Introduction
Cocoa (Theobroma cacao L.) is a crucial component of chocolate, a globally consumed commodity (WCF 2018, ICCO 2021) with a market value projected to attain USD 190 billion by 2026 (Voora et al. Reference Voora, Bermúdez and Larrea2019). The demand for cocoa led to a record production of 5 million tonnes worldwide in 2021 (ICCO 2021), with production systems varying widely from sustainable agroforestry to intensive monocultures (Schroth et al. Reference Schroth, Garcia, Griscom, Teixeira and Barros2016, Amfo & Ali Reference Amfo and Ali2020, Gama-Rodrigues et al. Reference Gama-Rodrigues, Müller, Gama-Rodrigues and Mendes2021). Native to the Amazon biome (Wood Reference Wood, Wood and Lass1985), cocoa thrives in shaded environments with suitable soil and the presence of beneficial fauna, such as pollinators and pest controllers (Schroth & Harvey Reference Schroth and Harvey2007, Toledo-Hernández et al. Reference Toledo-Hernández, Wanger and Tscharntke2017). Traditional intercropping with native plants is considered vital for sustainable agriculture (Gama-Rodrigues et al. Reference Gama-Rodrigues, Müller, Gama-Rodrigues and Mendes2021); however, the rising interest in cocoa farming has led to the expansion of more intensive farming practices, such as full-sun cultivation (Tondoh et al. Reference Tondoh, Kouamé, Martinez Guéi, Sey, Wowo Koné and Gnessougou2015, Wainaina et al. Reference Wainaina, Minang, Duguma and Muthee2021). This shift increases tropical deforestation and creates conflicts between cocoa production and forest conservation. Understanding how cocoa productivity is affected by changes in landscape configuration and composition is crucial for developing sustainable agricultural practices.
In Brazil, the seventh largest cocoa producer (Hernandes et al. Reference Hernandes, Efraim, De Andrade Silva and De Castilho Queiroz2022), cocoa is mainly grown in two distinct biomes: the Atlantic Forest and the Amazon Forest (Gama-Rodrigues et al. Reference Gama-Rodrigues, Müller, Gama-Rodrigues and Mendes2021). Such forest ecosystems provide benefits beyond biodiversity conservation, including essential services for agriculture such as pollination, climate regulation and nutrient cycling (Cassano et al. Reference Cassano, Schroth, Faria, Delabie and Bede2009, Toledo-Hernández et al. Reference Toledo-Hernández, Wanger and Tscharntke2017, WCF 2018). Forest conservation also offers socio-environmental benefits, such as the preservation of wildlife, water resources, cultural heritage and overall quality of life for local communities (Small et al. Reference Small, Munday and Durance2017, Hipólito et al. Reference Hipólito, Sousa, Borges, Brito, Jaffé and Dias2019).
In both biomes, agroforestry systems (AFSs) are the predominant method used for cocoa cultivation. AFSs involve intercropping cocoa trees with other tree species (natural or planted), which help maintain forest cover (Braga et al. Reference Braga, Domene and Gandara2019, Gama-Rodrigues et al. Reference Gama-Rodrigues, Müller, Gama-Rodrigues and Mendes2021), provide shade for cocoa trees (Melo et al. Reference Melo, Batista, Costa, Vilar, França and Augusto2017, Xavier et al. Reference Xavier, Nascimento and Chiapetti2021) and enhance soil nutrients (Toledo-Hernández et al. Reference Toledo-Hernández, Wanger and Tscharntke2017). In the Atlantic Forest biome, where c. 70% of national cocoa croplands are concentrated, c. 70% of cocoa farmers use a specific AFS locally known as cabruca (Oliveira et al. Reference Oliveira, Waleska, Sambuichi and Filho2011). This system involves removing only the forest understory, leaving most native canopy species to shade the cocoa trees (Schroth et al. Reference Schroth, Garcia, Griscom, Teixeira and Barros2016). With up to 80 native trees per hectare (Melo et al. Reference Melo, Batista, Costa, Vilar, França and Augusto2017), cabruca is crucial for conserving Atlantic Forest remnants (Cassano et al. Reference Cassano, Schroth, Faria, Delabie and Bede2009, Sambuichi et al. Reference Sambuichi, Vidal, Piasentin, Jardim, Viana and Menezes2012) and mitigating climate change impacts (Nogueira et al. Reference Nogueira, Roitman, Carvalho, Soldati and Jacobson2019, Heming et al. Reference Heming, Schroth, Talora and Faria2022). In the eastern Amazon state of Pará, where cocoa production expanded significantly during the 1970s (Melo et al. Reference Melo, Batista, Costa, Vilar, França and Augusto2017), similar AFSs are used for some 80% of Pará’s production (Grilli Reference Grilli2022). This method is economically viable for recovering degraded areas in the Amazon while benefitting the local economy (Schroth et al. Reference Schroth, Garcia, Griscom, Teixeira and Barros2016). However, regardless of the region, cocoa AFSs have faced increasing pressure, particularly on large farms, to convert to monoculture practices with clonal varieties cultivated in full sun (Oliveira et al. Reference Oliveira, Partelli, Cavalcanti, Gontijo and Vieira2019, Igawa et al. Reference Igawa, de Toledo and Anjos2022).
