Effects of demand-side restrictions on high-deforestation palm oil in Europe on deforestation and emissions in Indonesia

Demand-side restrictions on high-deforestation commodities are expanding as a climate policy, but their impact on reducing tropical deforestation and emissions has yet to be quantified. Here we model the effects of demand-side restrictions on high-deforestation palm oil in Europe on deforestation and emissions in Indonesia. We do so by integrating a model of global trade with a spatially explicit model of land-use change in Indonesia. We estimate a European ban on high-deforestation palm oil from 2000 to 2015 would have led to a 8.9% global price premium on low-deforestation palm oil, resulting in 21 374 ha yr−1 (1.60%) less deforestation and 21.1 million tCO2 yr−1 (1.91%) less emissions from deforestation in Indonesia relative to what occurred. A hypothetical Indonesia-wide carbon price would have achieved equivalent emission reductions at $0.81/tCO2. Impacts of a ban are small because: 52% of Europe’s imports of high-deforestation palm oil would have shifted to non-participating countries; the price elasticity of supply of high-deforestation oil palm cropland is small (0.13); and conversion to oil palm was responsible for only 32% of deforestation in Indonesia. If demand-side restrictions succeed in substantially reducing deforestation, it is likely to be through non-price pathways.


Introduction
Tropical deforestation and forest degradation emit 1.8-11.4 GtCO 2 yr −1 (Olsson et al 2019). The largest direct cause of tropical deforestation is the conversion of forested land to produce globally traded 'forestrisk commodities,' e.g. palm oil, beef, soy, and paper (Curtis et al 2018, Pendrill et al 2019. Consumer countries are increasingly attempting to combat tropical deforestation and climate change by distinguishing whether forest-risk commodities were produced with high or low levels of recent deforestation, and restricting demand for commodities associated with high deforestation. Public demand-side restrictions on high-deforestation commodities have been enacted in the United Kingdom (DEFRA 2015), Netherlands (PIANOo 2017), Norway (Norwegian Parliament 2018, Rainforest Foundation Norway 2018, and France (République Française 2018), introduced in the United Kingdom (BBC 2020), vetoed in California (California Legislature 2020, Office of the Governor 2021), and proposed in the European Union (European Parliament 2020;Bager et al 2021) and United States (United States Congress 2021) (table SI1 available online at stacks.iop.org/ERL/17/014035/mmedia).
The recent public policy focus on restricting demand for high-deforestation commodities represents a shift in approach from positive incentives to keep tropical forests standing through international carbon payments via markets or funds (i.e. REDD+; Seymour and Busch 2016). It potentially complements private corporate commitments to eliminate deforestation from commodity supply chains (Lambin et al 2018, NYDF Assessment Partners 2019, Hargita et al 2020. The goals of governments that restrict demand for high-deforestation commodities include both ending domestic consumption of products associated with deforestation as well as contributing to reducing deforestation abroad (e.g. République Française 2018). Meanwhile governments of tropical commodity-producing countries have accused consumer countries of using climate concerns as thinly guised protectionism for their own farmers (e.g. Reuters 2019). However, despite the controversial and potentially central role of demand-side restrictions on high-deforestation commodities as a climate policy, their impact on reducing tropical deforestation and emissions has yet to be quantified.
Here we model the effects of one important bloc of consumer countries (Europe) restricting demand for the high-deforestation type of one important forest-risk commodity (palm oil) on deforestation and emissions in one important tropical country (Indonesia). We do so by integrating a global trade model (GTAP-BIO; Taheripour et al 2019) with a spatially explicit model of land-use change in Indonesia (OSIRIS; Busch et al 2015). We use spatial data on oil palm expansion (Austin et al 2017) and forest cover loss in Indonesia from 2000 to 2015 (Hansen et al 2013) to distinguish whether oil palm expansion occurred on land deforested after 2000 ('high deforestation') or before 2000 ('low deforestation'). We estimate how much larger impacts would be if other countries outside Europe also enacted demand-side restrictions. And we compare the impacts of demandside restrictions to the impacts of a benchmark alternative climate policy: a hypothetical domestic carbon price on emissions from deforestation in Indonesia.
