Land use diversification may mitigate on‐site land use impacts on mammal populations and assemblages

Land use is a major cause of biodiversity decline worldwide. Agricultural and forestry diversification measures, such as the inclusion of natural elements or diversified crop types, may reduce impacts on biodiversity. However, the extent to which such measures may compensate for the negative impacts of land use remains unknown. To fill that gap, we synthesised data from 99 studies that recorded mammal populations or assemblages in natural reference sites and in cropland and forest plantations, with or without diversification measures. We quantified the responses to diversification measures based on individual species abundance, species richness and assemblage intactness as quantified by the mean species abundance indicator. In cropland with natural elements, mammal species abundance and richness were, on average, similar to natural conditions, while in cropland without natural elements they were reduced by 28% and 34%, respectively. We found that mammal species richness was comparable between diversified forest plantations and natural reference sites, and 32% lower in plantations without natural elements. In both cropland and plantations, assemblage intactness was reduced compared with natural reference conditions, but the reduction was smaller if diversification measures were in place. In addition, we found that responses to land use were modified by species traits and environmental context. While habitat specialist populations were reduced in cropland without diversification and in forest plantations, habitat generalists benefited. Furthermore, assemblages were impacted more by land use in tropical regions and landscapes containing a larger share of (semi)natural habitat compared with temperate regions and more converted landscapes. Given that mammal assemblage intactness is reduced also when diversification measures are in place, special attention should be directed to species that suffer from land use impacts. That said, our results suggest potential for reconciling land use and mammal conservation, provided that the diversification measures do not compromise yield.


| INTRODUC TI ON
Biodiversity is decreasing worldwide due to human-induced environmental change (IPBES, 2019).On average across the globe, local terrestrial biodiversity intactness has declined by almost half compared with conditions without human pressures, and 25% of comprehensively assessed animal and plant species are threatened with extinction (Díaz et al., 2019;Schipper et al., 2020).Habitat loss and degradation are the main drivers of terrestrial biodiversity loss worldwide (Arneth et al., 2019;Díaz et al., 2019;Maxwell et al., 2016), threatening ~40% of terrestrial plant and animal species (IUCN, 2022).Already 12% and 2% of the ice-free land is used for cropland and forest plantations, respectively (Arneth et al., 2019).
Given the need for agriculture and forestry to meet the increasing demands of a growing human population (Winkler et al., 2021), and in view of various global biodiversity targets (CBD, 2022;UN, 2019), it is imperative to optimise land use practices towards minimizing impacts on biodiversity and safeguarding the provisioning of ecosystem services (IPBES, 2019).
Land use diversification measures have been proposed as one of the strategies to reduce the impacts of agriculture and forestry on the abundance and diversity of nondomesticated species (Bommarco et al., 2013;Kremen et al., 2012;Kremen & Merenlender, 2018;Pretty et al., 2018).Land use diversification can be defined as the intentional enhancement of functional biodiversity in agricultural and forestry landscapes to maintain ecosystem services (Kremen et al., 2012).Generally, land use diversification measures increase vegetation diversity, reduce the contrast between cultivated land and natural habitat, and enhance the biological resources in humanmodified landscapes available to species (Kremen et al., 2012).Land use diversification encompasses a variety of measures, including the embedding of (semi)natural elements (e.g.native understories, uncultivated field strips and hedgerows) and the diversification of the cultivated species, such as in polyculture cropland or mixed-age distributions of perennials and mixed tree species stands (Kremen et al., 2012).
Recent studies have confirmed the potential of such land use diversification measures to reduce the negative effects of land use on biodiversity.The inclusion of hedgerows and trees has been shown to enhance biodiversity in agricultural landscapes and to improve conditions for pollinators (Barrios et al., 2018;Ponisio et al., 2016).In addition, crop heterogeneity has been shown to enhance biodiversity (Sirami et al., 2019).Recent reviews highlight positive effects of diversified cultivation (e.g.intercropping and cultivar rotation), maintaining understory vegetation, and mixing natural forest elements with agriculture (Beillouin et al., 2019;Oakley & Bicknell, 2022;Rosa-Schleich et al., 2019).Correspondingly, several meta-analyses reveal that the abundance and richness of nondomesticated plant and animal species are generally higher in diversified farming systems compared with conventional farming systems (Batáry et al., 2011;Gonthier et al., 2014;Lichtenberg et al., 2017;Sánchez et al., 2022;Tamburini et al., 2020).
Although these studies provide evidence for the biodiversity benefits of land use diversification measures, no study has assessed nondomesticated biodiversity in land use sites with or without diversification measures in comparison to natural reference conditions.Hence, it is not known to what extent land use diversification measures may compensate for the negative impacts of agricultural and forestry land use.Understanding how natural ecological communities respond to land use and diversification thereof is key for an adequate evaluation of the actual biodiversity benefits of diversification measures.In addition, change of natural communities is an important measure in global biodiversity assessments (e.g.IPBES, 2019; Leclère et al., 2020;Schipper et al., 2020;WWF, 2020), yet existing databases and models underlying these assessments (e.g.PREDICTS, GLOBIO) generally do not distinguish land use diversification measures (Hudson et al., 2017;Schipper et al., 2020).
Here, we quantify the influence of land use diversification measures in cropland and forest plantations on mammal populations and assemblages relative to natural reference conditions.We focus on mammals, which are largely missing or underrepresented in field studies or meta-analyses of land use diversification impacts (Oakley & Bicknell, 2022;Sánchez et al., 2022;Sirami et al., 2019), thereby acknowledging that further work is needed to address other taxonomic groups too.We synthesised data from 99 papers spanning 571 mammal species from 17 orders and all biogeographical realms except Antarctica.We quantified individual species abundance, species richness and assemblage intactness in cropland and forest plantations under varying land use diversification measures relative to natural reference sites.We grouped land use diversification measures to reflect (i) embedding of natural elements and (ii) diversified cultivation (i.e.cultivation of multiple crop or tree species or mixedage tree stands in cropland or forest plantations).In addition, we studied the extent to which the responses of mammal species or assemblages to these diversification measures are modified by species characteristics or the environmental context.

