Environmental assessment of the Ecuadorian cocoa value chain with statistics-based LCA

Life cycle assessment (LCA) has been widely applied in the context of interdisciplinary value chain analyses (Dabat et al. 2018a, b) and/or applied to study whole value chains (Hellweg and Milà i Canals 2014; Meinrenken et al. 2020). This is particularly true for agri-food value chains (Basset-Mens et al. 2021), including the cocoa and chocolate sector (Vesce et al. 2016; Recanati et al. 2018; Bianchi et al. 2020; Boakye-Yiadom et al. 2021).

Despite the publication of various Ecuadorian cocoa value chain studies (CEPAL 2014; Vassallo 2015; Acebo 2016; Ríos et al. 2017; Guilcapi 2018; Henry et al. 2018; Barrera et al. 2019; INIAP 2019; FAO and BASIC 2020), applying different methodologies and exploring in varying degrees of depth different elements of the value chain functioning, no comprehensive environmental assessment has been published to date. Existing partial analyses include, for instance, a recent energy assessment focused on climate change (Pérez Neira 2016) and various studies on cadmium (Chavez et al. 2015; Barraza et al. 2017; Argüello et al. 2019), as well as studies on biodiversity and carbon sequestration associated with cocoa plantations (e.g. Jadán et al. 2015; Samaniego et al. 2017).

This study presents a detailed statistics-based life cycle assessment (LCA) of the Ecuadorian cocoa value chain, produced in the context of the latest available value chain study (Avadí et al. 2021). The value chain study performed a functional analysis as a starting point for socio-economic and environmental analyses, including a detailed mapping of the chain, which facilitates detailed environmental assessments, as discussed in Acosta-Alba et al. (2022). The use of statistical data to inform life cycle inventories is not uncommon (Moreau et al. 2012; Weymar and Finkbeiner 2016; Avadí et al. 2017; Pradeleix et al. 2022), even for the construction of life cycle inventories databases such as AGRIBALYSE (Koch and Salou 2016; Asselin-Balençon et al. 2020) and Agri-footprint (Blonk Consultants 2019).

Cocoa is one of the main crops grown in Ecuador. The agricultural area dedicated to cocoa (601 954 ha in 2019, or 4% of total land use) represents the largest area dedicated to a permanent crop the country: 38% in the period 2014–2019, followed by oil palm and banana with 18% and 12%, respectively. Dry bean production, which reached 283 680 t in 2019, has grown at an average annual rate of 15% since 2014. Such growth has been achieved mainly by improving yields and by replacing failed oil palm areas with cocoa, as demonstrated in Avadí et al. (2021) (see key historical surface and yield statistics in the Supplementary Material). Several varieties of cocoa are grown in Ecuador, but production is dominated by two main varieties: “national” or Cacao Fino y de Aroma (CFA, 43% of area and 28% of production in 2017) and clonal varieties, chiefly CCN-51 (57% of area and 72% of production in 2017). Cocoa is mainly produced on the Ecuadorian Coast (but also in the Highlands and Amazonia), and its value chain is highly complex (Fig. 1). Cocoa systems are dominantly monocrop plantations, but 13% of the cocoa surface consists of associated systems, mostly cocoa trees associated with food crops, but to a lesser extent (2–3%), agroforestry systems, mostly in the Amazonia (Avadí et al. 2021).

The functional analysis presented in Avadí et al. (2021) identified several key value chain structural features, namely, a typology of actors (Supplementary Material) and the existence of five sub-chains.

Among intermediaries, several strategies were identified, which are closely related with the sector’s main sub-chains. These strategies consist of commercial brokering of unfermented beans dried on farm or along roadsides, and two types of post-harvest performed at collection centres (quality-oriented and volume-oriented). The former usually ferments wet beans in wooden crates and dries them under a plastic tunnel, often on wooden drawers or on concrete surfaces. The latter allows the wet beans to ferment for a short time on woven plastic bags and dries those using thermal means (gas, diesel).

