Protecting tropical forests from the rapid expansion of rubber using carbon payments

Expansion of Hevea brasiliensis rubber plantations is a resurgent driver of deforestation, carbon emissions, and biodiversity loss in Southeast Asia. Southeast Asian rubber extent is massive, equivalent to 67% of oil palm, with rapid further expansion predicted. Results-based carbon finance could dis-incentivise forest conversion to rubber, but efficacy will be limited unless payments match, or at least approach, the costs of avoided deforestation. These include opportunity costs (timber and rubber profits), plus carbon finance scheme setup (transaction) and implementation costs. Using comprehensive Cambodian forest data, exploring scenarios of selective logging and conversion, and assuming land-use choice is based on net present value, we find that carbon prices of $30–$51 per tCO2 are needed to break even against costs, higher than those currently paid on carbon markets or through carbon funds. To defend forests from rubber, either carbon prices must be increased, or other strategies are needed, such as corporate zero-deforestation pledges, and governmental regulation and enforcement of forest protection.


Map of forest inventory locations. Forest inventories were obtained from six landscapes (F01 -F06) in Cambodia (Supplementary
); dense forest was sampled in each landscape, and open forest in three landscapes. Two landscapes are managed for biodiversity conservation (F01 and F02), two are partly managed by communities (F04 and F05) and two are not under formal management (F03 and F06). Some selective logging had taken place in all landscapes prior to data collection, as has occurred across most of the region (Supplementary Note 1). F03 is represented by a single marker as the inventory comprised a single 60 ha plot, and F04 is represented by a single marker as individual plot locations were not available. Dense   Sensitivity analysis assessing the effect of commodity prices on breakeven carbon prices. Sensitivity analyses explored the impacts of increasing and decreasing a) timber, b) agricultural and c) resin prices on carbon breakeven price, using a price index randomly drawn from a uniform distribution between 0.3 and 2 (i.e. from less than a third, to two times the mean price), following Gilroy et al 6 . This index range encompassed all annual producer farm gate price fluctuations for natural rubber 2000 -2014 inclusive. For each input variable examined (timber, agricultural or resin price), carbon equilibrium prices are shown in relation to the price index. For agricultural price sensitivity analysis, we only show sensitivity of carbon breakeven prices under the "No timber logged" scenario for each crop. Increases in timber and agricultural prices could have a strong effect on the opportunity costs of conservation. In contrast, resin prices would need to increase by more than two times their current value in order to reduce carbon equilibrium prices. Grey lines on the panels represent real world carbon prices (Supplementary Table 10): dotted = $5 tCO2 -1 (indicative of voluntary market forest carbon sales and non-market carbon fund prices), short dash = $13 tCO2 -1 (indicative of compliance market prices) and long dash = $36 tCO2 -1 (indicative of the social cost of carbon).

Supplementary Figure 7
Consequence of an alternative 5% discount rate on opportunity costs and breakeven carbon prices needed to protect forests from logging and conversion to rubber and other cash crops. Opportunity costs (OC; a -j) include forgone profits from logging (for luxury class timber, trees ≥10 cm DBH, other classes of timber, trees ≥40 cm DBH), and/or conversion to agriculture, offset by resin revenue, except where resin trees (class II) are logged out (in the "luxury, I, II logged" and "all timber logged" scenarios). Breakeven carbon prices (BCP; k -t) are the prices needed to offset opportunity costs, REDD+ setup costs and implementation costs. Costs are shown separately for dense and open forests. Time-averaged post-deforestation land use carbon stocks partially offset forest carbon losses. Grey lines on BCP panels represent real world carbon prices: dotted = $5 tCO2 -1 (indicative of voluntary market forest carbon sales and non-market carbon fund prices), short dash = $13 tCO2 -1 (indicative of compliance market prices) and long dash = $36 tCO2 -1 (indicative of the social cost of carbon). Outliers (more than 1.5x the interquartile range) are not displayed to improve the clarity of the figure; the value shown above each box-whisker gives the n outliers excluded out of 10,000 modelled results.

