A global assessment of animal community responses to agricultural management

Abstract

Transitions in agricultural management through crop and noncrop diversification on intensively managed croplands or pastures have benefited biodiversity. However, the extent to which agricultural management benefits species communities present in undisturbed ecosystems remains largely unclear. We quantified the animal biodiversity associated with a range of agricultural systems varying in crop and noncrop diversity and compared these with biodiversity associated with undisturbed reference ecosystems (mostly forests). We added data on agricultural management to three large databases—PREDICTS, GLOBIO, and a database compiled by Kuipers et al.—and grouped 107,386 observations from 151 studies into one of nine agricultural classes. We evaluated the animal biodiversity associated with the agricultural class, including the presence of threatened species, based on four biodiversity metrics (intactness, relative richness, compositional similarity, relative abundance). Biodiversity of monoculture annual and perennial croplands was least like that of reference ecosystems across biodiversity metrics. We found small biodiversity benefits of crop diversification and the presence of sparse trees on farmland. Biodiversity in agroforests and silvopastures was the most similar to reference ecosystems, probably due to a high density of trees, resulting in similar vegetation structure and diversity to undisturbed forests. Over time, biodiversity increased in agroforests, whereas the biodiversity of perennial croplands remained stable. Overall, our results suggested that the extent to which species associated with undisturbed ecosystems find refuge in agricultural areas is influenced by agricultural management, but different types of agricultural systems produced varying benefits. The outcomes of our study highlight the potential of introducing agricultural policies that aim to enhance agricultural management through tree planting and crop diversification to accommodate species that inhabit undisturbed ecosystems.

INTRODUCTION

Over one third of terrestrial land is used for agriculture (FAO, ²⁰²⁴; Foley et al., ²⁰⁰⁵), and human population growth coupled with increasing affluence is expected to intensify the demand for agricultural land (Marques et al., 2019). Agricultural expansion and intensification are major drivers of biodiversity loss (Newbold et al., 2015; Schipper et al., ²⁰²⁰). Recent modeling studies suggest a need for transitions to more biodiversity‐friendly agriculture to bend the curve of biodiversity loss (Leclère et al., 2020). The Kunming–Montreal Global Biodiversity Framework (GBF) acknowledges this need, setting a target to enhance biodiversity‐friendly agricultural practices by 2030 (CBD, ²⁰²²). However, current global models fail to distinguish between conventional and biodiversity‐friendly management, and only broadly differentiate between extensive and intensive land management (Alkemade et al., 2022). This limitation hampers our understanding of the effectiveness of transitions to biodiversity‐friendly agriculture and the potential trade‐offs it may produce, such as increased land needs due to decreased yield (López Rodríguez et al., 2024). Consequently, it constrains our ability to develop clear policy pathways to curb biodiversity loss. To address this knowledge gap, it is crucial to incorporate the impact of different types of agricultural management into biodiversity models (López Rodríguez et al., 2024). This requires the calculation of biodiversity estimates for both conventional and biodiversity‐friendly practices (López Rodríguez et al., 2024).

Agricultural practices that benefit biodiversity include planting of a variety of crops and noncrop species, such as forest trees or hedges within fields (Beillouin et al., 2021; Tamburini et al., ²⁰²⁰); maintaining landscape diversity (Estrada‐Carmona et al., 2022); and reducing the use of fertilizers and pesticides (Beaumelle et al., 2023; Kleijn et al., ²⁰⁰⁹; Sánchez‐Bayo & Wyckhuys, ²⁰¹⁹). Enhancing the diversity of crops in agricultural fields increases biodiversity because it enhances the variety of food available to species (Mala et al., 2020). The inclusion of trees and hedges in agricultural fields helps maintain animal biodiversity (Barrios et al., 2018; Lecq et al., ²⁰¹⁷; Siebert, ²⁰⁰²) because vegetation complexity from trees and hedges provides diverse food sources, microhabitats, and shelter for birds and small mammals (Auad et al., 2015; Silva et al., ²⁰²⁰). Higher tree densities mimic natural land biodiversity, whereas lower densities resemble abandoned land (Neita & Escobar, 2012). Furthermore, agricultural ecosystems with mature trees and hedges typically harbor higher biodiversity compared with those with younger growth stages (Gibbs et al., 2016; Klaa et al., ²⁰⁰⁵; Staton et al., ²⁰¹⁹). Reductions in fertilizer use, herbicides, and pesticides enhance mainly the abundances and occurrences of species affected by these inputs (Tscharntke et al., 2021). Finally, enhancing landscape diversity may support greater biodiversity by enhancing the number of habitats, food resources, and shelter at a landscape scale (Dunning et al., 1992; Estrada‐Carmona et al., ²⁰²²).

Most meta‐analyses assessing the biodiversity benefits of agricultural practices compare areas with increased crop and noncrop diversity with a simplified reference (i.e., a monoculture cropland or managed pasture without trees in a homogeneous agricultural landscape and a conventional level of fertilizer and pesticide use). Such meta‐analyses show that enhancing crop and noncrop diversity significantly increases biodiversity in agricultural land (Beillouin et al., 2021; Sánchez et al., ²⁰²²; Tamburini et al., ²⁰²⁰; Yahya et al., ²⁰²²). However, comparisons with a simplified reference do not clearly indicate whether enhancing crop and noncrop diversity helps increase the number and abundance of species specifically adapted to undisturbed ecosystems, such as pristine forests and natural grasslands. Prior studies suggest that this is not always the case; for instance, enhancing crop diversity provides little conservation value for vulnerable animal species that are adapted to undisturbed ecosystems (Petit & Petit, 2003). Yet, there are also cases in which endangered species find habitat in agricultural areas but not in adjacent undisturbed ecosystems (Nugent et al., 2022). For instance, endangered species may find habitat on pastures with low livestock densities, but not in adjacent undisturbed shrublands (Nugent et al., 2022). Therefore, it is vital to understand the extent to which species associated with undisturbed ecosystems and species from elsewhere find habitat in different types of farmlands. This requires a detailed biodiversity assessment in which different aspects of biodiversity, including species community composition and the numbers of species, are compared between areas differing in agricultural practices and undisturbed reference ecosystems.

