Assessing the potential additionality of certification by the Round table on Responsible Soybeans and the Roundtable on Sustainable Palm Oil

Our spatial unit of analysis is all countries with RTRS (n = 8 regions) and RSPO (n = 12) certified production from January 2013 through December 2014, as well as countries that had no certified production but were among the top ten global producers in terms of soy (n = 2) and oil palm (n = 2) in 2014 [1]. The countries considered in our analysis composed >95% of 2014 global soy and oil palm production. We also apply a subnational approach in Brazil and Indonesia, where subnational data on the focal crop and total agricultural areas are available, to examine the extent to which spatial scale influences our conclusions. In this analysis, we exclude administrative units with no Indonesian oil palm area in 2010 and no Brazilian soy production in 2014. Since means and significance tests presented in the results are weighted by the proportion of total global certified area in a country, these statistics only consider countries with some certification; unweighted means are presented in tables. Statistical tests were performed in R.

Potential additionality with respect to land cover (A, equation (1)) is a function of: (1) required protection of natural vegetation under the certification standard relative to existing policies and enforcement (relative stringency, S); and (2) business as usual land cover dynamics (BAU):

$$\begin{eqnarray}&&A=S\;\times \;{\rm{BAU}}.\end{eqnarray} \tag{ 1 }$$

Stringency is a basic requirement for additionality; if a standard has zero stringency, then additionality will also be zero. The relative stringency (S) of any standard in a particular region depends on the standard's absolute stringency (the natural vegetation protection required by standard in all regions, s), land cover policies in each region (p) and the degree of enforcement of such policies in each region (e):

$$\begin{eqnarray}&&S=1-p\times e/s,\\ &&S=0\;{\rm{where}}\;s\lt p\;{\rm{or}}\;s=0.\end{eqnarray} \tag{ 2 }$$

We multiply $p\times e$ since enforcement is a necessary complement to any policy. We divide the combined numerator by absolute standard stringency (s) to examine how close existing policies duplicate standard protections. Since relative stringency (S) diminishes as $p\times e$ approaches s, this fraction is subtracted from 1.

To evaluate p and s we use three criteria: clear cutting restrictions (X); riparian buffer requirements (Y); and endangered species protections (Z) on agricultural properties (table S1). The extent to which a policy satisfies these three criteria is ranked on a scale of 1 to 3, with 1 being the lowest. For example, for X a score of 1 indicates no restrictions on clear cutting, 2 indicates partial restrictions, and 3 indicates full restrictions (no clearing allowed). Values of 1, 2, and 3 are weighted according to the type (t) of restriction/rule: specific-mandatory (full weight), procedural-mandatory (75% weight), or specific-voluntary (75% weight). Since clear cutting restrictions are most directly related to forest protection, this category is weighted most heavily (three times the weight of Y and Z). The raw score is then scaled between 0 to 1 by subtracting the minimum score of 5 and dividing the total by 10. The final equation for calculating the raw stringency score is as follows:

$$\begin{eqnarray}&&p,s=(3\times {X}_{t}+{Y}_{t}+{Z}_{t}-5)/10.\end{eqnarray} \tag{ 3 }$$

To capture different enforcement levels in each region (e), we combine the Transparency International Corruption Perception Index (www.transparency.org/cpi2014/results) and Forest Cover Monitoring Capacity metrics produced by by Romijn et al [42]. We convert each to a 0–1 scale, and multiply the two metrics to generate e.