A small percentage of beans that undergo fermentation and drying (i.e. < 10%) are further processed by artisanal or industrial means. Industrial and artisanal processing differs, rather than in terms of technology or scale, in terms of the interconnectivity between the different processes. Cocoa processing is broadly divided into two stages: primary and secondary processing. Often, both stages are integrated.

Primary processing consists of roasting the beans, grinding (with shell separation), crushing to obtain liquor, and as an optional step the separation of the liquor into its components butter and powder. These products are collectively referred to as “semi-processed”. Secondary processing consists of combining the products of primary processing with other ingredients (milk, sugar, and possibly aromatic ingredients in small quantities) to produce chocolate. There are different types of chocolate, but broadly speaking, we speak of industrial chocolate (e.g. couverture) or consumer chocolate (dark, milk, white). In addition to the raw material, cocoa processing consumes energy (usually electricity) and very little water. The processing technologies and processes applied in Ecuador are rather standard, and have been described in numerous publications (Pérez Neira 2016; Recanati et al. 2018; Abad Merchán et al. 2020; Bianchi et al. 2020; Ramos-Ramos et al. 2020), including artisanal processing (Aguilar Jaramillo 2005).

The five identified key sub-chains consist of various interlinkages of different types of actors and actors’ strategies across the value chain (Supplementary Material):

The LCA-based environmental assessment of the cocoa value chain and sub-chains in Ecuador aims to determine the potential impacts of its current functioning on the three classical areas of protection. To estimate these impacts, life cycle inventories representing the various types of systems in each link of the value chain (i.e. the various types of farming systems, artisanal and industrial processing and distribution), were constructed in terms of representative production units. To do so, mostly secondary data were collected for the most representative system types, as defined in the actor typologies (Supplementary Material). Scarce primary data were also obtained via field visits and surveys and used mainly for confirmation of trends. Secondary data, from which agricultural inventories were built, were obtained from various sources:

The scope (boundaries) of the study includes, as limited by data availability and quality, (i) all types of producers, further segregated by region and cocoa variety; (ii) two types of post-harvest operations; iii) primary and secondary transformation; and (iv) separate analyses for the identified sub-chains (from agricultural production to export).

Two main functional units were retained for the agricultural phase: 1 t of cocoa (in dry bean equivalents) and 1 ha of cocoa production, at farm gate. For semi-processed and processed products, 1 kg of product was retained, at factory gate.

The allocation of impacts among co-products was not necessary, since, from an environmental point of view, it is not necessary to separate the co-products of primary processing (liqueur, butter, and powder; see Supplementary Material), when they are considered together as the output of the process (data paucity prevented disaggregating these co-products). When modelling cocoa systems under cultural associations, the data used were sufficiently segregated to identify the specific agricultural inputs to cocoa in these associations.

Life cycle inventories were constructed using as the main data source (for the agricultural phase) the ESPAC 2019 database, complemented by other sources used to inform specific details, approximations, and assumptions. The ESPAC 2019 data were treated as follows:

The inventories of post-harvest processes were modelled on the basis of primary data (10 datasets). The inventories of the transformation processes, including average formulations of processed cocoa products, were modelled on the basis of primary data (3 datasets) and secondary data (4 datasets: Ntiamoah and Afrane 2008; Pérez Neira 2016; Recanati et al. 2018; Boakye-Yiadom et al. 2021), basically in terms of energy consumption. Processing does not consume chemicals, and water consumption is marginal. Ingredients other than cocoa derivatives (sugar, milk powder), as well as complementary/auxiliary processes (production of electricity, packaging materials, fertilisers, pesticides and fuels, combustion, infrastructure), were obtained from LCI databases (Koch and Salou 2016; Wernet et al. 2016; Nemecek et al. 2020). The formulations of three types of chocolate were established from primary and secondary data (Table 1).

