Introduction

Forest is being converted to Hevea brasiliensis rubber plantations across Southeast Asia, resulting in the loss of forest carbon stocks and substantial declines in biodiversity1,2,3. Eighty five per cent of global rubber area occurs in Southeast Asia, where expansion has driven northward into Cambodia, China, Laos, Myanmar and Vietnam (hereafter termed mainland Southeast Asia, but also known as the Indo-Burma biodiversity hotspot), replacing forest and traditional swidden cultivation1,2. Despite its massive extent (8.6 million ha in Southeast Asia in 2014, equivalent to 67% of oil palm extent4) and comparable negative consequences for biodiversity and ecosystem services5, conversion of forest to rubber monoculture has not faced the same public scrutiny as oil palm. Here, we analyse carbon outcomes and opportunity costs of forgoing forest conversion to rubber, including profits from timber extraction, and ask whether permitting selective logging could reduce these opportunity costs to improve the likelihood of success for forest carbon finance.

Where climatic conditions are suitable for both oil palm and rubber, they can generate similar profits per unit land area, but oil palm provides better returns if labour supply is restricted5. However, rubber can tolerate a wider range of climatic conditions and soil types, permitting its expansion into mainland Southeast Asia2, although reduced yields and tree mortality are reported from many northern parts of its range1. Recent expansion of rubber has mostly occurred in areas unsuitable for oil palm1,6.

Demand for natural rubber continues to grow, predominantly driven by the tyre industry, and plantations are predicted to expand by 4.3–8.5 million ha within a decade2. This expansion not only threatens forest carbon stocks, but also has serious implications for biodiversity conservation. The forests of mainland Southeast Asia are globally unique ecosystems, supporting numerous threatened and endemic animal, bird, invertebrate and plant species7,8, including exceptionally valuable luxury timbers (eg, rosewoods, Dalbergia spp.).

Rubber prices are currently relatively low (Supplementary Fig. 1), offering a lull in expansion, and an opportunity to develop strategies for future planting that minimise negative outcomes for climate and forests. Stemming rubber expansion onto biodiversity-rich forest could reduce carbon emissions and achieve conservation gains simultaneously, making efficient use of limited funds9,10. However, the effectiveness of any forest carbon finance scheme will depend on the number of carbon credits generated, the perceived costs of conserving forest and the price offered for carbon credits.

Costs of forest conservation can be broken down into practical costs of conservation activities (transaction and implementation costs) and the costs of forgone economic activity (opportunity costs); these are borne at local, sub-national or national scales. In contrast, the economic costs of climate change are largely borne globally. Estimates of damage costs caused by carbon emissions are termed the “social cost of carbon”, and are used in the design of policies to regulate carbon emissions11. Such estimates have been made using various methodologies, and vary by an order of magnitude11,12,13,14,15,16, but perhaps the most policy-relevant is that developed for the government of the United States ($36 per tCO2 in 201511,13,17), which we use to discuss the context of our results.

We analysed the carbon outcomes and opportunity costs of forgoing forest conversion to rubber, including profits from timber extraction, and asked whether permitting selective logging could reduce opportunity costs and improve the likelihood of success for forest carbon finance. We modelled scenarios of protecting either intact forest (“No timber logged”), or forest degraded by permitting selective logging (felling only trees ≥40 cm DBH at three intensity levels, depending on timber royalty/value classes: (1) “Luxury logged”; (2) “Luxury, I, II logged”; (3) “All trees logged”), from subsequent conversion to rubber (Fig. 1; Supplementary Note 1). We calculated rubber profits based on typical monocultural plantation systems, containing high-yielding clonal varieties of rubber planted at densities of 400–550 stems per ha18. Such systems are ubiquitous across mainland Southeast Asia, within both smallholdings and larger estates1,19,20,24.

