Skip to main content
Dryad logo

Trade-offs between biodiversity and agriculture are moving targets in dynamic landscapes

Citation

Macchi, Leandro et al. (2020), Trade-offs between biodiversity and agriculture are moving targets in dynamic landscapes, Dryad, Dataset, https://doi.org/10.5061/dryad.msbcc2fvt

Abstract

  1. Understanding how biodiversity responds to intensifying agriculture is critical to mitigating the trade-offs between them. These trade-offs are particularly strong in tropical and subtropical deforestation frontiers, yet it remains unclear how changing landscape context in such frontiers alters agriculture-biodiversity trade-offs.
  2. We focus on the Argentinean Chaco, a global deforestation hotspot, to explore how landscape context shapes trade-off curves between agricultural intensity and avian biodiversity. We use a space-for-time approach and integrate a large field dataset of bird communities (197 species, 234 survey plots), three agricultural intensity metrics (meat yield, energy yield and profit), and a range of environmental covariates in a hierarchical Bayesian occupancy framework.
  3. Woodland extent in the landscape consistently determines how individual bird species, and the bird community as a whole, respond to agricultural intensity. Many species switch in their fundamental response, from decreasing occupancy with increased agricultural intensity when woodland extent in the landscape is low (loser species), to increasing occupancy with increased agricultural intensity when woodland extent is high (winner species).
  4. This suggests that landscape context strongly mediates who wins and loses along agricultural intensity gradients. Likewise, where landscapes change, such as in deforestation frontiers, the very nature of the agriculture-biodiversity trade-offs can change as landscapes transformation progresses.
  5. Synthesis and applications. Schemes to mitigate agriculture-biodiversity trade-offs, such as land sparing or sharing, must consider landscape context. Strategies that are identified based on a snapshot of data risk failure in dynamic landscapes, particularly where agricultural expansion continues to reduce natural habitats. Rather than a single, fixed strategy, adaptive management of agriculture-biodiversity trade-offs is needed in such situations. Here we provide a toolset for considering changing landscape contexts when exploring such trade-offs. This can help to better align agriculture and biodiversity in tropical and subtropical deforestation frontiers.

Methods

Bird sampling

We combined data from three extensive bird surveys in the study region (Decarre, 2015; Macchi et al., 2013; Mastrangelo & Gavin, 2012, Fig. 1). Birds were surveyed at 234 sites between 2009 and 2013 (see Appendix S1 in supporting Information). Each site was sampled using point counts, where all individual birds were identified to the species level. Sites were arranged along a gradient of remaining natural vegetation. This allowed us to apply a space-for-time approach in which landscapes with different shares of remaining natural vegetation can be interpreted as temporal stages along a deforestation trajectory. This is justified for the Chaco, as the region was dominated by natural vegetation until the 1990s, but has since turned into a global deforestation hotspot (Baumann et al., 2017; Kuemmerle et al., 2017).

Bird sampling sites were placed in all the main land systems in the region: 79 were located in natural woodlands, 16 in natural grasslands, 32 in subsistence ranching systems, 39 in silvopastoral systems, 27 in pasture systems, and 41 in croplands. Sites located in the same type of land system were spatially separated by at least 0.5 km (Fig. S1). Sampled land systems were distributed across the study area, and despite some were clustered regionally, land systems were generally well interspersed (see Methods in Decarre, 2015; Macchi et al., 2013; Mastrangelo & Gavin, 2012). The unbalanced number of sample units roughly represents the distribution of these land systems in the Dry Chaco (Baumann et al., 2017). It is important to note that currently in the Dry Chaco natural grasslands are very scarce and only a few species occurred exclusively there in our study. Bird assemblages of woodlands and natural grasslands were thus considered as baseline communities, as they both share the condition of being largely undisturbed by human activities. Species accumulation curves suggest that all land systems and their associated bird assemblages were adequately represented in our dataset (Fig. S2).

 

Agricultural intensity metrics

Adequately addressing agricultural intensity and its impact on the environment requires considering the different dimensions of intensification (Kehoe et al., 2017). We measured agricultural intensity at all bird sampling sites using three metrics (1) meat yield [kg / ha*year]; (2) energy yield [GJ / ha*year], and (3) profit [United State dollars USD / ha*year]. While meat and energy yields are descriptors of agricultural production, profit is the variable more directly related to the farmers’ expectations and decision making (Jobbágy & Sala, 2014).

We collected data on forage production along rainfall gradients for each land system, and estimated meat yield (secondary production) considering parameters of livestock systems destined to meat production in the Dry Chaco (Murray, Baldi, von Bernard, Viglizzo, & Jobbágy, 2016). In the case of croplands, we considered soybean production, which was transformed to pork live weight [kg / ha*year] using a specific 5:1 conversion ratio (Smil, 2013). Next, we converted meat from livestock systems and grain from croplands into energy yield following standard conversion metrics (USDA, 2011). Finally, we applied a net-return econometric model (Murray et al., 2016) to translate meat production per land system to profit, considering the corresponding production and transportation costs related to each land system (for a full description of the estimation of the agricultural intensity metrics see Appendix S2).