Supplementary Figure 8
Consequence of an alternative 15% discount rate on opportunity costs and breakeven carbon prices needed to protect forests from logging and conversion to rubber and other cash crops. Opportunity costs (OC; a -j) include forgone profits from logging (for luxury class timber, trees ≥10 cm DBH, other classes of timber, trees ≥40 cm DBH), and/or conversion to agriculture, offset by resin revenue, except where resin trees (class II) are logged out (in the "luxury, I, II logged" and "all timber logged" scenarios). Breakeven carbon prices (BCP; k -t) are the prices needed to offset opportunity costs, REDD+ setup costs and implementation costs. Costs are shown separately for dense and open forests. Time-averaged post-deforestation land use carbon stocks partially offset forest carbon losses. Grey lines on BCP panels represent real world carbon prices: dotted = $5 tCO2 -1 (indicative of voluntary market forest carbon sales and non-market carbon fund prices), short dash = $13 tCO2 -1 (indicative of compliance market prices) and long dash = $36 tCO2 -1 (indicative of the social cost of carbon). Outliers (more than 1.5x the interquartile range) are not displayed to improve the clarity of the figure; the value shown above each box-whisker gives the n outliers excluded out of 10,000 modelled results.

Supplementary Figure 9
Frequency distribution of stem diameters for each timber royalty class within each forest inventory. Data only includes stems ≥30cm in dense forest (a, i -vi) and open forest (b, i -iii). Bars show the relative frequency of trees within DBH size categories, while curves represent a smoothed density distribution. Frequency distribution of tree sizes is remarkably similar across all forest landscapes.    7 13.00 18.00 Corporate internal carbon prices 8 36.00 US government social cost of carbon 9 36.00 40.00 World Bank social cost of carbon 8 19 Supplementary Table 5 Data sources for agricultural net present value calculation. For each parameter, price and cost estimates from different sources were adjusted to US$ 2013 using an averaged CPI for Thailand, Cambodia and Vietnam 10 before calculation of means, variance or range. FAO producer prices are compared to price estimates from other sources, including GEM commodity prices, in Supplementary Figure 1. For each parameter an explanation of data treatment and value selection is given; resulting cropspecific parameter values are provided in Supplementary Table 8. Rubber yield 1,12,13 Cambodia Yield data for each production year of the 25-year management cycle of a large plantation were extracted from each study. Minimum and maximum yields for each year, across all studies, were used as minimum and maximum input parameters for the simulation model (n = 3). Yield estimates that appeared to use unclear units which could correspond to wet latex yield rather than dry rubber yield, were excluded from analysis. For each model iteration an annual dry rubber yield curve (t ha -1 yr -1 ) over 25 years was generated as a proportion of the maximum attainable yield. For each iteration (i), a proportion (propi range 0 to 1) was selected randomly from a uniform distribution. Then for each production year (t, range 0-24) the run-specific yield (Yi,t) was calculated applying this proportion to the interval between reported minimum (Ymint) and maximum (Ymaxt) yields, following the equation:

Rubber -large scale monocultural plantation crop
The run-specific proportion used to create the yield curve was generated independently for each crop, for each iteration. Two examples, for propi = 0.25 and 0.87, are shown in Supplementary Figure 10. For open forests, a yield penalty was applied to account for predicted slower tree growth in dry conditions by delaying the onset of tapping until 10 years after planting, rather than the usual 6 years 16 .
Rubber production cost 1,[12][13][14] Cambodia Production cost data for each year of the 25-year management cycle for large monocultural plantations were extracted from each study (n = 2 referring to large plantations), standardised to USD ha -1 yr -1 in US$ 2013. Costs included: land clearance (using bulldozer or tractor, year 1 only, n = 1), land preparation (annual), planting material, annual inputs, annual labour costs. We do not include the costs of felling rubber trees at the end of the 25-year management cycle as this cost is associated with replanting and the start of the next yield production cycle.
For each model iteration an annual cost curve ($ ha -1 yr -1 ) over 25 years was generated as a proportion of the maximum cost. For each iteration (i), a proportion (Cpropi range 0 to 1) was selected randomly from a uniform distribution. Then for each production year (t, range 1-25) the run-specific cost (Cit) was calculated applying this proportion to the interval between reported minimum (Cmint) and maximum (Cmaxt) costs, following the equation: Ci,t = Cmint + (Cmaxt-Cmint) * Cpropi. The run-specific proportion used to create the cost curve was generated independently for each crop, for each iteration.