We explored how agricultural management can support animal species inhabiting undisturbed ecosystems (i.e., undisturbed forests or grasslands). We also assessed the impact of agricultural management on biodiversity; undisturbed ecosystems were the reference. To do so, we compared the effects of nine agricultural management classes on global animal biodiversity, including vertebrates and insects, with undisturbed forests and grasslands as references. These classes encompassed crop combinations and varying densities of noncrop trees in croplands and pastures, as outlined by López Rodríguez et al. (2024). We employed four biodiversity metrics, including intactness, similarity, relative species richness, and relative total abundance, to assess biodiversity responses to agricultural management. We anticipated lower biodiversity values across all classes compared with undisturbed ecosystems. However, we expected classes with high tree density to exhibit higher biodiversity than those with fewer or no trees. Additionally, we expected classes with high crop diversity to exhibit higher biodiversity than conventional monoculture classes. We sought to provide biodiversity estimates that could be used to parameterize global biodiversity models to simulate effects of transitions to biodiversity‐friendly agriculture on biodiversity over space and time. Such simulations could help identify effective policy pathways to mitigate biodiversity loss, ultimately guiding policy actions that aim to reverse the trend of biodiversity decline (Leclère et al., 2020).

METHODS

To assess the effects of agricultural management on animal biodiversity, we first compiled biodiversity databases with data on the effects of land use on biodiversity. Second, we enhanced the biodiversity databases with information on agricultural management, ultimately differentiating different types of agricultural systems. These agricultural systems (hereafter classes) were defined based on crop and noncrop diversity: minimal (e.g., monoculture croplands), planting of trees (e.g., agroforests), planting of multiple crops in one field (e.g., multicultures), and crop type (e.g., annual or perennial crops). The classes were derived from the diversity gradient of an agricultural management classification that combined diversity and intensity gradients and was specifically developed to assess the effects of management on biodiversity for modeling purposes (López Rodríguez et al., 2024). Following the classification of agricultural classes, we applied linear mixed‐effects models to four biodiversity metrics to assess the effects of the agricultural classes on animal biodiversity. For some classes, we assessed biodiversity changes over time. Our methods resemble those of van ‘t Veen et al. (2025) on the effects of forest management on biodiversity.

Data collection and classification

We combined three established biodiversity databases, each containing datasets that compare the abundance of species found in different land‐use categories with their abundance in original, undisturbed ecosystems (combined database available at https://doi.org/10.5281/zenodo.15282094). The databases were PREDICTS (Hudson et al., 2017), GLOBIO (Alkemade et al., 2009), and a database compiled by Kuipers et al. (2023). Each database provided species‐ to genus‐level abundance data derived from individual empirical field studies conducted worldwide. The data in the databases were originally extracted from published scientific papers or reports. Each scientific paper or report represented a unique source, which may encompass one or more studies. These studies were differentiated based on the location, year, and land use under which the data were collected.

From the combined database, we selected the sources on agricultural land use, which concerned animal species, particularly amphibians, birds, insects, mammals, and reptiles. We chose these animal taxonomic groups because they are most often modeled in global and regional biodiversity models and because biodiversity assessments across these taxonomic groups were conducted priorly (Alkemade et al., 2009). This allowed us to compare our outcomes with those of previous studies and produce animal biodiversity estimates that could be used to parameterize biodiversity models. We decided to omit plants from this study due to data limitations; few studies in the databases compared monoculture annual and perennial croplands and natural reference sites. Likely there was limited incentive for researchers to compare such intensive agricultural land with undisturbed reference ecosystems because plants are actively removed and inhibited in the process of intensive crop production.

Each of the studies selected from the databases included data gathered in areas affected by agricultural land use and data gathered in an undisturbed reference site. The reference sites were defined as terrestrial ecosystems, such as forests or grasslands, where no human land use had occurred over at least 80 years. The authors of papers from which the data were retrieved referred to the sites as reference sites or primary, natural, undisturbed, or pristine forests or grasslands. The reference site occurred in the vicinity of the site subject to land use and harbored an ecosystem that would have occurred in the agricultural sites had they remained untouched by humans. Most studies in the database included reference sites that harbored forest ecosystems. Croplands and pastures in these studies were, therefore, mainly converted forests.

The three databases did not yet include specific data on the agricultural management of the agricultural sites. To address this, we revisited the original sources and supplemented the databases with information on the agricultural classes outlined by López Rodríguez et al. (2024). We identified nine agricultural classes: agroforests, sparse trees on cropland, silvopastures, sparse trees on pastures, monoculture annual crops, monoculture perennial crops, multiculture annual crops, multiculture perennial crops, and pastures (Table 1). Agroforests, silvopastures, sparse trees on cropland, and sparse trees on pastures could be classified as types of agroforestry, which was defined as the presence of trees on agricultural land. Agroforests were, therefore, defined as a specific type of agroforestry, which concerns the cultivation of crops under tree canopies (see Table 1 for a more extensive definition). For perennials and agroforests, we recorded the time since establishment if authors provided this information. We hypothesized that biodiversity increases with both an enhanced diversity of crops and the presence of trees. We hypothesized that sites where crop or livestock production occurs under the canopy, as seen in agroforests and silvopastures, harbor biodiversity that is more similar to that of reference sites than sites where crop diversity is enhanced. We expected this because trees provide shelter, and because layers in the canopy produce a complex vegetation structure, which is associated with higher biodiversity (August, 1983). Table 1 includes more detailed hypotheses.

TABLE 1.

Agricultural management classes applied based on López Rodríguez et al.’s (2024) classification and associated hypotheses.