Using cost benefit analysis and net present value, we find that the carbon prices needed to match the costs of protecting intact or selectively-logged forests in mainland Southeast Asia from conversion to rubber are $29.86–$37.48 per tCO2 for dense forests, or $30.93–$51.11 per tCO2 for open forests. Rubber forms a far greater proportion of opportunity costs than timber, and breakeven carbon prices are much lower under scenarios of logging alone. Our calculated prices are higher than those currently paid on carbon markets (~$5–$13 per tCO2), or through carbon funds (~$5 per tCO2), but are close to, or below, policy-relevant estimates of the social cost of carbon ($36 per tCO2). The mismatch between predicted social costs of carbon at the global scale, carbon market prices and the local costs of avoiding carbon emissions from deforestation, poses a serious challenge for forest carbon finance in tackling global climate change. Given our findings, additional strategies beyond carbon finance will be needed to prevent forest conversion to rubber. However, risk aversion and non-economic values of forest could reduce the carbon payments needed to leverage protection, relative to our calculations.

Results

Timber volume and carbon stocks

Harvestable wood volume of all tree species, assuming a minimum harvestable DBH of 40 cm (≥10 cm for luxury timber; Supplementary Note 1), was 49.4 ± 0.5 m3 per ha in dense forest, but just 13.6 ± 0.3 m3 per ha in open forest (mean ± SE; Table 1). Timber of royalty classes I and II accounted for 54% of volume in dense forest and 69% in open forest. Luxury timber was rare, contributing only 1.1 ± 0.0 m3 per ha in dense forest (2%) and 1.3 ± 0.1 m3 per ha in open forest (10%).

Table 1 Mean carbon stock and wood volume held in harvestable stems of each timber royalty class in dense and open forests
Full size table

Mean forest carbon stocks, measured as combined above-ground and below-ground biomass (AGB and BGB) of all stems ≥10 cm DBH, were 194 ± 1.2 tC per ha in dense forest (123–284 tC per ha among individual landscapes) and 104 ± 0.8 tC per ha in open forest (60–157 tC per ha among landscapes; Table 1, Supplementary Fig. 3). Lower carbon stocks in open forest reflected both a greater proportion of smaller stems (Supplementary Fig. 4) and lower stem density (mean across inventories 213–311 per ha in open; 415–589 per ha in dense; Supplementary Data 1), despite similar wood density (weighted mean 0.713 g cm−3 in open; 0.630 g cm−3 in dense).

Forest carbon stock changed minimally following selective logging of luxury timber (“Luxury logged” scenario), reducing by 1% in dense forest and 2% in open. Additional logging of classes I and II (“Luxury, I, II logged”) reduced carbon stocks by 20% in dense forest (to 175 ± 1.1 tC per ha) and 26% in open forest (to 78 ± 0.7 tC per ha; Table 1; Fig. 2). Even the removal of all trees ≥40 cm DBH (“All trees logged”), left 60% (dense), and 66% (open) of original carbon stock. Therefore, substantial forest carbon is retained before conversion to rubber, even after logging of valuable timber or removing all large trees.

Fig. 2
figure 2

Effects of selective logging on dense and open forest carbon stocks. Carbon stock (tC per ha) of a dense and b open forest in initial state (“No timber logged”) and under scenarios of selective or complete logging (following Fig. 1), and c time-averaged carbon stocks (taCs) of rubber plantations in Southeast Asia3. Central bar shows median, box shows upper and lower quartiles, whiskers extend to 1.5× the inter-quartile range, and outliers are presented as dots

Full size image

Rubber plantations are considered forest cover in FAO Forest Resources Assessments25, while the USAID LEAF Atlas maps rubber-dominated landscapes as forest (such as Southern Thailand), but protected open forests in Eastern Cambodia as non-forest26. However, we find that even assuming high post-deforestation time-averaged carbon stocks (taCs) for rubber of 52.5 tC per ha (from Cambodia, Thailand and Indonesia3, likely greater than will be achieved in seasonal open forest environments), conversion of intact forest to rubber would still generate net losses of 141.5 ± 1.2 tC per ha in dense forest and 51.5 ± 0.8 tC per ha in open forest (Fig. 2). Even conversion of degraded logged open forest would generate net emissions, as well as biodiversity loss. Additionally, although we do not account for changes in soil organic carbon (SOC) because the Intergovernmental Panel on Climate Change (IPCC) tier 1 carbon outcome calculation method assumes no SOC change with conversion to perennial tree crops, conversion of lowland forest to rubber plantations does generate SOC emissions27, which, if included, would increase net emissions.

