Where are commodity crops certified, and what does it mean for conservation and poverty alleviation?

Voluntary sustainability standards have expanded dramatically over the last decade. In the agricultural sector, such standards aim to ensure environmentally and socially sustainable production of a variety of commodity crops. However, little is known about where agricultural certification operates and whether certified lands are best located for conserving the world's most important biodiversity and benefiting the most vulnerable producers. To examine these questions we developed the first global map of commodity crop certification, synthesizing data from over one million farms to reveal the distribution of certification in unprecedented detail. It highlights both geographical clusters of certification as well as spatial bias in the location of certification with respect to environmental, livelihood and physical variables. Excluding organic certification, for which spatial data were not available, most certification of commodity crops is in tropical regions. Certification appears to be concentrated in areas important for biodiversity conservation, but not in those areas most in need of poverty alleviation, although there were exceptions to each of these patterns. We argue that the impact of sustainability standards could be increased by identifying places where it would be most beneficial to strengthen, consolidate, and expand certification. To achieve this, standards organizations will need to undertake more rigorous collection of spatial data, and more detailed analysis of their existing reach and impacts, with attention to potential trade-offs between different objectives. Efforts to promote spatial prioritization will require new partnerships to align specific conservation aims with the interests and capabilities of farmers.


51
Improving the environmental and social sustainability of agriculture is an ongoing challenge 52 worldwide (Tilman and Clark, 2015). Governments have responded to this challenge by developing 53 legislation and initiatives such as agri-environment schemes (Batáry et al., 2015). Alongside these 54 government-led initiatives, the work of multiple stakeholders has led to the creation and promotion 55 of voluntary sustainability standards systems, also referred to as certification schemes (Potts et al., 56 2014;Rueda et al., 2017). These standards typically define the practices of sustainable agriculture, 57 and identify actions producers must take to be certified as environmentally and socially responsible 58 (Milder et al., 2015). Over the last decade, there has been a near-exponential increase in area 59 managed under certification (Tayleur et al., 2016). Certification is often promoted as a way for 60 individual consumers to make more ethical purchasing decisions (Dauvergne and Lister, 2010). It is 61 also proposed as a way to mitigate negative impacts of commodity production and improve the 62 wellbeing of farmers and farm workers in the developing world (Lenzen et al., 2012). Many multi-63 national companies now use certification to help achieve and demonstrate progress towards public 64 sustainability commitments (Dauvergne and Lister, 2012;Levin and Stevenson, 2012). Land under 92 What influences the spatial distribution of certification?

93
To the extent that spatial targeting of certification can be said to have occurred to date, it has largely 94 been a by-product of the management of specific supply chains (Garrett et al., 2016;Getz and 95 Shreck, 2006;Renard, 2010;Vellema et al., 2015). Companies that have committed to responsible 96 practices have worked to ensure that those producing the agricultural commodities they use are 97 certified. Some of these efforts have been reactive, responding to civil society campaigns, regulatory 98 requirements, or anticipation of campaigns or regulations. Others have been more proactive, aiming 99 to increase the security or quality of commodity supply, or reputational benefits to a company's 100 brand. Such efforts reflect to some extent the imperative to target certification to places of greatest social and environmental risk. For instance, civil society campaigns have highlighted egregious 102 instances of deforestation and infringements of community rights. Another mechanism is the use of 103 certification as a policy proxy by governments. For instance, the US state of Pennsylvania obtains FSC 104 certification for its state forests, and some government procurement policies preference or require 105 responsibly sourced products, including certified products (Steering Committee of the State of 106 Knowledge Assessment of Standards and Certification, 2012). Although indirect and often reactive, 107 both supply chain commitments and procurement policies therefore offer some opportunities to 108 effect spatial targeting. The creation of sustainability standards focused on specific crops implicated 109 in environmental and social problems has also resulted in spatial targeting at a very coarse scale (it is 110 notable that all of the certification schemes for which we obtained data are concentrated in tropical 111 countries).