The intensification of cocoa farming and the expansion of other economic activities, such as soybean and cattle production, can exacerbate negative effects on landscapes (Amaral et al. Reference Amaral, de Souza Ferreira Filho, Chagas and Adami2021, Skidmore et al. Reference Skidmore, Moffette, Rausch, Christie, Munger and Gibbs2021, Haddad et al. Reference Haddad, Araújo, Feltran-barbieri, Perobelli, Rocha and Sass2024). Although the benefits of natural forests for cocoa crops are well recognized (Baah et al. Reference Baah, Anchirinah and Amon-Armah2011, Utomo et al. Reference Utomo, Prawoto, Bonnet, Bangviwat and Gheewala2016, Toledo-Hernández et al. Reference Toledo-Hernández, Wanger and Tscharntke2017, Kongor et al. Reference Kongor, Boeckx, Vermeir, Van de Walle, Baert and Afoakwa2019), the role of landscape configuration in cocoa productivity is less well understood (Asubonteng et al. Reference Asubonteng, Pfeffer, Ros-Tonen, Verbesselt and Baud2018). Landscape structure, including the configuration and composition of crop and non-crop habitats, affects ecosystem services that are crucial for agriculture, such as pollination (Potts et al. Reference Potts, Biesmeijer, Kremen, Neumann, Schweiger and Kunin2010), pest control (Grab et al. Reference Grab, Danforth, Poveda and Loeb2018), soil fertility (Zhu & Meharg Reference Zhu and Meharg2015) and water quality (Cana Verde et al. Reference Cana Verde, Brandão, Souza and Silva2023).
We investigated historical land-use changes in Brazil’s primary cocoa-producing states and analysed cocoa productivity at small and large scales over the same periods and areas. We first used historical data on land-use changes and cocoa productivity at the municipal level to understand how changes in landscape structure have been associated with historical trends in cocoa productivity. We then used current landscape data (with detailed information on productivity for different size classes of farms) to explore which metrics of landscape structure better predict recent cocoa productivity for both small- and large-scale farms. Given cocoa’s dependence on ecosystem services and microclimatic conditions (Ruf Reference Ruf2007, Schroth et al. Reference Schroth, Läderach, Martinez-Valle and Bunn2017, Amfo & Ali Reference Amfo and Ali2020), we expected that increases in forest cover and decreases in fragmentation would have been highly beneficial for cocoa productivity. Additionally, we anticipated higher productivity on small-scale farms due to their typically more heterogeneous landscapes, which facilitate access to species that contribute to fruit production (Michalski et al. Reference Michalski, Metzger and Peres2010, Martin et al. Reference Martin, Seo, Park, Reineking and Steffan-Dewenter2016).
Materials and methods
Study areas
This study focuses on Brazil’s two primary cocoa-producing states, Pará and Bahia, which together account for 97% of the country’s cocoa production (IBGE 2021). These regions exhibit distinct landscape characteristics. Approximately 77% of Pará’s territory is covered by the Amazon rainforest (MapBiomas 2023), receiving annual precipitation of between 1500 and 3000 mm (Bastos et al. Reference Bastos, Pacheco and Figueiredo2002). In contrast, Bahia’s territory is covered by a mix of the Atlantic Forest, Cerrado and Caatinga biomes, with 49% of the state encompassing these areas (MapBiomas 2023), and annual precipitation ranges from 600 to 2000 mm, varying across the different biomes (Santos et al. Reference Santos, Santos, Santos, Santos and Lacerda2008, Sambuichi et al. Reference Sambuichi, Vidal, Piasentin, Jardim, Viana and Menezes2012). Both states experience a tropical climate according to the Köppen classification (Alvares et al. Reference Alvares, Stape, Sentelhas, De Moraes Gonçalves and Sparovek2013). The regions also differ in their cocoa production histories. Cocoa has been cultivated in Bahia since the mid-seventeenth century (Walker Reference Walker2007), whereas Pará’s cocoa production has expanded only within the last 50 years (Melo et al. Reference Melo, Batista, Costa, Vilar, França and Augusto2017).