The agricultural intensity gradient showed marked differences in the yield values when production increased from natural to intermediate and on to highly-intensified production systems. For all the intensity metrics, croplands were the highest-yielding system (mean values of meat: 260.2 ± 51.3 standard deviation (SD) [kg / ha*year]; energy: 33.7 ± 19.6 SD [GJ / ha*year]; and profit: 189.9 ± 55.2 SD [USD / ha*year]), followed by silvopasture and pasture, and lastly the subsistence ranching system. Natural woodland and grassland without livestock were assumed to have zero yields in all metrics (Fig. 1 and Table S1).

 

Landscape composition and environmental conditions

We used covariates that reflected variation in landscape composition and environmental conditions within our study region. Considering the importance of habitat availability for determining species’ occurrence (Fahrig, 2013), we calculated the woodland extent within buffers of 6 km and 10 km around each sampling site (i.e. landscape spatial scale of c.100 km2 and c.320 km2 respectively). To do this, we used forest cover maps from the Global Forest Change datasets (30 m resolution, Hansen et al., 2013) for the bird sampling corresponding year.

Water availability is one of the main environmental constraints for agriculture in the Chaco (Houspanossian et al., 2016). Thus, we calculated mean annual rainfall (hereafter: rainfall) and an aridity index (hereafter: aridity) for all bird sampling sites based on weather stations data by the National Institute of Agricultural Technology (INTA). We interpolated mean annual rainfall using geo-statistics (semi-variograms and kriging). The aridity index was calculated dividing rainfall by evapotranspiration. Our covariates were generally only weakly correlated (Fig. S3).

Occupancy modelling

To assess the trade-off between avian biodiversity and agricultural intensity, we fitted trade-off curves between the two (Phalan et al., 2011). As a proxy for biodiversity, we estimated the probability of occupancy per species using a Bayesian framework (MacKenzie et al., 2006). Yet, occupancy models do not account for species’ abundance as occupancy is solely based on whether or not a species is detected at a site. Occupancy models have some important advantages in this context. First, occupancy models account for imperfect detection, as some bird species may be more common or easier to detect than others. Second, occupancy models control for different sampling effort, observer identity, and sampling period among surveys as part of the detectability model (MacKenzie et al., 2006). Third, occupancy models allow for the consideration of multiple covariates when assessing the biodiversity vs. agricultural intensity relationship (Kéry & Royle, 2016). Finally, a Bayesian modelling framework allows for the simultaneous assessment of agriculture-biodiversity curves for the entire community and for each individual species.

We employed multi-species occupancy models under a Bayesian framework (Kéry & Royle, 2016). To build the detection history, we used spatial replicates per site (9 for Macchi et al., 2013, 6 for Mastrangelo & Gavin, 2012, and 4 for Decarre, 2015). The detectability model included the covariates survey (i.e., data source), and openness (i.e., categorical variable for the habitat type at the sampling site: woodland or open vegetation). We assessed bird occupancy as the response variable with seven possible covariates: three agricultural intensity metrics (meat yield, energy yield, or profit), two landscape composition covariates (woodland extent for 6 km or 10 km buffers), and two environmental covariates (rainfall or aridity). In addition, each model included the interaction between the respective agricultural intensity metric and either woodland extent, rainfall or aridity, leading to a total of 24 different model combinations (Table S2). The interaction term allowed us to understand how the response of bird occupancy to agricultural intensity (e.g. meat yield) varied in relation to our environmental covariates (e.g. woodland extent). We fitted the models using only uncorrelated covariates (Fig. S3), and considered a covariate to have a strong effect when the 95% credible interval of the parameter estimate (CRI) did not overlap zero. We compared all models using Watanabe-Akaike information criterion (WAIC; Watanabe, 2010), which is a useful model selection criterion for hierarchical models (Broms, Hooten, & Fitzpatrick, 2016). Once we identified the best fitting model for the entire bird community, we examined this model in detail regarding responses at the species level. Specifically, we assessed the response curves for the 37 most common bird species (naïve occupancy of at least 10%) that showed a strong effect of the meat yield*woodland extent interaction. More details on the occupancy modelling are provided in the Supporting Information, Appendix S3.

Funding

Bundesministerium für Bildung und Forschung, Award: 031B0034A

DFG, Award: KU 2458/5-1

Fondo para la Investigación Científica y Tecnológica, Award: PICT 2006-1693

Instituto Nacional de Tecnología Agropecuaria, Award: PNNAT 1128053

Alexander von Humboldt-Stiftung