Cashew -smallholder monocultural plantation crop
Cashew producer price (see also Supp. Fig. 1 Six sets of yield data, for different management cycle lengths, were extracted from ACI (2005 12 ) for small farms, covering four provinces. Some yields seemed surprisingly high (up to 5.5 t ha -1 ), given that average yields from a large scale plantation in Cambodia were reported to be 0.8 t ha -1 1 , and a report on the cashew industry in Cambodia states that the maximum known yields from Cambodia are 2.0 t ha -1 , with exceptional yields of 2.5 t ha -1 reported from Vietnam 17 . We therefore excluded any datasets reporting average yields >2.5 t ha -1 over a 25-year plantation cycle.
Yield curves reported for management cycles of less than 25 years (e.g. 10 or 15 years) were extrapolated according to the proportional yield declines reported for 25-year management cycles. Minimum and maximum yields for each production year, across all six yield curves, were used as minimum and maximum input parameters for the simulation model and an iteration-specific yield curve was simulated, as for rubber. The run-specific proportion used to create yield curve was generated independently for each crop, for each iteration.

Cashew production cost
12 Cambodia Production cost data for each year of the 25-year management cycle were extracted from each of the six datasets in ACI (2005 12 ), standardised to USD ha -1 yr -1 (adjusted to 2013). Costs included: land clearance (using manual labour, year 1 only), land preparation (year 1 only), planting material, annual inputs and labour costs (including imputed family labour costs). Where management cycles reported were for less than 25 years, costs were extended to year 25 based on final year costs in each dataset. Variation in reported labour requirements among the six datasets was high (total annual labour days ranged from 25 to 267 days ha -1 ). On inspection of the budgets, this was accounted for by a ten-fold greater number of annual labour days reported for harvesting in one smallholder budget (220 -300 days ha -1 ). In three other smallholder datasets, annual harvesting labour days ranged from 16 -30 days ha -1 , while a maximum of 35 days ha -1 were reported from a survey of 140 smallholder cashew farmers in Ghana. We therefore considered this large estimate of labour input to be an error, and limited the maximum annual harvesting days to 35 days ha -1 .
Minimum and maximum costs for each production year, across all six datasets, were used as minimum and maximum input parameters for the simulation model, and an iteration-specific cost curve was simulated, as for rubber. The run-specific proportion used to create the cost curve was generated independently for each crop, for each iteration.

Cassava -smallholder monocultural annual crop
Cassava producer price (see also 11 Thailand/ Vietnam Producer prices for raw cassava are published for Cambodia on FAOSTAT 11 , but they are much higher than any of the following price estimates for raw cassava, and therefore appear to be incorrect: producer prices for Thailand and Vietnam; farm gate prices for Cambodia reported in ACI As for rubber, we suggest that small farmers may receive lower prices where access to markets is limited. There is also variation in price among provinces depending on the degree of harvesting, drying and transportation costs paid by the farmer versus middlemen; we base our NPV estimates on the most commonly used system in Cambodia, in which farmers sell raw cassava to traders, and traders pay harvesting, drying and transport costs 14  2007 22 ), suggesting that these bounds are appropriate. Yields were assumed to remain constant over 25 years of repeated annual planting. This assumption is supported by data from Cambodia, that show no difference in yield between plots continuously cropped with cassava for 25 years without fertilisation and soils that had been cropped for <10 years 21 . Supporting this, yields of 13.3 t ha -1 yr -1 are reported from fully exhausted soils in Colombia 23 . However, cassava is potassium limited, and potassium-depleted soils can produce yields as low as 5 t ha -1 yr -1 in India, but if soils contain potassiumproducing minerals, depletion does not occur even in absence of fertiliser application 23 . Cassava is tolerant to low soil fertility conditions 21,23,24 . Evidence from Cambodia suggests that soil type does not affect cassava yield, although farmers self-identified areas of higher soil quality where cassava yields were increased within individual farms 21 . We thus do not modify our yield estimates for open and dense forest areas, despite predicting that soil types may differ under each forest type. Yield curves were created for 25 consecutive years of cultivation, as for rubber and cashew, using the same proportion of maximum yield for each year. The run-specific proportion used to create yield curve was generated independently for each crop, for each iteration.

