinggi.
Hilangnya tutupan hutan yang berdekatan dengan perkebunan karet di provinsi Jambi pada tahun 1991–2018 memiliki korelasi positif dengan luas perkebunan, keterpencilan (waktu tempuh ke kota), dan jarak ke kawasan lindung terdekat, mengindikasikan bahwa perkebunan dengan letak terpencil berisiko menyebar ke dalam hutan rentang tahun 2000–2018. Demikian pula, berkorelasi positif dengan luas perkebunan, waktu tempuh ke kota, dan jarak ke kawasan lindung.
Kenaikan harga karet akan meningkatkan laju deforestasi dan degradasi hutan setiap tahunnya, sedangkan harga yang lebih rendah mempunyai dampak sebaliknya.
Dengan memantau perubahan harga dan identifikasi perkebunan yang berdekatan dengan perbatasan hutan yang bukan kawasan lindung dapat memberikan deteksi awal mitigasi deforestasi.
Plain language summary
enRubber agriculture significantly contributes to tropical deforestation and biodiversity loss. This research focuses on understanding the socio-economic characteristics of rubber plantations that increase their likelihood of contributing to forest loss and degradation. By analyzing survey data from questionnaires with smallholder rubber farmers in Jambi, Indonesia, a pattern emerged suggesting that larger, remotely located plantations are more likely to be linked with forest loss and degradation in their surrounding areas from 2000 to 2018. The global rubber price also influences deforestation rates, with higher prices in the previous year leading to more deforestation and degradation and lower prices having the opposite effect. These findings highlight the need for careful monitoring of rubber plantation characteristics and market prices to help mitigate their impact on natural forests.
简明语言摘要
zh橡胶种植业严重加剧了热带森林砍伐和生物多样性丧失。本研究的聚焦于了解那些会导致森林丧失和退化可能性增加的橡胶种植园的社会经济特征。通过分析印尼占碑省橡胶小农户的问卷调查数据，研究发现从2000年到2018年间，面积越大、位置越偏远的种植园其周边的森林丧失和退化面积越大的一种模式。全球橡胶价格也会影响森林砍伐率，上一年较高的天然橡胶价格会导致更多的森林砍伐和退化，而与较低的价格相对的则是较低的森林砍伐和退化。这些研究结果突出表明，有必要密切监测橡胶种植园的特征和天然橡胶价格，以减轻它们对天然林的影响。
Ringkasan sederhana
idPertanian karet secara signifikan berkontribusi terhadap deforestasi dan hilangnya keanekaragaman hayati di hutan tropis.
Penelitian ini bertujuan untuk memahami karakteristik sosio-ekonomi perkebunan karet yang menjadi faktor pendorong hilangnya sebagian besar tutupan hutan alam.
Dengan menganalisa data kuesioner dengan petani karet skala kecil di Jambi, Indonesia, muncul sebuah pola yang menunjukkan bahwa perkebunan skala besar dan terpencil menjadi penyebab hilangnya tutupan hutan pada tahun 2000–2018.
Kenaikan harga karet dunia juga meningkatkan laju deforestasi serta degradasi hutan, terutama ketika kenaikan harga pada tahun sebelumnya, sedangkan harga karet rendah mempunyai efek sebaliknya.
Temuan ini menyoroti perlunya pemantauan yang cermat terhadap karakteristik perkebunan karet dan harga pasar karet untuk mendukung mitigasi dampak negatifnya terhadap hutan alam.
Practitioner points
en
-
Larger and more remotely located rubber plantations are more likely to contribute to forest loss and degradation.
-
Fluctuations in global rubber prices may affect rubber-associated deforestation and forest degradation rates.
-
Early identification of plantations at high risk of contributing to deforestation and degradation and implementing strategies to mitigate these threats could help curb future forest loss due to rubber agriculture.
实践者要点
zh
-
面积越大、位置越偏远的橡胶林越有可能导致森林丧失和退化。
-
全球天然橡胶价格的波动可能会影响与橡胶林相关的森林砍伐和森林退化率。
-
尽早识别导致森林砍伐和退化的高风险橡胶种植园，并采取措施缓解这些威胁，有助于遏制未来因橡胶种植业造成的森林丧失。
Ringkasan Praktisi
id
-
Fluktuasi harga karet global dapat mempengaruhi laju deforestasi dan degradasi hutan alam.
-
Semakin luas dan terpencilnya perkebunan karet berisiko tinggi menyebabkan hilangnya tutupan hutan alam.
-
Identifikasi dini perkebunan karet yang berisiko menyebabkan deforestasi dan degradasi hutan alam merupakan strategi preventif yang dapat membantu mengurangi hilangnya tutupan hutan alam yang disebabkan oleh perluasan perkebunan karet dimasa yang akan datang.