Previous studies have examined the distal impacts of demand-side policies on palm oil production (e.g. Britz and Hertel 2011, Elliott 2015, Jafari et al 2017, Taheripour et al 2019, Hsiao 2021. However, these studies considered the effects of policies on palm oil in aggregate; they did not differentiate between highdeforestation and low-deforestation palm oil. Disaggregating the palm oil market into high-deforestation and low-deforestation segments, as we do here using spatial data on the location of oil palm expansion, lets us better represent the recent wave of public policies, and is the principal innovation of this paper. Our paper advances several other strands of literature as well. Previous studies have investigated the environmental impacts of bans, e.g. on shark fin (Ferretti et al 2020), ivory (Sosnowski et al 2019), and rhino horn (Crookes and Blignaut 2015). But due to the illicit nature of these markets, reliable data on production, trade, and consumption is scarce, which hampers the calibration of sophisticated policy models, as we undertake here using readily available data on the international palm oil market.
Previous studies have researched the effects of eco-labels for products that are interchangeable in consumption but vary in the environmental friendliness of their production, e.g. dolphin-safe tuna (Kirby et al 2014), Forest Stewardship Council (FSC) certified timber (Blackman et al 2018), and Roundtable on Sustainable Palm Oil (RSPO) certified palm oil (Carlson et al 2018). But in the case of eco-labels, price premiums are driven by voluntary consumer preferences and voluntary producer adoption. We are aware of only one previous study of mandatory national restrictions on products that do not comply with eco-certification, for tropical timber (Cole et al 2021).
Finally, our study contributes to literature estimating the effectiveness of initiatives to reduce deforestation and emissions in Indonesia, e.g. through protected areas (Brun et al 2015, Shah andBaylis 2015), debt-for-nature swaps (Cassimon et al 2011), tenure security and REDD+ projects (Resosudarmo et al 2014), reduced-impact logging , FSC certification (Miteva et al 2015), community forestry (Santika et al 2017(Santika et al , 2019, community land titling (Kraus et al 2021), multifunctional landscapes (Santika et al 2015), and a moratorium on concessions for oil palm, timber, and logging (Busch et al 2015, Chen et al 2019. It also contributes to a growing effort to integrate global economic models with finescale analysis of land-use change .

Methods
We modeled the effects of restrictions on demand for high-deforestation palm oil in consumer countries on deforestation and emissions in Indonesia. We integrated the GTAP-BIO global trade model and the OSIRIS model of land-use change in Indonesia in a three-step process (figure SI1). First, we developed the land-use change model to estimate how the conversion of forests to oil palm plantations (which produced 'high-deforestation' palm oil) and the conversion of non-forests to oil palm plantations (which produced 'low-deforestation' palm oil) responded to changes in prices (i.e. 'price elasticities of supply'). Then, using these price elasticities implemented in the GTAP-BIO model, we modeled how prices, quantities, and trade flows of high-deforestation and low-deforestation palm oil would change globally if consumer countries restricted demand for highdeforestation palm oil. Finally, we mapped global price changes back to spatially explicit changes in land use and emissions in Indonesia. We selected palm oil as the commodity of interest, Indonesia as the study region, Europe as the policy focus, and 2000-2015 as the study period for reasons elaborated in the supplementary information (scope of inquiry).
We modeled the conversion of forested and nonforested land to oil palm plantation from 2000 to 2015 as a function of palm oil price and other geographic factors (slope, elevation, distance from roads and provincial capitals, logging, timber, and oil palm concessions, and strict and multiple-use protected areas) using the OSIRIS land-use change model, as elaborated in the supplementary information (determinants of oil palm expansion). OSIRIS has been used to model policy decisions including reference levels for REDD+ (Busch et al 2009), incentives to reduce deforestation in Indonesia (Busch et al 2012), and carbon prices for pan-tropical reforestation (Busch et al 2019).