Logging and conversion to rubber

Total opportunity costs of protecting intact forest (“No timber logged + forgone rubber” scenario), calculated as the profit from a single offtake of all commercial timber (Supplementary Table 2) plus the 25-year net present value (NPV; 10% discount rate) from subsequent rubber plantations (Supplementary Table 3), were $16,841 (median, interquartile range $12,118–$21,397) per ha in dense forest and $7,674 ($4,581–$11,250) per ha in open forest (Fig. 3). Intervening after removal of luxury timber (“Luxury timber logged + forgone rubber” scenario) reduced opportunity costs of logging by 38% in dense forest and 56% in open; however, total opportunity costs (including rubber) were only reduced to $15,097 ($10,738–$19,390) per ha and $5,956 ($3,341–$8,663) per ha, respectively. After all valuable timber had been logged out (“Luxury, I, II, logged” or “All trees logged”), the opportunity cost of rubber alone was $12,570 ($8,436–$16,698) per ha in dense forest, and $5,089 ($2,532-$7,764) per ha in open forest. Rubber therefore forms the majority of total opportunity costs (75% of median in dense forest, 66% in open forest). The substantially lower NPV of rubber in open forest arises from the delay of tap** from 6 years to 10 years after planting, due to slower tree growth in drier conditions. Using a discount rate of 8% to allow comparison, our estimate of rubber NPV in dense forest areas ($16,533 per ha, $11,403–$21,732 per ha) is similar to estimates from lowland ** indicators of economic value and biodiversity loss for rubber plantations in ** indicators of economic value and biodiversity loss for rubber plantations in ** a social cost of carbon for US regulatory analysis: a methodology and interpretation. Rev. Environ. Econ. Policy 7, 23–46 (2013)." href="/article/10.1038/s41467-018-03287-9#ref-CR13" id="ref-link-section-d35698549e1615">13,17. In contrast, a 2011 review calculated a mean of $177 per tCO2 (SD $293 per tCO2)16 and subsequent analyses gave estimates of $103–$138 per tCO214 or $220 per tCO215. Considering these higher estimates, preventing forest conversion to rubber represents cost-effective action to reduce emissions, even for intact open forests. However, the social cost of carbon is based on the premise of a global cost-benefit exercise, comparing the economic cost of carbon emissions to the cost of controlling emissions12. A key issue is that, even where these cost estimates align, in the absence of a global carbon pricing mechanism the path from a calculated social cost of carbon to compensation for the local-scale costs of avoided emissions is unclear.

There are indications that carbon market prices could be raised much closer to the social cost of carbon in some sectors to meet climate change targets: for instance, if recent proposals to set a price floor of €20–€30 (~$23–$34) per tCO2 on the EU Emissions Trading Scheme come to fruition, a new global benchmark could be set32. Recent research investigating the incentives required to decarbonise the energy sector to meet the Paris Agreement targets (i.e., limiting global temperature rise to below 2 oC) similarly found the need to introduce carbon prices to the industry and power sectors of all countries, with prices of $20 per tCO2 by 2020 and $120 per tCO2 by 2030 in OECD countries, $10–$90 per tCO2 in major emerging economies, and $5–$30 per tCO2 elsewhere33. Carbon finance for dense forests could then be nearly as lucrative as conversion to rubber in the near-term, while in the longer-term, prices could be high enough to defend even open forests, notwithstanding the risk of unintended market feedbacks arising from restricted supply (see below)34. We also note that even if full opportunity costs of conservation are not met, smaller financial incentives to conserve forest may be an attractive option if there is existing social or political pressure to conserve, or where non-market values are recognised.

Open forests tend to be drier with poorer sandier soils35 and thus may have lower agricultural potential. We incorporated the effect of delayed maturation of rubber under drier open forest conditions, but robust data on rubber yields in such conditions are notably lacking36. Poorer growth could also result in lower rubber carbon stocks1,3, reducing the breakeven price needed to match rubber profits in open forest areas, and improving the prospects for protection using carbon finance. Robust data on the relative yield and carbon stocks of rubber plantations established on former open or dense forests are urgently required3. However, where land proves unfavourable for rubber, the high potential NPV of cassava, known to grow successfully in less favourable environments, at least in the short term (Supplementary Table 5), reinforces the importance of other mechanisms to curb forest conversion.