1 INTRODUCTION
Anthropogenic land-use change and forest disturbances pose significant threats to natural ecosystems and biodiversity while also contributing to climate change through carbon emissions (Barlow et al., 2016; Moffette et al., 2021; Vancutsem et al., 2021). Despite efforts and commitments to reduce commodity-driven deforestation, it remains driven by various types of agriculture—permanent, shifting, commercial, and subsistence (Curtis et al., 2018; Houghton, 2012; Jayathilake et al., 2021). A considerable portion of this deforestation is due to boom crops, such as rubber, which replaced over 5 million hectares of forested lands between 2001 and 2014 in mainland Southeast Asia alone (Hurni & Fox, 2018). Forest degradation also requires more attention, given its negative consequences, which include reduced carbon stocks (Pearson et al., 2017) and biodiversity loss. The alignment of rubber plantations with four biodiversity hotspots in Southeast Asia emphasizes the urgency to assess land-use changes associated with rubber expansion and its implications on endemic and threatened species (Wang et al., 2020; Warren-Thomas et al., 2015).
The demand for rubber products remains high, with total consumption expected to continue increasing (IRSG, 2018; Kenney-Lazar et al., 2018), leading to further deforestation risks. Predictions suggest that over 4 million ha of additional land area will be required to meet this demand in the near future (Warren-Thomas et al., 2015). However, with the massive expansion of rubber plantations in China (Zhang et al., 2019) and subsequent changes in demand-supply dynamics and reduced prices, rubber production has declined in some Southeast Asian countries such as Malaysia (Ali et al., 2021) and Thailand (Sowcharoensuk, 2019). Nevertheless, rubber-driven deforestation continues to occur in the region, in countries like Indonesia (Otten et al., 2020) and Cambodia (Grogan et al., 2019). Such expansion into bio-climatically suitable tropical forests could result in significant biodiversity loss, increasing species extinction risks (Wang et al., 2020).
The underlying socio-economic and spatial drivers of plantation expansion are complex and interrelated, posing challenges for regulation (Meyfroidt et al., 2020). Understanding these drivers is essential for formulating efficient land-use planning policies and strategies (Velasco et al., 2020). Economic profitability is a key determinant of agricultural land use (Clough et al., 2016), with multiple management factors such as technical guidance and available capital influencing productivity and plantation expansion (Agwu, 2007; Meyfroidt et al., 2014). Land tenure and governance factors also play a role in shaping land use decisions (MacUra et al., 2015). In addition, accessibility to cities through road networks affects deforestation and degradation rates (Barber et al., 2014), as transportation costs can limit commercial agricultural activities (Dang et al., 2019), reducing disturbances to forests via selective logging or fire. The establishment of protected areas (PAs) helps mitigate land-use spillovers, but deforestation continues to occur in areas without protection (Delacote et al., 2016), highlighting the complex nature of deforestation control and the need for comprehensive strategies that address these diverse factors.
Global commodity prices also drive agriculture-related deforestation (Angelsen & Kaimowitz, 1999; Verburg et al., 2014). For instance, the expansion of oil palm plantations in Indonesia is linked to palm oil prices, with higher prices likely leading to increased deforestation (Gaveau et al., 2019). Similarly, in the Amazon, escalated deforestation has been associated with high soy and corn prices (Harding et al., 2021). In Southeast Asia, the price of rubber has also been associated with increased deforestation in countries like Cambodia (Grogan et al., 2019) and Laos (Junquera et al., 2020). However, these inferences are based on correlations between overall deforestation and commodity prices without accounting for price in the context of plantation-related management factors or other spatial characteristics.
Building deforestation-free rubber supply chains is a challenging task that requires tracing produce back to plantations with the potential to expand into natural habitats. Source locations of products are usually obscured by the involvement of various aggregators and distributors, complicating the identification of product origins (Curtis et al., 2018). However, due to consumer demand for sustainable commodities, producers are increasingly interested in implementing certification schemes. These schemes aim to ensure products meet social and environmental sustainability criteria, potentially allowing producers to benefit from price premiums similar to those seen in the oil palm industry (Heilmayr et al., 2020). In the rubber industry, the supply chain often involves multilevel intermediaries before products reach manufacturers and global markets (Ma et al., 2014), making it difficult for large conglomerates to pinpoint the exact locations of their suppliers. This complexity hinders their ability to fulfil commitments to deforestation-free products and to guarantee sustainability throughout their supply chains (Curtis et al., 2018).
This study aims to enhance the understanding of rubber plantation expansion by examining the relationship between rubber-driven deforestation and degradation, fluctuations in rubber prices, spatial context, and associated socio-economic drivers in smallholder rubber plantations in Jambi Province, Indonesia. Globally, Indonesia has the largest area dedicated to rubber plantations (FAO, 2021) and concurrently has one of the highest rates of deforestation (Margono et al., 2014). This is driving severe risk of species extinction (Hoffmann et al., 2010), making it crucial that these threats are mitigated.