To determine the fraction of tree cover loss attributable to soy or oil palm expansion (BAU) we assess the proportion of total tree cover loss (L_tot, ha) due to expansion of all crops (L_crops, ha), and the contribution of soy or oil palm expansion (E_focal, ha) to total crop expansion, (E_tot, ha). To calculate the contribution of crop expansion to tree cover loss, we consider two scenarios, similar to the approach applied by Koh et al [43]. Our low bound estimate assumes that crop expansion first occurs in already-cropped lands, represented by crops that experienced net contraction during the time period of interest, followed by clearing of natural ecosystems. Our high-bound estimate assumes that crops first expand into areas of natural vegetation, followed by existing croplands. We compute the mean of low- and high-bound estimates to represent L_crops. This metric does not account for crop expansion into land uses besides forests and croplands (e.g., pastures, grasslands), and is therefore most robust in regions dominated by forests and croplands (e.g., Indonesia). We assume BAU of zero in cases with focal crop contraction or net tree cover gain:

$$\begin{eqnarray}&&{\rm{BAU}}=\;{E}_{{\rm{focal}}}/{E}_{{\rm{tot}}}\times {L}_{{\rm{crops}}}/{L}_{\mathrm{tot},}\\ &&{\rm{BAU}}=0\;{\rm{where}}\;{E}_{{\rm{focal}}}\leqslant 0\;{\rm{or}}\;{L}_{{\rm{tot}}}\leqslant 0.\end{eqnarray} \tag{ 4 }$$

Ideally, BAU would represent the fraction of total natural vegetation loss converted (directly or via displacement) to the crop in question [44]. Yet, such conversion data are rarely available [18].

For subnational analysis, individual metrics of crop expansion and contraction across all crop types were unavailable; we were only able to obtain data on net cropland change. Therefore, we calculate BAU as the average of low and high estimates of crop contribution to deforestation (C_focal, ha):

$$\begin{eqnarray}&&{\rm{BAU}}=\;{C}_{{\rm{focal}}}/{L}_{{\rm{tot}}},\\ &&{\rm{BAU}}=0\;{\rm{where}}\;{C}_{{\rm{focal}}}\leqslant 0\;{\rm{or}}\;{L}_{{\rm{tot}}}\leqslant 0.\end{eqnarray} \tag{ 5 }$$

The high bound estimate of C_focal is focal crop expansion, or total deforestation if focal crop expansion is greater than forest loss. If net cropland change is less than zero, the low bound estimate is the sum of net cropland change and focal crop expansion. If net cropland change is greater than zero and focal crop expansion is greater than net cropland expansion, the low bound estimate is net crop expansion. Otherwise, the low bound estimate is equal to the high bound estimate.

We compute regional adoption rates (U) by comparing certified crop area within a region (C_cert) to total crop area in that region (C_all):

$$\begin{eqnarray}&&U={C}_{{\rm{cert}}}/{C}_{{\rm{all}}}.\end{eqnarray} \tag{ 6 }$$

Finally, we multiply adoption (U) and additionality (A) to compute the potential impact of certification (I) in a given region:

$$\begin{eqnarray}&&I=U\times A.\end{eqnarray} \tag{ 7 }$$

To quantify regional distribution of roundtable certification (${C}_{{\rm{cert}}})$ we downloaded tabular summaries of certified area and production from roundtable websites (www.responsiblesoy.org, www.rspo.org). Data on total crop area in each region (${C}_{{\rm{all}}})$ were obtained from FAOSTAT [1]. In most cases, we examine adoption in 2014, since this is the last year for which certification data and crop area are both available. Since the United States had RTRS certified production only in 2013, in this case we compare 2013 overall production to 2013 certified production. For subnational analysis, we use publically available audit reports to identify the location of certified production at the province (Indonesia) or state (Brazil) level for 2014, and compare this to total oil palm area in 2010 (Indo-DAPOER), or total soy area in 2013 (IBGE). While these dates do not align, more temporally resolved data are not available to refine this analysis.

To examine whether certification is targeting smallholders (T), we compare mean certified farm size in 2014 $({F}_{{\rm{cert}}})$ to mean country farm size $({F}_{{\rm{all}}})$:

$$\begin{eqnarray}&&T={F}_{{\rm{cert}}}/{F}_{{\rm{all}}}.\end{eqnarray} \tag{ 8 }$$

Audit reports contain data on individual farms for individual and multi-site certifications. For RTRS group certifications, when the details for each individual farm are not listed, average farm sizes are provided. For RTRS multi-site certifications, we divide total certified area by the number of certified sites to obtain mean farm area. We include only productive area rather than total farm area, which may include conservation areas. For RSPO grower (i.e., corporate) certifications, we assume that planted area is equivalent to farm size. For RSPO group (i.e., smallholder) certifications, we divide the total planted area by the number of group members to generate farm size. Mean country farm sizes were collected from FAO's 2000 World Census of Agriculture [45] and the Brazilian Institute of Geography and Statistics' 2006 Agricultural Census [46].