Although rapid expansion of cassava for animal feed contributed to serious deforestation of northern Thailand in the late 20th century37, and in Cambodia cassava expanded at a similar rate to rubber within the last decade (see Methods), in the longer-term, cassava extent has been relatively stable across Asia (3.9 million ha in 1981, 4.1 million ha in 2014), although yields nearly doubled (11.9 t per ha per year in 1981, 21.9 t per ha per year in 201438). This contrasts strongly with Asian rubber extent, which almost doubled from 5.1 million ha in 1981 to 9.8 million ha in 2014, despite similar yield increases (0.7 t per ha per year in 1981, 1.2 t per ha per year in 2014)38. Thus, although we estimate median NPV to be greater for cassava than rubber, demand does not currently appear to be driving expansion of cassava onto forest to the same degree as rubber in Asia. In addition, Cambodian cassava growers currently face problems accessing markets and supply chains, and depletion of soil quality and declining yields with repeated cultivation are known issues39. We acknowledge that this may reduce the cassava NPV realised by farmers relative to our estimates, although we have been unable to quantify this (Supplementary Table 5). Nevertheless, use as a biofuel feedstock may further increase cassava demand within Asia and at the global level it is rapidly expanding40. Together with our high estimated NPV for cassava this suggests that, although not currently a major regional driver of deforestation in Asia, the importance of cassava may increase.

We have used a non-spatial approach to estimate breakeven carbon prices for dense and open forest in mainland Southeast Asia. We show that rubber NPV is likely substantially lower in open than in dense forest areas, and were able to incorporate geographic variability in timber and carbon stocks, which influenced breakeven carbon prices. Additional spatial variation in rubber NPV and carbon breakeven prices will be generated by distance to markets, regional farm gate prices, yields (affected by soil type and quality, topography, elevation, water availability, planting material, etc.), and labour costs, amongst others. However, spatially explicit data on potential rubber NPV are not currently available for most of mainland Southeast Asia, including Cambodia, not least because recently established plantations are not yet productive. This precluded meaningful marginal cost curve analyses to quantify the additional area of rubber expansion that could potentially be avoided with incremental increases in carbon price. Such spatial analysis has been conducted locally for ** indicators of economic value and biodiversity loss for rubber plantations in ** a national-level program64. A number of pilot REDD+ demonstration projects are underway, seeking funding from voluntary carbon markets65.

Modelling opportunity costs and carbon breakeven prices

Opportunity costs were defined as forgone direct profits from logging, the net present value (NPV) of rubber in large plantations or of other cash-crop agriculture (cassava, cashew, and sugarcane, in large plantations or smallholdings), and cash income from collecting dipterocarp tree resin, which is forgone once dipterocarp trees are felled (Supplementary Note 2). This traditional livelihood activity directly conflicts with logging, because resin-producing species also have valuable timber66, being listed as class II species (Supplementary Data 1). Total opportunity cost was thus based on forgone profits from logging and/or rubber, offset by lost resin revenue where class I and II trees are logged out. We do not estimate other non-timber forest product benefits (Supplementary Note 2).

We did not distinguish between legal and illegal activities, as we wished to understand the underlying economic drivers of forest conversion, while acknowledging that when designing actual incentive mechanisms there may be good reasons for treating legal revenue streams differently from illegal ones, so as to avoid indirectly rewarding illegal behaviour. However, timber revenue was calculated as a farm-gate price and we did not consider legal downstream benefits that accrue to the wider economy, nor those benefits of doubtful legality that accrue mainly to non-local actors, including agro-industrial companies and elite logging “tycoons”48,53,62, such as the sale of high value timber on the international market (Supplementary Table 8, although export of logs and most sawn timber is illegal; Supplementary Note 1). These exclusions are likely to lead to an under-estimate of opportunity cost, but are appropriate given the absence of any robust data on these benefits, and the need for any forest carbon finance scheme to operate transparently and legally. Hence, our calculations provide a minimum estimate of the economic challenge that a forest carbon finance scheme may face in influencing land use decisions.