- 1.
To quantify the amount of forest loss occurring within rubber plantations and in their surrounding areas.
- 2.
To identify key plantation characteristics and other socio-economic variables that affect deforestation and degradation resulting from rubber expansion.
- 3.
To evaluate how annual fluctuations in rubber prices, alongside plantation management practices, correlate with deforestation and forest degradation.
2 MATERIALS AND METHODS
2.1 Study site
The study was conducted in Jambi Province, Indonesia, an area that has been experiencing a rapid loss of tropical rainforests (Kubitza et al., 2018). Agriculture in Jambi is dominated by crops like rubber and oil palm, making this province the third-largest rubber-producing region in Indonesia (Schwarze et al., 2015). Rubber production in Jambi is primarily conducted by smallholder farmers, with only a few large-scale companies involved (Kubitza et al., 2018). In this province, the rubber production cycle involves a seed season around February to March and a grafting season from June to October (Penot et al., 1998). Unlike other regions, Jambi's high rainfall means the wintering season, a period when rubber tapping typically halts, is less defined, allowing for continued tapping operations throughout the year (Kelfoun et al., 2003).
2.2 Data collection and study design
Socio-economic data were collected using a semi-structured survey questionnaire, implemented on the ground by Agridence (formerly HeveaConnect), an organization working on digitalizing the global natural rubber supply chain to create a technology-powered, data-enriched ecosystem (https://rubber.agridence.com/). The surveys were conducted over a period of 12 months between 2018 and 2019, gathering data on farmer demographics, plantation characteristics, and cultivation practices (Supporting Information S1: Table S1). The interviews were conducted in Bahasa, Indonesia by four interpreters surveying different areas of Jambi Province (Figure 1a). For the same plantations, additional information on land tenure and agricultural advice were also added using RubberWay, a digital risk mapping tool built to identify sustainability risks in the upstream rubber supply chain (https://rubberway.tech/). Altogether, a total of 1103 smallholder rubber plantations across eight regencies in Jambi were surveyed and geographically located using the Global Positioning System (GPS) (Figure 1b). Since the plantations had not been spatially delineated, recorded GPS locations were taken as centre points and circular buffers were created with the same area of the respective plot sizes (information about plot size was collected through interviews with farmers) to represent each plantation (referred to as ‘plantation buffers’) using ArcGIS Pro. Further, for every plantation, circular buffers of 500 m and 1 km from these central points, excluding the previously created plantation buffers representing plot size, were created to identify possible plantation expansion.
Plantation-level variables were complemented by four spatial variables: travel time to cities (Weiss et al., 2018), land cover type in 2009 (GlobCover) (ESA, 2010), distance to PAs (UNEP-WCMC & IUCN, 2020), and bioclimatic suitability for rubber (Ahrends et al., 2015). Google Earth Engine (Gorelick et al., 2017) and ArcGIS Pro were used to extract data from the different datasets for the respective rubber plantations and buffer areas.
2.3 Data analyses
2.3.1 Quantifying tree cover loss, deforestation, and degradation
To quantify deforestation and degradation within and outside plantations, combined data on forest loss (‘tree cover loss’; Hansen et al., 2013) and deforestation and degradation of undisturbed tropical moist forests (Vancutsem et al., 2021) were extracted for plantations and buffer areas. Here, ‘tree cover loss’ refers to the complete removal of the tree canopy (Hansen et al., 2013), ‘deforestation’ refers to the long-term transformation of tropical moist forests into non-forested areas, and ‘degradation’ refers to disturbances in the canopy of tropical moist forests that are detectable from satellite imagery over a short period of time, typically less than 2.5 years (Vancutsem et al., 2021). By combining tree cover loss data (Hansen et al., 2013) and deforestation and degradation data of tropical moist forests (Vancutsem et al., 2021), this study aimed to provide a comprehensive view of forest dynamics related to rubber plantations. This approach was necessary because the Hansen data set may not effectively distinguish between cyclical rubber replacement and actual deforestation (Kou et al., 2015), nor does it capture short-term forest degradation, such as selective logging. On the other hand, the Vancutsem data set focuses specifically on the deforestation of tropical moist forests and may underrepresent other deforestation activities (Vancutsem et al., 2021). Both datasets were employed in the analysis to determine whether the trends identified were robust and consistent across different measures of forest loss, deforestation, and degradation.
Data were extracted for ‘Hansen tree cover loss’ for each plantation and the surrounding 500 m and 1 km buffer areas outside plantation areas from 2000 to 2017. ‘Vancutsem deforestation’ that occurred in undisturbed tropical moist forests, specifically due to tree plantations (Vancutsem et al., 2021), was also calculated for plantations and buffer areas using Google Earth Engine from 1991 to 2018. Similarly, using ‘Vancutsem degradation data’, the area of forests that underwent degradation within plantations and buffer areas between 1991 and 2018 was also quantified (Supporting Information S1: Figure S1).