We defined palm oil as 'high-deforestation' if it was produced on land that was converted from forest after the year 2000, and 'low-deforestation' if it was produced on land that was converted from forest before the year 2000. We calculated the price elasticity of supply of high-deforestation and low-deforestation oil palm cropland using the observed relationship between palm oil price and conversion, as elaborated in the supplementary information (price elasticity of supply of oil palm).
We modeled global demand and supply of palm oil using GTAP-BIO, as elaborated in the supplementary information (a global market for two types of palm oil). GTAP is a widely used global Computable General Equilibrium model (Hertel 1997). For this study we used the GTAP 9 data base which represents the global economy in 2011. This data base disaggregates global economic activity and bilateral trade into 57 sectors and 140 regions (Aguiar et al 2016). We used the GTAP-BIO model (Taheripour et al 2017), which further disaggregates oilseeds into four subsectors (oil palm, soybeans, rapeseed, and other oil crops), as well as four oilseed crushing industries, in order to capture the potential for demandside substitution among these products. GTAP-BIO aggregates the full data base to seven regions (Indonesia and Malaysia; European Union; United States; China; Brazil; Other South America; Rest of World) for computational and analytical tractability. GTAP-BIO models the market for palm oil in three successive, vertically integrated stages: oil palm cropland; palm fruit; and palm oil. Our main extension and improvement of the GTAP-BIO model for this paper was to disaggregate demand for and supply of palm oil into two distinct commodities, high-deforestation palm oil and low-deforestation palm oil, allowing different prices, quantities, and trade flows for each.
In our base scenario, all Indonesian palm oil exports were treated equally by consumers and therefore exhibited a single global price. In our primary policy scenario, consumer countries in Europe imposed a ban on the use of high-deforestation palm oil, as elaborated in the supplementary information (restricting demand for high-deforestation palm oil).
In secondary policy scenarios, we sequentially expanded the breadth of participation across regions. Starting with Europe (the destination for 12.0% of palm oil produced in Indonesia and Malaysia, in GTAP-BIO), we expanded to Europe plus the United States (13.7%), Europe plus the United States plus China (23.1%), and all countries outside Indonesia and Malaysia (56.0%).
After running the base and policy scenarios in GTAP-BIO to generate changes in prices of oil palm cropland, we used the Indonesia land-use change model described above to calculate corresponding changes in quantities of conversion. We calculated greenhouse gas emissions due to conversion of forest to oil palm plantation by multiplying the area of conversion of forest to oil palm in each cell by the cellspecific emission factors used in Busch et al (2015), as elaborated in the supplementary information (landuse change and greenhouse gas emissions).
We used cost curves from Busch et al (2015) to identify the hypothetical carbon price at which an equivalent reduction in emissions could be achieved, using either a mandatory carbon price (payments for emission reductions plus taxes on emission increases) or a voluntary carbon price (payments for emission reductions only).
To assess uncertainties associated with model parameterization, we generated 95% confidence intervals around trade volumes and producer prices for each scenario, using a systematic sensitivity analysis in GTAP-BIO, as elaborated in the supplementary information (sensitivity analysis). To assess uncertainties associated with emissions from conversion, we tested alternative forest carbon data sets and emission factors for peat and mineral soil.