112
Despite these examples, there do not yet appear to have been coordinated strategic efforts to 113 systematically identify the places where the impact of certification could be greatest. There are 114 considerable opportunities to do so, to identify priorities for future civil society campaigns, 115 corporate efforts, and government interventions. Currently, at the country level, agricultural 116 certification has poor representation in the world's 31 poorest countries (those classified by the 117 World Bank as low income) and for staple crops of low export value (Tayleur et al., 2016).

118
Analogously, within the forestry sector, certification has been criticized for failing to protect tropical 119 forests that are most at risk, with the majority of certified wood coming from temperate developed 120 countries (Gullison, 2003). Without a more strategic approach to strengthening, consolidating, and 121 expanding agricultural certification, there is a risk that it may not reach those areas and producers 122 where the greatest additionality can be gained.

123
Spatial prioritization as a conservation and poverty alleviation tool 124 While global coverage of certification is still limited, its rapid uptake by producers of some of the 125 most environmentally-damaging commodity crops indicates its potential to contribute to conservation and development. Given sparse resources, certification, like other voluntary incentive 127 schemes, should be prioritized to where its introduction could have most additional beneficial 128 impact (Wünscher et al., 2008). One of the few studies to explore how well standards are targeted 129 found that adoption of two schemes (the Round Table on Responsible Soy (RTRS) and the 130 Roundtable on Sustainable Palm Oil (RSPO)) was better directed towards places where they could 131 reduce deforestation in some countries but less so in others, and that the standards were 132 disproportionately adopted by large producers rather than smallholders (Garrett et al., 2016). While 133 there has been some targeting of high-risk commodities for certification such as palm oil and 134 soybeans, little is known about whether certification reaches those areas of greatest conservation 135 and poverty alleviation need within the global ranges of these crops. Although the areas of greatest 136 need are not always those where certification can have most impact -because supporting 137 conditions for certification also vary, and alternative interventions may sometimes be more effective 138 -identifying such areas provides an initial basis for spatial targeting.

139
We aimed to: (1) develop the first detailed global map showing where certification is located,

140
synthesizing data from all of the main standards for which data were available; and (2) characterize 141 biodiversity and poverty in landscapes in which certification currently operates, globally, regionally 142 and within countries, using as case studies crops for which sufficient data exist. We use these 143 analyses to illustrate methods for identifying priority areas that could be targeted to maximize the 144 incremental benefits of improving, consolidating, and expanding certification, and outline how doing 145 so could increase the contribution of certification to global sustainability. We have assumed that the 146 expansion of certification has been too recent and limited to have yet had a detectable influence on 147 the biodiversity and poverty datasets we used, and our analysis should thus be interpreted as an aid 148 to priority-setting, rather than implying any causal influence of certification on these variables.

151
Obtaining spatial data on certified producers Data on the spatial location of certified farms were obtained through publicly available datasets and 153 via direct approaches to standards bodies (see Supplementary Materials for details). We sought data 154 from all major standards and codes of practice covering the certified commodity crops with the 155 highest levels of certification: banana, cocoa, coffee, cotton, tea, soybean, sugar, and palm oil (Potts 156 et al., 2014). The scope of the data search was not limited to any particular geography, but the 157 standards for which data were available operate primarily in tropical countries. Not all schemes 158 were able or willing to provide data (see Supplementary Materials for details). In some cases,

159
permission was granted only on condition that data were used in aggregate with other standards so 160 that the specific locations for individual schemes and producers could not be identified. To meet this 161 requirement our maps are at the resolution of 30 km × 30 km cells, after first standardizing all data 162 by converting them into point localities. The format of data available from each standard varied 163 considerably: while most were able to provide a coordinate for each certificate, a few schemes had 164 postal address data only. RSPO was the only standard that routinely collects polygon data outlining 165 plantation boundaries. Usable spatial data were not available for certified cotton, so this commodity 166 was excluded from further analyses.