2.3.2 Identifying factors associated with tree cover loss, deforestation, and degradation
Linear mixed-effects models were utilized to identify the factors contributing to tree cover loss, deforestation, and degradation in rubber plantations, incorporating both socio-economic and environmental variables as predictors. These variables, based on existing knowledge, could act as predictors of plantation expansion. Here, three dependent variables were considered separately: Hansen tree cover loss, Vancutsem deforestation of undisturbed tropical moist forests, and degradation of undisturbed tropical moist forests (Table 1).
| Dependent variables | Description | Data source | |
|---|---|---|---|
| 1 | Tree cover loss | Map of global tree cover loss between 2000 and 2017 at 30 m spatial resolution. | Hansen et al. (2013) |
| 2 | Deforestation of undisturbed tropical moist forests | Changes in undisturbed tropical moist forest cover from 1991 to 2018 at a 30 m spatial resolution. For this study, the deforestation that occurred because of tree plantations is considered. | Vancutsem et al. (2021) |
| 3 | Degradation of tropical moist forests | Degradation of tropical moist forests yearly from 1990 to 2018 at 30 m resolution. | Vancutsem et al. (2021) |
| Independent variables | Description | Theoretical rationale | |
|---|---|---|---|
| 1 | Plot size | Size of the plantation (hectares) | Actors behind varying land holding sizes are known to be driving different commodity crop expansion patterns (Meyfroidt et al., 2014). |
| 2 | Age of trees | Age of the rubber trees (years) | Age of rubber trees is an important consideration (Miyamoto, 2006) as rubber output falls drastically beyond a certain age, which may drive further expansion. |
| 3 | Disease presence | Whether rubber trees are affected by diseases | Diseases could affect productivity and management factors (Mazlan et al., 2019), and farmers may expand to meet the reduced output of disease-stricken plantations. |
| 4 | Income | Monthly income from rubber (USD) | Capital, investment opportunities (Meyfroidt et al., 2014), and availability of cash (Junquera et al., 2020) could be linked to increased crop expansion. |
| 5 | Land title | Type of land ownership documents (Official title/Community letter/Family heritage/None) | Land tenure insecurity is associated with deforestation and forest encroachment (Wannasai & Shrestha, 2008). |
| 6 | Advice frequency | How often/frequency of receiving agricultural advice (Annually/more often/once in the past 3 years/none in the past 3 years) | Agricultural extension services could affect plantation management (Agwu, 2007). A better understanding of rubber cultivation practices enables farmers to increase yields, reducing the need to expand. |
| Spatial data | Source and rationale | ||
|---|---|---|---|
| 7 | Minutes to city | Travel time in minutes to the nearest urban centre (contiguous area with 1500 individuals per km2 or built environment with over 50,000 people) | Source: Weiss et al. (2018) Higher accessibility could increase deforestation (Barber et al., 2014). |
| 8 | Suitability | A map of habitat suitability for rubber developed using species distribution modelling of 31 climatic, topographic, and substrate-related variables | Source: Ahrends et al. (2015) Plantations are likely to expand in areas with higher suitability. |
| 9 | Distance to PA | Distance to the nearest protected area from the midpoint of the rubber plantation | Source: UNEP-WCMC & IUCN (2020) Protected areas face fewer deforestation threats (Delacote et al., 2016). |
| 10 | Land cover type | The type of land cover in 2009 (cropland/cropland mosaic/forest/vegetation/urban/water) | Source: Globcover 2009 (ESA, 2010) This is to identify the type of land cover that had undergone tree cover loss. |
The following plantation characteristics were included as independent variables: plantation/plot size (hectares), age of rubber trees (years), disease presence, monthly income from rubber (USD), availability of land titles, and frequency of agricultural advice received. The following spatial variables from the surrounding landscape were also included as independent variables: distance to the nearest PA (km), travel time to cities (minutes), land cover type in 2009, and bioclimatic suitability for rubber trees. Data sources and the reasoning for variable selection are presented in Table 1. The R statistical environment (R Core Team, 2020) was used for all statistical analyses.
For the model selection process, 170 linear mixed-effects models were constructed (Supporting Information S1: Table S2), each incorporating predictor variables identified as important drivers of plantation expansion and consequent deforestation. These variables encompassed plantation management factors such as land tenure status, crop age, and disease presence, together with landscape characteristics such as proximity to urban centres and PAs (Table 1). The models varied in their combinations of explanatory variables, including an intercept-only model and possible interaction terms. The selection of the most suitable model was based on the Akaike Information Criterion (AIC), where the model with the lowest AIC value was considered to attain the best balance between explanatory power and simplicity. To address spatial autocorrection, which can influence the independence of observations, a Monte Carlo test with 999 permutations was employed to calculate Moran's index. If spatial autocorrelation was detected, a ‘rational’ spatial autocorrelation structure was used, which produced the best AIC values among different correlation structures considered (except for one instance when an ‘exponential’ spatial autocorrelation structure was used). The spatial autocorrelation structure is determined by the correlation matrix for each model, which assigns correlation weights to different pairs of geographic observations (Mets et al., 2017). ‘Here, the ‘rational’ quadratic spatial autocorrelation structure determines the correlation between two observations at an r‘’ distance and a d’ range as 1/(1 + (r/d)^2).