To calculate carbon breakeven prices, we incorporated the following parameters: timber profit (assuming a single offtake; $ per ha), forest carbon stock (tC per ha), post-deforestation land-use carbon stock (tC per ha), 25-year discounted resin revenue (10% discount rate; $ per ha) and 25-year NPV of rubber or cash crops (10% discount rate, $ per ha). All input costs and prices were adjusted to 2013 US$ before analysis; all output values are thus in 2013 US$.

When calculating each parameter, to account for both uncertainties within, and variance between, data sources, values were resampled for each of 10,000 model iterations from either a normal distribution defined by the mean and standard error of the mean (SE; used to resample timber volume, carbon stock values and agricultural farm gate prices, for which the distribution of values was known), or a uniform distribution bounded by minimum and maximum estimates (used to resample agricultural yields, agricultural input costs, timber prices and timber extraction costs, for which the underlying distribution of parameter values was unknown; Supplementary Table 9).

All parameter values were resampled independently at each iteration, with the exception of timber volume (m3 per ha) and carbon stock (tC per ha). For these, a single forest inventory was randomly selected for each model iteration, in order to capture geographic variation without bias from relative inventory extent (Supplementary Table 1), and thereby avoiding the need for weighting. For each forest inventory, the mean and SE of timber volume, carbon stock and stem density (Supplementary Data 1) were calculated per tree species, and thus each royalty class. Where trees of smaller size classes were sampled from subplots nested within main plots (Supplementary Table 1), standardised values (per ha) still allowed mean and SE to be calculated per royalty and size class and, as the numbers of subplots and main plots were equal within each such inventory (Supplementary Table 1), no weightings were required. From the selected inventory, timber volumes and carbon stocks were simultaneously sampled from the same point in their distribution relative to the mean, as values of timber volume and carbon stock were likely to be correlated.

Stem-specific timber volume and carbon stock estimates (derived separately from DBH) were negatively skewed across all forest inventories. To address this, timber and carbon densities per plot were square-root transformed before calculating the mean and SE for each forest inventory, reducing the influence of infrequent plots with exceptionally high timber and carbon density. Timber and carbon values resampled from the resulting normal distribution were back-transformed before use in the model. Carbon stocks and wood volumes are presented in results as the unweighted mean and 95% confidence interval across all 10,000 iterations.

Agricultural yields, costs and farm-gate prices for all crops were sampled independently for each iteration, but the position within the distribution for each parameter was held constant across each crop type for each iteration. Resin production and prices were also sampled independently for each iteration. By sampling yields, input costs and farm-gate prices independently, we underestimate variability in NPV relative to that potentially arising if variation in these parameters were correlated, for example due to site location. Similarly, although parameter values varied between model iterations, we assumed a constant temporal yield-curve over the 25 years and did not incorporate inter-annual stochastic variability.

Opportunity costs of logging and rubber (or other cash-crops; offset by lost resin revenue; $ per ha) and breakeven carbon prices needed to offset the opportunity costs, plus setup (transaction) and implementation costs (based on emitted forest carbon offset by post-deforestation land-use carbon stocks; $ per tCO2) were then generated for each scenario. Results are presented as the median and interquartile range across 10,000 model iterations. Indicative real-world carbon prices (Supplementary Table 4) are shown in relation to breakeven carbon prices.