Intermediation and transport—to collection centres all over the country, to processing in Quito where cocoa semi-processing and processing plants are concentrated, and to the port of Guayaquil from where most export originates—were modelled on the basis of average distances transported (Amazonia to Guayaquil: 450–650 km, Amazonia to Quito: 200–300 km, Highlands to Guayaquil: 360 km, Highlands to Quito: 110 km, Coast to Guayaquil: 180–370 km, Coast to Quito: 230–510 km), inspired by the literature (Pérez Neira 2016).

A series of aggregate processes (e.g. weighted averages between regions, between regions and cocoa varieties, national average) were calculated for the analysis:

The life cycle impact assessment methods recommended by the European Community’s Product Environmental Footprint (PEF) initiative (EC 2013) were applied. The updated list of methods presented in the recent Product Environmental Footprint Category Rules Guide (Version 6.2—June 2017, (Zampori and Pant 2019)) was used, as available in SimaPro v9.2 (method: EF v3.0). The EF 3.0 method does not take into account carbon sequestration in biomass (Carbon dioxide, in air), whereas other methods, e.g. ILCD 2011 Midpoint + V1.11 (EC-JRC 2012), do take into account such sequestration (incorrectly for annual crops, but correctly for perennials). Therefore, the EF 3.0 method was modified to replace the characterisation factor of “Carbon dioxide, in air” from 0 to − 1. This logic complies with the consensual definition of “carbon sequestration”, which implies a net removal of C from the atmosphere to be stored in a long-term reservoir, be it soil and biomass. (Agostini et al. 2015; Chenu et al. 2019).

This list of midpoint indicators was complemented with ReCiPe endpoint indicators (2.2 Endpoint World H/A (Hierarchy/Average)). ReCiPe was chosen because it presents endpoint indicators in all three areas of protection, based on many relevant impact categories (Huijbregts et al. 2016). The hierarchical perspective (H) was chosen because it is based on the most common normative principles with respect to timeframe and other issues and is therefore often found in scientific models (Goedkoop et al. 2013).

Moreover, key considerations, such as the contribution of C sequestration in biomass (C sequestration in plant biomass is considered in the climate change impact category as negative emissions) and soils to climate change, and land use change impacts on biodiversity, were explicitly modelled.

Biomass and thus C accumulation in biomass by different types of cocoa systems were determined from the literature, as described in Section 2.2. SOC sequestration associated with the different systems was estimated using the RothC model (Coleman and Jenkinson 2014), based on initial soil C content, annual inputs of organic matter (residual above- and belowground biomass, such as pruning and harvest residues, and necromass, plus organic fertilisers), and local pedoclimatic conditions (precipitation, evapotranspiration, temperature, soil density, soil clay content, etc.), according to the method and data sources described in Albers et al. (2022). The R script and CSV files with input data are included in the Supplementary Material. Two types of global agroecological zones (GEZ) (Fischer et al. 2012) were considered for the coastal region. Simulations had durations of 10 years for CCN-51 systems and 20 years for CFA systems (national average plantation ages reported in ESPAC 2019 are 8 and 24, respectively). Soil erosion by rainfall was considered, using the RUSLE2 model (Foster 2005), as soil erosion implies SOC losses (Lugato et al. 2016). The most dominant soil types in the different regions were chosen, according to two contrasting sources (FAO/IIASA 2009; Quesada et al. 2011).

A recent report (Deteix 2021) compared different theoretical frameworks for estimating biodiversity impacts associated with agricultural activities. Based on this report, UNEP recommendations (UNEP 2016, 2019), and CIRAD recommendations on LCA practice in developing and transition economies (Basset-Mens et al. 2021), the method described in Chaudhary and Brooks (2018) was retained. This method provides characterisation factors for land use (and land use change) impacts by country and ecoregion (Olson et al. 2001), expressed in terms of potential disappeared fraction of species per unit area (PDF·m⁻²), which includes five taxa (plants, mammals, birds, amphibians, and reptiles).

Water resource depletion was not explored, because water for agricultural use is relatively abundant in Ecuador (21% access to irrigation, yet only 20 out of > 4 000 sampled farms declared losses due to water deprivation), and several infrastructure projects facilitate access, although not without problems (Salmoral et al. 2018; Mohammadpour et al. 2019).