Regencies within Jambi Province were used as random intercepts in the models. The analysis confirmed the absence of any multicollinearity (variance inflation factor < 3) between independent variables. Following this, model averaging was performed on the 95% confidence model set, defined as the minimum set of top-performing candidate models that cumulatively account for at least 95% of total Akaike weight (Burnham & Anderson, 2002; Grueber et al., 2011). The land cover type was not included as an explanatory variable in models focused on deforestation and degradation of undisturbed tropical moist forests, given that the areas calculated for these dependent variables specifically reflect changes in undisturbed tropical moist forests. Instead, for these analyses, 150 linear mixed-effects models were considered. After listing them based on their AIC values, model averaging was conducted on models that cumulatively accounted for at least 95% of Akaike weights. The variances explained by fixed and random effects were determined using the marginal and conditional R2 (Nakagawa & Schielzeth, 2013).
Linear mixed-effects modelling was extended to incorporate the 500 m and 1 km buffer areas surrounding the plantation perimeters. Hansen tree cover loss, Vancutsem deforestation of undisturbed forests, and degradation of forests for 500 m and 1 km buffer areas were modelled against the explanatory variables described above. An information-theoretic model selection process was used to identify factors associated with tree cover loss, deforestation, and degradation in plantation-adjacent buffer areas. Top models were selected using AIC values, and model averaging was performed for models amounting to 0.95 of Akaike weights.
2.3.3 Identifying the effects of annual rubber prices on deforestation and degradation
To identify whether global rubber prices were linked to on-the-ground deforestation, data on rubber market prices were collected from the International Monetary Fund's primary commodity prices database (IMF, 2021). Monthly prices (USD) were averaged to reflect annual global rubber prices from 1990 to 2018. Yearly deforestation and degradation rates for undisturbed tropical moist forests were quantified for the same period using the annual change data set available (Vancutsem et al., 2021). Following this, the annual time series trends for deforestation, degradation, and price were plotted and observed. Hansen tree cover loss was not considered here as the available time span (2000–2017) did not fully align with the rubber price data set.
Since annual data were not available for explanatory variables on plantation management and surrounding spatial characteristics, these data were assumed constant throughout the study period for each year. The selection of fixed effects was based on their strong associations with deforestation and degradation, as determined in top-performing models from previous analyses. These included plot size, land title presence, age of trees, travel time to cities (minutes to city), and distance to PAs. The year was incorporated as a random slope, with regency and plantation ID serving as nested random intercepts to account for spatial and temporal variability. Due to the large quantity of ‘zero’ values present in annual deforestation and degradation data, area estimates (in hectares) were transformed into count data (number of hectares) to facilitate modelling with generalized linear mixed-effects models employing a negative binomial distribution and a log link function. This analysis focused solely on 500 m and 1 km buffer areas since plantation factors were repeated for the entire timeline, assuming these characteristics were established in 1990 and remain unchanged. In total, 60 different models were proposed and evaluated using an information-theoretic approach, with the best models identified based on their AIC values. Model averaging was then applied to those models accounting for 95% of Akaike weights, ensuring a comprehensive assessment of factors influencing deforestation and degradation in relation to global rubber prices.
3 RESULTS
3.1 Rubber-associated deforestation and degradation
Deforestation and degradation linked to rubber cultivation were observed both within the plantations themselves and in 500 m and 1 km buffer areas. Analysis of tree cover loss revealed that within plantation areas, as well as 500 m and 1 km buffer areas, 25%, 27%, and 26% of the areas experienced tree cover loss, respectively (Supporting Information S1: Table S4). From the total area of plantations, 12% had experienced deforestation of undisturbed tropical moist forests between 1991 and 2018. The deforestation in the 500 m and 1 km buffer areas was 14% and 15%, respectively (Supporting Information S1: Table S4). Forest degradation within these buffer areas reached 36% and 34% (Supporting Information S1: Table S4).
Temporal patterns of rubber-associated deforestation and degradation showed a noticeable trend. Of the total tree cover loss that occurred in 500 m buffer areas, 59% occurred between 2001 and 2010, while 41% occurred between 2011 and 2018 (Figure 2, Supporting Information S1: Table S4). A similar pattern was observed for deforestation of undisturbed tropical moist forests where 20%, 44%, and 36% of the deforestation occurred between 1991–2000, 2001–2010, and 2011–2018, respectively (Figure 2, Supporting Information S1: Table S4). The percentage area of degradation was 36%, 37%, and 27% for each respective period. Trends in 1 km buffer areas were very similar (Figure 2, Supporting Information S1: Table S4). Overall, the highest proportions of rubber-associated deforestation and degradation were observed between 2001 and 2010.