To define BAU we collate data on tree cover loss and crop expansion in the 3 year period prior to initial issuance of certificates in each country or subnational region. If an administrative area did not contain a certified farm or plantation in 2014, we calculate loss and expansion from 2010 to 2013. In Indonesia, subnational oil palm extent was available only through 2010. Therefore, we estimate BAU from 2007 to 2010 in cases with no certification, or when initial certification occurred after 2010. Country crop areas were obtained from FAOSTAT [1]. For subnational Brazil analysis, we obtained soy and total perennial and annual crop area from the annual Municipal Agricultural Surveys from the Brazilian Institute for Geography and Statistics [47]. In Indonesia, we obtained province-scale oil palm area from Indo-DAOPER [48], and annual crop data from Indonesia's Central Bureau of Statistics [49]. To examine the sensitivity of potential additionality to the BAU baseline we also present results for the 2 to 7 year periods prior to initial certification.

Tree cover change data were derived from Hansen et al [21]. We consider 30% tree cover in year 2000 to represent natural forest or savanna vegetation, while reduction of tree cover to ∼0% constitutes tree cover loss. This approximation of natural vegetation is prone to both over and underestimation depending on the ecosystem, since it excludes savanna and grassland ecosystems, but includes tree plantations [50].

RSPO and RTRS are being adopted in regions with similar potential additionality with respect to native vegetation conversion (RSPO—weighted mean A = 0.14; RTRS—weighted mean A = 0.15, table 1). These differences are not significant (weighted linear regression, p = 0.73).

RTRS and RPSO have identical standard stringency with respect to native vegetation conversion (table S2). Standard stringency (s) exceeds policy stringency for native vegetation conversion (p) in all regions included in the study (table S2). The difference between s and p is smallest in the Brazilian Amazon, China, and Uruguay, where a combination of policies strictly protects native forests from conversion to agriculture and also safeguards riparian zones. However, additional protections for biodiversity habitat in these countries are minimal [51–53]. Argentina (Cordoba, La Pampa, and Santa Fe), India (Madya Pradesh), Indonesia, Malaysia, and Paraguay have some combination of moderate to high restrictions on clear cutting, and low to moderate protections for riparian buffers. In other regions legal frameworks typically prohibit clearing forests in state-controlled protected areas and forest estates, but allow clearing on private land, and/or fail to protect riparian buffers. Few countries have strong habitat protections for biodiversity with the exception of Canada (Ontario) and the United States. Yet, these two regions have weak protections for forest conversion on private agricultural properties.

RSPO countries have slightly higher enforcement levels (weighted mean e = 0.41) than RTRS countries (weighted mean = 0.37, table 2). Canada and the United States rank highest in terms of enforcement (e > 0.70), with very good monitoring capacity and low corruption. Honduras, Nigeria, and the Solomon Islands all score <0.1 on our enforcement index. As a result, RSPO has lower relative stringency levels than RTRS (RSPO—weighted mean S = 0.79; RTRS—weighted mean S = 0.85, table 1).

Across oil palm countries, our proxy metric suggests that oil palm expansion composed 0%–32% of net tree cover loss, while soy expansion totaled 0%–24% of tree cover loss (table 3). Thus, RSPO has slightly greater BAU (weighted mean = 0.18) than RTRS (weighted mean = 0.17). In the soy sector, Brazil receives the highest BAU score (0.20) because soy accounted for 24% of all crop expansion, and 63%–100% of tree cover loss is associated with crop expansion either directly or through displacement (table 3, figure 1). Soy area contracted in China, so this country receives a BAU score of zero. For oil palm, Indonesia had high BAU (0.23), with 25% of total crop expansion attributed to oil palm, and 84%–100% of tree cover loss potentially due to crop expansion. During the BAU period from 2006 to 2009, Indonesia had high levels of oil palm expansion (1.3 M ha) and tree cover loss (4.4 M ha). Malaysia also stands out with a BAU of 0.16. The Brazilian Amazon had negligible oil palm expansion compared to tree cover loss rates, with the lowest non-zero BAU in the oil palm sector.