Results
Of the 11.8 million hectares of oil palm plantation in 2015, we find that 6.4 million hectares (54%) was planted after 2000 on land that had been forested in 2000. A 1.0 million hectares (8.7%) was planted on land that was not forested in 2000. 4.4 million hectares (37%) was planted before 2000; the forested status of this land before 2000 could not be determined from our data (figure SI2). Use of an alternative map corresponding to the Indonesia Ministry of Forestry's primary and secondary forest cover types in Indonesia (Global Forest Watch 2019) suggests that just 2.1 million hectares of oil palm (18%) was planted after 2000 on land that had been forested circa 2000. Our estimate of the share of post-2000 oil palm expansion that occurred on forested land, 86%, is higher than estimates of previous analyses (at least 56% from 1990 to 2005, Koh Gaveau et al 2021; and 38% in a multi-period metastudy, Meijaard et al 2020). This is because our forest cover base map (Hansen et al 2013) included more areas of tree cover (e.g. secondary forests; non-forest tree cover) than forest cover base maps used in other studies (e.g. Margono et al 2014, Ministry of Environment andForestry 2015), and because our definition of conversion allowed for a longer time increment between forest cover and plantation establishment than some other studies (up to 15 years in our study rather than up to 10 or 5 years in Gunarso et al 2013, Austin et al 2017. More conversion of forest and non-forest to oil palm plantation between 2000 and 2015 occurred, all else equal, on land that was flatter, lower, nearer to roads, nearer to provincial capitals, and had greater estimated potential revenue from palm oil (tables SI2 and SI3). Conversion to oil palm was less likely within strict and multiple-use protected areas. Conversion to oil palm was more likely within lands designated as oil palm concession (kebun) or timber concession (HTI) circa 2010. Within lands designated as logging concessions (HPH), forest-to-palm conversion was more likely, but non-forest-to-palm conversion was less likely. Our finding that lower slope, lower elevation, and less distance to large cities were determinants of oil palm expansion is consistent with Shevade and Loboda (2019).
We estimated the price elasticity of forest-topalm conversion to be 0.13, meaning that every 1% increase in the price of palm oil was associated with a 0.13% increase in forest-to-palm conversion. Meanwhile price elasticity of non-forest-to-palm conversion was 0.17, meaning that every 1% increase in the price of palm oil was associated with a 0.17% increase in non-forest-to-palm conversion (table SI4). These numbers compare with an estimate by Gaveau et al (2021) that a 1% decrease in the price of palm oil was associated with a 1.08% decrease in industrial oil palm expansion in Indonesia from 2001 to 2019.
The price elasticity of supply for highdeforestation oil palm cropland was equal to the price elasticity of forest-to-palm conversion, i.e. 0.13, meaning that every 1% increase in the price of palm oil was associated with a 0.13% increase in high-deforestation oil palm cropland. The price elasticity of supply for low-deforestation oil palm cropland was smaller than the price elasticity of nonforest-to-palm conversion because a substantial area of low-deforestation oil palm cropland had already been established by the year 2000; this area was unaffected by later price changes. As a result, we estimated the price elasticity of supply of low-deforestation oil palm cropland to be 0.033, meaning that every 1% increase in the price of palm oil was associated with a 0.033% increase in low-deforestation oil palm cropland. Combined, the price elasticity of supply for all Indonesian oil palm cropland was 0.086, meaning that every 1% increase in the price of palm oil was associated with a 0.086% increase in oil palm cropland.
A European ban on consuming highdeforestation palm oil decreased exports of highdeforestation palm oil from Indonesia and Malaysia to Europe by 99% by design of our experiment. We simulated that a European ban would have increased exports of low-deforestation palm oil from Indonesia and Malaysia to Europe by 31.1% (table 1). About half (52%) of the high deforestation trade flow that would otherwise have been exported to Europe shifted to other regions (figure 1), resulting in a 9.5% decrease in the volume of high-deforestation palm oil exported from Indonesia and Malaysia and a 1.4% increase in the volume of low-deforestation palm oil exported from Indonesia and Malaysia (table 1). The global share of palm oil exports from Indonesia and Malaysia consumed by Europe would have fallen from 21% to 14% (table SI5).
A European ban on consuming highdeforestation palm oil would have produced an 8.9% price premium for low-deforestation palm oil, composed of a 7.5% price decrease for highdeforestation palm oil and a 0.7% price increase for low-deforestation palm oil (table 1). The quantity of high-deforestation palm oil produced in Indonesia would have decreased by 6.3%, while the quantity of low-deforestation palm oil would have increased by 0.3% (table SI6). Global production of palm oil would have decreased by 1.32%, with shifts in production among oilseeds resulting in a decrease in global production of edible oils of 0.44% (table SI7).