3.2 Factors associated with deforestation and degradation
The area of tree cover loss that occurred from 2000 to 2017 showed significant associations with several key variables (Supporting Information S1: Table S5). Plot size, minutes to city, and the interaction between these two variables were found to positively correlate with tree cover loss within plantation buffers (Supporting Information S1: Figure S2a). The tree cover loss adjacent to plantations, within 500 m buffer areas, was positively associated with plot size, minutes to city, and distance to PAs (Figure 3, Supporting Information S1: Table S5). For instance, when the distance to PA increases by 1 km, the tree cover loss observed increases by 0.1 ha (unstandardized estimates). The variance explained by fixed effects was 3.5%, while the variance explained by random effects was negligible (marginal R2 0.037 and conditional R2 0.037, Supporting Information S1: Figure S3). Within the extended 1 km buffer area, tree cover loss showed a positive relationship with plot size, distance to PAs, and minutes to city (Figure 3, Supporting Information S1: Figure S2b–d, Table S5). Furthermore, a negative relationship was observed between tree cover loss and a lack of land titles (Figure 3).
The analysis of deforestation of undisturbed tropical moist forests within plantation buffers showed poor correlations with the independent variables (Supporting Information S1: Figure S4a, Table S6). Similar relationships were observed for deforestation within extended 1 km buffer areas (Supporting Information S1: Figure S5a,b, Table S6). Therefore, these models were not considered further. Here, the highest random intercept was observed for the West Tanjung Jabung region (Supporting Information S1: Figure S6), with random effects explaining 17.4% of the variance (marginal R2, 0.001 and conditional R2, 0.175).
Forest degradation within plantation boundaries showed positive associations with plot size and minutes to city (Supporting Information S1: Figure S4b, Table S7). In 500 m buffer areas, it was positively related to plot size, minutes to city, and distance to PAs (Figure 4c, Supporting Information S1: Table S7). Fixed effects explained 6.7% of the variance, while random effects explained 13%, with the highest random intercept observed for the Batang Hari region (marginal R2, 0.067 and conditional R2, 0.197) (Supporting Information S1: Figure S6d–f). Similar relationships were observed for the degradation of forests within extended 1 km buffers (Figure 4d, Supporting Information S1: Table S7). For instance, forest degradation observed increases by 0.7 ha when travel time to cities increases by 1 min (unstandardized estimates).
3.3 The effect of rubber price on annual deforestation and degradation
Time series data for annual deforestation and degradation from 1991 to 2018 followed similar trends within and around rubber plantations (Supporting Information S1: Figure S7). There were instances when annual deforestation and degradation showed the same pattern as the rubber price of the previous year (Supporting Information S1: Figure S7c).
The annual deforestation within 500 m buffers was positively associated with the interaction between rubber price and plot size (Figure 5a, Supporting Information S1: Figure S8a). Minutes to city, plot size, and lack of land titles also showed positive relationships (Figure 5a, Supporting Information S1: Table S8). Similar relationships were observed for 1 km buffer areas as well (Figure 5b, Supporting Information S1: Table S8). In the model for the 1 km buffer area, fixed effects explained 4.5% of the variance, while random effects explained 44% (marginal R2, 0.045 and conditional R2, 0.486, Supporting Information S1: Figure S9).
Annual degradation within 500 m buffer areas was positively associated with rubber price (i.e., higher rubber prices were associated with increased degradation and lower rubber prices with reduced degradation), minutes to city, and lack of land titles (Figure 5c, Supporting Information S1: Table S9). In 1 km buffer areas, annual degradation showed a positive relationship with plot size, minutes to city, and lack of land titles (Figure 5d, Supporting Information S1: Table S9). Fixed effects explained 21% of the variance, while random effects explained 45% in the best model for the 1 km buffer area (marginal R2, 0.21 and conditional R2, 0.66, Supporting Information S1: Figure S9). Model diagnostic plots were used for visual inspection of the top models explaining tree cover loss, deforestation, and degradation in the 1 km buffer areas (Supporting Information S1: Figure S10).
4 DISCUSSION
Identifying plantation-level socio-economic drivers of deforestation and degradation poses a challenge but is important for ensuring a deforestation-free supply chain within the rubber industry. This study's findings reveal that in Jambi, rubber-associated deforestation and degradation have significantly impacted natural ecosystems over the last 30 years. Spatial variables such as travel time to cities and distance to the nearest PA have shown positive relationships with increased forest loss and degradation. Additionally, high rubber prices in the preceding year were positively associated with increased deforestation and degradation around rubber plantations, especially for larger plantations, whereas lower prices correlated with reduced deforestation and degradation. These outcomes indicate the effects of global market dynamics and socio-economic characteristics of rubber plantations, underscoring the need for rigorous transboundary collaborative efforts towards sustainability.