The resulting agricultural inventories are in the form described in Table 2 and presented in the Supplementary Material. The post-harvest and processing inventories are presented in Tables 3 and 4. The large difference in input intensity between mean and median-based inventories is noticeable and is further explored in Section 3.3.

Values presented consists of means followed by medians (in parenthesis).

The assessment of cocoa production impacts, disaggregated stepwise by producer type, region, cocoa variety, and system type, suggests several overlapping dynamics.

For instance, impacts per ha of cocoa increase along the gradient of producer types (from small subsistence to large producer, although medium producers feature slightly higher impacts than large ones), although yields per ha also increase along the same gradient (Fig. 2). This is due to the levels of intensification associated with the different producer types, which overcompensate for the economies of scale that are achieved. The per tonne weighted impacts of large and medium producers are slightly higher than those of small producers, implying that the high yields combined with intensification of the former do not represent an environmental benefit over the low yields and extensification of the latter. The single scores are dominated by climate change and land use. The impacts per ha are mainly on human health for large and medium producers (and for the national average) and on ecosystems for small producers (Supplementary Material).

When disaggregating production impacts only by producing region, it can be noticed that cocoa from Andean origin has lower impacts, both per ha and per t, than cocoa from other origins (Fig. 2). This is partly due to the absence of large commercial producers, and to the characteristics of Andean systems, which are generally extensive with low input intensities. When disaggregating impacts only by cocoa variety, impacts of CFA are considerably lower than those of CCN-51 only per t, but not per ha (Fig. 2).

Disaggregating results by type of producer x cocoa variety, the impacts per ha of cocoa of the large producers of CFA and CCN-51, as well as those of the medium producers of CCN-51, are considerably higher than those of the other combinations (Fig. 3). Climate negative emissions play a role on these differences, particularly for smallholders producing CCN-51, as the combination of very small inputs pressure and the accumulation of carbon in biomass contribute to larger climate negative emissions.

Further disaggregating by producing region (type of producer x variety x region), the impacts per kg of cocoa from large CFA producers on the Coast are considerably higher than for the other combinations (Fig. 4). This is due to the large contribution to the impacts of fertilisation (~ 45%) and irrigation (~ 25%) for CFA cocoa from large producers on the Coast. The net impact of CFA from medium-sized farmers in the Highlands is low, but the contribution of land use is very high (offset by C sequestration in biomass), due to the very low average yield of these farmers (0.07 t·ha⁻¹, based on a sample of two farms, one of which is extensive). Apart from these exceptions, at this level of disaggregation (by type of producer, by variety and by region), there appear to be major differences across impacts, with certain combinations clearly underperforming (e.g. small producers’ Amazonian CFA) as long as the particularities of these systems—i.e. multifunctional agroforestry systems—are not included in the assessment.

When disaggregating by type of producer x variety x type of farming system (monoculture, associated with other crops, agroforestry system), it is observed that there are large differences between the impacts per t of cocoa of different types of associations, which is due to the fact that, although relative yields do not vary, C sequestration in biomass varies significantly (i.e. it is much higher in agroforestry systems than in other associated systems) (Supplementary Material). In addition, impacts per ha of monoculture are systematically higher than those of associated systems, despite lower yields, due to the lower input intensity of systems in cultural association (Fig. 5).

The processing impacts (post-harvest, processing; Fig. 6) do not include the cocoa supply and the preceding processing steps. It is notable that thermal drying has higher impacts than solar drying, and that milk chocolate production has considerably higher impacts than the other types of chocolate, due to the impacts inherited from the other ingredients (milk powder, sugar).

If the impacts of chocolate production (including cocoa supply) are disaggregated by region, they do not vary by cocoa origin, except for chocolate based on Amazonian cocoa from agroforestry systems (Fig. 7). This is mainly due to the relative distances between production areas and Quito (the country’s capital, located in the northern Highlands), where most (if not all) cocoa processors are located. Aggregated impacts (i.e. cocoa + transport + processing) vary, but C sequestration in biomass varies as well, generally in the same proportion, resulting in similar net impacts between origins, varieties, and producer types.

