Transparency, traceability and deforestation in the Ivorian cocoa supply chain

We calculated the volume of cocoa exported by distinct traders and imported by distinct countries using data on export shipments from 2019 (SI table S1). As no information is publicly-available on the location of farms supplying each trader, we map the supply chain up to the first buyer, and not to the farm. For traders that are transparent about their 'direct' suppliers, we identified the list of cooperatives from which they source from, and the department in which they are located, using Mighty Earth's 'Cocoa Accountability Map'. This map includes information on 4451 cooperatives and lists, for 710 of them, companies that are known to source from them (SI figure S2(A), table S2). The latter comes from self-disclosures in 2019 and 2020 by major cocoa traders and processing companies. A smaller volume is purchased from 'acheteurs agréés'; buyers licensed to source cocoa from specific departments (SI table S3). Their supply comes from an approximate known origin, much like a cooperative. When reporting the results, they are included under the term 'sourcing traced to cooperatives'.

The volumes sourced by companies from individual cooperatives are estimated from the number of farmers per cooperative (provided for 592 cooperatives, SI figure S2(B)), and production volumes per farmer. For the remaining 119 cooperatives with a known trading company but no number of farmers provided, we simulated this number using 1000 Monte Carlo estimates (random sampling with replacement) drawn from the set of disclosed cooperative sizes. Next, we used recent production data (kg of dry beans/farmer/year) collected from 441 farms across the country (Bymolt et al 2018) (SI figure S3). We assigned a production volume to each cooperative-company connection using 1000 Monte Carlo estimates of this dataset, keeping the confidence interval. This calculation assumes that traders purchase the entirety of the cocoa of each of their supplier farmers (rather than each farmer selling to multiple traders), which provides an upper-bound estimate of the volumes traced to cooperatives. The latter are assigned to the department where the cooperative is located—thus assuming that the cocoa sourced by a distinct cooperative mainly comes from the department where it is registered.

We then estimated each trader's proportion of indirect sourcing by subtracting the volume traced to cooperatives from the total exported volume. This provides an estimate of the level of traceability for those traders that disclosed their sourcing. For untransparent trading companies that did not disclose any information on their sourcing (43 companies of 73 in total), the total exported volumes are marked as 'unknown sourcing'. This provides an estimate of the level of transparency in the sector.

As an additional analysis, we compiled the number of mapped farms reported in CFI signatories' 2018–2020 progress reports and estimated the total number of farms that these companies are sourcing from based on their traded volume in 2019 (SI text S1.5, table S4).

To calculate cocoa deforestation, we overlaid a remote-sensing cocoa land use map for the years 2019–2021 at 10 m resolution (Kalischek et al 2022) (hereafter, the '2019 cocoa map') with a map of Tropical Moist Forest (TMF) produced by European Commission's Joint Research Centre (JRC), depicting annually from 1990 to 2020 the extent of undisturbed tropical moist forest and its changes at 30 m resolution (Vancutsem et al 2021) (SI figure S5).

We created a map of the land-use dynamics over the last 20 years including deforestation and degradation, their association with cocoa expansion, and the remaining undisturbed and degraded TMF cover. A degraded forest is defined by the JRC TMF dataset as a forest that has been temporarily disturbed (<2.5 years), which can include regrowth, but excludes tree plantations (SI text S1.6.1). Here, 'deforestation' corresponds to pixels that switched from undisturbed TMF to another land cover between 2000 and 2019, and 'degradation' to pixels that switched from undisturbed to degraded TMF. From those, pixels detected as cocoa were classified as 'cocoa deforestation' and 'cocoa degradation', the latter also including cocoa pixels detected as undisturbed forest by the TMF map in 2019 (0.3 Mha). Cocoa degradation could be forest converted to cocoa but with a relatively dense remaining shade canopy cover, or shaded cocoa plantations that were too recently established to be classified as deforestation (requiring 2.5 years of detected disturbance).