Class	Description	Hypothesis
Monoculture annual crops	Agricultural land on which only one annual crop species is planted (e.g., corn); crop rotation may occur; no trees are present.	The biodiversity levels across biodiversity metrics are lowest for monoculture annual crops because of the homogenous vegetation structure and diversity and the annual disturbance of land to prepare it for new production rounds.
Monoculture perennial crops	Agricultural land on which only one perennial crop species is planted (e.g., sugarcane, oil palm); no other trees are present.	Biodiversity is slightly higher in monoculture perennial crop areas than annual crop areas because crops grow over multiple years, resulting in lower levels of disturbance of the land. The vegetation structure and diversity of croplands do not become more similar to those of reference sites. Thus, there is no increase in biodiversity over time.
Multiculture annual crops	Agricultural land on which two or more annual crop species are planted at the same time (e.g., corn and beans); crop rotation may occur; no trees are present.	Biodiversity in multiculture annual crop is slightly higher than in monoculture annual crop areas because crop variety increases the diversity of vegetation, resulting in more food resources for species and a wider food web.
Multiculture perennial crops	Agricultural land on which two or more perennial crop species are planted at the same time (e.g., banana combined with coconut).	Biodiversity in multiculture perennial crops is slightly higher than monoculture perennial crops because crop variety increases the diversity of vegetation, which results in more food resources for species and a wider food web. The structure and vegetative diversity of the croplands do not become more similar to those of reference sites. Therefore, there is no increase in biodiversity over time.
Sparse trees on cropland	Cropland where trees are present at low densities; trees are used for timber, as windbreaks, or for tree crop production; unlike agroforests, this setup lacks distinct layers of crops and canopy because insufficient trees occur for crops to grow under the canopy.	Sparse trees on agricultural land slightly enhance biodiversity in croplands because they provide habitat and shelter for additional species.
Agroforests	Agricultural land where crops are grown under a canopy of cultivated trees (e.g., for timber, food, energy, or other products or to enhance the growth of crops); can involve selective removal of certain trees from naturally regenerated forests, allowing crops to grow beneath and alongside the remaining predominantly native species; can be established by planting or seeding either native or non‐native trees directly onto farmland to create an agroforest; characterized by its layered structure, featuring multiple strata of crops and trees and a layered canopy.	Crop production under the canopy produces layers in the canopy, resulting in a more complex vegetation structure and diversity than in annual and perennial tree crops and sparse trees on agricultural land. This results in significantly higher levels of biodiversity than found in other agricultural classes. As agroforests mature, vegetation structure and composition become more similar to those of reference sites.
Pasture	Grasslands on which livestock graze (e.g., cows) either temporarily or permanently; grass is planted or grows naturally; grass species may be non‐native, native, or mixed; no differentiation is made between intensities of livestock grazing or types of livestock.	Variation in biodiversity is high because different intensities of grazing affect biodiversity in varying ways, where heavy grazing affects biodiversity more than light grazing, and biodiversity in pastures is higher than in monoculture annual and perennial crops because pastures are more similar to naturally occurring ecosystems (i.e., natural grasslands) than croplands.
Sparse trees on pastures	Trees used for timber, tree crop production, or shade; livestock graze under the scattered trees temporarily or permanently; grass may be planted or naturally occurring, including species that could be non‐native, native, or a mix of both; unlike for silvopastures, insufficient trees occur for livestock to be fully grazing under a canopy of trees.	The few trees on pasture land slightly enhance biodiversity in pastures because they provide habitat and shelter for additional species.
Silvopasture	Designated for the planting or seeding of one or more tree species, whether native or non‐native, intended for wood or energy production; livestock graze in pastures under and between trees.	Biodiversity in silvopastures is higher than in monoculture and multiculture crop areas and areas with sparse trees on agricultural land because silvopastures resemble forests. Trees provide food and shelter for species, and the vegetation structure is more complex than in conventional croplands.

Apart from the general assessment on animal biodiversity, we assessed the effects of agricultural management on endangered species. To assess endangered species, we retrieved the IUCN Red List data for all taxonomic groups included in our study (available at https://www.iucnredlist.org/, accessed 29 August 2025). The Red List data provide the binomial scientific names of species along with their extinction risk status (IUCN, ²⁰²⁴). We integrated these Red List data with our biodiversity database by aligning the binomial names of the species in our database with those listed on the IUCN Red List. We included this assessment because it provided additional information on the types of animal communities occurring in undisturbed ecosystems that can inhabit agricultural land and the extent to which changes in agricultural management can enhance habitat provisioning for animal species. As one of the main threats to species is agriculture (Balmford et al., 2025), we assumed larger impacts on endangered species alone compared with all species, even though some endangered species depend on extensive agriculture and may compensate for this effect (Nugent et al., 2022).

Biodiversity metrics

We evaluated animal biodiversity with four distinct metrics: biodiversity intactness (IN), relative richness (SR), compositional similarity (SI), and relative total abundance (TA). These metrics compare different elements of species composition between agricultural sites and reference sites. The formulas used to calculate the biodiversity metrics were provided by van ‘t Veen et al. (2025).

Biodiversity intactness (IN) indicates how closely the biodiversity of a given agricultural class resembles the reference sites, given the types and abundances of species present. IN was quantified using the Mean Species Abundance (MSA) of originally occurring species (Schipper et al., 2020). First, we produced response ratios by dividing the abundances of species in agricultural sites by those in reference sites. If a species occurred in an agricultural site but not in a reference site, we removed the species from the analysis. We did this because IN should reveal how similar the species composition of an agricultural site is to an undisturbed site. If the abundance of a species in the agricultural site was higher than the abundance of that same species in the reference site, the response ratio was set at 1 (i.e., maximal biodiversity intactness). This was done to avoid overestimating IN due to high abundances of generalist or opportunistic animals in agricultural sites. The resulting ratios were subsequently averaged per study and divided by the total number of species occurring in reference sites. Therefore, IN purely captures how similar the species compositions of agricultural sites are to that of reference sites, revealing the extent to which agricultural land under different agricultural practices can provide habitat for species adapted to undisturbed ecosystems. As the databases mainly included reference sites harboring forest ecosystems, biodiversity intactness in our study mainly captured the extent to which agricultural land under different management practices provided habitat for species adapted to undisturbed forests.

