Using land‐use history and multiple baselines to determine bird responses to cocoa agroforestry

Abstract Agroforests can play an important role in biodiversity conservation in complex landscapes. A key factor distinguishing among agroforests is land‐use history – whether agroforests are established inside forests or on historically forested but currently open lands. The disparity between land‐use histories means the appropriate biodiversity baselines may differ, which should be accounted for when assessing the conservation value of agroforests. Specifically, comparisons between multiple baselines in forest and open land could enrich understanding of species’ responses by contextualizing them. We made such comparisons based on data from a recently published meta‐analysis of the effects of cocoa (Theobroma cacao) agroforestry on bird diversity. We regrouped rustic, mixed shade cocoa, and low shade cocoa agroforests, based on land‐use history, into forest‐derived and open‐land‐derived agroforests and compared bird species diversity (species richness, abundance, and Shannon's index values) between forest and open land, which represented the 2 alternative baselines. Bird diversity was similar in forest‐derived agroforests and forests (Hedges’ g* estimate [SE] = ‐0.3144 [0.3416], p = 0.36). Open‐land‐derived agroforests were significantly less diverse than forests (g* = 1.4312 [0.6308], p = 0.023) and comparable to open lands (g* = ‐0.1529 [0.5035], p = 0.76). Our results highlight how land‐use history determined the conservation value of cocoa agroforests. Forest‐derived cocoa agroforests were comparable to the available – usually already degraded – forest baselines, but entail future degradation risks. In contrast, open‐land‐derived cocoa agroforestry may offer restoration opportunities. Our results showed that comparisons among multiple baselines may inform relative contributions of agroforestry systems to bird conservation on a landscape scale.


PALABRAS CLAVE
In agroforestry research, different baselines -various kinds of forest, perennial monoculture, and open land -are commonly applied (Mupepele et al., 2021), but rarely in combination within the same study (Martin et al., 2020). In this context, considering multiple baselines may be particularly beneficial because agroforests can differ in land-use history (Martin et al., 2020), meaning they originate from different baselines (forests or open lands) (Fig. 1). A nonquantitative review highlights the importance of land-use history for ecosystem services and biodiversity in tropical agroforests (Martin et al., 2020). Authors of this article suggest that forest-derived agroforests typically degrade forests, whereas open-land-derived agroforests typically restore open lands. This path-dependency leads to contrasting outcomes for ecosystem services and biodiversity. Taking the land-use history of focal agroforests and multiple baselines into account may thus enrich understanding of the value of agroforests for biodiversity and ecosystem services.
One crop commonly farmed in agroforestry systems is cocoa, the most important ingredient of chocolate. Practiced across multiple tropical biodiversity hotspots (FAO, 2020), cocoa agroforestry has value for biodiversity (Bisseleua et al., 2009;Jarrett et al., 2021) and ecosystem services (De Beenhouwer et al., 2013). This value has been recognized in quantitative syntheses on biodiversity (Bennett et al., 2022;Maney et al., 2022) and ecosystem services (Niether et al., 2020) across various types of cocoa agroforestry systems. Nonetheless, cocoa agroforest expansion into forest is a key driver of forest loss in West Africa (Tutu Benefoh et al., 2018) and contributes to forest degradation in Latin America and Southeast Asia (Rice & Greenberg, 2000). But cocoa agroforest can also be established on historically forested open land. For example, on Sulawesi, Indonesia, 50% of cocoa plantations were established on open lands and 50% inside forests (Rice & Greenberg, 2000). Land-use history may also affect biodiversity (Kessler et al., 2009;Maney et al., 2022), ecosystem services (Nijmeijer et al., 2019), and labor requirements (Ruf, 2001) in cocoa agroforestry systems and might be itself influenced by policy (Orozco-Aguilar et al., 2021). Importantly, benefits of open-land-derived agroforests would likely turn into trade-offs if agroforests were established on naturally open lands, such as savannas. However, given the climatic niche of cocoa (Schroth et al., 2016), encroachment into forests appears to be a far greater risk than encroachment into naturally open lands (Tutu Benefoh et al., 2018).
In this light, a recent meta-analysis by Bennett et al. (2022) makes an important contribution to the understanding of bird responses to cocoa agroforestry. Their synthesis brings together data from 23 peer-reviewed articles in a comparison of rustic cocoa, mixed-shade cocoa, low-shade cocoa, and annual monoculture with a forest baseline, thereby combining studies with space-for-time designs (De Palma et al., 2018) and a single baseline (i.e., forest). Bennett et al. (2022) compared species richness, abundance, and Shannon's index values before refining their analysis for various functional guilds. The authors also looked at how various habitat features and landscape composition influence bird communities in cocoa agroforests.
We reanalyzed Bennett et al.'s (2022) data to demonstrate how considering land-use history and multiple baselines enriches understanding of the conservation value of cocoa agroforests for birds.