167
Validation and standardization of spatial certification data 168 Several factors influenced the accuracy of the spatial data: 1) In some standards, multiple farms (e.g.

169
within a co-operative) are represented by a single certificate and coordinate, often referred to as a 170 'group certificate'; 2) Occasionally the coordinate for a certificate is associated with an 171 administrative office rather than the actual farm; 3) Some farms hold multiple certifications, e.g.

172
Rainforest Alliance/SAN and Fairtrade, but because spatial data are often imprecise, many certified 173 farms are small, and common identifiers are not used across standards, such overlaps cannot be 174 identified by spatial coincidence of points. We converted address data into point locations using the 175 ESRI Online World Geocode service which identified coordinates for 23% of all addresses entered.

176
We tested the sensitivity of our results to the inclusion of these data by repeating analyses with and without them. As certification patterns did not change significantly, we report the results including 178 the geocoded data.

179
To improve accuracy, we undertook a number of data cleaning steps. First we checked whether the 180 coordinate location corresponded to the country named in the accompanying metadata. Where 181 points were not located in the correct country, simple transformations (swapping latitude and 182 longitude, and hemisphere) were attempted. If this did not locate the coordinate in the correct 183 country, the point was discarded. Points that did not fall within the relevant crop growing area as 184 defined by our crop map (see Crop Cover) were also discarded. Excluding the geocoded address data 185 mentioned above, 93% of the data provided met these validation requirements.

186
To account for spatial inaccuracies and to protect the privacy of individual producers, we 187 summarized data using 30 km × 30 km grid cells created with the Fishnet tool in ArcMap 10.2 using 188 an equal-area projection. Each grid cell was classified as either containing certified land or not.

228
We chose three variables for which global spatial data at a fine-scale resolution were available. The 229 first was mean travel time to closest city of >50,000 people as calculated by Nelson (2008) in his 230 global map of accessibility, which we used as a proxy for market access. Secondly, we calculated the 231 mean percentage of the population in poverty for each 30-km grid cell using the global poverty map 232 created by Elvidge et al. (2009) from satellite data on night-time lighting. Finally we calculated mean 233 field size for each 30-km grid cell as calculated by Fritz et al. (2015). Field size has been shown to 234 correlate with farm size (Levin, 2006) and so we used grid cells with small field sizes as a proxy for 235 the presence of smallholder farmers.

236
Other variables

237
To investigate other factors that might characterize or influence the location of certified crops we 238 also calculated mean altitude and slope from the global SRTM dataset (USGS, 2004).

240
We used bootstrap resampling tests to examine patterns in those grid cells containing certification 241 versus those that did not, for a number of different variables. Because data were summarized at the 242 30-km scale, covariate values within each grid cell could not be attributed directly to certified farms, 243 so our tests examined how the local landscapes in which certification exists differ compared to non-244 certified landscapes, without implying causation. To run the resampling tests we first defined our 245 certified sample as all the 30-km grid cells containing certified farms for each crop. The test statistic 246 was then calculated as the mean of covariate values from the certified sample. To create our test 247 distribution we then obtained a random sub-sample without replacement from non-certified grid 248 cells of the same size as our certified sample and calculated the mean for the sub-sample. We 249 sampled without replacement as we were using a finite population. We weighted the probability of 250 a grid cell being included in the random sample by the proportion cover of the commodity crop of 251 interest. This allowed us to generate the values that might be expected for each variable if certification was located randomly within the distribution of each crop. We ran our resampling 253 routine using the wrswoR package in R (Müller, 2016). We repeated the resampling procedure 254 10,000 times in order to create our test distribution and then calculated the quantile in which our 255 test statistic fell. Our test was two-tailed as we had no prior expectation as to whether certified 256 values would be higher or lower than non-certified, so we considered anything below 2.5% or above 257 97.5% significant.