Compositional similarity (SI) measures the number of species that agricultural and reference sites have in common and is highest when classes have more species in common and less species different. We assessed SI with the Sørensen similarity metric (Sørensen, 1948), which reaches its peak when the category has the same types of species present in the reference as the agricultural sites (formula in van ‘t Veen et al. [2025]).

Relative species richness (SR) measures the response ratio of the number of species occurring in agricultural sites to that in reference sites. Total abundance (TA) measures the response ratio of the sum of the abundances of all species in agricultural sites to that in reference sites. To correct for a right‐skewed distribution, the SR and TA response ratios were log‐transformed.

We compared the four biodiversity metrics because each captured a different aspect of biodiversity. By comparing IN and SI, we could ascertain whether reductions in intactness stem from a decline in species uniquely suited to reference sites or from a decrease in their abundance. Furthermore, analyzing IN and SI alongside SR helped us determine whether species adapted to reference environments were being replaced by others or if there was a notable overall change in species communities in response to agricultural management. Lastly, examining IN in conjunction with TA provided insight into whether declines in abundance were confined to species suited to reference sites or if changes in management resulted in a wider decrease in species populations.

We generated a Spearman correlation matrix to evaluate the relationships between biodiversity metrics. The results of this analysis helped us determine the degree of complementarity and redundancy among the biodiversity metrics.

Statistical analyses

We conducted statistical analyses with R 4.3.1 using the glmmTMB and lme4 packages (R Core Team, ²⁰²³). Spatial data distributions were visualized in QGIS (QGIS Development Team, ²⁰²³).

Assessing agricultural management effects on animal biodiversity

We adopted a similar approach to van ‘t Veen et al. (2025) to test the effects of agricultural management on animal biodiversity. We analyzed the effects of each agricultural class by constructing separate generalized linear mixed‐effects models for each of the biodiversity metrics, with the metrics as response variables and the classes as fixed effects. To account for variability among datasets and potential data nonindependence, we included nested random effects for source–study combinations (i.e., 1|Source/Study). Given that the predicted values for both IN and SI are constrained between 0 and 1, we employed beta‐regression models with a logit‐link function and used the glmmTMB package for these two metrics (Brooks et al., 2017; Ospina & Ferrari, ²⁰¹²). Because beta‐regression cannot handle values of exactly 0 or 1, we applied the Smithson–Verkuilen transformation (Smithson & Verkuilen, 2006) to adjust IN and SI metrics to fall within the (0,1) range. To estimate SR and TA, we utilized Gaussian linear mixed‐effects models and the lmer function of the R package lme4 because the two metrics can have values exceeding 1 and our data conformed to a Gaussian distribution. We performed Tukey's post hoc tests with Bonferroni correction to test for significant differences in biodiversity between agricultural classes. We conducted the tests with the glht function from the R package multcomp (Hothorn et al., 2024). For the presentation of results, we back‐transformed the log‐response ratios and logit‐link transformations. We assessed whether agricultural classes significantly explained variation in the data as an explanatory variable by conducting Omnibus tests. These likelihood‐ratio chi‐square tests were applied to the database with all the classes together.

We conducted an additional set of analyses focusing on threatened animal biodiversity by applying the same modeling approach outlined above but used SR, SI, and TA of threatened animal species as our response variables. We developed distinct models for various combinations of conservation status categories defined by the IUCN Red List to examine whether agricultural management effects differed among these categories. The category combinations analyzed were species classified from least concern to critically endangered, species from near threatened to critically endangered, and species from vulnerable to critically endangered.

Temporal changes in biodiversity for agricultural classes with perennials

For agricultural classes with perennials (i.e., agroforests, monoculture perennials, and multiculture perennials), we assessed the influence of time since the establishment of perennials on biodiversity outcomes. We constructed similar generalized linear mixed‐effect models as those described in Section “Assessing agricultural management effects on animal biodiversity”, adding the years since the establishment of the perennials as an additional explanatory variable instead of the classes. We built one model for agroforests and one for perennials (both monoculture and multiculture perennials because insufficient data were available to analyze these classes separately). We built these models for these classes because we only obtained sufficient temporal data for these classes (more than five data points). For the presentation of results, we back‐transformed the log‐response ratios and logit‐link transformations.

Robustness analyses

We assessed the sensitivity of the models to changes in the data pool by randomly omitting 10–40% of the data sources from the database 20 times and measuring the resulting variance in the predicted estimates for each biodiversity metric. Larger variance indicated larger sensitivity. Furthermore, we tested whether geographical location influenced our modeling results. To do so, we repeated the modeling exercise of Section “Assessing agricultural management effects on animal biodiversity” and added the continent on which the data were gathered as additional fixed effect besides the nine classes. We then used analysis of variance (ANOVA) to compare the model with only the classes as fixed effects and the model with both the classes and continent as fixed effects. We used the Bayesian information criterion (BIC) derived from ANOVA to assess the added value of including continents as an explanatory variable. This provided insights into whether spatial variability in the data contributes to explaining variation in biodiversity beyond the effects of agricultural management.