Timber profits

Forest inventories (Supplementary Fig. 2) were used to estimate timber volumes. Five inventories used fixed sampling areas (3.1–60 ha total per landscape; Supplementary Table 1) while inventory F05 used variable radius plots67. Within forest types, sites had similar relative distributions of size classes for all trees (Supplementary Fig. 4, Supplementary Table 10) and within royalty classes (Supplementary Fig. 9). In each, all trees ≥ 10 cm DBH were measured and identified to species level; nomenclature was standardised across datasets following The Plant List Version 1.1 (2013)68. Vines were not recorded. Tree species were assigned to one of five royalty (value) classes: luxury, I, II, III, or non-classified (NC; Supplementary Data 1) based on a government list of timber species69. The 14 luxury timber species included Burmese Rosewood (Dalbergia oliveri, commonly called D. bariensis in Cambodia (Supplementary Data 1), IUCN EN, CITES Appendix II), Siamese Rosewood (D. cochinchinensis, VU, CITES Appendix II) and Burmese Padauk (Pterocarpus macrocarpus, unassessed). All Dipterocarp species were class I or II and included popular timber species (e.g. Dipterocarpus alatus, EN and Anisoptera scaphula, CR); other popular timber species in class I were Sindora siamensis, LC, and Heritiera (Tarrietia) javanica, unassessed63).

Logging revenues depend on the minimum tree size commercially harvested, which may differ according to royalty class. Luxury species are exceptionally valuable and even small amounts are harvested70; we therefore assumed all luxury trees ≥ 10 cm DBH would be logged. For class I and II species, we assumed a minimum harvestable DBH of 40 cm, but also explored the effect of reducing minimum harvestable DBH to 30 cm (Supplementary Note 1; Supplementary Table 2). Class III trees (that are used for local construction or fuelwood) and non-classified timbers (assumed to be only useful as fuelwood) ≥ 40 cm DBH were initially assumed to have market value (Supplementary Table 8). However, extraction of class III and non-classified timbers was found to incur a net cost in open forest (−$57, −$456 to −$21 per ha) and in dense forest (−$1,336, −$2,096 to −$850 per ha), despite assuming relatively low timber extraction costs, that involved selective logging activity by local people in Cambodia in a “business-as-usual” scenario with no formal logging concessions, inventories, management plan, or demarcation of logging areas. Costs included: wage labour, food, motorbike fuel, ox-cart transportation to the roadside/village and chainsaw maintenance, but excluded the capital cost of the chainsaw. If class III and non-classified classes were to be extracted, this would likely be for firewood or construction locally (with extraction costs subsumed within non-market subsistence livelihood activities), or through destructive clearance during land preparation for agriculture (already considered in establishment costs for scenarios of agricultural conversion; Supplementary Table 11). Therefore, maximum potential timber profit accrued solely from logging luxury, class I and class II timber, and we assumed that the opportunity cost of protecting forests from further logging after the “Luxury, I, II timber logged” scenarios was reduced to zero (Fig. 3, Supplementary Table 2).

The timber profit remaining in the forest (R x ; $ per ha) in each logging scenario x, representing the opportunity cost of intervening at that point to protect timber remaining within the forest from further logging, was calculated as shown in equation (1):

$$R_x = \mathop {\sum}\limits_{{{a}} = 1}^5 {V_a \times \left( {p_a - C} \right)}$$
(1)

Where V a is the timber volume (m3 per ha) of timber royalty class a (five classes; Supplementary Data 1), estimated using Cambodian government standard timber equations (Supplementary Table 12) reduced by 20% to account for wastage71, p a is the timber price ($ per m3) for that royalty class (Supplementary Tables 8 and 13), and C the cost ($ per m3) of extraction to the roadside or village (Supplementary Table 14). Timber prices were estimated at the local (roadside or village) level in the absence of formal timber markets.

Forest carbon stocks

As for timber volume, forest inventories (Supplementary Table 1) were used to estimate forest carbon stocks for all stems ≥ 10 cm DBH, per royalty class, per forest plot, quantifying AGB and BGB22. AGB per plot and per royalty class (t per ha) was calculated using the Chave D Moist forest equation, using DBH only72, verified via destructive sampling for a REDD + pilot project in Cambodia22. Species-specific wood density was used where possible73; other species were assigned the mean wood density for the genus within the same region or, if no values were available for the genus, the mean wood density of all tree species across all inventories. BGB was estimated as 24% of AGB per plot74. AGB and BGB were summed, and multiplied by 0.5 to give estimated carbon content (tC per ha)72. Deadwood and SOC pools were not estimated; deadwood accounts for only 3% of emissions reductions from avoided deforestation, and SOC stocks are assumed not to change when land use conversion is to perennial crops according to the Intergovernmental Panel on Climate Change (IPCC) tier 1 carbon outcome calculation methods, and the carbon accounting methodology used for the REDD + pilot project in Cambodia22.