12,14
Cambodia Annual production cost data for raw cassava were extracted from each of six datasets and standardised to USD ha -1 yr -1 units, adjusted to US$ 2013. Costs in all studies included: land clearance (using manual labour, year 1 only), land preparation (annual), planting material, annual inputs, annual labour costs (including imputed family labour costs), but did not include drying or transportation costs. Cost curves were created for 25 consecutive years, as for rubber and cashew, using the same proportion of maximum cost for each year. The runspecific proportion used to create the cost curve was generated independently for each crop, for each iteration.

Sugar -large scale monocultural annual crop
Sugar producer price (see also Supp. Figure  1) 11 Thailand / Vietnam Producer prices are published for Cambodia on FAOSTAT 11 , but are much higher than FAOSTAT prices for Thailand and Vietnam and much higher also than Cambodian farm gate prices reported in ACI 12 . Prices were checked against global prices, and EU minimum prices, as Cambodia receives preferential pricing to export sugar to the EU through the "Everything but Arms Treaty" 25 . However, Cambodia's Producer Prices were substantially lower than either of these indicators. We therefore used the mean and standard error of the mean (SE) of annual producer prices across Thailand and Vietnam for years 2003 -2012 ($34.34 t -1 ), but as for rubber and cassava, suggest that small farmers may often receive lower prices where access to markets is limited.

Sugar yield 12 Cambodia
Annual yield data for large scale plantations from two regions of Cambodia were extracted from ACI 12 . Minimum and maximum reported yields were used as input parameters for the simulation model and were assumed to remain constant over 25 years of repeated annual planting. Yield curves were created for 25 consecutive years of cultivation, as for rubber and cashew, using the same proportion of maximum yield for each year. The run-specific proportion used to create yield curve was generated independently for each crop, for each iteration.

Sugar production cost
12 Cambodia Annual production cost data were extracted, standardised to USD ha -1 yr -1 , adjusted to US$ 2013 prices. Costs included: land clearance (using tractor or bulldozer, year 1 only), land preparation costs (annual), planting material, annual inputs, annual labour costs. Cost curves were created for 25 consecutive years, as for rubber, using the same proportion of maximum cost for each year. The run-specific proportion used to create the cost curve was generated independently for each crop, for each iteration.

Inflating US$ to 2013
Consumer price index (CPI) 10 Cambodia/ Thailand/ Vietnam The World Bank Consumer Price Index (CPI) is available at the country level. For parameters obtained from multi-country studies, CPIs for Cambodia, Thailand and Vietnam were averaged and used to deflate all input costs and prices to 2013.
Supplementary Table 6 Post-deforestation land-use carbon stock estimates. Time-averaged carbon stocks (taCs) of above-ground biomass (AGB) and below-ground biomass (BGB) were estimated as either 50% of the carbon stock of a crop/plantation at the maximum rotation length 26 , or for rubber, as the carbon stock as calculated by a regression equation at the median rotation length 27 .

22.32
Out estimate of taCs for a cashew plantation on a 10-year plantation cycle (22.32 tC ha -1 ) were generated based on field data from Cambodia 28 . AGB of cashew plantations for each year of a 10-year plantation cycle were extracted from the data; BGB was assumed to be 24% of AGB 29 , and AGB + BGB carbon stock was assumed to be 50% of biomass. We calculated taCs to be 50% of this value 27 . Where field data were not provided for a given year, the value for the next oldest year was used, generating a conservative estimate.

2.5
Carbon stock for "annual cropland" in dry and seasonal areas of Asia reported as 5 tC ha -1 ; time-averaged carbon stock is 50% of this value 26 .