RESULTS

Data

In total, the database included 151 studies, obtained from 133 sources, where agricultural classes could be identified, with in sum 107,386 observations of 6523 species. Each study could include data on multiple classes and on different taxonomic classes. Most studies were from Africa (40), followed by South America (36), North America (34 studies, mainly from Central America), Asia (22), Oceania (12), and Europe (7). Not all agricultural classes were represented on every continent. North America featured studies on nine classes, followed by Africa with eight, South America, Oceania, and Asia with six, and Europe with five (Figure 1). The representation of taxa varied, with mammals being the most frequently studied group (79 studies), followed by insects (58), birds (23), reptiles (22), and amphibians (21). The high representation of mammals is largely due to the mammal‐centric meta‐analysis database by Kuipers et al. (2023). From the classes identified by López Rodríguez et al. (2024), crop–livestock integrated classes and landscape diversity effects on biodiversity could not be assessed due to lack of data. Additionally, data constraints prevented us from assessing threatened species in agroforests, sparse trees in cropland, and monoculture of perennial crops classes.

FIGURE 1.

(a) Distribution of empirical, animal biodiversity studies by agricultural class and the number of studies per country (pie charts) in three large datasets (see text) (black circles, scaling reference) and number of animal diversity studies per agricultural class by (b) continent and (c) taxonomic class.

Biodiversity metrics associated with different agricultural classes

Agricultural class significantly explained variability in the data for intactness, similarity, and relative richness, indicating that agricultural management was associated with biodiversity. For TA, the classes did not explain variability in the data, indicating that the total abundance of animal species did not vary substantially in response to agricultural management. All biodiversity metrics were positively correlated with each other, with correlations ranging between 0.1 and 0.9 (Appendix S1). The high correlation (0.9) was found between IN and SI because both focus on the common species in agricultural and reference ecosystems.

Intactness (IN)

Our analysis revealed that IN was significantly lower in all agricultural classes compared with reference sites (95% CI 0.12–0.71) (Figure 2a; Appendix S2). Among the classes, monoculture annual (95% CI 0.12–0.31), monoculture perennial crops (95% CI 0.23–0.44), and pastures (95% CI 0.24–0.45) exhibited the lowest levels of IN, and agroforests (95% CI 0.48–0.71), silvopastures (95% CI 0.25–0.65), and multiculture perennial crops (95% CI 0.27–0.63) showed the highest levels relative to reference sites. We found slightly lower IN in monoculture annual than monoculture perennial crops. Agroforests demonstrated significantly greater IN compared with monoculture annual (p = 0.0001), monoculture perennial (p = 0.0293), and pasture classes (p = 0.0453) (Appendix S2).

FIGURE 2.

Impacts of agricultural class on animal (a) intactness, (b) compositional similarity, (c) relative species richness, and (d) relative total abundance (classes described in Table 1) (points, estimates; whiskers, 95% confidence intervals; n, number of studies for each class; rr, number of response values used to determine intactness; o, number of species shared between agricultural and reference sites; s, total number of species present in agricultural and reference sites; a, number of abundance values used to calculate relative total abundance; dotted line, level of biodiversity in undisturbed ecosystems used as a reference). Each study may encompass different agricultural classes. Separate biodiversity values were calculated for each class.

Compositional similarity (SI)

Monoculture annual (95% CI 0.36–0.71) and monoculture perennial (95% CI 0.55–0.81) crops exhibited the lowest levels of SI, whereas agroforests (95% CI 0.86–0.95), silvopastures (95% CI 0.61–0.94), and multiculture annual crops (95% CI 0.60–0.97) showed the highest SI levels (Figure 2b; Appendix S2). We found slightly lower SI in monoculture annual than monoculture perennial crops. Agroforests demonstrated significantly higher SI than monoculture annual (p = 0.0002), monoculture perennial (p = 0.0095), and pasture classes (p = 0.0303) (Appendix S2). We only assessed SI for monoculture perennial crops (95% CI 0.26–0.86), agroforests (95% CI 0.87–0.99), and sparse trees on cropland (95% CI 0.43–0.94) for vulnerable to critically endangered animal species (Figure 3; Appendix S2). None of the other conservation status categories showed differences with reference sites.

FIGURE 3.

Impacts of agricultural class on threatened animal (a) relative richness, (b) compositional similarity, and (c) relative total abundance (classes described in Table 1) (points, estimates; whiskers, 95% confidence intervals; n, number of studies for each class; s, total number of species found in agricultural and reference sites; o, number of species shared between agricultural and reference sites; a, number of abundance values used to calculate relative total abundance; dotted line, level of vulnerable to critically endangered biodiversity in undisturbed ecosystems used as reference).

Relative species richness (SR)

We observed notably lower SR in monoculture annual crops (95% CI 0.53–0.94), monoculture perennial crops (95% CI 0.52–0.80), and pastures (95% CI 0.61–0.94) than in reference sites, whereas we found no significant differences in SR between other classes and reference sites (Figure 2c; Appendix S2). Regarding differences among classes, we only found significantly lower SR in monocultures of perennial crops than in agroforests (p = 0.027, 95% CI 0.88–1.42) (Appendix S2). Monoculture perennial crops (95% CI 0.45–0.98) showed lower SR of vulnerable to critically endangered animal species than reference sites, and the same accounted for sparse trees on cropland (95% CI 0.15–0.47) (Figure 3; Appendix S2). None of the other Red List categories significantly differed between agricultural and reference sites. SR of endangered species is much lower in sparse trees in cropland (0.27 on average), than for all species in the same category (0.67). However, only two studies were involved.

Relative total abundance

Overall, we found a similar TA between classes and reference sites, except for multiculture annual crops (95% CI 0.16–0.91), which had significantly lower TA compared with reference sites (Figure 2d; Appendix S2). We observed no significant differences in TA across different classes (Appendix S2). Monoculture perennial crops (95% CI 0.11–0.37) also showed significantly lower TA of vulnerable to critically endangered animals compared with reference sites (Figure 3; Appendix S2).