258
We carried out our bootstrap resampling tests at the global level to examine broad biases in the 259 spatial distribution of certification. To examine regional and within-country spatial bias we then   (Table 1). For some crops, cells with certification coincided with higher 281 importance for biodiversity conservation than was typical of cells with the same crop without 282 certification. Certified coffee, tea, and cocoa all occurred in cells with higher importance for birds, on 283 average, than that in non-certified cells. The distribution of coffee, both certified and not, included 284 areas of particularly high conservation importance for birds ( Figure 2). Certified tea occurred in cells 285 with higher importance for amphibians, while the soy production cells with highest amphibian value 286 were less likely to be certified. For mammals, coffee certification occurred in cells with higher 287 conservation importance than that in coffee cells without certification. However, for all other crop-288 taxon combinations, there were no significant differences between cells with certification and 289 without, in respect to their importance for birds, amphibians or mammals.

290
Certified tea occurred on average in grid cells with greater protected area coverage, while certified 291 oil palm and coffee occurred in cells with less protected area coverage than non-certified cells. Cells 292 with certified tea coincided to a greater extent with IBAs than non-certified cells, while cells with 293 certified cocoa had less overlap with IBAs than non-certified cells. There were differing patterns 294 between crops with respect to rates of tree cover loss. Cells with certified soy, oil palm, or cocoa 295 coincided with higher rates of loss, while cells with certified coffee or tea coincided with lower rates 296 of loss compared to cells growing uncertified crops of the same type.

297
For most crops, grid cells with certification tended to have larger fields, be closer to market towns, 298 and have a lower percentage of the population in poverty than the distribution of the crop more 299 generally. Although cells with certified soy tended to have larger field sizes, they were also further 300 from towns, and in poorer areas. Certified cocoa was found in cells with smaller field sizes, although still closer to towns, and in wealthier areas than non-certified cocoa. Physically, certified crops often 302 occupied cells with significantly different (higher, lower, or similar, depending on the crop) altitude, 303 slope, and crop cover compared to the crops' global distributions (Table 1).

304
Case study: Cocoa in West and Central Africa

305
We explored the extent to which these global patterns persist at regional and national scales, 306 focusing on three data-rich case study areas. Across the West and Central African cocoa-growing 307 region, Cameroon, Cote d'Ivoire, Ghana, Nigeria, Sierra Leone, and Togo all grew certified cocoa, 308 although certification was restricted to only a single grid cell in Togo and two in Sierra Leone. Across 309 the region as a whole, cells with certified cocoa had similar importance for birds to cocoa cells 310 without certification. The global-level pattern of higher importance for birds in certified cells was 311 reflected in some countries (Côte d'Ivoire, Ghana and Cameroon), but not in Nigeria (Table 2). For 312 amphibians, certified cells had higher importance in some countries, but not globally or across the 313 West African cocoa-growing region as a whole. For mammals, cells with certified cocoa had higher 314 conservation value at a regional level in West Africa, and in most of the cocoa-growing countries 315 within it, whereas at a global level there was no difference from cells without certification.

316
In West Africa as a whole, grid cells with certified cocoa did not have significantly greater cover of 317 either protected areas or IBAs but were closer to market towns and had lower levels of poverty.

318
When examining patterns in individual countries, cells with certification tended to have higher 319 conservation value and to occur closer to towns and in areas of lower poverty than cocoa-growing 320 cells without certification. Landscapes with certification tended to be in grid cells with lower levels of 321 cocoa cover than the control. Patterns at the country level were not always reflected at the regional 322 (West Africa) level. For example, cells with certified cocoa had higher importance for birds and 323 amphibians for three of the four countries examined in Table 2, but no significant relationship was 324 found at the regional level, likely because of variation within and between countries.