We tested the sensitivity of our findings to different remote sensing products by repeating our analyses using two other published cocoa maps (Vivid Economics 2019, Abu et al 2021) and two other deforestation datasets (Hansen et al 2013, BNETD, FAO and SEP REDD+ CI 2017). We used primarily the cocoa map and JRC-TMF maps because of their superior accuracy (SI table S5) and publicly-available methodologies.

To link to the 2019 trade flows, we calculated the 'cocoa deforestation' as the area that lost its undisturbed TMF between 2000 and 2015 and is detected as cocoa in 2019. Thus, here, cocoa degradation and deforestation are merged into 'cocoa deforestation'. Cocoa trees need three to five years to become fully productive, so deforestation events after 2015 are not expected to have contributed to the 2019 cocoa harvest.

To link cocoa sourcing to cocoa deforestation, we calculated the relative cocoa deforestation per department, by dividing the annualised cocoa deforestation area by the 2019 cocoa production per department, to obtain annual hectares deforested per tonne of cocoa produced. In the absence of official publicly available data on subnational cocoa production in Côte d'Ivoire, we generated a map of cocoa production per department by weighting the 2019 cocoa map by a map representing the suitability to grow cocoa per pixel (Schroth et al 2016) (SI figure S6). We then allocated the national cocoa production (ICCO 2021) to each department according to their weighted cocoa area.

We calculated the 'cocoa deforestation exposure' of each trader or importing country, which provides a measure of the exposure of actors to cocoa deforestation based on their sourcing patterns. We multiplied the relative cocoa deforestation by the volume traced to cooperatives of each actor, per department, to obtain the cocoa deforestation exposure linked to their traced sourcing. We also calculated actors' annual relative cocoa deforestation exposure (i.e. number of hectares deforested per tonne of cocoa traced to cooperatives, divided by 15 years). The relative cocoa deforestation for untraced cocoa was estimated as an average of the relative cocoa deforestation per department, weighted by the proportion of total untraced volume supplied by each department. We multiplied this value by actors' untraced (unknown or indirectly-sourced) volumes to calculate each actor's exposure to cocoa deforestation linked to their untraced sourcing.

We trace 43.6% (95% CI: 42.6%–44.7%) of Côte d'Ivoire's exports back to a specific cooperative and department, i.e. 875 468 tonnes (95% CI: 855 360–896 993 tonnes) of 2006 406 tonnes exported in total in 2019. The majority of cocoa (over 55%) is thus untraced, either indirectly sourced from local intermediaries by major traders (23.9%, 95% CI: 22.9%–24.9%), or exported by untransparent traders (32.4%), i.e. 'unknown' sourcing (figure 2). Untraced sourcing dominates in the North–East and along the western border of the cocoa producing region of Côte d'Ivoire (80%–100% untraced sourcing) (SI figure S7).

Zoom In Zoom Out Reset image size

In 2019, eight companies (Cargill, Barry Callebaut, Olam, Touton, Sucden, S3C, ECOM, and Africa Sourcing) handled 60% of exports (SI figures S8 and S9). The share of indirect sourcing among traders varied between 25% (95% CI: 10.4%–34.4%) (ECOM) and over 70% (95% CI: 66.4%–75.1%) (Sucden), posing strong traceability challenges. There are also gaps in transparency. A third of the cacao was exported by untransparent traders, among which S3C and Africa Sourcing, as well as many of the smaller companies, who disclose no information about their suppliers. Even traders who did disclose their supplying cooperatives can greatly strengthen their level of transparency: their disclosed data is scattered, incomplete, irregularly or never updated, and non standardised. The Accountability Framework Initiative provides unified guidance to companies on reporting and disclosure practices to increase the credibility of their claims.

Looking at importing markets, European Union countries imported 60% of Ivorian cocoa beans in 2019, of which only 45.6% (95% CI: 44.5%–46.8%) were traceable to cooperatives (figures 3(A) and (B)). These data detail the country where cocoa was first imported—the Netherlands, for example, appears as the principal destination (27% of exports) because the port of Amsterdam is the main hub for cocoa imports into Europe, with cocoa then being re-exported or processed for consumption elsewhere. The rate of untraced imports among major importers varied between 36% (95% CI: 33.3%–39.8%) (France) and 66% (95% CI: 66.2%–67.1%) (Malaysia).