A contribution analysis identifies the dominant sources of impacts and their relative importance by producer type, including production-weighted national average tonne of dry beans and hectare of cocoa plantation (Fig. 8):

The impacts of the sub-chains are ordered, from highest to lowest, as: Quality > Premium > Organic > Volume > Semi-processed (Fig. 9). This is mainly due to the differences in yields and negative direct field emissions among the different types of producer and cocoa varieties that feed the sub-chains. Transport contributes marginally to the impacts, as does primary processing in the case of the semi-processed sub-chain.

The identified sub-chains have different impact intensities, due to differences in yields between the different types of cocoa producers and varieties that feed the sub-chains. Transport contributes marginally to the impacts, as does primary processing in the case of the semi-processed products sub-chain.

All disaggregated midpoint results are included in the Supplementary Material.

The choice of median over mean values when building inventories has a noticeable effect on impact results (Fig. 10). Means-based impacts are considerably higher, but less representative of the actual practices of cocoa producers.

The variability of impacts associated with the variability of key contributing factors defining each combination of producer type, region, and variety—yield, intensity of fertiliser and plant protection product use, and even the method of calculating direct emissions—was explored. Pairwise comparisons, propagating uncertainty with Monte Carlo (1000 runs, 95% confidence), indicate, for example, that:

The variability observed across systems is large and must be considered for comparisons to be meaningful. Impacts are highly sensitive to yield and the amount of C sequestration in biomass (which is a function of variety, type of system, type of cultural association, and age of the plantation).

Monte Carlo uncertainty propagation did not consider the biodiversity impact category, as it is not included in EF v3.0, and thus was computed separately. All described Monte Carlo results are presented in the Supplementary Material.

Using the modified EF 3.0 method, the impacts of dry grain and chocolate from Ecuador were compared with equivalent processes available in ecoinvent, WFLDB and AGRIBALYSE, in terms of climate change (Tables 5 and 6). It is observed that the impacts of Ecuadorian beans are comparable to those of Brazil, and the impacts of both countries are considerably lower than those of the other exporting countries. The reasons are multiple, and include pressure on water use, and relative phytosanitary, fertilisation, and transport intensities, as well as energy use intensity and origin (i.e. nature of the energy mix) across origins, especially for chocolate products.

Land conversion plays a dominant role: for example, in Côte d'Ivoire's systems, land conversion accounts for > 90% of climate change impacts, while in Ecuador, because cocoa has been planted for more than 20 years in disturbed areas, the net sequestration of C in biomass is so significant that climate change impacts are low or negative (Table 7).

Ecuadorian yields have increased in recent years compared to the world average (Supplementary Material). The increase in cocoa production in Ecuador, and more generally in the Americas, follows a different strategy than in other continents (yield improvement vs. area expansion). It has been observed that Africa, the main cocoa producing continent in the world, increases its production by incorporating 1.23 million ha, with increases of less than 5% in yields (t·ha⁻¹), while Asia reduces its share in world production by 97 000 t, despite incorporating more than one million new hectares to production. In the other hand, the countries of the Americas increase their production through productivity increases of 84.2% and the incorporation of 124 000 ha of new production (representing an 8% increase in the area sown), in a more sustainable model than that used by the other continents (Arvelo Sánchez et al. 2017).