Case study: Coffee in Central America
Grid cells containing certified coffee are most prevalent in several Central American countries (Costa 327 Rica, El Salvador, and Guatemala), outnumbering non-certified coffee-growing cells. In the remaining 328 countries, certification presence is still high with the exception of Panama where it is absent. The 329 general pattern for the both the Central American region and the individual countries was for 330 certification to occur in cells with higher levels of conservation importance compared with non-331 certified coffee-growing cells ( Table 3). Rates of tree cover loss were lower in most cells with 332 certification, while the incidence of certification in cells with protected areas varied by country. In

333
Central America overall, certified cells tended to be closer to markets, while poverty levels in 334 certified cells were higher than in non-certified cells in Honduras and Nicaragua, and lower in 335 Mexico. Certified cells consistently occupied regions of higher altitude, slope and crop cover, 336 perhaps due to greater suitability of these conditions for high-quality shade-grown coffee, which is 337 more likely than sun-grown coffee to be marketed as a premium product to consumers for whom 338 certification has resonance.

339
Case study: Oil Palm in Southeast Asia

340
Certified oil palm in Southeast Asia (SE Asia) was found solely in Malaysia and Indonesia and tended 341 to be located in grid cells with lower than average importance for bird conservation than non-342 certified oil palm and in areas with lower coverage of IBAs and protected areas ( Table 4). Rates of 343 tree cover loss were higher in certified cells in SE Asia. From a livelihoods perspective, certified cells 344 were closer to towns and had lower levels of poverty. Cells with certified oil palm were also in areas 345 with lower altitudes and slope and higher percentage of crop cover, suggesting that these might be 346 more favourable crop-growing areas. Patterns at the SE Asia regional level appeared primarily 347 influenced by patterns of certification in Malaysia. Certified oil palm cells in Indonesia appeared to 348 have few differences compared with non-certified cells, although they were perhaps located in more 349 favourable, intensively-farmed agricultural areas, as altitude and slopes were lower but field size and 350 percentage crop cover were higher.

355
We developed the most detailed global map of commodity crop certification yet produced. It shows 356 that certification for each crop is concentrated in certain geographical areas, and largely absent from 357 others (Fig. 1). According to available spatial data, most commodity crop certification is in tropical 358 countries, although this is a pattern that would change if spatial data were available for organic 359 schemes (Tayleur et al., 2016). Our analysis quantified biases in each crop's certified locations 360 compared with gradients of conservation importance, tree cover loss and poverty (Table 1). Patterns 361 varied on a crop-by-crop and country-by-country basis, but overall, certification appears to be 362 concentrated in areas that are important for biodiversity conservation, relatively close to markets, 363 and with lower poverty levels (Figs 2, 3; Tables 1-4). These patterns suggest that existing standards 364 may be well-positioned to have a conservation impact if they promote the right practices, but are 365 less well-positioned to assist the very poorest farmers. However, there were exceptions to each of 366 these patterns, and relationships between certification and other variables were less consistent 367 (Tables 1-4). Some of the patterns found when data were pooled at global or regional levels persist 368 within individual countries, while others do not (Tables 2-4). This underlines the importance of 369 selecting the most appropriate decision-relevant scale for analysis of spatial patterns.

370
Explaining patterns of certification 371 Some of the patterns likely reflect geographical differences in growing practices, some of which are 372 more amenable to certification than others. For example, shade-grown coffee is more likely to meet 373 requirements of speciality coffee buyers and many certification standards, and growers may be 374 more likely to seek certification, compared with sun-grown coffee (Takahashi and Todo, 2014). The 375 higher conservation value of certified coffee cells in Central America might be because shade-grown coffee, and thus certified coffee, is more common in remote, high altitude locations with steep 377 slopes (Table 3): locations where many restricted-range species could be expected to occur. Other 378 patterns are more difficult to explain, such as higher rates of tree cover loss in cells with certified 379 cocoa, palm oil, and soy. In the case of palm oil and soy especially, halting deforestation is a key 380 objective for certification standards. It may be that certification is reaching these crops in recent 381 frontiers, while being associated with more established areas of cultivation for other crops, such as 382 tea and coffee. If land at high risk from forest clearance is becoming certified, this could be good 383 news for conservation, as long as certification proves effective at preventing deforestation (e.g.