Production data
We retrieved data on cocoa production (kg) and production area (ha) at the municipality level per year from the Municipal Agricultural Production database provided by the Brazilian Institute of Geography and Statistics (PAM/IBGE 2023) to estimate cocoa productivity (kg/ha). We used data on cocoa production and landscape from all cocoa-producing municipalities in the two selected regions for which data were available (161 in total: 101 from Bahia and 60 from Pará). This represents an area of cocoa cultivation of 551 636 ha (93% of the total cocoa production area in Brazil; PAM/IBGE 2023; Fig. 1 & Table S1).
Figure 1. Study area showing the two regions where the selected municipalities are located (shaded areas) in the states of Pará and Bahia, Brazil. The land-use and land-cover map are based on 2023 MapBiomas (https://mapbiomas.org).
Data were retrieved for two 3-year time periods (1985–1987 and 2019–2021) selected to align with the temporal range of spatial landscape data provided by MapBiomas (described below) up to the time of this study. For each municipality, we calculated the variation in productivity by dividing the average productivity for the recent years (2019–2021) by the average productivity for the earlier years (1985–1987; hereafter, ‘cocoa productivity ratio’).
To compare the effects of landscape structure metrics (see below) on recent cocoa productivity of small- and large-scale farms, we used data from the latest Brazilian Agricultural Census (Censo-Agro 2017). Based on information in the literature (Altieri et al. Reference Altieri, Funes-Monzote and Petersen2012, Castro & Teixeira Reference Castro and Teixeira2012, Scalco et al. Reference Scalco, Pigatto and Souza2017), we categorized rural properties with an area of 10 ha or less as ‘small’ and those with an area greater than 10 ha as ‘large’. Then, for each municipality size class, we calculated the productivity class as the average productivity of small and large farms (hereafter, ‘recent cocoa productivity’).
Landscape metrics
We retrieved land-use and land-cover maps for specific years (1985, 1986, 1987, 2015, 2016, 2017, 2019, 2020 and 2021) from MapBiomas (MapBiomas 2023), and for each municipality and year we extracted the following annual landscape metrics: forest cover, forest edge density, landscape diversity, forest fragment level, forest area and municipal area (this last metric was from IBGE 2021; Table 1). For each municipality and year, we also extracted information from the Municipal Agricultural Production database (PAM/IBGE 2023) on the percentage of land used for agriculture and the percentage of local agricultural land used for growing cocoa (ratio between cocoa-growing area and total agricultural area within the municipality).
Table 1. Description of the predictor variables used in the data analysis. The analyses were run separately for the two states. All predictors considered for statistical analysis had correlation values lower than 0.6. For the first objective (how changes in landscape structure have been associated with historical trends in cocoa productivity), landscape diversity change was excluded from the analyses due to the high collinearity with forest fragmentation change, and municipal area was initially considered but later excluded due to collinearity with recent forest percentage. For the second objective (metrics of landscape structure predicting recent cocoa productivity for both small and large farms), total cropland cover was excluded from the analyses due to the collinearity with cocoa cropland importance.
a Predictors not included in the analysis due to high collinearity. For more detailed information on collinearity analyses, see Table 2.
To understand how changes in landscape structure have been associated with historical trends in cocoa productivity, we estimated how much each of the landscape metrics changed over time in each municipality by calculating the ratio between the annual mean from the 2019–2021 period and the annual mean from the earlier 1985–1987 period. In this way, a value of 0.5 represents a reduction of 50%, and a value of 1 represents no change. After assessing collinearity between all landscape variables, we selected the following variables for this part of the analyses: forest-cover change (FCC), total cropland change, recent forest cover (RFC; 2019–2021), forest fragmentation change (FFC) and municipality area.
