High-resolution maps show that rubber causes substantial deforestation

The study addresses how much deforestation in Southeast Asia is attributable to the expansion of natural rubber, a major commodity produced largely by smallholders and challenging to detect via remote sensing due to spectral similarity with forests. Existing global assessments for commodities often rely on model-based approaches (for example, land balance models) that are not spatially explicit and carry large uncertainties. For rubber, published estimates of deforestation impacts diverge more than fivefold, contributing to debate about whether rubber should be included in deforestation regulations (e.g., EU Deforestation Regulation, UK Environment Act secondary legislation). There is a pressing need for spatially explicit, high-resolution evidence to calibrate policy, guide targeted interventions, and enable monitoring of compliance with zero-deforestation commitments. The authors aim to provide such evidence by directly mapping rubber extent and associated deforestation across Southeast Asia, which accounts for over 90% of global natural rubber production.

Prior work has mapped deforestation drivers for a few commodities (notably oil palm and soy), but spatially explicit maps for many others, including rubber, have been lacking. Rubber’s impact on forests has been estimated via models combining national crop expansion statistics with generic tree cover loss, which do not spatially link crop expansion to forest loss and may misattribute drivers. The widely used land balance dataset has suggested rubber is a minor deforestation driver compared to soy and palm oil, influencing policy focus; however, these figures are uncertain and can under- or overestimate commodity roles due to statistical inconsistencies and the inability to track spatial displacement. Other studies using spatial data report much higher rubber-related deforestation: WRI estimated 2.1 million ha (2001–2015) across seven countries; Hurni & Fox estimated >5 million ha (2003–2014) in continental Southeast Asia; and several subnational or country studies have documented strong links between rubber expansion and forest loss (e.g., Cambodia). The literature also notes rubber’s smallholder dominance, phenological characteristics complicating detection, and the potential environmental and biodiversity impacts of expansion.

• Study area and scope: Southeast Asia, covering the vast majority (>90%) of global natural rubber production; China included for main production areas (Xishuangbanna and Hainan).
• Rubber extent mapping (reference year 2021): Used Sentinel-2 imagery at 10 m resolution and Google Earth Engine. Mapped mature rubber plantations by exploiting rubber’s distinct phenological signature: synchronized leaf fall and regrowth windows allow discrimination from evergreen and deciduous forests. Due to frequent cloud cover, multi-year image composites were used. The algorithm was run separately for mainland vs. insular Southeast Asia to account for climatic heterogeneity affecting phenology. ESA WorldCover 10 m 2020 (tree cover ≥10%) was used as a mask to constrain mapping to tree-covered areas. A disturbance mask was developed to exclude pixels with disturbances mimicking rubber phenology within the defoliation/refoliation window. Only mature rubber (approximately >5 years old) was mapped.
• Deforestation detection for mapped rubber: For all pixels mapped as rubber in 2021, deforestation timing was assessed using Landsat time-series and the LandTrendr spectral-temporal segmentation algorithm. The largest breakpoint in normalized burn ratio (NBR) was used to detect sudden transitions from tree cover to bare/burnt ground. Only the first major breakpoint since the early 1990s was counted, to minimize counting plantation rotations. Pixels were considered deforested only if pre-break NBR exceeded 0.6 to ensure prior dense tree cover; relaxing this threshold yields higher deforestation totals. Deforestation dates were aggregated into 1993–2000 and 2001–2016.
• Accuracy assessment and area estimation: Mapping accuracy for rubber extent and deforestation timing was evaluated using reference data and sample-based area estimation following good practices (Olofsson et al. 2014). Reported metrics include overall accuracy (OA), user’s and producer’s accuracy, and 95% confidence intervals. Sample-based area estimates were used to quantify uncertainty and to provide alternative area figures (asterisked) where applicable.
• Data products and visualization: Resulting 10 m rubber extent and 30 m deforestation maps were aggregated to 500 m for visualization; original-resolution maps are made available via an online app. Country-level summaries and overlays with Key Biodiversity Areas (KBAs) were produced using the BirdLife KBA database.