| Data collection
We performed a systematic literature search to obtain empirical data on mammal species populations and assemblages in land use sites and corresponding natural reference sites.We performed the search agriculture, biodiversity, cropland, forest plantations, forestry, individual species abundance, intactness, land management, mean species abundance, species richness using ISI Web of Science in January 2021 (see Supporting methods S1.1 for the search string employed).The search yielded 6516 unique publications, of which we selected 99 papers (published between 1993 and 2020) that contained at least one pairwise comparison of individual mammal species abundance (per species or genus) or species richness (per order or class) in a land use site (cropland or forest plantation with or without diversification measures) and a natural reference site (Figure S1).
We structured the data by source (i.e.publication), study, site and species (group).A single data source may contain several studies if the observations were performed in multiple regions or seasons, each characterized by a corresponding reference site.If a study contained several reference sites (i.e.sites with natural vegetation), we used the most pristine site as the reference (i.e. the site with lowest human disturbance in terms of, for example, hunting or previous logging activities), based on the descriptions in the corresponding source.We extracted several measures of abundance and richness, including overall number of individuals or species, population density (e.g.number of species/ha or number of individuals/ha) and physical trapping or camera trapping rates (number of species captured per trapping effort or number of individuals captured per trapping effort).
In addition, we compiled data on various context variables that may affect mammal population or assemblage responses to cropland and forest plantations.Specifically, we included four species characteristics that may affect population-level responses and three variables related to the environmental context, relevant for both the population-and assemblage-level responses (Table 1).We obtained the input data either from the primary studies or from external data sources, which we linked to the observations in our data set based on species identity, study location and year of the study (for details see Table 1).
Our final data set included 166 unique cropland sites from 80 publications, and 81 forest plantation sites from 31 publications.

| Land-use diversification measure classification
We characterised each land use site based on aggregated land use diversification measure categories (Table 2), due to insufficient data to distinguish more specific agricultural management measures.Specifically, we differentiated cropland and forest sites by (i) the in/ exclusion of natural elements and (ii) the in/exclusion of cultivation of diversified crop/tree types (i.e.four possible combinations of in/ excluding land use diversification measures).

| Effect sizes
We analysed the responses of mammal populations and assemblages to land use (with and without diversification measures) based on a complementary set of indicators covering both abundance and richness, which are independent dimensions of biodiversity (Schipper et al., 2016).We included the following indicators: (i) relative individual species abundance (RIA), measured as the individual species abundance in land use sites compared to natural reference conditions; (ii) relative species richness (RSR), measured as the species richness in land use sites compared to natural reference conditions; and (iii) assemblage intactness, measured as the mean species abundance (MSA) (Midolo et al., 2019;Schipper et al., 2020; Table 3 for details and Figure 1 for the spatial distribution of the effect sizes).Effect sizes are calculated as the natural log-transformation of relative individual species abundance (lnRIA), the natural log-transformation of the relative species richness (lnRSR), and the logit-transformation of mean species abundance (logitMSA).MSA is logit-transformed because its values fall between zero and one.Relative individual species abundance and relative species richness can be higher than one if species abundance or richness is higher in the land use site compared to the corresponding natural reference condition.In some cases, we observed zero abundance or richness in either a land use site or the natural reference site, precluding calculation of the effect sizes.We therefore transformed all effect measures with the Smithson-Verkuilen transformation (Smithson & Verkuilen, 2006) (see Table 3 for details and Figure S2 for an overview of the effect sizes).