Zoom In Zoom Out Reset image size

Traceability to farm lags further behind traceability to cooperatives. For the seven trading companies who disclose their farm-level mapping in Côte d'Ivoire as part of the CFI—handling 52% of the total traded volume—we estimated that 69% (95% CI: 64%–74%) of the volumes they source 'directly' from cooperatives was mapped to farm-level (343 207 farms of the estimated 495 000 farms supplying them 'directly' through cooperatives (95% CI: 465 K–533 K)). For comparison, this is consistent with the self-reported number across all CFI companies, which was 74% in 2020 and 72% in 2021 (CFI 2021). Yet, when including their indirect sourcing, this only represents 40% of the total 850 000 farms supplying these traders (95% CI: 845 K–854 K), which in turn only represents 22% of all cocoa exports in 2019. Approximately 579 000 farms (95% CI: 575 K–583 K) supply the EU market through these CFI traders (figure 3(C)), compared to the 343 207 farms that they had georeferenced in 2020 (465 400 reported in 2021). This shows that current levels of farm-level traceability are insufficient to meet the EU's future due-diligence legislation, expected to require geolocation of the product's origin. Overall, 838 000 hectares of deforestation over 2000–2015 were associated with 2019 EU imports (figure 3(D)). Nor is the EU the only market moving towards due-diligence for deforestation-free imports: similar legislation is also tabled in the United States, UK, Canada and Japan (OECD 2021)—the destination of 11.6%, 4.6%, 3.2% and 0.1% of Ivorian cocoa exports, respectively.

We found 2.4 Mha of cocoa deforestation and degradation over 2000–2019 (0.125 Mha per year), of which 1.25 Mha is cocoa degradation. This does not include the deforestation of already degraded forest in 2000. The remainder is linked to other agricultural crops such as palm oil, rubber or food crops, logging and mining (BNETD, FAO and SEP-REDD+ CI 2017). Over that period, 5.2 Mha of the initial 6.7 Mha of undisturbed TMF in 2000 was lost (i.e. 77%), either degraded or deforested, leaving the country with less than 1.5 Mha of undisturbed tropical moist forest in 2019, representing 4.7% of its territory. This percentage increases to 9.4% when accounting for the 1.5 Mha of degraded TMF (figure 4(A)). The latest national flora and fauna survey gives a similar total natural forest area of 2.5 Mha in 2020 (without wooded savanna) (IFFN 2021).

Zoom In Zoom Out Reset image size

Some limitations inherent to spatial datasets may affect the results. For instance, the removal of shade trees or the rehabilitation of old abandoned cocoa farms may be categorised as cocoa deforestation or degradation, because highly shaded farms in 1990 (the starting point of the JRC TMF time series) may have been classified as undisturbed TMF. On the other hand, young cocoa farms are, to some extent, not detected by the cocoa map—thus very recent deforestation for cocoa is certainly classified as non-cocoa deforestation or degradation.

Drivers of land use changes are complex and deforestation or forest degradation rarely have a single underlying cause (Pendrill et al 2022b). It is possible that other land uses occurred in the period between deforestation or degradation and the detection of cocoa in 2019—such as other crops on deforested land, or timber extraction or the planting of highly shaded food crops on forests detected as degraded. We report, however, 'cocoa deforestation' and 'cocoa degradation', because cocoa was the major expanding land use during this time period, occurring on 45% of land deforested or forest degraded between 2000 and 2019, and other studies also find cocoa as the main driver of deforestation in Côte d'Ivoire (BNETD, FAO and SEP REDD+ CI 2017, Higonnet et al 2017, Ruf and Varlet 2017, Barima et al 2020). Cocoa was identified as the major export-oriented crop expanding in Sub-Saharan Africa in 2000–2013, at an annual rate of 132 000 ha, with Côte d'Ivoire being one of the countries most at risk of deforestation because of its low available cropland outside forest areas (Ordway et al 2017). Pantropical studies estimated annual cocoa deforestation rates in Côte d'Ivoire of 33 000 ha y⁻¹ (Goldman et al 2020a) and 13 000 ha y⁻¹ (Pendrill et al 2022a) over 2001–2015, while a perception-based study estimated the role of cocoa in deforestation at 23% (the highest of all drivers) between 1990 and 2015 (Nitidae 2016). These are considerably lower than our 125 000 ha y⁻¹ of cocoa deforestation and degradation over 2000–2019.