• Rubber extent (2021): 14.2 million ha of mature rubber plantations mapped in Southeast Asia; over 70% located in Indonesia, Thailand, and Vietnam. Additional substantial areas in China (Xishuangbanna and Hainan), Malaysia, Myanmar, Cambodia, and Laos. The mapped estimate is conservative; sample-based estimates suggest potentially larger total area. Country-level mapped areas (ha): Indonesia 4,745,921; Thailand 3,744,139; Vietnam 1,606,594; China 1,097,213; Malaysia 985,335; Myanmar 779,717; Cambodia 618,135; Laos 574,035. Rubber within KBAs totals ~1,072,800 ha (about 8% of mapped rubber area), with Cambodia having 19% of its rubber in KBAs.
• Accuracy: Rubber mapping OA = 0.95 ± 0.02 (95% CI). Mainland Southeast Asia OA > 0.99 ± 0.01; insular Southeast Asia OA = 0.85 ± 0.06 due to weaker seasonality and cloud cover. User’s accuracy ~0.99; producer’s accuracy ~0.95 (point-based) but ~0.57 under area-weighted estimation because rubber occupies a small fraction of tree cover and omission errors have larger area influence. Deforestation date classification OA = 0.85 ± 0.09.
• Deforestation attributable to rubber: Total of ~4.1 million ha of forest cleared for areas that were rubber in 2021 during 1993–2016; nearly three-quarters (≈3 million ha) occurred since 2001. Sample-based estimate since 2000: 2.5 ± 0.35 million ha (95% CI). Using a relaxed NBR threshold (removing the 0.6 pre-break threshold) increases estimated loss to almost 6 million ha.
• Annualized rates (Table 2; 2001–2016, Southeast Asia): 186 thousand ha yr−1 from mapped data; sample-based estimate 156 ± 22 thousand ha yr−1. Country per-year contributions: Indonesia 66, Thailand 39, Malaysia 20, Cambodia 15 (thousand ha yr−1).
• Country patterns: Indonesia, Thailand, and Malaysia together account for over two-thirds of rubber-related deforestation since 2001. Cambodia experienced substantial recent deforestation linked to rubber, with >40% of rubber plantations associated with deforestation (2001–2016) and 19% of rubber area located within KBAs.
• Underestimation in policy-used datasets: Model-based land balance estimates used in EU/G7/UK policy place rubber-related deforestation at <0.7 million ha (2005–2017) or <1 million ha (2005–2018) across ~135 countries, far below this study’s conservative, spatially explicit estimates for Southeast Asia alone. Even the lower 95% CI of this study exceeds those model-based values, indicating policy-relevant underestimation by factors of two to three or more, with extreme discrepancies for certain countries (e.g., several hundredfold for Cambodia in earlier versions).

The results demonstrate that natural rubber has driven several million hectares of forest loss in Southeast Asia since the early 1990s, contradicting prevailing model-based assessments that have minimized rubber’s role relative to other commodities. By directly linking mapped rubber plantations to prior forest loss using satellite time series and a conservative first-break approach, the study provides spatially explicit evidence of the scale, timing, and geography of rubber-related deforestation. This addresses the core research need for accurate, commodity-specific deforestation mapping to inform policy, compliance monitoring, and supply-chain risk assessments. The findings imply that earlier models underestimate rubber’s impact, likely due to limitations of national statistics, displacement effects (e.g., oil palm moving into traditional rubber areas with compensatory rubber expansion elsewhere), and unattributed deforestation in land balance frameworks. Policy implications include strong support for the inclusion of rubber in deforestation regulations (e.g., EU Deforestation Regulation) and the necessity for high-resolution monitoring to enforce legal and voluntary zero-deforestation commitments. Comparisons with other commodities suggest rubber’s deforestation footprint exceeds that of coffee and cocoa (contrary to some previous assumptions) but remains below oil palm, albeit by a smaller factor (approximately 2.5–4) than previously claimed. Challenges remain around supply-chain traceability and ensuring smallholders—who produce about 85% of natural rubber—are not disproportionately burdened, but the concentration of rubber consumption in a few tire manufacturers and existing initiatives (GPSNR, FSC group certification) present opportunities for improved governance and support.

This study delivers the first Southeast Asia-wide, high-resolution maps of rubber plantations and associated deforestation back to the early 1990s, revealing at least 4.1 million ha of rubber-driven forest loss—several times higher than model-based estimates commonly used in policymaking. With more than 1 million ha of rubber in Key Biodiversity Areas and continued forest conversion after 2000, rubber’s environmental impacts are substantial and ongoing. The work underscores the necessity of using spatially explicit, remotely sensed evidence to guide regulation, corporate commitments, and conservation priorities. Future research should: expand coverage beyond Southeast Asia to regions with recent rubber growth (e.g., West and Central Africa); integrate plantation age, rotation, and land-use history to refine attribution; improve detection in cloudy equatorial zones and for younger, agroforestry, or diseased stands; and link spatial deforestation patterns to supply chains to enable traceable, verifiable compliance and targeted interventions that safeguard smallholder livelihoods.

• Conservative scope and baseline: Deforestation was assessed only for areas mapped as rubber in 2021; earlier conversions to rubber that were subsequently replaced by other land uses are missed. Using the first detectable deforestation event since ≈1993 minimizes counting plantation rotations but excludes rotations after that first event. The NBR >0.6 threshold likely underestimates deforestation in drier or fire-prone areas; removing it increases totals substantially.
• Detection constraints: Only mature rubber (>~5 years) was mapped; younger plantations, agroforestry/jungle rubber with subdominant canopy, and diseased or atypically defoliating stands are prone to omission. ESA WorldCover tree cover mask may exclude some rubber if tree cover was misclassified.
• Regional heterogeneity and cloud cover: Insular Southeast Asia has weak seasonality and persistent clouds, reducing phenological signal and data availability; omission errors and uncertainty are higher there. About 7–10% of Indonesia/Malaysia lacked clear Sentinel-2 images in parts of the period.
• Attribution ambiguity: Some deforestation detected before rubber establishment may have initially been for other land uses (e.g., industrial forestry) before conversion to rubber, particularly in areas of recent expansion at climatic margins.
• Accuracy and area estimation: Although overall accuracies are high, area-weighted producer’s accuracy for rubber is lower due to class imbalance; sample-based estimates provide uncertainty bounds but residual errors remain.