To explore which metrics of landscape structure better predicted recent cocoa productivity for both small- and large-scale farms, we estimated each metric of the municipality’s landscape by calculating the annual average from 2015 to 2017. We decided to use landscape data from the 2 years before the production data from the 2017 Census to account for a probable time-lag in the response of agricultural production to landscape changes (Ruf & Zadi Reference Ruf and Zadi1998; details on these metrics are given in Table 1). After assessing collinearity between all landscape variables, we selected the following variables for this part of the analyses: scale (small versus large), forest cover, cocoa cropland importance, forest fragmentation and landscape diversity.
We processed the 30-m-resolution annual maps of the regions provided by the MapBiomas Project (MapBiomas 2023) using QGIS software version 3.22 (QGIS 2021). The landscape metrics were calculated using the ‘landscapemetric’ package (Hesselbarth et al. Reference Hesselbarth, Sciaini, With, Wiegand and Nowosad2019) in R version 4.1 (R Team Core 2018).
Statistical analysis
All analyses were performed using R version 4.1 (R Team Core 2018). For each proposed objective, we used general linear models and applied model selection. For each response variable, we created a global model that included all of the selected predictor variables and their interactions (Table 1) and confirmed its validity by inspecting the normality and heteroscedasticity of its residuals. We selected the most parsimonious model(s) based on the Akaike information criterion (AIC) by using the ‘dredge’ function from version 1.43.17 of the ‘MuMIn’ package (Barton & Barton Reference Barton and Barton2020). We considered model(s) that had a a ΔAIC < 4 as ‘best’ model(s). This approach allows for the derivation of coefficients for important parameters that would probably be excluded when only considering the single best model (Burnham & Anderson Reference Burnham and Anderson2002). We also tested the spatial autocorrelation of the residuals of the full models with Moran’s I correlation values using the ‘spline.correlog’ function from version 1.2-9 of the ‘ncf’ package (Bjornstad & Cai Reference Bjornstad and Cai2020).
Changes in cocoa productivity over time
To test the effect of landscape changes on the cocoa productivity change (cocoa productivity ratio) during the analysed time period, we used general linear models (Gaussian distribution) with the predictors defined in Table 1. The cocoa productivity change was log-transformed to normalize the residuals. To minimize the influence of some municipalities with extreme values, total agriculture change was also log-transformed. We also considered the interaction between the levels of RFC and all landscape change metrics, as the effects of landscape change may differ. For example, these effects depend on how degraded the initial (and final) conditions were (e.g., a 50% decline in forest in a region that originally had 1% forest may have little impact, while 50% decline is substantial for an area that had 80% forest). Additionally, areas that currently had high forest cover were historically likely to have received less attention from producers. Since we found spatial autocorrelation in the data of municipalities from Bahia (Fig. S1), we added an autocovariate at a scale of 100 000 km in the global model state, calculated using the ‘autocov_dist’ function of version 1.1-8 of the ‘spdep’ package (Bjornstad & Cai Reference Bjornstad and Cai2020). Before running the analyses, we scaled and centralized the predictor variables to allow comparison of the regression coefficients. We then ran model selection (based on the AIC) to define the most parsimonious model(s).
Effect of farm scale on current cocoa productivity
To identify which parameters best predict recent cocoa productivity in small- and large-scale farms, we used a linear mixed model with the predictor variables described in Table 1. We normalized our response variable (recent cocoa productivity) through log-transformation. To test whether the effect of landscape on cocoa productivity depends on the scale of farming, we included the two-way interaction between scale and all other predictors in the global model. As we had two values per municipality (small-scale versus large-scale), we included the municipality as a random effect variable. Models were run using the ‘lmer’ function of version 1.1-24.1 of the ‘lme4’ package (Bates et al. Reference Bates, Maechler, Bolker, Walker, Christensen and Singmann2021). As different municipalities have different sizes and the extent of forest cover may strongly influence the potential of natural areas that provide ecosystem services, we used municipality area (ha) as a covariate. We included all municipalities that had at least one of the scale categories (large and small scales), resulting in 156 selected municipalities. We also scaled and centralized the predictor variables to allow comparison of the regression coefficients. We then ran model selection (based on the AIC) to define the most parsimonious model(s).
Results
The model selection for both cocoa productivity change and recent cocoa productivity yielded a group of models with equivalent explanatory power (ΔAIC > 4) in the two Brazilian states analysed (Pará and Bahia; Tables S2 & S3). We reported the results considering the effect of the cocoa predictor variables according to the group of best models instead of significance values.