Côte d'Ivoire has experienced several waves of cocoa expansion, first in the East, then expanding in the Southwest in the 1970s. The most recent belt of expansion in the Northwest, along the Guinea and Liberia borders (Sanial 2017), has heavily impacted protected areas (figure 4(A)) where the few remaining forests have been concentrated (BNETD, FAO and SEP REDD+ CI 2017). This is also mirrored by the concentration of cocoa deforestation in the western departments, with Sassandra, San-Pédro, Tabou and Guiglo amounting for 28.5% of the total cocoa deforestation over 2000–2015 (figure 4(B)). Over the 2000–2019 period, we found that protected areas (including classified forests) in the cocoa growing region lost in total over 60% of their undisturbed TMF of 2000 (representing 1.3 Mha), with 62% of this deforestation and degradation being due to cocoa (SI table S5). In 2019, 60% of the remaining protected undisturbed TMF was sheltered by the Taï National Park—the best preserved protected area of the country. 25% of the cocoa area in Côte d'Ivoire was located within protected areas and the area under cocoa production has still been expanding (FAO 2020, Vivid Economics 2020). Accounting for higher yields on recently cleared land (Ruf 1987), cocoa in protected areas very likely represents more than 25% of the national production of the world's largest cocoa producing country.

Multiple factors can explain the high rate of cocoa deforestation and degradation, including reduction in soil fertility and yields, migration, the lack of law enforcement, corruption, the lack of financial means and the 2000's political instability (Ruf et al 2015, Barima et al 2016, Andrieu et al 2018, Ongolo et al 2018). However, beyond all these factors, cocoa deforestation is rooted in the increasing international demand for cocoa.

To link cocoa deforestation, here corresponding to all areas of undisturbed TMF lost to cocoa and thus also including areas detected as cocoa degradation, to 2019 trade flows, we consider the 2000–2015 period, during which 55% of the 2000 undisturbed TMF area was lost (from 6.7 Mha to 3 Mha). This is substantially more than the loss of forest cover reported by the National REDD+ Secretariat, i.e. from 4.7 Mha to 2.7 Mha (BNETD, FAO and SEP REDD+ CI 2017). This difference may be due to differing methodologies and forest definitions (see SI table S6). In total, 1.5 Mha were replaced by cocoa over that period (40.7% of the undisturbed TMF lost), at a rate of 100 000 ha per year. The relative cocoa deforestation (ha/tonne produced) is quite homogenous across the country, with slightly larger values in the West (figure 4(B)).

57.7% of the total cocoa deforestation exposure is attributed to untraced sourcing (24.5% to indirect and 33.2% to unknown sourcing), which, in relative terms, represents a slightly higher risk of being exposed to deforestation than sourcing traced to cooperatives (47.9 compared to 45.4 ha/kton.y⁻¹) (figure 5(A)). Cargill, Barry Callebaut, Olam and Touton each have 100–180 000 ha of cocoa deforestation exposure linked to their 2019 sourcing, but the exposure per tonne traced to cooperatives is similar for all traders (figure 5(B)). The exception is Touton whose higher relative cocoa deforestation exposure is due to its sourcing pattern concentrated on few departments among which some faced high cocoa deforestation, e.g. Guiglo—from which 18% of Touton's sourcing traced to cooperatives came from. The eight CFI traders, handling 55% of the total exports, encompass 53.4% of the total cocoa deforestation exposure. Their sourcing traced to cooperatives has a slightly lower relative exposure than non-CFI signatories' (44.3 compared to 48.6 ha/kton.y⁻¹ for non-CFI traders). As the CFI was signed in 2017, this difference should be considered as a baseline and not as the impact of the agreement (SI figure S10).