These changes in the market for palm oil would have resulted in 320 609 hectares (5.03%) less conversion of forest to oil palm plantations in Indonesia from 2000 to 2015 (21 374 ha yr −1 ), with a resulting reduction in associated emissions of 315.9 million tCO 2 (21.1 million tCO 2 yr −1 ; 5.86%) (table 2). This would have reduced Indonesia's total deforestation of 20.0 million hectares from 2000 to 2015 by 1.60%, and its total emissions from deforestation of 16.6 billion tCO 2 by 1.91% (figure 2). Considering 11.2 billion tCO 2 in emissions from sectors other than land-use change and forestry (Climate Watch 2020), this would have reduced Indonesia's emissions from all sources by 1.14%. The deforestation emission reductions of 1.91% from a European ban would have been equivalent to a hypothetical mandatory Indonesia-wide price on carbon emissions from deforestation of $0.81/tCO 2 , or a voluntary price of $1.57/tCO 2 in 2013 USD (table 2).
A ban on high-deforestation palm oil by a combined bloc of Europe and the United States would have reduced deforestation by 365 015 hectares (24 334 ha yr −1 ; 5.72% of deforestation associated with conversion of forest to oil palm plantation; and 1.83% of all deforestation) and reduced emissions

Discussion
European demand-side restrictions on highdeforestation palm oil from 2000 to 2015 would have resulted in a moderate price premium for low-deforestation palm oil (8.9%), but only minor impacts on deforestation (−1.60%) and emissions from deforestation (−1.91%) in Indonesia (table 2). The impact of demand-side restrictions would be small because: about half (52%) of the highdeforestation palm oil that would otherwise have been exported to Europe would be absorbed by increased consumption in non-participating countries, including domestic consumption in Indonesia; the price elasticity of supply of high-deforestation oil palm cropland is small (0.13), meaning relatively large changes in price result in relatively small changes in quantity; and conversion to oil palm was responsible for only 32% of deforestation in Indonesia. A benchmark policy of a mandatory Indonesiawide price on deforestation emissions would have been as effective at $0.81/tCO 2 in 2013 USD (table 2). This price is one-to-two orders of magnitude lower than carbon market prices circa April 2021 of $9/tCO 2 in the Regional Greenhouse Gas Initiative, $16/tCO 2 in Korea, $18/tCO 2 in California and Quebec, $26/tCO 2 in New Zealand, and $50/tCO 2 in the European Union (World Bank 2021).
More consumer countries implementing demand-side restrictions would result in greater emission reductions. By following Europe's lead, other countries would avoid first-mover costs of designing policies. However, even under very optimistic assumptions about participation by consumer countries, demand-side restrictions on highdeforestation commodities would still have only modest effects on deforestation and emissions; e.g. a ban on high-deforestation palm oil by all countries outside Indonesia and Malaysia from 2000 to 2015 would have reduced deforestation in Indonesia by just 3.77% and emissions from deforestation in Indonesia by just 4.49% (table 2).
As modest as our estimates are, they likely overstate the impacts of demand-side restrictions on palm oil for several reasons. Actual demand-side policies have been narrower in scope than we modeled, covering only the portion of palm oil used in biofuels (e.g. Norway), or only the portion of palm oil used in public procurement (e.g. Netherlands, France), or only the portion of commodity-associated deforestation that is illegal (e.g. US, UK; Dos . On the supply side, the definition of deforestation used by RSPO is also narrower in scope than we modeled, covering deforestation only after 2005 rather than 2000, and only primary forest rather than all forest. Importantly, we assumed that restrictions on palm oil use would be fully effective. We did not consider, for example, the ability of producers of highdeforestation commodities to circumvent restrictions by disguising their product as low-deforestation (e.g. Reuters 2020). We also did not consider implementation or transaction costs of verifying or certifying palm oil as low-deforestation, which can be considerable and can pose disproportionate challenges to smallholders (Brandi et al 2015).