We detected changes in cocoa productivity between 1985–1987 and 2019–2021, but with different variations between states (Table S2). While in Bahia most cocoa-producing municipalities experienced losses in cocoa productivity (change on logarithmic scale < 0), in Pará most municipalities experienced increases in cocoa productivity (change on logarithmic scale > 0; Fig. 2).
Figure 2. Effects of change in forest cover (a,d), total cropland change (e,h) and forest fragmentation (i,l) on cocoa productivity (kg/ha) over time in municipalities with low (10%) and high (90%) recent forest cover values for the states of Bahia (a,b,e,f,i,j; blue) and Pará (c,d,g,h,k,l; red). Forest-cover change, total cropland change, forest fragmentation change and cocoa productivity change over time were calculated as the ratios between average annual productivity (kg/ha) in 2019–2021 and 1985–1987. The values lower than 1 represent losses in forest cover (a,d). The positive values on the y-axes (above the dashed lines) represent increases and the negative values (below the dashed lines) represent declines in the cocoa productivity ratio. The positive values on the x-axes (right of the dashed lines) represent increases and the negative values (left of the dashed lines) represent declines in current forest cover. Data account for 161 municipalities in Pará and Bahia, Brazil. Estimates were extracted from the full model, and points represent the partial residuals (see statistical details in Supplementary Table 2a,b).
Do changes in landscape structure over time influence cocoa productivity?
In both states, cocoa productivity changes were positively related to FCC, but the effect was lower for Bahia despite the greater forest expansion over time (FCC > 0; Fig. 2a,b). In Pará, although no expansion of forest cover over time was detected (maximum FCC value = 0), higher cocoa productivity values were found in municipalities that conserved their forests at values close to the previous monitoring period (FCC ≃ 0). In this state, the positive relationship was more accentuated in municipalities that currently have low forest cover (i.e., an interaction effect was detected between FCC and RFC; Fig. 2c,d & Table S2).
Total cropland changes had a low effect on cocoa productivity over time in municipalities in Bahia (Fig. 2e,f & Table S2). However, the municipalities in Pará with low availability of RFC had experienced positive changes in cocoa productivity associated with total cropland expansion (Fig. 2g,h).
The effect of FFC on cocoa productivity was also associated with the level of RFC harboured by municipalities (an interaction was detected between FFC and RFC for both states; Table 2). However, similar to the other metrics, changes in forest fragmentation had minimal effect on cocoa productivity in Bahia (Fig. 2i,j). In Pará, municipalities with low levels of RFC experienced a negative impact on cocoa productivity due to increased forest fragmentation over time (Fig. 2k,l).
Table 2. Pearson’s correlation coefficients between the predictor variables considered for cocoa productivity ratio: forest-cover change (FCC), recent forest cover (RFC), total cropland change (TCC), forest fragmentation change (FFC), landscape diversity change (LDC), forest density change (FDC) and municipal area (MUA); and recent cocoa productivity: forest cover (FC), forest area (FA), cocoa cropland importance (CCI), total cropland (TC), forest fragmentation (FF), forest density (FD) and landscape diversity (LD). Results based on 161 cocoa-producing municipalities evaluated in the states of Pará and Bahia, Brazil. Correlated variables (coefficient ≥ 0.6) are indicated in bold.
Does the influence of landscape structure on current cocoa productivity differ between large and small farms?
The influence of landscape structure on recent cocoa productivity differed between small and large farms, but this was only evident for certain landscape metrics (Fig. 3 & Table S3). For example, the effect of forest fragmentation was negative for both the small and large farms of Bahia (Fig. 3e,f). In contrast, no negative effects were detected in Pará; instead, positive trends were observed in municipalities where cocoa cropland plays a significant role in agriculture (Fig. 3g,h). Similarly, landscape diversity had a positive effect on productivity of small- and large-scale farms in Bahia (Fig. 3i,j), but negative effects were detected in municipalities of Pará where cocoa cropland was important (Fig. 3l). The main difference detected between small and large farms related to the effect of forest cover in municipalities where cocoa production did not make up a substantial proportion of agriculture. In such municipalities, forest cover was positively related to the productivity of large-scale farms and negatively related to the productivity of small farms.