| Analysis
We established linear mixed-effects models to estimate effects of cropland and forest plantation land use diversification measures on mammal individual species abundance, assemblage intactness, and species richness.To account for nonindependence in the data due to multiple effect sizes per study, we specified study identity nested within source as random intercept.Because the Smithson-Verkuilen (2006) transformation introduces small and large effect size values for zero abundance or richness in land use or natural reference sites, respectively, and because such effect size values (in log-space) may skew back-transformed average effects (in normalspace), we weighted the effect sizes in the richness and abundance regression models to correct for this.We designed the weighting such that the mean of a biodiversity effect measure of zero (no biodiversity in the land use site; X t = 0) and one (equal biodiversity in the land use and natural reference site; X t = X c ) was equal to 0.5, and that the mean of a biodiversity effect measure related to biodiversity of zero in the natural reference size (X c = 0) and an effect measure of one equals 2 (in normal space) as follows (Equation 1, where n is the number of effect sizes; see We weighted the effect sizes for MSA based on the number of species included in the estimate, reasoning that the larger the number of species sampled, the more representative the indicator of community level intactness.We used the square-root of the number of species to correct for the positive skew in the data (Schipper et al., 2020).
We performed model selection in two steps.In the first step, we considered only the land use diversification measures (Table 2), in view of the aim of our study and the scope of our literature search.
For each biodiversity indicator and land use type (cropland and forest plantation), we fitted a set of models considering all possible combinations of diversification measures and their interactions (as fixed effects)-that is, we considered both independent and combined effects of land use diversification measures-and selected the bestsupported model based on Akaike's Information Criterion (corrected for small sample sizes; AICc).In a next step, we explored the extent to which the responses of individual species abundance, species richness, and assemblage intactness to the cropland and forest plantation practices are modified by possibly relevant context variables (Table 1).
To that end, we used the best-supported models from the first step as a starting point, extended these with the context variables (also considering possible interactions), and selected again the best-supported model based on AICc.We considered the species-specific context variables only in the individual species abundance models.In both model selection steps, we allowed for interactions only if the corresponding main effects were also included (Nelder, 1998).
We performed omnibus tests to evaluate whether the bestsupported models explained significantly more variance than the models that did not consider land use diversification measures (null-models; see Table S2 using chi-squared tests) and whether the best-supported models explained significantly more variance than unexplained variance in the response data (using F-tests).In addition, we performed F-tests to test the significance of different levels of the fixed effects (i.e. the in-vs.exclusion of land use diversification measures, see Table S6).
We executed the analyses in R using the lme4 and lmerTest packages (Bates et al., 2015;Kuznetsova et al., 2017;R Core Team, 2023).
We used maximum likelihood estimation for the model selection and refitted the best supported models with restricted maximum likelihood for coefficient estimation.