Conversely, there are also reasons our estimates could understate policy impacts. Many of the demand-side policies that have been passed or are under consideration cover multiple forest-risk commodities, and thus would have an increased effect on deforestation. Furthermore, the presence of 'stickiness' in commodity supply chains (Dos Reis et al 2020) could result in smaller geographical shifts in consumption, and thus larger changes in prices and greater reductions in deforestation.
Several limitations with our data and model introduce uncertainty to our estimates of reductions in deforestation and emissions from deforestation. As elaborated in the supplementary information, these include the inclusion of tree cover other than forest in our land-cover change data; the exclusion of many smallholder plantations from our oil palm coverage data; and the inability to disaggregate Indonesia from Malaysia in our global trade model. Future data improvements could lead to more accurate estimates. However, we do not expect that improving these data would alter our central finding that the effects of trade restrictions would be small due to the three factors we have identified.
The effects of demand-side policies on deforestation in other regions, commodities, and time periods will depend on the same three factors: share of global consumption comprised by the countries participating in a ban; price elasticity of supply for the high-deforestation variety of the commodity; and share of deforestation comprised by the commodity. For example, the effects of European demandside restrictions on oil palm-driven deforestation in Indonesia would likely be smaller in the present (e.g. 2021-2030) than for the earlier 2000-2015 time period that we analyzed. This is because two of the three factors influencing impact have decreased: the European share of global palm oil consumption has declined (USDA 2019), and the share of Indonesia's deforestation due to oil palm expansion has diminished (Austin et al 2019, Gaveau et al 2019. If demand-side restrictions are to succeed in substantially reducing deforestation, therefore, it would have to be through non-price pathways that are beyond the scope of our analytical model. These could include social pathways (e.g. raising consumer awareness, imposing stigma, changing social norms); industrial organization pathways (e.g. commodity producers or supply chain intermediaries applying European standards to exports to all markets and not just exports to Europe); or synergistic effects with other domestic, bilateral, or corporate forest conservation actions. Conversely, there are also nonprice pathways that could increase deforestation. For example, we did not consider whether import restrictions constitute protectionism for farmers in wealthy countries in violation of World Trade Organization rules. Backlash to perceived protectionism has resulted in some tropical countries enacting compensatory domestic price support policies, e.g. biofuel mandates in Indonesia (Elliott 2015). Such non-price effects, both positive and negative, could have a larger impact on land-use changes than direct price effects.
Our findings imply that demand-side restrictions on high-deforestation commodities alone are insufficient to substantially counteract distal deforestation. Consumer countries seeking to reduce deforestation in commodity-producing countries should support more direct forest and climate protection policies such as carbon payments, which even at a very low price could meet or exceed the effects of demand-side restrictions.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.

Acknowledgments
J B and O A are grateful for funding from the Norwegian Agency for Development Cooperation. J B is grateful for funding from the Ann and Tom Friedman Fellows for Science program. T H acknowledges funding from the National Science Foundation (CBET-1855937, SES-1463644, OISE-2020635 and DEB-1924111) and U.S. Department of Agriculture, National Institute of Food and Agriculture (IND01053G2). F T acknowledges partial funding by the National Biodiesel Board and also by U.S. Federal Aviation Administration Office of Environment and Energy through ASCENT, the FAA Center of Excellence for Alternative Jet Fuels and the Environment, Project 001 through FAA Award Number 13-C-AJFE-PU under the supervision of Anna Oldani. Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the FAA. We are grateful to William Baldwin-Cantello, Fanny Gauttier, David Kaimowitz, and two anonymous referees for helpful comments on earlier versions of this manuscript. Much of this work was completed while J B was at Earth Innovation Institute.