Zoom In Zoom Out Reset image size

Deforestation exposure trickles down to importing countries, with a large share reaching the Netherlands and its cocoa processing facilities (figure 5(C)). In Germany, 72% of chocolate confectionery was certified according to sustainability standards in 2019 (79% in 2021) (BDSI 2021), but we show that the country imported 101 000 ha of 2000–2015 deforestation. Here, we mapped physical flows of cocoa imports, whereas certification credits for several certification standards are decoupled from physical flows, leaving room for deforestation to enter the market.

Tracing cocoa and linking sourcing to deforestation comes with some qualifications. We tested the sensitivity of our findings by repeating our analyses using different remote sensing products (SI text S2, tables S6, S7 and figures S11–S16). The combination of the ETH and JRC-TMF maps, which are the most accurate maps available for deforestation and cocoa plantations, shows a higher rate of cocoa deforestation than other datasets as ETH detects more cocoa and JRC-TMF more forest loss—notably because it includes forest degradation (figure 4(C)). But, importantly, regardless of the magnitude of cocoa deforestation, the relative attribution of the exposure to specific companies and international markets remains stable (figures S13–S16). Given the consistent patterns across datasets, uncertainty in remote sensing products should not be a barrier to corporate and government action on cocoa-driven deforestation. Yet, current approaches to attributing deforestation to cocoa and other crops, which intersect cocoa and forest maps developed in isolation, could be improved by developing harmonised land use products which simultaneously map multiple land covers and uses, and transitions over time, as Mapbiomas does in South America and Indonesia (Alencar et al 2020). Further, when attributing deforestation exposure to international markets, it is noteworthy that our results cover the country of import, rather than consumption. Consumption-based attribution studies, on the other hand, also introduce further methodological assumptions and modelling uncertainties and are agnostic as to the identity of which companies handle products along the supply chain (West et al 2022). Finally, though deforestation happens on farms, we map sourcing and deforestation exposure at the department-level, which is the finest resolution possible in the absence of publicly-available data about the location of farms supplying each company.

Companies, as well as the forthcoming EU due-diligence legislation, count on farm-level traceability to reach zero-deforestation supply chains. If traceability offers a unique opportunity to improve the living conditions of farmers, through better traced payments and a clearer cooperative system (INADES et al 2022), several loopholes may undermine its effectiveness in curbing deforestation.

Company-led traceability is challenging and could be improved, but is not sufficient to address deforestation. Large parts of the supply chain are not mapped; only 44% of the Ivorian cocoa is traced to cooperatives, and 22% is traced to farm under the CFI. Companies in the CFI reported lower traceability in 2021 (72%) vs 2020 (74%) in their 'direct' supply chain from cooperatives. This decline highlights the challenge of traceability, not least the year-on-year changes in farmer membership of cooperatives and company programs requiring frequent updates of the supply base data (Barry Callebaut 2020). As company-led traceability only covers a partial share of the whole supply chain, it leaves gaps in which deforestation is left unmonitored. Buyers less exposed to reputational risks and less committed to sustainability can continue sourcing from these regions—and their non-compliant cocoa can end up in the indirect sourcing of committed actors. Similarly, as the EU due-diligence only applies to volumes sold in the EU, there is a risk of market bifurcation, with non-compliant cocoa sold to non-EU markets (40% of exports). Considering the market dominance of the EU and the similar due-diligence legislations under development in other importing markets, this risk of leakage is limited compared to other contexts, such as the Brazilian cattle sector (le Polain de Waroux et al 2019, Zu Ermgassen et al 2020), but still remains. If traceability does not cover the whole supply chain, deforestation leakages risk undermining its conservation additionality (Lambin et al 2018, Gardner et al 2019, Garrett et al 2019, Gollnow et al 2022).