419
To improve standards on certified farms, for example, it might be worthwhile for coffee certification 420 standards to incorporate stronger protection for wild species and their habitats in landscapes 421 identified as having especially high importance for conservation, such as those in Honduras (Table 3).

422
This could be achieved by incentivising farmers to 'step up' from entry-level schemes, such as the 4C 423 coffee standard, to more comprehensive standards, such as Rainforest Alliance/SAN. It could be 424 fostered by varying scheme requirements geographically, demanding compliance with key 425 biodiversity criteria in relevant areas or by ensuring more frequent or more thorough audits of 426 practices relevant to biodiversity. Audit data, in combination with spatial biodiversity data, could be used to identify as high priorities for intervention any farms that are performing poorly against 428 environmental criteria in areas of conservation importance; the same analyses could be used to 429 reward farmers performing well in priority areas. Training programmes aimed, for instance, at 430 reducing specific threats such as hunting, or at habitat management for threatened species, could be 431 targeted towards producers in areas identified as being of especially high value for biodiversity 432 conservation. There might be specific opportunities for NGOs to engage with producers: for

456
The RSPO is adopting some of these approaches in an attempt to increase smallholder uptake. 457 Comparing regional with country-level patterns of importance for birds and mammals suggests that 458 certification in West and Central Africa misses some of those cocoa-growing areas that are most 459 important for biodiversity. Extending certification to cocoa-growing countries it has barely reached, 460 such as Sierra Leone and Togo, while strengthening biodiversity-related criteria, could play a role in 461 conservation efforts. However, expansion would need to be linked to an appropriate market, 462 because while some certified products such as coffee and cocoa now have mainstream markets -463 40% and 22% of production respectively -demand has tended to lag supply. For example, less than 464 one third of certified coffee was sold as such in 2012, which may limit future expansion and financial 465 benefits for farmers (Potts et al., 2014). Efforts to expand certification can also go further to

471
The accuracy of our analyses was limited by data quality. Many schemes have not yet developed 472 rigorous protocols for the collection and/or sharing of spatial data. As a result, spatial data were 473 often available for only a subset of the certificates within each standard. For some standards, no 474 spatial data were available. For example, we contacted more than 200 organizations that certify 475 organic agriculture, but received few positive responses covering only a handful of producers. For 476 some crops (cotton, and in some cases sugarcane) certification locations referred to processing mills, not to farms. Other schemes were only able to provide addresses. The use of non-standard address 478 formats and non-Roman alphabets meant that the success rate of geocoding was low and those 479 coordinates that were created could not be ground-truthed. For our analysis we summarized data at 480 the 30-km scale. This was primarily to ensure farmer confidentiality, but also reduced the impact of 481 imprecise spatial coordinates and farms with multiple certifications. A disadvantage of aggregation 482 at this scale is that a large proportion of land within each cell is likely not certified. Our decision to 483 use the Monfreda map, clipped with the GAEZ map, was also an imperfect representation of crop 484 distribution for our 'control' distributions, but these were the best global data available. Finer-485 resolution analyses would be preferable in order to reflect the true spatial patterns for individual 486 standards. It is important to recognize that our analyses show only correlation, and not causation, 487 but correlations are useful for identifying gaps and priorities.

488
Our difficulties in locating and assembling a spatial database of certification lead us to recommend 489 that greater resources be invested by certification organizations in collecting and organizing such 490 information. While during the course of this study we found that spatial data were often lacking and

492
Improving the provision of spatial data is consistent with the commitments of certification 493 organizations to transparency and traceability. Challenges remain, such as ensuring that the right to 494 privacy of producers is respected, and that commercially-sensitive data are handled appropriately.