METHODS
To separate bird diversity estimates between forest-and openland-derived agroforests, we gathered information on the land-use history of focal agroforests in the introduction and method sections of 16 papers underlying the comparison of 3 bird biodiversity metrics (richness, abundance, Shannon's index values) in the meta-analysis by Bennett et al. (2022). Additionally, we extracted information on the human influence (e.g., selective logging, secondary vs. primary forest, fragmentation) on forest baselines from the introduction and methods sections of the same studies (Appendix S1). We renamed the land-use category "annual monoculture" (from Bennett et al. [2022]) as open land, in line with Martin et al. (2020). According to the underlying articles, the open land category includes predominantly annual crops, but also plantain (Harvey & González Villalobos, 2007) and pasture (Estrada et al., 1997;Estrada & Coates-Estrada, 2005)) (Appendix S1).
The separation based on land-use history revealed that 10 studies compared forest-derived agroforests with forests and 4 studies contrasted open-land-derived agroforests with forests. Two studies directly compared forest-and open-land-derived cocoa agroforests (Kessler et al., 2009;Reitsma et al., 2001). We used Bennett et al.'s (2022) data to provide additional results when land-use history and multiple alternative baselines are considered.
We excluded 2 studies (Schulze et al., 2004;Waltert et al., 2011) in which the same underlying data as in other studies were used (Waltert et al., , 2005 because their use of these data (Bennett et al. 2022) was pseudoreplication (Appendix S1). We also excluded Reitsma et al. (2001) because the study encompasses forest-and open-land-derived agroforests without separating the 2 during data collection and analysis, preventing the calculation of separate effect sizes. Furthermore, 2 studies took place at the same sites but with different data (Estrada et al., 1997;Estrada & Coates-Estrada, 2005), 1 of which included only Neotropical migrants (Estrada & Coates-Estrada, 2005). In this case, we followed Bennett et al. (2022) and included both. We also excluded 3 studies in which diversity measures were not applied to the entire bird community. This left us with 10 studies (Appendix S1).
To directly compare open-land-derived agroforests with open lands, we calculated Hedges' g* for this comparison of 2 effect sizes of different metrics from the same study ) (Appendix S2). We also calculated Hedges' g* of effect size for the 2 types of cocoa agroforests and the open lands relative to the available forest baselines. We operationalized this with the same methods and R scripts as in Bennett et al. (2022).
Before fitting Hedges' g* into a model, we ran a test of the heterogeneity of the data of the full community in the metacont function of R package meta 5.0.2 (Balduzzi et al., 2019). In line with Bennett et al. (2022), we found significant heterogeneity between studies for the comparison of all land systems with forests (Appendix S3). Thus, we built a linear mixed effect model to determine the difference between the 3 land systems (forest-derived agroforest, open-land-derived agroforest, and open land) and forests with the metareg function in the R package metafor 3.0.2 (Viechtbauer, 2010) with the study key as a random effect. We did not find significant heterogeneity for the comparison of open-land-derived agroforest with open land (Appendix S3). Therefore, to compare open-land-derived agroforests with open land, we used a simple linear model.  (Fig. 2a, Appendix S5).

Forest
The assessment of forest baselines in underlying studies in Bennett et al. (2022) revealed that only 3 studies compared agroforests with near-primary forests or mature forests, whereas 13 studies compared agroforests with fragmented, selectively logged, disturbed, used, or secondary forests (Appendix S1).