Figure 3. Effect of forest cover (a–d), forest fragmentation (edge density of natural forest; e–h) and landscape diversity (i–l) on recent levels of cocoa productivity (kg/ha) of small (≤10 ha, red lines) and large (>10 ha, red lines) farms from Bahia and Pará, Brazil. Data are shown for municipalities with low (a,c,e,g,i,k; cocoa makes up 10% of the farmland within the municipality) and high (b,d,f,h,j,l; cocoa makes up 90% of the farmland within the municipality) importance of cocoa farmland within the municipality. Data represent 156 municipalities in 2019. Estimates were extracted from the full model, and points represent partial residuals (see statistical details in Supplementary Table 3a,b).
Discussion
The extensive spatially explicit historical and recent cocoa productivity data provide evidence that forest loss and fragmentation have negatively impacted cocoa farm productivity. These effects were particularly pronounced on large-scale farms and in regions with more intensive cocoa management practices (such as Bahia) compared to regions where agroforestry cocoa production was more common (such as Pará). While our approach does not account for local environmental factors – such as soil quality, differences in cocoa varieties and cultivation and management techniques (Monroe et al. Reference Monroe, Gama-Rodrigues, Gama-Rodrigues and Marques2016, Gama-Rodrigues et al. Reference Gama-Rodrigues, Müller, Gama-Rodrigues and Mendes2021, Reges et al. Reference Reges, Maia, Sarmento, Silva, Santos and Damaceno2021) – which also play a significant role in agricultural profitability, the patterns we identified provide valuable insights for future management plans.
Do changes in landscape structure over time influence cocoa productivity?
The effect of landscape structure metrics on cocoa productivity was less pronounced in the municipalities of Bahia (Atlantic Forest) compared to those of Pará (Amazon Forest). This difference is probably attributable to the contrasting histories of cocoa farming in these two states. While there was an increase in forest cover during the study period in Bahia, the availability of natural forests remains critically low due to a long history of degradation. Currently, the Atlantic Forest in Bahia accounts for only 11.1% of its original extent (INPE 2022), in contrast to the 67% of remaining Amazon rainforest in Pará (MapBiomas 2023). In the south-eastern region of Bahia, where the majority of the state’s cocoa production is concentrated (Delabie et al. Reference Delabie, Argolo, Jahyny, Cassano, Jared and Mariano2011), primary forest cover was already very low in 1985 (the earliest period considered for comparison in this study), comprising just 8.88% of the total land cover (Landau et al. Reference Landau, Hirsch, Musinsky and WW2008). Consequently, municipalities that showed declines had limited losses compared to the reference year. In contrast, municipalities that saw increases in forest cover during the study period were those where the original forest cover was nearly non-existent. As a result, these areas showed very limited gains in forest cover, with the state accumulating a loss of 3000 ha over the study period (MapBiomas 2023). In contrast, in Pará (Amazon rainforest), where many municipalities still had extensive forest areas in 1985, nearly all municipalities faced declines in forest cover, with many experiencing reductions of 20% or more, and these relative changes were associated with mean losses per municipality of 140 000 ha of forest cover (MapBiomas 2023). The limited changes in forest area detected in Bahia probably resulted in insufficient variation during the study period to detect effects comparable to those observed in Pará, explaining the differences detected in our study. Further analyses with data prior to 1985 would be important to evaluate the effects of forest loss in Bahia, but unfortunately historical data availability on cocoa production is limited.