Temporal changes in biodiversity for agricultural classes with perennials

Temporal changes in biodiversity were assessed for agroforests and perennial crops (both mono‐ and multiculture). For agroforests, we applied a log transformation to the time since establishment as we observed a logarithmic pattern in biodiversity metrics suggesting that biodiversity initially increases rapidly following management interventions but then slows down over time. We found that IN (χ ² _1, 15 = 77.127, p ≤ 2.2 × 10⁻¹⁶) and SI (χ ² _1, 15 = 40.482, p = 1.98 × 10⁻¹⁰) rapidly increased in the first years since establishment in agroforests and then increased more slowly around 30 years post‐establishment. In contrast, we observed stable IN (χ ² _1, 11 = 1.5149, p = 0.2184) and SR (χ ² _1, 11 = 0.9544, p = 0.3286) over a 40‐year period following the establishment of perennial crops, and a slight, but statistically significant, decline in SI (χ ² _1, 11 = 3.8845, p = 0.04873) over this period (Figure 4; Appendix S2).

FIGURE 4.

Relationships between time since establishment of perennials and biodiversity metrics (relative total abundance, intactness, relative richness, and compositional similarity) for (a) agroforests and (b) a combination of mono‐ and multiculture perennial tree crops for which sufficient data were available to develop generalized linear mixed‐effect models (classes defined in Table 1) (N.sp, number of observations per biodiversity metric for each time point since establishment of perennial vegetation; circle size, relative to N.sp; shading, 95% confidence intervals).

Robustness analyses

Randomly excluding 10–40% of the sources resulted in minimal variation in the effect sizes for animal biodiversity metrics, with variances below 0.02 for IN, 0.05 for SR, and 0.06 for SI (Appendix S1). The impact of source removal varied among metrics; for example, IN showed the largest variance in multiculture annual crops, whereas SI was most variable in “multiculture perennial crops” and sparse trees on pastures. SR was most sensitive to source removal, particularly in monoculture annual crops and sparse trees on pastures. Models incorporating continent as an additional explanatory variable, alongside agricultural class, did not significantly enhance the biodiversity models with only class as explanatory variable (Appendix S2).

DISCUSSION

We assessed the effects of in‐field crop and noncrop diversity on animal biodiversity relative to undisturbed reference ecosystems. We found that monocultures and pastures have the lowest biodiversity levels, followed by multicultures and agroforestry classes. Overall, we observed highest biodiversity levels in agroforests, including a high richness of endangered fauna.

Effects of agricultural management on animal biodiversity

We observed lowest levels of intactness (0.32 and 0.2, respectively) and richness (0.65 and 0.71, respectively) in monocultures of perennial and annual crops among all agricultural classes, aligning with our expectations. This observation, consistent with findings by Haddad et al. (2011), may be attributed to the limited crop diversity and absence of trees and other natural elements, resulting in a homogeneous vegetation structure and diversity. Supporting our hypothesis and the results of others (Fan et al., 2024), we found slightly higher intactness and similarity under monoculture perennial crop production than annual crop production, likely due to a lower frequency and intensity of anthropogenic disturbances throughout perennial crop production (e.g., less frequent tillage). Although we found that perennial monocultures hosted some endangered animal species, their relatively high richness (0.66), combined with moderate similarity (0.59) and low abundance (0.2), suggests that the species composition differed substantially from that of reference ecosystems.

In line with our hypotheses, we observed that multicultures of perennials and annuals supported animal communities that were more similar in species composition to undisturbed ecosystems than monocultures, though the difference in biodiversity was not statistically significant. This result likely stems from the greater diversity of crops, which enhances the diversity of food and potential shelter for animals (Fan et al., 2024). Our finding aligns with prior studies comparing multicultures and monocultures, which consistently report significantly higher biodiversity in multicultures (Beillouin et al., 2021; Cozim‐Melges et al., ²⁰²⁴; Tamburini et al., ²⁰²⁰), irrespective of whether the crops are annual or perennial (Fan et al., 2024; Priyadarshana et al., ²⁰²⁴). Our findings, therefore, suggest that transitioning from monocultures to multicultures can be beneficial for biodiversity, provided it does not involve converting natural land into agriculture. Unlike monocultures, we observed limited differences in biodiversity between multiculture perennial and annual crop production. This could be due to the substantial data limitations on multicultures, which may obscure differences between perennials and annuals. However, it is also possible that a diversity of annual crops has a more pronounced positive impact on biodiversity than a transition from annuals to perennials.

We observed similar biodiversity in pastures as in monoculture and multiculture croplands, generally displaying low biodiversity levels and hosting animal species distinct from those found in reference ecosystems. Our low estimates for pastures were surprising because a prior study demonstrated significant variations in MSA estimates for different types of pastures, ranging from 0.3 for heavily managed pastures to 0.6 for those with low livestock densities (Alkemade et al., 2013). Furthermore, overall biodiversity in pastures is greater when vegetation is diverse (Lok & Fraga, 2008; Scherber et al., ²⁰¹⁰), is natural rather than seeded (Alkemade et al., 2009, 2013), and increases over time since the last disturbance (Peeters et al., 2014; Yoshihara et al., ²⁰¹⁶). It is likely, therefore, that our databases mainly included intensively managed pasture sites. However, our databases may also have included more studies that compared pasture to forest ecosystems than to grassland ecosystems. Comparisons with grassland reference sites would likely result in higher biodiversity values than comparisons with forest reference sites because pastures likely provide more similar habitats to natural grasslands than to forests. In the future, it is crucial to further differentiate between types of pastures and between different types of the reference sites (e.g., whether forest or grassland ecosystem) to avoid underestimating biodiversity in pastures that replace natural grasslands or that are managed with low intensity.

We found similar biodiversity in sparse trees on cropland or pasture, as in multiculture croplands. This indicates that increasing noncrop diversity by planting sparse trees could enhance biodiversity, and that species compositions can become more like those found in reference sites. This finding is in line with those of previous studies. For instance, Galle et al. (2017) found substantially higher biodiversity in pastures on which sparse trees were planted than similarly managed pastures without trees. Therefore, our finding further emphasizes that the introduction of natural elements in agricultural areas can promote biodiversity.