Post-deforestation land-use carbon stocks

We estimated the AGB and BGB carbon stocks of post-deforestation land-use classes which may partially offset forest carbon emissions. Time-averaged carbon stocks (taCs) were estimated for each crop type, which give the mean C stock over a plantation cycle from planting to harvesting3,75. This approach allows carbon stock estimates to be scaled up from a single plot to the landscape level, comparison of land uses with different rotation lengths, accommodates clearance and carbon release at the end of the crop rotation, and better reflects the net carbon outcomes and long term climate impact of a transition from one steady-state land use to another than a time series of carbon fluxes76,77. The taCs approach is consistent with the IPCC Good Practice Guidelines78 and the carbon accounting methodology used for the REDD + pilot project in Cambodia22.

The carbon stock estimate for rubber (52.5 tC per ha) was based on multiple estimates of taCs calculated either as the carbon stock in the median year of the plantation cycle using logistic or Gompertz models of growth, or 50% of the carbon stock in the final year of the plantation cycle assuming a linear biomass increase3. Time-averaged carbon stocks of other crops (cashew 22.3 tC per ha, sugarcane 6.8 tC per ha, cassava 2.5 tC per ha) were estimated as 50% of the carbon stock accumulated at the maximum rotation length (Supplementary Table 6). As for forests, soil carbon stocks were not considered, although there is strong evidence for soil carbon reductions when forest is converted to rubber3 or other tree cash crops27.

Resin revenue

The potential revenue generated by resin collection over a 25-year period (D; $ per ha; years 0–24 inclusive) was calculated following equation (2):

$$D = \mathop {\sum}\limits_{{{n}} = 0}^{24} {\frac{{\left[ {\left( {t \times i} \right) \times y} \right] \times p_R}}{{\left( {1 + r} \right)^n}}}$$
(2)

where t is resin tree stem density per ha (from forest inventories), i the likely proportion of productive trees (i.e., excluding non-starter or exhausted trees that do not yield), y the resin yield (litres per stem per year) and p R the resin price ($ per litre; Supplementary Table 15). Resin revenue is discounted over a 25-year timeframe n (years 0–24 inclusive) using a discount rate r of 10%, with no discount applied in year 0. All trees ≥ 30 cm DBH can potentially be tapped, with no identified relationship between resin yield and DBH; labour costs were not included in calculation of resin profits, as resin tends to be collected only when there are few or no alternative wage options66.

Agricultural net present value

We estimated likely farm-gate profits for rubber and other cash crops using region-specific data (Supplementary Tables 5 and 11). The 25-year discounted net present value (NPV, $ per ha; P b ) of each potential crop (b; rubber, cashew, cassava, sugar) was calculated as shown in equation (3):

$$P_b = \mathop {\sum}\limits_{{{n}} = 0}^{24} {\frac{{\left( {y_b \times p_b} \right) - C_b}}{{\left( {1 + r} \right)^n}}}$$
(3)

where y b is the year-specific yield (t per ha per year), p b the price ($ per t) and C b the cost of production ($ per ha per year). Profits are discounted over a 25-year timeframe n (years 0–24 inclusive) using a discount rate r of 10%, with no discount applied in year 0. In the case of annual crops, an “end-of-year” perspective is taken, whereby returns accrue in the same year as crop establishment.

A comparison of the spatial distribution of historically suitable environmental space for rubber1 and the spatial distribution of deciduous dipterocarp forest (DDF24; which shares many characteristics with our open forest category, although often in fine-grained mosaic with mixed deciduous and other forest types) shows most DDF lies outside the optimal zones for rubber cultivation. Reduced rubber yields were predicted (though the magnitude of reduction was not defined) in areas of drought risk, defined as <60 mm rainfall per month for >5 months per year, and/or <1,200 mm rainfall per year and/or <20 mm rainfall during the driest quarter1. This drought risk definition overlaps with the bioclimatic limits of DDF (1,000–1,500 mm rainfall per year with a defined dry season)24. Although rubber yield reductions due to drought have not been quantified, reduced dry season growth can delay the onset of tap** from the sixth to the tenth year after planting36. We therefore delayed the onset of tap** in the plantation cycle in open forest scenarios (Supplementary Table 11).