Carbon sequestration in agricultural systems, which takes place in plant biomass and soil organic carbon (SOC), is a function of several parameters: amount and permanence of biomass (above and below ground), amount and frequency of organic matter inputs, pedoclimatic conditions, and other agricultural practices (e.g. soil tillage, irrigation). The history of each agricultural plot is also a determinant of the impact on climate change, especially with regard to land use change (LUC), which is analysed in LCA over the last 20 years (e.g. Table 8). The underlying logic is that a change of land use from a more natural to a less natural environment (e.g. from forest to agricultural system; from agroforestry agricultural system to monoculture agricultural system) implies a loss of C (Brandão and Milà i Canals 2013; Koellner et al. 2013). The impact category LUC captures such dynamics, and there is even talk of “indirect LUC”, i.e. the indirect consequences on LUC dynamics in one system due to changes in another, albeit the validity/relevance of the concept is non-consensual (Finkbeiner 2013; Marvuglia et al. 2013). In the case of cocoa, for instance, a change in demand (from companies operating in Ecuador or from international markets) could imply additional LUC (e.g. deforestation or crop substitution). The exploration of such phenomena is part of the so-called consequential LCA (Zamagni et al. 2012); not adopted in this work.

In general, perennial systems sequester more carbon than annual systems, due to the persistence of the biomass and protection of the soil from erosion. For Ecuador, an aerial biomass accumulation curve (average between CCN-51 and CFA, Supplementary Material) was determined, from which the amount of C sequestered by cocoa plants in monoculture was derived.

The estimations of biomass and residual biomass accumulation for each type of system are summarised in Table 8.

The results (Supplementary Material) suggest a more interesting SOC sequestration potential in the CCN-51 systems than in the CFA systems (in terms of the respective annual sequestration rates), on the Coast. In the Highlands, CFA sequestration rates are higher than those of CCN-51. In Amazonia, too, sequestration rates are higher for CFA (especially in agroforestry systems) than for CCN-51, but in both cases, these rates are generally negative; i.e. there is a net loss of SOC. For Amazonian agroforestry systems, and only on Acrisol soil (among the different predominant soil types), sequestration rates of between 0.31 and 0.35 t·ha⁻¹·year⁻¹ were estimated. In contrast, Torres et al. (2014) suggest annual sequestration rates of 1.8 for CCN-51 and 0.47 for agroforestry CFA, but the authors did not indicate the soil type used in their calculations.¹ Cerri et al. (2006) report an annual SOC sequestration in Amazonian agroforestry systems of 0.83 t·ha⁻¹.

Only on the Coast, especially in drier areas, the net SOC (i.e. SOC at the end of the simulation minus total erosion) is positive, probably due to the intensity of rainfall erosion in the Highlands and in Amazonia. No estimate of SOC sequestration is sufficiently informative unless erosion is considered (Albers et al. 2022).

Agricultural expansion and other activities (mining, oil extraction, urban expansion) contribute to deforestation and environmental degradation in Ecuador. For example, in the period 1990–2000, 74 000 ha of forest were converted to other land uses annually (Cuesta et al. 2013). In the period 2008–2014, deforestation reached 47 500 ha per year (MAE 2015).

Monoculture agricultural production contributes to environmental degradation, including loss of biodiversity (Bonilla et al. 2016). This is particularly dangerous in the Amazonia, which has an extremely diverse and fragile ecosystem. In this region, the expansion of the agricultural frontier has historically been associated with colonisation (sometimes sanctioned by the state), extractive activities (oil, timber), and the expansion of the agricultural frontier (Viteri-Salazar and Toledo 2020) (Supplementary Material). In the Amazonia, there are already areas of overlap between different types of cocoa systems and protected areas. Nevertheless, native communities that produce via agroforestry systems (e.g. chakra system (Torres et al. 2014; Coq-Huelva et al. 2017)) are key actors for the preservation of natural and cultivated biodiversity.

The Ecuadorian ecoregions are Eastern Cordillera real montane forests (Highlands), Napo moist forests (Amazonia), and Western Ecuador moist forests (Coast). The corresponding characterisation factors are listed in the Supplementary Material. The impacts weighted by the different regional areas occupied by the different types of producers are presented in Fig. 11. It can be seen that, as the vast majority of the cocoa area is occupied by smallholders, their contribution to species loss is dominant, although these systems present a lower risk per ha. On the other hand, the potential impacts per ha are lower on the coast than in the highlands or Amazonia, largely due to the relative number of species between these regions.