DISCUSSION
Our results showed that considering the land-use history of focal agroforests along with multiple baselines offers an opportunity to draw nuanced conclusions about the bird conservation value of different cocoa agroforestry systems.
Our findings are in line with Bennett et al. (2022) for rustic and mixed-shade forest-derived agroforests -these systems hosted a bird diversity comparable to forest baselines (Fig. 2a, Appendix S3). However, the recommendation "implementing rustic and mixed shade agroforestry systems" (Bennett et al. 2022) is controversial because rustic agroforests are by definition forest-derived (Moguel & Toledo, 1999), so establishing new ones will contribute to forest degradation and associated species turnover -as documented by Bennett et al. (2022). Considering multiple taxa, a recent analysis by Maney et al. (2022) also demonstrates significant decreases in diversity under the conversion of primary forests to forest-derived cocoa agroforests.
The forest baselines in the articles we analyzed represented fragmented (Faria et al., 2006), disturbed (Davies et al., 2015), partly secondary (Reitsma et al., 2001;Van Bael et al., 2007) or selectively logged forests (Greenler & Ebersole, 2015;Harvey & González Villalobos, 2007) (list of all studies in Appendix S1). Such forests typically have lower bird diversity than less disturbed primary forests -which may themselves lose species (Stouffer et al., 2021) -suggesting shifting baseline syndrome and an overestimated value of forest-derived agroforests for bird diversity. Nonetheless, we agree with the recommendation of maintaining already established biodiverse forest-derived agroforests, in line with Martin et al. (2020) and Raveloaritiana et al. (2021).
For low-shade intensified cocoa, we found that when land-use history was not considered and the comparison was only with forests, as in Bennett et al. (2022), interpretation challenges resulted that should be considered. All low-shade intensified agroforests included in Bennett et al. (2022) were established on open lands (Appendix S1). Considering those agroforests as the last step of an intensification from forests via rustic and mixed-shade cocoa to low-shade intensified cocoa is thus inaccurate. Instead, these open-land-derived low-shade intensified agroforests could have rehabilitated the open lands on which they were established, leading to possible gains in biodiversity. One study  included 2 estimates of bird diversity and data on bird diversity in open lands (i.e., annual cropping in Bennett et al. [2022]), enabling a direct comparison with an alternative baseline. This comparison revealed a diversity in open-land-derived agroforests similar to diversity in open lands (Fig. 2b, Appendix S3), but estimates were uncertain given the small sample size. However, Waltert et al. (2004) shows species turnover between the 2 land uses and much lower diversity in cocoa compared with forests, suggesting distinct bird communities in open-land-derived agroforests.
The only study included in Bennett et al. (2022) that directly compared forest-and open-land-derived agroforests (Kessler et al., 2009) Reitsma et al. (2001) mention that focal agroforests differed in land-use history, but did not consider this difference in their analyses.
Extrapolating to the landscape scale, our results suggest that the benefits of cocoa agroforestry for bird conservation can be best harnessed under the consideration of land-use history. Seeing open-land-derived agroforests as a restoration opportunity (Martin et al. 2020), rather than habitat degradation (Bennett et al., 2022), may help improve management practices so that agroforests deliver for conservation and production goals. For example, this view could help identify historically forested but currently open lands as priority areas for agroforestry systems promotion (Martin et al., 2020) or steer programs to increase shade tree diversity in open-land-derived agroforests , which could benefit birds (Gordon et al., 2007). In contrast, forest-derived agroforests could serve as buffer zones around protected areas or could be maintained as biodiverse elements within agricultural landscapes (Tscharntke et al., 2011). Evaluating the benefits of agroforestry in response to principal baselines may help make agroforestry a key element of complex agricultural landscapes.
We argue that future analyses and meta-analyses on biodiversity and ecosystem services in agroforestry systems should consider land-use history and multiple baselines. Here, going beyond forest and open land as broad categories may offer an interesting research avenue. Specifically, comparing forestderived agroforests with old-growth forests as well as selectively logged or secondary forests could give a more nuanced picture of the value of agroforests for biodiversity, possibly showing that they are less diverse than old-growth forests but comparable to logged or secondary forests. However, for open land, we were already short on estimates, so a further differentiation in various kinds of open lands would require additional empirical studies in cocoa agroforests.
We conclude that open-land-derived cocoa agroforests should not be dismissed simply because they have a lower bird diversity than forest-derived cocoa agroforests. Rather, by being established on historically forested open lands, they will contribute to agricultural production within working landscapes without worsening the status quo for biodiversity. Moreover, while forest-derived cocoa agroforests have higher bird diversity, they should not be the preferred form of cocoa production, especially if this entails the further transformation of remaining forests. Considering alternative baselines thus allows for more nuanced policies in the cocoa sector.