4.1 Deforestation and degradation around rubber plantations
This study suggests that a quarter of the total area within buffers around the considered rubber plantations has undergone tree cover loss, with close to 60% of this loss occurring between 2001 and 2010. Indonesia has seen significant deforestation, with the country experiencing the highest increase in forest loss globally from 2001 to 2012 (Hansen et al., 2013). It is important to note, however, that the replanting of rubber also involves tree cover loss, which could be mistaken for deforestation (loss of forests) in global remote sensing datasets. Jambi Province is a prominent rubber-producing region in Indonesia (Tata, Rasnovi et al., 2008), and local studies corroborate high deforestation levels. Since 2005, the conversion of Indonesian forests to industrial plantations has increased rapidly (Gaveau et al., 2016). Specifically, Sumatra, where Jambi Province (the location of this study) is situated (Guillaume et al., 2016), has seen extensive deforestation, with less than 10% of forest cover remaining in 2012 (Margono et al., 2014). One of the main drivers of Sumatran deforestation has been rubber agriculture (Laumonier et al., 2010), where smallholder farmers account for the majority of rubber plantations (Guillaume et al., 2016). Therefore, the tree cover loss detected in this study likely represents the ongoing reduction of forest cover in Jambi. However, part of this loss may be attributed to the cyclical nature of rubber cultivation, where replanting is common (Tata, Rasnovi et al., 2008). Since the multispectral reflectance of mature rubber closely resembles that of tropical forests (Li & Fox, 2012), it is possible that some rubber tree loss may also have been inaccurately recorded as forest loss.
Analysis from this study detected that 15% of the area around rubber plantations (within a 1 km buffer) experienced deforestation (of tropical moist forest), while 34% experienced forest degradation (of tropical moist forest) between 1991 and 2018. These deforestation estimates refer to the loss of undisturbed tropical moist forests, particularly due to the expansion of tree plantations (Vancutsem et al., 2021). The relatively low deforestation percentage is not surprising given that by 1985, 50% of Sumatra's natural forests had already been converted to other land uses (Laumonier et al., 2010). During the period from 1991 to 2018, for which deforestation records were available for the present analysis, the extent of undisturbed tropical forests available was already low (Tata, Noordwijk et al., 2008). This was particularly true for Jambi Province, which underwent considerable transformations over this period (Guillaume et al., 2016). The higher percentage of forest degradation that occurred around rubber plantations is particularly noteworthy. This is important, as clearing of primary forests often follows initial degradation events (Margono et al., 2014), indicating that these areas may be at risk of deforestation in the near future.
4.2 Factors affecting deforestation and degradation
The size of the plantation (plot size) was positively associated with tree cover loss, deforestation, and forest degradation in the areas surrounding rubber plantations. Within the buffers created for plantation size, the deforestation of undisturbed tropical moist forests may indicate forest loss directly resulting from the establishment of these plantations. In plantation-adjacent buffer areas, this may indicate ongoing or potential expansion. The linkage between plot size and forest loss around plantations suggests that larger plantations have a greater tendency to expand into nearby forests. In Asia and Africa, smaller agricultural field sizes are generally associated with higher rates of deforestation (Dang et al., 2019), potentially due to the prevalence of smallholder plantations in these regions. Rubber plantations considered in this study were also smallholder-owned (sizes ranging from 0.1 to 5 ha). Among these, larger plantations appeared more likely to initiate further expansion (higher income was associated with plantations that were larger in size, according to the data), given that smaller plantations could be constrained due to limited access to credit and capital (Meyfroidt et al., 2014).
Accessibility, measured as travel time to cities (minutes to city), was positively linked to both tree cover loss and forest degradation around rubber plantations, particularly in more remote areas where forest cover is still extensive. However, this contrasts with previous studies suggesting that the lack of accessibility could deter deforestation activities (Junquera et al., 2020), while enhanced accessibility through road systems drives deforestation and forest degradation in regions like Sumatra (Poor et al., 2019). In the rubber plantations assessed, travel time interacted with plantation size, suggesting that larger plantations, which are more remote and have less access to urban centres, experienced greater forest loss. Since the remaining forest cover in Jambi is relatively low and competition for land, especially with oil palm, is intense, the spatial contagion effects of agricultural expansion may push plantation expansion into more remote areas (Lim et al., 2017). In addition, the nature of rubber production, involving semi-processed rubber sheets or slabs, makes rubber relatively easy to transport (Promme et al., 2017) and less dependent on immediate access to transportation systems. This factor may lessen the imperative for close proximity to infrastructure that is more critical for other crops like oil palm.