We observed higher intactness (0.44) and similarity (0.83) metrics in silvopastures compared with pastures (0.33 and 0.72, respectively), although these differences were not statistically significant. This is in line with our hypothesis because some silvopastures maintain a high density of native forest cover, which can support animal species found in reference ecosystems (Picasso & Pizarro, 2024). Our findings are also in line with prior studies that used treeless pastures as a reference and showed that silvopastures enhance biodiversity across various animal taxonomic groups (Beillouin et al., 2021), although the effects depended on the animal groups studied and the biodiversity metric employed (McAdam et al., 2007). Like pastures, our findings for silvopastures were likely influenced by the type of reference ecosystem used. Because we mainly compared silvopastures with forest ecosystems, our current results may be more applicable to scenarios where conversion occurs from forests to silvopastures than to those from natural grasslands to silvopastures. If natural grasslands are converted into silvopastures, species adapted to natural grasslands may be repelled, further underscoring the need to differentiate between types of reference ecosystems in future studies.

We found significantly higher intactness (0.61) and similarity (0.92) in agroforests than in monocultures and pastures. This aligns with results from previous studies pointing at agroforests as the most biodiverse form of agriculture (Alkemade et al., 2009; Beillouin et al., ²⁰²¹), likely due to a more complex vegetation structure and diversity than found in other agricultural classes (Santos et al., 2022). We found similar richness but lower intactness and similarity in agroforests to reference sites, which reveals that a loss of species adapted to reference ecosystems is compensated for by an inflow of species from elsewhere. This may indicate that conversion of conventional cropland into agroforests in fragmented forest landscapes could enhance overall biodiversity, corroborating prior findings (Asare et al., 2014; Haggar et al., ²⁰¹⁹; Oakley & Bicknell, ²⁰²²; van ‘t Veen et al., ²⁰²⁵). Finally, we found evidence for our hypothesis that biodiversity increases over time in agroforests, which may be explained by an increasingly similar vegetation structure and diversity to reference sites as agroforests mature (de Leijster et al., ²⁰²¹).

We observed no differences in richness and total abundance of vulnerable to critically endangered species between agroforests and reference ecosystems, highlighting the potential of agroforests to host threatened animals adapted to reference ecosystems. Nevertheless, we found dissimilarity of threatened species composition between agroforests and reference sites, which may reveal that agroforests do not provide habitats for all endangered species adapted to reference ecosystems. This is in line with prior studies that show that, although agroforests can host some threatened species, they are not a substitute for undisturbed forests (Singh et al., 2021). This emphasizes the importance of preserving intact ecosystems for biodiversity conservation on top of increasing biodiversity in agricultural land (Gibson et al., 2011; Laurance et al., ²⁰¹²; Laurence et al., ²⁰¹⁴).

Limitations

Our analyses revealed biases and limitations in data concerning certain agricultural and taxonomic classes. Notably, we found a substantial geographical bias. First, we found a regional concentration of certain agricultural classes in certain regions in the world (e.g., agroforests are more common in tropical regions [Zomer et al., ²⁰¹⁴]). As a result, species pools may be more taxonomically similar under some agricultural classes (e.g., agroforests) than others (e.g., monoculture croplands) because the types of crops and noncrops produced and agricultural practices adopted are likely more similar in the same region than between different regions. This challenges the comparison of their biodiversity estimates. Second, our analyses revealed an overrepresentation of data from North American sites, particularly Mexico, and an underrepresentation from Asian sites. There was also a bias toward tropical regions, likely due to the abundance of undisturbed reference ecosystems at risk of conversion to agricultural land in the tropics (Laurance et al., 2014), with agricultural impacts on biodiversity in temperate regions often related to agricultural intensification rather than expansion. As a result, most studies in temperate regions focus on enhancing biodiversity within intensively managed agricultural lands, rather than addressing the conversion threat to undisturbed ecosystems. This pattern is notable in Europe, where undisturbed ecosystems are rare and agricultural practices typically intensify on existing farmland instead of expanding into natural areas (van Vliet et al., ²⁰¹⁵). We advocate for more meta‐analyses on the effects of agricultural management in temperate regions to address the data gap. Finally, literature from areas such as Russia, central Asia, China, and parts of South America is frequently not published in English, which complicates identification of these studies and their inclusion in meta‐analyses.

We found little information on agricultural input use in the original source documents, which prevented an assessment of the effects of agricultural inputs on biodiversity. Because the type and intensity of agricultural inputs vary geographically (Wu et al., 2018), it is likely that biodiversity within studies carried out in the same agricultural class was affected in varying ways by agricultural inputs. Additionally, intensities of land use have changed over the past decades, particularly in developing regions (Wu et al., 2018). Consequently, agricultural areas several decades ago could have been exposed to different types and levels of agricultural input than the same areas presently. This challenges the aggregation of data from studies conducted in different periods. As a result, variation in agricultural input across regions and time periods likely partially explains the variability in biodiversity responses to different agricultural classes observed in our study. Overall, our inability to differentiate the effects of agricultural intensity is unfortunate because fertilizers (de Graaff et al., ²⁰¹⁹; Humbert et al., ²⁰¹⁶; van Vooren et al., ²⁰¹⁸), pesticides (Beaumelle et al., 2023; Quandahor et al., ²⁰²⁴), and irrigation (Cabodevilla et al., 2022; Gols et al., ²⁰²⁴; Melloul et al., ²⁰²⁴) contribute to biodiversity loss in agricultural areas. We recommend that future field studies on biodiversity in agricultural settings document and record data on agricultural input use to facilitate more comprehensive meta‐analyses.