| Individual mammal species abundance
In croplands without natural elements, individual mammal species abundance was reduced to 72% (confidence interval [CI] 56%-92%) of the abundance in a natural reference situation (Figure 2a; Table S7).In contrast, in croplands that include natural elements, abundance was similar to, or even higher than, natural reference conditions (131%; CI = 96%-179%).We did not find evidence for effects of crop diversification on individual mammal species abundance (Table S3).Individual species abundance in forest plantations was 68% (CI = 50%-93%) of the abundance in natural Mean species abundance (MSA), measured as compositional intactness of local communities (arithmetic mean of the relative abundance of species-truncated to 1-across all species present in undisturbed natural habitat).MSA is transformed (MSA′) based on the number of effect measures (n).Effect sizes are calculated as the logit-transformed MSA′ (logitMSA) a Effect measure transformation based on Smithson and Verkuilen (2006).
habitat (Figure 2b; Table S7), with no differentiation in response resulting from diversification measures (Table S3).
For both cropland and forest plantations, we found that the individual species responses depend on their habitat breadth (Figure 2; Table S3).In cropland without natural elements and in forest plantations, species with narrower habitat breadth (<5-7 habitat types) decline in abundance compared a natural reference, while habitat generalists increase.In cropland including natural elements, abundance of habitat specialists (one habitat type) is similar, whereas abundance of generalists is higher than natural reference conditions (Table S7).We did not find evidence for any of the other context variables influencing individual species abundance responses to cropland or forest plantation practices (Table S3).
We did not find evidence for any of the context variables influencing species richness responses to cropland, but we did find a positive linear relationship between (absolute latitude) and species richness responses to forest plantations (Table S4).We found that relative to natural reference conditions, species richness is lowest in forest plantations at low (i.e.tropical) latitudes and increases towards higher latitudes (until ±50°; Figure 3b).Species richness is consistently lower in forest plantations without natural elements compared with natural reference sites.In forest plantations with natural elements, species richness is larger than in natural reference sites above absolute latitudes of 30°.We did not find evidence for the influence of the habitat class of the natural reference site or the share of (semi)natural habitat in the landscape on mammal species richness in forest plantations relative to natural reference conditions (Table S4).
Intactness is lowest in cropland at low (i.e.tropical) and high latitudes, with an average intactness of 0.16 (CI = 0.06-0.37) in cropland without natural elements and 0.40 (CI = 0.17-0.70) in cropland with natural elements at 0° latitude.We did not find evidence for any of the other context variables influencing mammal assemblage intactness in cropland (Table S5).In forest plantations, we found a negative relationship between the share of (semi)natural landcover in the landscape and assemblage intactness (Figure 4b).This shows that intactness is higher in plantations when there is little natural habitat in the landscape.We did not find evidence for differences in mammal assemblage intactness in forest plantations at different latitudes or for differences in intactness when comparing forest plantations to different natural habitat classes (Table S5).

| DISCUSS ION
We estimated the effects of land use diversification measures in cropland and forest plantations on mammal populations and assemblages based on spatial comparisons of empirical data between these land use sites and corresponding natural habitat sites.Such spatial comparisons are useful to analyse responses of local biodiversity to environmental change given the limited availability of time-series data allowing for before-after comparisons (Davison et al., 2021;Newbold et al., 2015).Because biodiversity responses are highly dependent on the ecological context and are expected to differ from site to site, we tested for potential influence of various context variables, including the species' habitat breadth, diet diversity, Nevertheless, we found consistently lower mammal assemblage intactness in cropland and forest plantations than in natural reference sites, irrespective of land use diversification measures.
The differences in response between the different metrics confirm the relevance of considering multiple complementary indicators in order to capture biodiversity change (Santini et al., 2017).For example, local species richness is insensitive to changes in species population sizes and does not reveal changes in community composition in case of a net-balance of species colonisation and extinction (Dornelas et al., 2019;Hillebrand et al., 2018;Williams et al., 2017).
Hence, species richness alone has limited value for informing conservation.Our findings for intactness suggest that land use diversification measures do not prevent changes in assemblage composition.This, in turn, indicates that some species may profit while others suffer from anthropogenic disturbance (Dornelas et al., 2014(Dornelas et al., , 2019;;Leung et al., 2020;Newbold, Hudson, Hill, et al., 2016;Vellend et al., 2013).Indeed, we found that the abundance of habitat specialists is lower in nondiversified cropland and forest plantations than in natural habitats, whereas habitat generalists may thrive, similar to earlier findings reported by Newbold et al. (2014).Our findings further indicate that species assemblages in the tropics are more impacted by cropland and forest plantations than assemblages in subtropical and temperate zones.This latitudinal pattern may reflect that subtropical and temperate zones (25°-50° latitude) have a longer history of agricultural land use and forest management (Ellis et al., 2021;Watson et al., 2018), which in turn may have resulted in adaptations of the species pool, while tropical species might be more sensitive (Newbold et al., 2020).Alternatively, the latitudinal trend may reflect that natural reference sites might be more degraded in temperate as compared to tropical regions, for example due to pollution, road disturbance or habitat fragmentation, resulting in suppressed biodiversity responses (Newbold, Hudson, Arnell, et al., 2016).
We further found that assemblage intactness is higher in forest plantations within landscapes characterised by a low share of (semi) natural habitat.This may reflect that species in largely altered landscapes have adapted to human land use or that the species remaining in the little intact habitat are those that are less sensitive to human disturbance.Alternatively, it could indicate that species in largely altered landscapes use forest plantations as a refugium (Bhagwat  et al., 2013).Interestingly, we found no evidence for a modulating effect of landscape context on the biodiversity response to cropland diversification.This contrasts with the results of the meta-analysis by Batáry et al. (2011), who reported larger species richness increases in response to agri-environmental management of croplands in landscapes with less (semi)natural habitats.However, their study included other taxonomic groups than mammals.We also found no evidence for mammal species richness or abundance responses to the diversification of cultivated species and/or age of crops and trees in croplands and forest plantations.This is in contrast to the findings of Beillouin et al. (2019), who found predominantly positive effects of crop diversification on biodiversity.However, they analysed biodiversity-related effect sizes in a second-order metaanalysis without focussing on specific biodiversity indicators or species groups.
In general, a lack of response in our models does not necessarily reflect actual absence of effects but could also be due to small sample sizes or large heterogeneity in the data.For example, based on the primary studies, we could not identify the amount and type of noncrop habitat included in the cropland and forest plantations sites, while these aspects likely affect mammal species responses (Estrada-Carmona et al., 2022).Similarly, we could not capture whether forest plantations cultivate exotic or native tree species, which may have different effects on mammal assemblages causing unexplained variance (Kriegel et al., 2021).Data on field size were also mostly absent, while field size is known to have a large effect on noncrop biodiversity in farmlands (Clough et al., 2020).To be able to capture impacts of more specific or other land use diversification and management measures (such as maintaining or increasing natural hedgerows or understories or improving the soil quality), and thus better direct business and land management policy interventions, more information would be needed on biodiversity responses to specific land use diversification and management measures.
Protecting global biodiversity to meet the goals of a post-2020 Global Biodiversity Framework requires significant action, as most targets set within the Strategic Plan for Biodiversity 2011-2020 have not been reached (Mace et al., 2018;Nature Editorial, 2020).
As habitat loss is a key driver of biodiversity loss and the growing demand for food and other resources may exacerbate land demands, there is a key role for strategies to manage land in ways that help to protect and preserve biodiversity (IPBES, 2019).Here, we showed that embedding natural elements in croplands and forest plantations has the potential to improve local (i.e.on-farm) mammal abundance, richness and intactness, in line with previous studies (Kremen & Merenlender, 2018;Perfecto & Vandermeer, 2010;Phalan, Onial, et al., 2011).However, for the implementation to be effective, land use diversification measure impacts on yields should