High hopes are placed in national farm-level traceability systems, as currently under development by the Coffee and Cocoa Council in Côte d'Ivoire (Aboa 2022), but they remain challenging and have to be transparent to be effective. First, cooperatives play a crucial role in traceability but they are poorly structured (El Makhloufi et al 2018). Many are the result of traders or local leaders' business interest, with a membership 'on paper', rather than farmers deciding to work together (Ruf et al 2019). As a result, links between farmers and buyers are fluid with farmers selling to multiple buyers to cope with delays in payments (El Makhloufi et al 2018, BASIC and FAO 2020, Nitidae and EFI 2021) and cooperatives buying from non-member farmers to meet the needs of their clients (IDEF 2021, Stoop et al 2021). As much of Côte d'Ivoire's forest were cleared before the cut-off date proposed by the Commission (31 December 2020), with less than 10% of the territory covered by tropical moist forest in 2019, the due-diligence's traceability requirements will be the major barrier to access the EU's market for Ivorian farmers and cocoa traders. Support from government and trading companies is thus required to improve the governance of cooperatives at the benefit of farmers and farm workers (Meemken et al 2019). Second, as non-compliant cocoa is widespread, it is crucial that the national traceability system is externally verifiable, and thus transparent. At least one-fourth of cocoa is produced within protected areas, and without transparency there is a risk that this cocoa continues to be hidden, jeopardising sustainability initiatives' impact (Gardner et al 2019). Making this data public would also alleviate companies' concerns about disclosing their competition-sensitive data, allowing for greater collaboration. Further, equitable frameworks are required to address non-compliant flows, otherwise leading to the exclusion of farmers who will have no other option than selling at low price to intermediaries (Bakhtary et al 2020).

Pre-competitive collaboration among traders and chocolate companies, at the landscape level, is one of the most promising ways to increase the effectiveness of sustainable sourcing initiatives, by delivering consistent impact at scale (Sayer et al 2013, Wolosin 2016, Carodenuto 2019, Bakhtary et al 2020, IDEF 2021). Companies buy from the same landscapes (figure 6(C)), but rarely collaborate and implement their separate sustainability programmes with cooperatives that are part of their 'direct' sourcing—ignoring regions where untraced sourcing dominates (SI figure S7). This lack of collaboration translates into scattered projects sprinkled over the territory, and into activities sometimes overlapping and redundant. For example, certification—which has been a major sustainability strategy in the sector for the past decade (Thorlakson 2018), led to considerable duplication of effort, with many double or triple certification of cooperatives (SI figure S17). Furthermore, a lot happens outside companies' supply chains, with about half of deforestation in Côte d'Ivoire not directly linked to cocoa. Yet, as cocoa occupies large parts of agricultural land (Kalischek et al 2022), it drives deforestation indirectly by pushing other crops inside protected areas (Ajagun et al 2022, Kumeh et al 2022). To address deforestation, cocoa companies need to work beyond their individual supply chains. Landscape or jurisdictional approaches, defined as governance initiatives promoting sustainable resource use in a specific landscape (e.g. jurisdiction) through a formalised collaboration between all stakeholders (Reed et al 2016, von Essen and Lambin 2021), have the potential to internalise indirect sourcing and deforestation drivers beyond the focal commodity (Zu Ermgassen et al 2022). The CFI identified priority regions to establish multi-stakeholder landscape governance (CFI 2018), but progress is elusive with only a few pilot projects initiated (CFI 2022). The current CFI regions only capture 23% of the 2000–2019 cocoa deforestation and degradation and 32% of the remaining undisturbed TMF—these numbers could increase by 10% if other key departments were included (figures 6(A) and (B)). Greater effort is needed, requiring government institutions to build the necessary enabling environment and companies to collaborate, support broader regulatory efforts, and provide funding beyond their own projects, focusing on landscapes most at risk of deforestation.

Zoom In Zoom Out Reset image size

Beyond Côte d'Ivoire's borders, the above recommendations—i.e. implementing landscape approaches, regulatory policies, and a transparent national traceability system—also apply to highly forested countries, where cocoa is expanding (Sassen et al 2022).