Despite the low amount of forest cover in Bahia and our lack of ability to detect effects of forest degradation in this state, we know that the remaining forest patches surrounding cocoa farms are crucial for sustaining cocoa pollinators (Toledo-Hernández et al. Reference Toledo-Hernández, Tscharntke, Tjoa, Anshary, Cyio and Wanger2021), and the absence of these pollinators could lead to a potential reduction in cocoa production of up to 100% (Klein et al. Reference Klein, Vaissière, Cane, Steffan-Dewenter, Cunningham and Kremen2007). In fact, the availability of surrounding forest cover generally enhances the profitability of perennial crops within the Atlantic and Amazon forests (Gama-Rodrigues et al. Reference Gama-Rodrigues, Müller, Gama-Rodrigues and Mendes2021, González-Chaves et al. Reference González-Chaves, Metzger, Paulo, Carvalheiro and Garibaldi2022), as elsewhere (Karp et al. Reference Karp, Mendenhall, Sandi, Chaumont, Ehrlich and Hadly2013, Berecha et al. Reference Berecha, Aerts, Muys and Honnay2015). Apart from pollination services, forest patches can supply high-quality water and improve soil conditions for agriculture (Decocq et al. Reference Decocq, Andrieu, Brunet, Chabrerie, De Frenne and De Smedt2016), also enhancing natural pest control for crops (Medeiros et al. Reference Medeiros, Martello, Almeida, Mengual, Harper and Grandinete2019). Indeed, the absence of forest patches and associated ecosystem services probably contributed to the spread of witch’s broom disease, which significantly impacted the state’s cocoa production in the 1980s and 1990s (Aguiar & Pires Reference Aguiar and Pires2019). Although the adoption of technological solutions, such as disease-resistant clones, has advanced cocoa management, this resistance tends to diminish over time due to the continuous pressure from the pathogen. Furthermore, the region’s increasing rainfall irregularity has hindered the recovery of productivity, resulting in slow and limited progress in yield improvements (Melo et al. Reference Melo, Batista, Costa, Vilar, França and Augusto2017). In contrast, in Pará, cocoa productivity has increased significantly, reaching levels nearly three times higher than those in Bahia, according to PAM/IBGE (2023).
In spite of the increases in cocoa productivity detected over time for Pará, these increases were heavily constrained by the loss of forest cover (almost null increases in municipalities that have lost 50% or more of their cover; Fig. 2c,d). This reinforces the idea that regions with extensive remaining forest areas are highly suitable for cocoa cultivation due to their greater biodiversity values and provision of ecosystem services important for crop profitability (Sassen et al. Reference Sassen, Van Soesbergen, Arnell and Scott2022). In addition, the municipalities of Pará with greater expansion of cropland over time had higher increases in cocoa productivity, especially when their RFC was <10% of municipal area (Fig.2g). This unexpected result could be related to specific patterns of expansion in the region. Over the past 30 years, cocoa cultivation in Pará has followed the state’s general agricultural expansion pattern, with production area increasing by c. 300% (PAM/IBGE 2023). Notably, c. 70% of cocoa plantations have been established on previously degraded lands, such as pastures (Venturieri et al. Reference Venturieri, Oliveira, Igawa, Fernandes, Adami and Júnior2022). This expansion trend has facilitated the establishment of AFSs in medium- to high-fertility soils, which can achieve high productivity values (Mendes Reference Mendes2017). Additionally, the productivity loss associated with the increasing fragmentation of forested areas over time in municipalities with lower percentages of RFC is probably linked to the expansion of intensive agriculture and livestock farming. In 2021, the area dedicated to agriculture in the state reached 22 million ha, nearly double the values seen in the 2000s (MapBiomas 2023). In Côte d’Ivoire, the unregulated expansion of cocoa monoculture contributed to over 37% of deforestation in protected areas (Kalischek et al. Reference Kalischek, Lang, Renier, Daudt, Addoah and Thompson2023); it faces significant production challenges due to drought and pest attacks (Krumbiegel & Tillie Reference Krumbiegel and Tillie2024). This reinforces the notion that agricultural expansion only has a positive relationship with productivity if it is extensive and occurs while maintaining well-conserved, non-fragmented natural habitats.
Does landscape degradation equally affect the cocoa productivity of small and large farms?
The observation that there are minimal differences in productivity between small and large farms in municipalities where cocoa is a major agricultural sector suggests that the distinctions between these two types of cocoa farms are not very pronounced in the studied regions. According to Rada and Fuglie (Reference Rada and Fuglie2019), Brazilian incentive policies have led to increased support for smallholders in the cultivation of perennial crops, enabling their productivity to match that of large-scale farmers. In contrast, in municipalities where cocoa is not the primary agricultural sector, large farms tend to be more productive in areas with higher forest cover, while small farms show the opposite trend. Despite the benefits of forest ecosystems for agricultural production (Medeiros et al. Reference Medeiros, Martello, Almeida, Mengual, Harper and Grandinete2019, Toledo-Hernández et al. Reference Toledo-Hernández, Tscharntke, Tjoa, Anshary, Cyio and Wanger2021), municipalities with significant forest cover are often located in remote regions (Kleinschroth et al. Reference Kleinschroth, Rayden and Ghazoul2019), which can limit access to technology and information (Campbell et al. Reference Campbell, Carvalheiro, Maués, Jaffé, Giannini and Freitas2018). In these contexts, smallholder farmers – often constrained by limited capital (Vaast & Somarriba Reference Vaast and Somarriba2014) – face significant challenges in adopting modern agricultural practices, particularly in areas where cocoa cultivation is not well established. This may reduce the profitability of their cocoa crops, even with favourable conditions provided by forest cover. Conversely, large-scale farms generally benefit from greater technical support (Teixeira et al. Reference Teixeira, Bianchi, Cardoso, Tittonell and Peña-Claros2021, Hu et al. Reference Hu, Li, Zhang and Wang2022), which helps optimize productivity by leveraging the advantages of forest resources in the region.