6.75
Carbon stock for sugarcane in dry and seasonal areas of Asia reported as 13.5 tC ha -1 ; time-averaged carbon stock is 50% of this value 26 . Supplementary Table 8 Price estimates for timber royalty classes at various selling points in Cambodia. Minimum and maximum prices for each royalty class for the roadside/village were used as input parameters for simulations; other price points (i.e. forest, domestic or international market) were not used in the final analysis. In the absence of species-specific records from formal timber markets, all timber species in each royalty class were assumed to fetch the same price as those species from that royalty class that were explicitly named in source of prices (Supplementary Table 14). Prices shown are mean of all available data from 2007 -2014 inclusive, except for Non-Classified timber, for which we use the price of fuelwood reported from field study in Cambodia 30 . All prices were inflated to $US 2013 using a CPI specific to Cambodia. All price data were based on interviews with villagers or market traders. ¥ n price estimates within and across all studies # Royalty class I timber is classed as more valuable than class II 40 ; we thus assume the same minimum price for class I and II timber in simulation models ($90.81) *As for forest price ~A s for domestic market price Resampling input parameters. For each sampling iteration, values for each parameter were sampled from either a uniform distribution between the minimum and maximum bounds, or where the shape of the distribution was known to be normal, from a normal distribution defined by the mean and standard error (SE) of the mean.

Input parameter Units Bounds
Agricultural NPV     Table 14 Timber extraction cost estimates. The minimum and maximum timber extraction costs from this table were used as input parameters for simulating timber costs. These costings assume 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 include: wage labour, food, motorbike fuel, ox-cart transportation to the roadside/village and chainsaw maintenance but exclude the capital cost of the chainsaw (around US$350 30 ). Costs in table are inflated to US$2013 using a CPI specific to Cambodia.

Reference
Extraction Cost USD m -3 processed wood Notes 30 116.01 Cost of cutting wood in forest and ox-cart transport to village. Labour, food, fuel, chainsaw oil, 2-stroke oil, chain, chainsaw maintenance, excludes capital cost of chainsaw ($350 dollars, last 10 years), ox cart to village. 31

75.77
Cost of cutting wood in forest and ox-cart transport to village. Hired labour to cut tree, chainsaw fuel, ox cart to village. 1

82.36
Cost of partial cut and transport (to village).
Forest degradation through logging has a complex history in Cambodia. Forest governance institutions were lost during the political turmoil of the Khmer Rouge era (1975 -1980). Subsequently, nearly 70% of forested land was allocated for logging concessions in the 1990s, followed by widespread over-harvesting both within and outside concessions 49 . All formal logging concessions were halted in 2002, and many have since been designated as protected areas. Forested land is owned by the state and some annual logging coupes have been allocated; however, large tracts of forest have no clear management plan and illegal logging remains pervasive 36,38,[49][50][51] . Allocation of forested areas for Economic Land Concessions (ELCs), which allow conversion to plantation crops, is a key driver of forest clearance in Cambodia; much of Cambodia's current timber harvest is extracted within and around ELCs 52 . Much focus has been placed on the extraction of the highest-value Luxury class timber 50 , which can generate high levels of short-term income 33 . However, logging of other species (of lower royalty classes, especially classes I & II) is also pervasive 5 and forms the bulk of timber harvested when forested land is cleared from ELCs 53 .
The Forestry Administration grants transport licenses for logs ≥30cm DBH harvested from within ELCs, except for luxury timber 53 . Minimum harvestable limits for all commercial tree species range from 30 -60cm DBH (as defined by the Ministry of Agriculture, Forestry and Fisheries in Prakas 089 (2005) -a ministerial or inter-ministerial proclamation in Cambodian law), however the level of enforcement of these limits is not clear. According to a Ministerial Declaration, Prakas 089, harvest of all luxury class timber is illegal, as is harvest of resin trees (some Dipterocarpus spp, all royalty class II, Supplementary Table 2) utilised by local people, unless they have given consent and been compensated. This latter group includes some of the most commercially valuable dipterocarp species. However, there is evidence for routine and widespread harvest of both luxury and resin trees 54 . In the 1990s commercial logging focussed on trees ≥45cm DBH 55,56 , while minimum commercial harvestable DBH elsewhere in Southeast Asia is ≥40cm DBH 57 , and an assessment of logging in Cambodia that modelled unsustainable extraction rates assumed trees ≥40cm would be harvested 58 . For class I and II species, we therefore assumed a minimum harvestable DBH of 40 cm. Luxury species are exceptionally valuable and even small amounts are harvested 59 ; we therefore assumed all luxury class trees ≥10cm DBH would be harvested. Class III trees are used for local construction purposes (i.e. as timber) or as fuelwood. Non-classified trees are assumed to be only useful as fuelwood; non-classified and class III trees ≥40cm DBH were therefore assumed to have market value as fuelwood.