F
Spatial distribution of the effect size data points for individual species abundance, species richness, and assemblage intactness in cropland, and forest plantations.n indicates the total number of effect sizes for each biodiversity indicator and land use type combination.
origin and body mass, as well as absolute latitude of the study site, habitat class of the natural reference site, and share of (semi)natural habitat in the landscape surrounding the site.Our results suggest F I G U R E 2 Responses (including 95% confidence intervals) of individual mammal species abundance to land use practices for (a) cropland and (b) forest plantations in terms of relative individual species abundance (RIA).Responses are shown for the best-supported models including only land use diversification measures (left) and best-supported models including both land use diversification measures and a context variable (right).lnRIA < 0 indicates a decrease and lnRIA > 0 indicates an increase in abundance in the land use site relative to the natural reference site.elements in cropland may result in individual species abundance and species richness similar to natural reference levels, significantly improving local mammal biodiversity compared with cropland devoid of natural vegetation.Similarly, we found that the inclusion of natural elements in forest plantations may offset the negative effects of forest plantations on mammal species richness.

F
I G U R E 3 Response (including 95% confidence intervals) of mammal species richness to land use practices for (a) cropland and (b) forest plantations in terms of relative species richness (RSR).Responses are shown for the bestsupported models including only land use diversification measures (left) and best-supported models including both land use diversification measures and a context variable (right).lnRSR < 0 indicates a decrease and lnRSR > 0 indicates an increase in richness in the land use site relative to the natural condition.al., 2008;Kanowski et al., 2005;Malhi et al., 2022;Simonetti

F
Responses (including 95% confidence intervals) of mammal assemblage intactness (MSA) to land use practices for (a) cropland and (b) forest plantations.Responses are shown for the best-supported models including land use diversification measures (left) and bestsupported models including both land use diversification measures and a context variable (right), MSA < 1 indicates a change in assemblage intactness in the land use site relative to the natural condition.
Potential moderators of the responses of mammal species populations or assemblages to land-use diversification measures.
Table S1 for examples of effect sizes and weighting): TA B L E 1 Classification of land use diversification measures.
TA B L E 3 Overview of biodiversity indicators and corresponding effect sizes included in this study.