To accommodate change in annual yield across a 25-year production cycle, model iteration-specific yield curves were simulated separately for each crop (Supplementary Fig. 10, Supplementary Table 11), using a single iteration-specific randomly-generated yield index (proportionate between minimum and maximum values, uniform distribution). Crop-specific production costs were sampled from a uniform distribution between the minimum and maximum values available in the literature (Supplementary Table 11) and crop-specific farm-gate prices were used (Supplementary Table 5 and 11).

Breakeven carbon prices

The breakeven carbon price (E xb ; $ per tCO2) required to offset the opportunity cost of forest conservation ((R x –D) + P b ) and cover the costs of REDD + project setup (one-off transaction cost) and implementation (discounted over 25 years; G; $ per ha) for each scenario of logging (x) and crop (b, including the option of no crop) was calculated as shown in equation (4):

$$E_{xb} = \frac{{\left[ {\left( {R_x - D} \right) + P_b} \right] + G}}{{3.67 \times Z_x}}$$
(4)

where Z x is the residual carbon stock (tC per ha) of all trees ≥ 10 cm DBH remaining in each logging scenario (x) and 3.67 the conversion factor from tC to tCO228. Carbon stocks of post-deforestation land-uses were subtracted from Z x when exploring the impact of incorporating these stocks on carbon breakeven prices. Estimated PES project setup ($4.95 per ha) and implementation costs ($9.47–$13.09 per ha per year) were obtained from a multi-year spending history and projected management expenditure budget for a pilot REDD + project in Cambodia (Wildlife Conservation Society, unpublished data). These costs fell well within annual management and implementation cost estimates in existing literature, that range from $0.87–$20.01 per ha79. Annual implementation costs were discounted and summed across a 25-year timescale.

Finally, sensitivity analyses explored the impacts on carbon breakeven price of: increasing or decreasing timber, resin and agricultural commodity prices (Supplementary Fig. 6), non-availability of resin markets (Supplementary Table 7), reducing the threshold of commercially viable stem diameter on timber profits (Supplementary Table 2), and alternative discount rates of 5%, 8 and 15% as applied to agricultural NPV and resin revenue (Supplementary Figs. 7, 8).

Leakage

The costs of controlling for leakage of avoided deforestation for rubber, or forest degradation through selective logging, were not included in analyses. Ultimately, the need for land for rubber expansion will only be mitigated through reduction in global demand for natural rubber, which is reliant upon 1) global markets and demand for products such as vehicle and aircraft tyres, 2) development of alternatives to natural rubber, or 3) improvements in recycling methods. Similarly, demand for timber, within and beyond mainland Southeast Asia, would need to be met from well-managed sources before leakage of forest degradation or conversion could be effectively controlled. However, a robust rubber sustainability initiative, or corporate zero deforestation commitments, may displace rubber plantations to sites where land use conversion entails negligible net carbon emissions. We have therefore not attempted to incorporate the cost of controlling leakage in this analysis.

Data availability

A summary of the agricultural data used for this study are shown in Supplementary Table 11, with data sources detailed in Supplementary Table 5. Original forest inventory data (Supplementary Table 1, Supplementary Fig. 2) are not publicly available, and were made available for sole use in this study with permission of the following organisations and co-authors: F01 by Permian Global in collaboration with Ecometrica and Birdlife Cambodia; F02 and F03 by the Forestry Administration of the Royal Government of Cambodia, Wildlife Conservation Society Cambodia Program and Wildlife Conservation Society Global Conservation Program; F04 by Cambodia Development Resource Institute (CDRI); F05 by D.P. Bebber and F.H. Lambrick; and F06 by I. Theilade, J.F. Maxwell and P. Chhang. Derived data supporting the findings of this study, and all R scripts used to resample data and run the models, are available from the corresponding author on reasonable request.