Tree cover loss and degradation in plantation-adjacent buffer areas were positively related to the distance to the nearest PAs, suggesting that losses predominantly occurred away from PAs where deforestation activities are typically restricted (Bebber & Butt, 2017). Despite evidence that over 40% of forest clearings in Indonesia occur through prohibited or illegal deforestation (Margono et al., 2014) and instances where PAs in Cambodia have been repurposed for rubber cultivation (Warren-Thomas et al., 2015), the findings of this study suggest that PAs in Jambi appear to have been effective in mitigating such losses. Similarly, PAs in Laos have demonstrated success in curbing rubber plantation expansion (Junquera et al., 2020). Given the limited availability of PAs in Jambi province, caution is required when extrapolating these findings on PA effectiveness for other crops and provinces in Indonesia.
4.3 The effect of rubber price on deforestation and degradation
Rubber price from the preceding year was positively related to annual deforestation and degradation in buffer areas, both directly and through interactions with rubber plantation size. This indicates that larger plantations have a higher propensity to drive deforestation when economic conditions are favourable. Similar correlations between forest loss and global rubber prices have also been observed in Cambodia (Grogan et al., 2019) and Laos (Junquera et al., 2020), two frontiers of recent rubber expansion. In turn, larger plantations, possibly due to better financial capacity and resources, may have a greater ability to afford the costs associated with expansion.
This study has a few caveats. Firstly, it lacked data on the actual boundaries of the plantations, relying instead on buffer areas to approximate plot size. While this approach accurately represents plantation size, it may result in less precise estimates of the area affected by deforestation or degradation. However, given the smallholder nature of these plantations, the difference may not be high. Secondly, the absence of repeated annual socio-economic survey data reduced the study's ability to assess the impacts of standards and certification schemes on rubber expansion. To more accurately determine the effects of these schemes and how they intersect with agricultural intensification and fluctuating prices, there is a need for organizations to introduce annual data collection and analysis. Lastly, although the findings presented here represent direct deforestation and can offer insights into potential areas of future rubber expansion, they should be carefully interpreted when informing land use regulations and governance decisions.
5 CONCLUSIONS
Sustainability in rubber agriculture can only be achieved when plantations are separated from deforestation events. The analysis in this study suggests that from smallholder-owned plantations, the ones that are larger in size, are more likely to cause deforestation due to expansion. Moreover, it was observed that rubber-associated tree cover loss and degradation are predominantly associated with rubber cultivation in relatively remote areas located away from PAs. Recognizing these spatial patterns is important for developing sustainable land-use strategies in the rubber industry. In particular, these patterns advocate for the prioritization of standards and certification schemes in relatively larger, remotely located plantations first, as these may have a better economic capacity to get involved. Global rubber prices could also incentivize plantation-associated deforestation, increasing risks to remaining natural habitats, especially in countries with development-oriented land governance policies. This points to the need for a dynamic price premium for sustainable producers, discouraging deforestation practices. Global organizations pursuing rubber sustainability should involve local organizations, buyers, producers, and governments in introducing pricing schemes that reflect deforestation risk, encouraging adherence among farmers. In conclusion, these findings help identify high-risk plantations, underscoring the need to involve such plantations in more effective management and conservation schemes. Technical support and awareness programmes could help mitigate some of this risk and assist the rubber industry in making informed decisions in creating deforestation-free supply chains, a critical step in protecting biodiversity and slowing climate change.
AUTHOR CONTRIBUTIONS
H. Manjari Jayathilake: Conceptualization; data curation; formal analysis; methodology; visualization; writing—original draft; writing—review and editing. Cheng Zhi Wei: Methodology; project administration; resources; writing—review and editing. Gerald Tan: Methodology; project administration; resources. Côme de la Porte: Methodology; project administration; resources; writing—review and editing. L. Roman Carrasco: Conceptualization; methodology; supervision; writing—review and editing.
ACKNOWLEDGMENTS
We wish to acknowledge the Industry Relevant PhD Programme (NUS-IRP) scholarship of the National University of Singapore, awarded to H. Manjari Jayathilake for funding this research. We would also like to thank the collaborative support received from Agridence Rubber and Rubberway Pte. Ltd. We would also like to recognize the assistance given by rubber smallholder farmers by participating in the surveys. The study is funded by the Faculty of Science, National University of Singapore, and NUS IRP Scholarship.
6 CONFLICT OF INTEREST STATEMENT
H. Manjari Jayathilake reports that rubber farmer surveys used for the study were conducted by Agridence Rubber Pte. Ltd. and Rubberway Pte. Ltd. Cheng Zhi Wei and Gerald Tan report a relationship with Agridence Rubber Pte. Ltd. that includes: employment. Come de la Porte reports a relationship with Rubberway Pte. Ltd. that includes: employment.
Open Research
DATA AVAILABILITY STATEMENT
The authors do not have permission to share survey data.