In this study, we combined data from various taxonomic groups to create an average biodiversity metric for all animal and threatened animal taxonomic groups together because we found insufficient data to differentiate across taxonomic groups. As a result, we analyzed varied data combinations from different taxonomic groups across agricultural classes, sometimes favoring insects or mammals, depending on data availability. This has implications for the comparability and robustness of our biodiversity estimates because prior evidence suggests that some animal classes respond more strongly to ecosystem changes than others (Burel et al., 1998). For instance, vertebrates react less intensely to land‐use change than insects (Gonthier et al., 2014). Therefore, our estimates smooth taxon‐ and species‐specific responses, likely resulting in smaller differences between biodiversity estimates across classes and wider confidence intervals. Further smoothening of responses likely occurs due to the large variety in sampling protocols used to collect biodiversity data (e.g., different plot sizes, sampling approaches), each with their own detection biases. In the case of threatened species, we did not specifically select for studies focusing on the detection of threatened species but rather identified the threatened species in the databases. As threatened species often occur at lower densities and most studies were not designed to explicitly detect them, their inclusion in the databases may be affected by chance. This could explain the uncertainty we observed in the effects of agricultural classes on threatened species. Despite observing only general patterns, our assessment offers an initial overview of agricultural management impact on biodiversity compared with undisturbed ecosystems. Future research can further explore differences in responses across taxonomic groups.

Although we produced biodiversity estimates for agricultural classes with clear distinguishing features (Table 1), the classification of data within each of the classes was largely based on qualitative interpretation of the original sources. For example, to distinguish agroforests from silvopastures, we relied on the word choice of the authors (e.g., using the words agroforest or silvopasture), as well as any additional information the authors provided, such as the presence of livestock and the types of crops. Because most studies did not provide any quantitative information (e.g., tree densities) that we could have used to improve definitions and robustness of our classification, the qualitative interpretation of agricultural classes likely produced some uncertainty in our classification. For example, in the case of sparse trees on cropland and pastures, we relied heavily on authors mentioning the occurrence of sparse trees. It could be that sparse trees occurred on the croplands and pastures examined in other studies but that these were not mentioned. This may explain the overlap we observed between the cropland and pasture classes, and sparse trees on cropland and pastures.

Understanding of how agricultural classes affect biodiversity is limited by the influence of landscape features. Diverse landscapes, with varied land use, management, and natural features, support greater biodiversity by providing a range of habitats, food resources, and shelter for different species (Dunning et al., 1992; Estrada‐Carmona et al., ²⁰²²). Under similar agricultural management, biodiversity is typically higher in landscapes with seminatural habitats compared with homogeneous ones (Arnold, 1983; Kormann et al., ²⁰¹⁵). As our study did not assess landscape characteristics, our understanding of their impact on biodiversity is constrained (Sánchez et al., 2022). We recommend future research to examine the effects of landscape diversity on biodiversity in agricultural land to help distinguish the effects of agricultural management from those of landscape features.

Model applications

The insights gained from this study can be used to improve global biodiversity models, which often use undisturbed ecosystems as references (e.g., GLOBIO, BILBI, cSAR‐iDiv, cSAR‐IIASA‐ETH, and PREDICTS) (Chaplin‐Kramer et al., 2024; Kim et al., ²⁰¹⁸). Our study provides the necessary biodiversity estimates to parameterize models and simulate the effects of different agricultural management classes on global biodiversity. Modeling biodiversity responses to agricultural class could help identify biodiversity‐friendly agriculture options to mitigate global biodiversity loss. Such modeling studies could also highlight potential trade‐offs between local biodiversity rises in response to transitions in agricultural management and increased agricultural land needs because of potential yield reductions associated with enhanced crop and noncrop diversity. This understanding may inform whether biodiversity loss is better mitigated by transitioning to intensive agricultural management with low crop and noncrop diversity but high yields to prevent agricultural expansion or by enhancing agricultural crop and noncrop diversity to reduce biodiversity loss within agricultural lands. Although previous modeling studies (Kok et al., 2020) explored these dynamics, they likely underestimated the positive effects of agricultural management transitions because they were unable to differentiate between management classes.

We evaluated the impact of agricultural management on animal communities, with undisturbed ecosystems as a reference. Our findings reveal that agricultural management influences the ability of species associated with undisturbed ecosystems to find refuge in agricultural areas, with notable differences across classes. Introducing multiple crops on farmland resulted in slightly more similar animal communities to those of undisturbed ecosystems than found in monocultures, as did the introduction of sparse trees. Agroforests exhibited the most similar animal communities to undisturbed ecosystems among all studied agricultural classes. Agroforests also inhabited many vulnerable and critically endangered species found in undisturbed ecosystems. The biodiversity benefits of agroforests may stem from their complex and diverse vegetation structure, characterized by a layered canopy of crops and trees. This vegetation complexity likely increases as agroforests mature, explaining the increase in biodiversity we found over time in this agricultural class. The biodiversity estimates from this study can be used to parameterize global biodiversity models, allowing for a thorough examination of the effects of agricultural management transitions on biodiversity across different scenarios. This would allow for a better understanding of the benefits and trade‐offs of enhancing crop and noncrop diversity for biodiversity over space and time and enable the differentiation of policy pathways aimed at achieving global biodiversity targets.

AUTHOR CONTRIBUTIONS

Susana López Rodríguez: Conceptualization; investigation; validation; writing—original draft; writing—review and editing. Hanneke van ‘t Veen: Conceptualization; data curation; formal analyses; investigation; methodology; validation; visualization; writing—original draft; writing—review and editing. Solen Le Clec'h: Writing—review and editing. Rob Alkemade: Conceptualization; funding acquisition; methodology; resources; supervision; writing—review and editing.

Supporting information

Appendix S1: Effects of agricultural variety class on animal biodiversity

Supplementary Appendix S2

Rodríguez, S. L. , van ‘t Veen, H. , Clec'h, S. L. , & Alkemade, R. (2026). A global assessment of animal community responses to agricultural management. Conservation Biology, 40, e70299. 10.1111/cobi.70299