Potential for land cover additionality from adopting certification varies greatly among countries. Paraguay and Argentina have the greatest potential additionality when adopting RTRS because of high rates of forest to cropland conversion (BAU) and high relative stringency of certification. In contrast, Uruguay has relatively low additionality compared to its BAU score due to extremely low relative standard stringency. Considering countries with RSPO certified production, only Indonesia, Malaysia, and the Solomon Islands have potential additionality >0.10, driven largely by high rates of forest to cropland conversion. China and the United States (RTRS), and Ghana and Nigeria (RSPO) have zero potential additionality due to declining area under soy and oil palm cultivation.

In 2014, RTRS-certified production totaled 1.4 M tons ((table 4) or <1% of global soy production. Brazil accounted for ∼50% of total certified soy area, followed by Argentina with 33%. By 2014, RSPO had certified 10.6 M tons or ∼20% of global crude palm oil production (figure 2), with ∼90% of the certified area concentrated in Indonesia and Malaysia. While RTRS adoption is extremely low in all countries (<1%), RSPO adoption rates range from 1.1% in Ecuador to 74% in Papua New Guinea.

Zoom In Zoom Out Reset image size

Examining relationships between potential land cover additionality and adoption by region, we note that certification has been adopted in all countries with higher levels of additionality >0.10–0.15, (table 1, figures 3 and 4). At lower levels of additionality zero adoption occurs more frequently. Due to high adoption rates, RSPO has much higher potential impact (weighted mean I = 0.029) than RTRS (weighted mean I = 0.0012). This difference is significant at a 99% confidence level (weighted linear regression, p < 0.01, table 1). RSPO impact is particularly high in Papua New Guinea and the Solomon Islands.

Zoom In Zoom Out Reset image size

Certification is predominantly adopted by large producers (table 4, figure 5). The median farm size for RSPO certified producers is 8427 ha compared to a median farm size of 3.6 ha in oil palm producing countries. The median farm size for RTRS certified farms is 2 512 ha compared to a median farm size of 104 ha in the soy producing countries. Only India and Uruguay have certified farm sizes that approach country-wide farm sizes. In RSPO countries, smallholder targeting tends to be lower in places with higher potential land cover additionality (figure 6). In RTRS countries targeting and additionality are uncorrelated.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To assess the sensitivity of our findings to the choice of BAU period, we generate BAU baselines from 2 to 7 years prior to initial certification. We find that relative additionality between countries depends strongly on the choice of baseline years (figure S1). For RSPO, the coefficient of variation (CV) across all potentially additionality scores (n = 5) by country ranges from 9 (Solomon Islands) to 245 (Ghana). For the RTRS, the CV ranges from 8 (Paraguay) to 146 (United States).

We present results from our subnational analysis in supplementary tables 3–4. In Brazil, existing policy stringency and BAU forest to cropland conversion rates vary substantially between states. Although states within the Legal Amazon have substantially more restrictive existing policies than states in other ecological biomes, Mato Grosso and Tocantins (both within the Legal Amazon) have very high BAU scores >0.70 and therefore have relatively high potential additionality >0.30. Nevertheless, high potential additionality is not associated with RTRS certification in Brazil, as two states (Piauí and Paraná) with the highest levels of potential additionality had zero certification. In Indonesia, subnational variation is driven entirely by BAU conditions, since policy stringency is identical across all provinces. In contrast to the Brazilian case, we find that high levels of potential additionality are associated with RSPO certification; the six provinces with the highest potential additionality >0.25 all contained some certified plantations.

