Alternative futures for Borneo show the value of integrating economic and conservation targets across borders

Balancing economic development with international commitments to protect biodiversity is a global challenge. Achieving this balance requires an understanding of the possible consequences of alternative future scenarios for a range of stakeholders. We employ an integrated economic and environmental planning approach to evaluate four alternative futures for the mega-diverse island of Borneo. We show what could be achieved if the three national jurisdictions of Borneo coordinate efforts to achieve their public policy targets and allow a partial reallocation of planned land uses. We reveal the potential for Borneo to simultaneously retain ∼50% of its land as forests, protect adequate habitat for the Bornean orangutan (Pongo pygmaeus) and Bornean elephant (Elephas maximus borneensis), and achieve an opportunity cost saving of over US$43 billion. Such coordination would depend on enhanced information sharing and reforms to land-use planning, which could be supported by the increasingly international nature of economies and conservation efforts.

This shows the uncertainty of allocating a planning unit to the final landuse zone. This is a combination of the classification uncertainty from multiple runs with the same input parameters, along with the variation in input parameters. There is no uncertainty surrounding zoning in scenario 1, as this scenario is based on implementing the existing land-use allocations. Removing the requirement for RIL had only a minor reduction in the opportunity cost for each scenario, whereas removing the species or forest targets resulted in larger opportunity cost savings. Scenario 1 was not included as the land-use allocation cannot be altered, therefore changing the targets does not have an impact. Scenario 4b was also excluded, as this scenario was already a variation on the targets in Scenario 4a.

Logging Profit
The estimated profit from timber harvesting was obtained from data on timber yields, costs and revenues for CL and RIL (Supplementary Table 6).The mean value per hexagonal 10 km 2 grid cell varied, depending on the forest condition and havestable area.  (62) , and so to allow for buffering of watercourses smaller than this threshold, we applied a uniform reduction factor of 12.2% to the remaining harvestable area in each hexagonal 10 km 2 grid cell (based on the required area for buffering small watercourses in three reserves in Sabah with moderate rainfall 63 ).
The profit per hectare harvested (Supplementary Table 7

Plantation Profit
Oil-palm suitability was estimated by classifying a variety of biophysical properties of land units into five categories based on their suitability for oil-palm production (Supplementary Table 5a). If any given pixel had at least one of the biophysical properties classed as 'not at all suitable', it was excluded from further analysis. The remaining pixels were summed into a cumulative suitability map, which was then tertiled into 3 suitability classes (with 1 being the most suitable). The average annual profit for oil-palm production was derived from industry specific finance models 69 based on state averages (for Sabah, Sarawak, and Kalimantan) of production per hectare of fresh fruit bundles and based on a crude palm oil price of $800 per tonne (Supplementary Table 5b). Different scenarios of yield (full yield, 25% less, and 50% less) were applied to the 3 suitability classes to produce a Borneo-wide layer of potential revenue from oil-palm production (which was summarised at the planning unit level and used as R i for oil-palm in equation 1). Oil palm is particularly well adapted to the humid tropics, which combined with growing demand, means revenues are likely to continue well into the future 70 . Oil-palm production was therefore measured in productive hectare equivalents (i.e. one hectare of oil-palm planted on land with 50% productive capacity equates to half a hectare of oil-palm production).
The In addition to revenues from oil-palm or industrial timber production, significant additional revenue can arise from timber harvest during conversion from forest to plantations 74 . This was a once-off profit attributable to year 0 (i.e. it was not discounted).
Timber revenues from clear-felling before conversion to oil-palm were estimated from logging revenues for each state and forest type (intact or previously logged), as given in the description of timber harvesting profits, combined with estimates of the percentage of additional timber that could be obtained from clear-felling rather than selective logging (Supplementary Table 7

Protected area costs
The average annual management costs for protected areas (per hectare) was based on the optimal management of large Indonesian terrestrial national parks (approx. 120,000 ha) 75 .
This value (of 2004 US$6.17 ha -1 yr -1 ) was similar to other estimates (e.g. Wilson et al. ) and was adjusted to 2009 US$ ($7.01 ha -1 yr -1 ) . The estimate includes field and administrative staff, equipment and infrastructure maintenance 75 . This 'loss' forms R i for protected areas in equation 1. Additional start-up costs arise when a new protected area is established, which was estimated at $50 ha -1 (76) . Were applicable -$50 ha -1 forms C i for protected areas in equation 1.

Probability of deforestation
We employed a tree cover loss map for the period 2000-2010 (60x60m grid cell size) as the base dataset for modelling the probability of deforestation 78,79 . In this dataset 'tree cover' is defined as areas of trees (≥5m height) with >25% canopy cover and 'tree cover loss' as the removal of tree stands. We restricted our analysis to losses of intact forest cover that The spatial layers of each of these variables were weighted by their respective coefficient to produce a relative probability map of deforestation.

Carbon
We evaluated the change in carbon stock for each scenario relative to the current landuse plan (scenario 1). We calculated potential CO 2 emissions as the difference in time averaged CO 2 relative to a simple baseline scenario in which any area of existing forest is converted to oil-palm. Emissions from this conversion are assumed to equate to the extant aboveground carbon 83 and including peat carbon if on peat soil. Carbon was converted to CO 2 e using an emissions factor of 3.67 84,85 . Peat soil carbon net emissions were estimated using net CO 2 fluxes for a 25-year period 86 , which considers all inputs and outputs (and a single fire during forest clearance), giving an estimate of 1503 Mg CO 2 e ha -1 over a 25-year time horizon. Below-ground carbon was not considered for mineral soils, due to a lack of data for all land-use transitions, and the comparatively small changes in time-averaged carbon stocks on most mineral soil types (e.g. converting primary forest to oil-palm would emit 32.0 Mg CO 2 e ha -1 over 49 years on mineral soils 87 , compared to 1503 Mg CO 2 e ha -1 for the same conversion over 25 years on peat soils).
We assumed protected areas would retain extant aboveground and peat carbon, and sequester carbon through natural regeneration. For degraded forest and forest regrowth with extant aboveground carbon contents less than intact forest, we assumed regeneration would increase aboveground carbon stocks to equal that of the average for intact forest. For severely degraded logged forests, we assumed protection would only increase the stock of carbon by 5%. Most of this class is in East Kalimantan Province and these forests were severely burned twice, in March-April 1983 and March-April 1998 (i.e. during the two most intense El Niño fire pulses on record, also declared national disasters in Indonesia 88 ). Because of further burning, these areas have exhibited limited natural regeneration, showing high levels of cover by invasive grass species, and are unlikely to regain significant quantities of forest cover or biomass without active restoration 89 . Active restoration was not considered in these analyses (i.e. we assumed no carbon benefits from protection of lands that currently have no forest cover).
RIL was assumed to result in a reduction of 30% of above ground carbon, relative to intact forest, and CL a reduction of 60% 32 , relative to intact forest. CL was also assumed to emit approximately 347.5 Mg CO 2 ha -1 if on peat soils due to soil disturbance 86 . Plantations (for industrial timber or oil-palm) were assigned no net change when planted on non-forest areas (0 Mg CO 2 ha -1 ), because the carbon sequestered in industrial timber and oil-palm plantations is ultimately released when trees are harvested. For the "other non-forest" landuse class, we assumed worst case carbon emissions (i.e. that of oil-palm).