495
However, these challenges are surmountable, and putting certified producers on the map also has 496 several benefits. Transparency can be used to deflect criticism: for example, open RSPO data have 497 been used to show that most fires are not on RSPO concessions (http://www.rspo.org/news-and-498 events/news/rspo-statement-on-the-indonesian-forest-fires). Good spatial data are essential for 499 demonstrating and auditing compliance with some criteria, such as adherence to restrictions on 500 deforestation (Tayleur and Phalan, 2016). Being able to cross-reference spatial data from different 501 standards could help to identify overlaps and streamline audit processes. Bodies such as the ISEAL 502 Alliance, which supports the sustainability standards community to define and implement best practices, could request minimum transparency guidelines for membership, and define best practice 504 for spatial data management and dissemination.

507
Certification is an increasingly ubiquitous tool, promoted by both the private sector and civil society 508 as important for improving the conservation and socio-economic impacts of agriculture. Our global 509 data synthesis revealed a number of concentrations of certification, both geographically and also 510 with respect to gradients of biodiversity, tree cover loss and poverty. While certification appeared to 511 coincide with areas important for biodiversity, it showed less overlap with areas of greatest poverty.

512
These results suggest either a mismatch between the objectives of sustainability standards studied 513 here and their potential to achieve them, or a greater emphasis on environmental than social 514 sustainability. Regional and country-level crop-specific analyses demonstrated different spatial 515 patterns, highlighting specific opportunities for increasing the impact of standards.

516
We describe three types of activities that could be targeted using spatial analyses to improve the 517 outcomes of certification: strengthening standards on certified farms, consolidating the coverage of 518 farms in already-certified landscapes, and expanding certification into new priority areas. As a 519 market-driven mechanism, certification will require support from a range of actors in the private and 520 public sector to enable spatial targeting. This would require private companies to consider 521 alternative and potentially riskier sourcing locations, financial institutions to strengthen the 522 environmental and social components of their lending criteria, NGOs to effectively advocate for 523 those areas that would benefit most and, finally, governments to provide favourable conditions and 524 requirements for sustainable production and trade. Better targeting in future would also be 525 facilitated by improved collection of spatial data, benchmarking across standards, and a renewed 526 commitment to transparency. 527 528 531 Award. We sincerely thank all the certification schemes that provided data for the project. We thank 532 Taís Pinheiro, Margaret Arbuthnot and Jack Robinson for assistance with data collation, Emilja Emma 533 for support in contacting certification schemes, and Alison Johnston for invaluable statistical advice.

545
http://www.birdlife.org/datazone/species (accessed 11.11.14). 548 Soc. Conserv. Biol. 25, 1176-85. doi:10.1111/j.1523-1739.2011 Table 3 Results from bootstrap resampling tests comparing the distribution of certified coffee grid cells versus non-certified coffee growing cells. Where the value for certified cells was significantly lower than for non-certified cells, the results are shown in light grey, while significantly higher certified values are shaded in dark grey. The values represent the significance value calculated as the number of non-certified values smaller or larger than the certified test statistic divided by the number of permutations (10,000). As tests were two-tailed, the significance threshold was set at 0.025. The fraction of certified to non-certified cells is given under the crop name. We did not have spatial data on IBAs in Mexico or Honduras, hence the N/As in the IBA column.  Table 4 Results from bootstrap resampling tests comparing the distribution of certified oil palm grid cells versus non-certified oil palm growing cells. Where the value for certified cells was significantly lower than for non-certified cells, the results are shown in light grey, while significantly higher certified values are shaded in dark grey. The values represent the significance value calculated as the number of non-certified values smaller or larger than the certified test statistic divided by the number of permutations (10,000). As tests were two-tailed, the significance threshold was set at 0.025. The fraction of certified to non-certified cells is given under the crop name.  where the value was not significantly different from the global distribution, or a solid triangle when it was significantly different. Each box plot represents 10,000 random sub-samples, equal in area to our certified sample, drawn without replacement from non-certified grid cells.                   The value for certified grid cells is signified by an open triangle where the value was not significantly different from the global distribution, and by a solid triangle when they were significantly different. Each boxplot represents 10,000 random sub-samples, equal in area to our certified sample, drawn without replacement from non-certified grid cells.