When examining RTRS and RSPO certification patterns, we find that countries with higher potential additionality >0.10–0.15 typically have some certification adoption. This result is somewhat surprising given that low levels of existing governance and high crop expansion into natural ecosystems likely inflate the costs of meeting certification requirements [33]. However, benefits of certification may also be greater in regions with high potential additionality. Multinational oilseed traders operating in regions with rapid deforestation are under substantial pressure from environmental groups to eliminate deforestation from their supply chains [54, 55]. Producers in such high-risk regions may seek certification to differentiate themselves and reduce the likelihood of market exclusion [19, 56]. Yet, our highly aggregated analysis is unable to examine the precise locations of certified and non-certified producers, nor their contributions to natural vegetation clearing. Even within regions where crops are expanding into natural ecosystems, producers seeking certification may be those who cleared land prior to RTRS and RSPO cutoff dates for forest clearing, and their costs of meeting the certification requirements would be relatively lower.

RSPO certification has been adopted at much higher rates than RTRS. While this effect is likely related to the age of these roundtables (RSPO issued its first certificates in 2008, RTRS in 2011), elevated adoption in the oil palm sector may also be explained by differences in soy and oil palm supply chains. Soy production occurs on thousands of individual farms, and each producer has little or no brand identity [57, 58]. If soy traders and retailers want to reduce reputational risk associated with tropical deforestation, they can source more soy from temperate regions. In contrast, palm oil plantations are frequently controlled by large multinational corporations, for whom brand identity and deforestation risk are important [59]. Moreover, the vast majority of palm oil is grown in just two tropical countries with high deforestation rates (Indonesia and Malaysia) [60]. Therefore, oil palm companies must directly address tropical deforestation on their own lands in order meet sustainability demands. The high rate of RSPO adoption is particularly remarkable given oil palm's high level of processing, since product visibility is expected to influence certification uptake [12].

High levels of adoption may be more important than high additionality for conserving natural ecosystems. Due to uniformly low adoption rates, RTRS is likely to have minimal effect on land cover dynamics despite adoption in regions with similar potential additionality. For example, soy production is increasing rapidly in the tropical Chaco forests of Argentina as investors flee more stringent environmental regulations in other regions [20, 61]. Here, RTRS could have large impact if adopted at a wide scale, yet current RTRS adoption stands at <1% of Argentina's soy area. In contrast, RSPO may be having a substantial effect especially in Papua New Guinea (93% adoption), despite medium potential additionality in this region.

Based on large differences between mean certified producer size and mean country farm size, our analysis shows that neither roundtable is successfully targeting small farmers. An exception is India, where most certified farms are <2 ha due to the strong presence of social NGOs aimed at improving livelihoods for small farmers [62]. Inability to reach small farms is a common failure among certification programs [27, 33–36], due to high fixed costs and complexity [30, 37, 38]. One exception is Fairtrade International, which aims to increase equity across all global commodity chains and traditionally has offered preferential certification to small farms [63, 64]. Restricting RTRS and RSPO certification access to any group of producers would not support the goal of protecting natural ecosystems through certification. Yet, smallholders are playing an increasing role in land cover change in both Brazil [40] and Indonesia [65] and should therefore be included in these certifications, despite the fact that large capitalized producers account for the vast majority of deforestation across tropical soy and oil palm sectors [40, 65, 66]. Both roundtables now offer group certification and smallholder funding programs [67], which could lead to higher smallholder certification rates via lowered transaction costs and informational barriers [68–70]. Such programs have already been implemented by organic certifiers and the Sustainable Agriculture Network Certification, with some success in increasing smallholder participation and benefits [71, 72].

RTRS and RSPO certifications are being adopted primarily in less-developed countries (figure 4). This pattern is distinct from dynamics observed in forest and fish certification programs, which tend to have higher adoption rates in developed countries [27, 73, 74], but consistent with coffee, cocoa and banana certification programs that are concentrated in tropical countries. In the soy sector, the United States and Canada produce 37% of global soy [1] but <2% of RTRS certified soy. Low RTRS adoption in Canada and the United States might be related to higher costs of adoption relative to Brazil and other South American countries, or to reduced targeting by NGOs focused on tropical ecosystem preservation. RSPO adoption in less-developed countries is expected, since oil palm is only grown in the tropics.