Despite detecting similar effects of forest fragmentation in small and large farms, the effects’ direction differed between states. In Bahia, which has the second highest deforestation rate of Atlantic rainforest in Brazil (INPE 2022), productivity was lower in municipalities with higher fragmentation, while in Pará no effect or weak positive trends were detected. Although cabruca – the main cocoa producing system – provides relatively high-quality conditions for biodiversity (Cassano et al. Reference Cassano, Schroth, Faria, Delabie and Bede2009), such benefits depend on the quantity, quality and spatial distribution of remaining native forest habitats (Schroth & Harvey Reference Schroth and Harvey2007). As mentioned above, conversion of forests to agriculture in this region was very intense, and farming systems depend heavily on chemical inputs. Due to the highly modified landscapes in their municipalities, cabrucas alone do not seem sufficient to retain the crucial services needed for cocoa farming in Bahia.
In Pará, where patches of Amazon forest are still vast in many municipalities, the fragmented forests are associated with a greater diversification of land-use activities, which is important for local biodiversity conservation that is relevant to agriculture (Benjamin et al. Reference Benjamin, Reilly and Winfree2014, Hipólito et al. Reference Hipólito, Boscolo and Viana2018, Aguilera et al. Reference Aguilera, Roslin, Miller, Tamburini, Birkhofer and Caballero-Lopez2020). Crop diversification is a strategy to address long-term declines in crop performance (Nelson et al. Reference Nelson, Patalee and Yao2022), helping maintain essential ecosystem functions (Tscharntke et al. Reference Tscharntke, Klein, Kruess, Steffan-Dewenter and Thies2005, Aguilar et al. Reference Aguilar, Gramig, Hendrickson, Archer, Forcella and Liebig2015, Weigel et al. Reference Weigel, Koellner, Poppenborg and Bogner2018). Indeed, the positive effect of fragmentation on productivity was detected only in small farms that were probably associated with more diverse landscapes where cocoa cultivation was less widespread. Conversely, in municipalities where local agriculture depended heavily on cocoa, increased landscape diversity was associated with greater productivity losses. This is probably due to the significant conversion of landscapes into simplified land-use systems, such as soybean monocultures and pastures, which are expanding across the state (Haddad et al. Reference Haddad, Araújo, Feltran-barbieri, Perobelli, Rocha and Sass2024), reducing environmental quality.
Conclusion
Our results demonstrate that cocoa production benefits from the conservation of forest cover. The continued expansion of conventional agriculture into natural areas and the change from traditional farming systems to full-sun cocoa plantations is likely to result in lower crop yields in the medium to long term for both small- and large-scale production. Sustainable agricultural practices such as AFSs, which support the maintenance of various environmental layers, including soil, microclimate and biodiversity, are essential for ensuring the long-term profitability of cocoa. The benefits of these practices are further amplified when they are integrated with the conservation of the region’s natural forest remnants. Given the high demand for cocoa in the international market and its economic significance for smallholders and vulnerable communities, cocoa AFSs offer an excellent model for sustainable agriculture of this kind in tropical countries. They demonstrate that agricultural development and forest conservation can indeed be mutually reinforcing.
Supplementary material
For supplementary material accompanying this paper, visit https://doi.org/10.1017/S0376892924000304.
Acknowledgements
We thank Carolina da Silva Carvalho for her suggestions at the beginning of this study and the Brazilian National Council for Scientific and Technological Development (CNPq, PQ307625/2021-4) for funding LGC.
Financial support
This work was supported by the Instituto Tecnológico Vale (grant number R100603.CA).
Competing interests
The authors declare none.
Ethical standards
Not applicable.
Open access