While RTRS and RSPO standards are substantially more stringent than land-cover policies in most soy and oil palm producing regions, certification standards do not extend full protection to non-primary-forest vegetation types, such as secondary forest, savannas and grasslands. RTRS defines native forest as areas of native vegetation >1 ha with canopy cover >35% and where >9 trees per ha reach 10 m in height, including forests in the humid tropical Amazon basin, the Atlantic Forest biome, and arid Chaco and Chiquitano regions of South America. This definition may fail to protect non-forest ecosystems and secondary forests. RSPO defines primary forest as forest that has never been logged and that has developed following natural disturbances and processes, or one that is used 'inconsequentially' by indigenous communities. Since Southeast Asia's extensive logged forests and agroforests do not qualify as primary forest, converting these land covers could be permissible by the RSPO. Thus, RTRS and RSPO relative standard stringency could be greatly increased by including land covers that are largely unprotected in soy and oil palm producing regions.

Several uncertainties limit the empirical conclusions of our study. While the additionality metric was developed from a critical reading of the literature, results depend on the metric's functional form and predictor variables, as well as weights given to each variable. Leakage is not considered in our analysis, but could reduce additionality through displacement effects. The broader effects of certification programs, such as induced changes in national policies, are not considered here.

Using countries as our basic units of analysis allows us to present an initial global overview of the potential additionality of roundtable certifications with respect to land cover change. However, our highly aggregated analysis is unable to account for variation in land cover dynamics, producer behavior, and policy enforcement at the farm or plantation scale. As evidenced by our subnational results, trends identified at the aggregate scale may not match those at finer scales. By applying a national scale analysis using a proxy dataset representing natural ecosystems, we also generate high uncertainty around rates of natural ecosystem conversion to the target crop. Since the 30% tree cover loss threshold excludes native grasslands and some savannas, and includes highly managed tree plantations, applying this threshold to regions with high rates of grassland or savanna conversion to croplands will underestimate BAU; applying it to regions with tree-like land cover conversion will generate artificially high BAU levels. Thus, our metric is problematic in regions such as the Brazilian Cerrado, rubber plantations in Indonesia, and North American grasslands [50]. Subnational analyses using refined land cover datasets would represent significant improvements to this initial evaluation. Given that our potential additionality metric is highly sensitive to the selection of BAU baseline years, we also recommend that future analyses pay careful attention to the time period over which BAU is calculated. To aid comparisons between studies, authors should minimize differences in baseline years.

Since the area certified by RTRS is quite small compared to global soy area, small increases in adoption could substantially alter the adoption and additionality patterns identified here. The positive trend between additionality and adoption could intensify if RTRS and RSPO follow the patterns of forestry certification, with increasing adoption in temperate-developed regions over time, since most temperate countries included in our analysis have very weak native vegetation protections on agricultural properties. However, further refinement of the BAU vegetation conversion metrics might also reveal that temperate-developed regions have experienced very low rates of direct forest conversion to cropland in recent years. In this case, greater adoption in temperate-developed countries would weaken the positive additionality-adoption relationship.

Although previous studies have identified the potential for corruption among certifiers [64], in this study we do not measure the degree of enforcement of RTRS and RSPO standards. Interviews with auditors and analysis of land cover and management practices before and after certification are required to assess standard enforcement. If data on standard enforcement were available, degree of enforcement could be multiplied by standard stringency to further refine our additionality metric.

Critically, results with respect to additionality of land cover change should not be taken as an indicator of the contribution of RSPO and RSTS to societal well-being. RSPO and RTRS are subject to intense criticism because the interests of groups directly affected by the expansion of certified crops (i.e., traditional land users) and by the standards themselves (i.e. smallholders and landless workers) have not been fully incorporated in the standard-setting procedures [12, 75–78]. Our analysis does not examine these outcomes.