Introduction

Around 40% of the global ice-free and desert-free land is used for agriculture, either as cropland or pasture (Springmann et al. 2018). While primarily used for the production of food, fuel and fibre, rural landscapes also host considerable levels of biodiversity. The majority of global biodiversity is actually found in these landscapes shared with people, and not necessarily in protected areas alone (Rodrigues et al. 2004; Baudron and Giller 2014). Most rural landscapes are complex and intricate mosaics comprised of areas of cropland and rangeland interspersed by patches of primary and secondary forest. Globally, more than 40% of agricultural land possesses at least 10% tree cover (Zomer et al. 2014), often in designed agroforestry systems (agroforestry is defined in its broad sense as ‘land use systems and practices in which woody perennials are deliberately integrated with crops and/or animals on the same land management unit’; Leakey 1996). Furthermore, 70% of forest (defined as land areas with at least 30% tree cover) are within 1 km of a forest edge and are thus highly fragmented (Haddad et al. 2015).

The increased demand for food, fuel, and fibre by an increasingly affluent and growing global population—projected to reach 9 billion by the middle of this century—drives agricultural expansion in some regions and agricultural intensification in others (Vandermeer et al. 2018). Both can negatively impact the ability of landscape mosaics to conserve biodiversity and deliver essential ecosystem services (Baudron and Giller 2014; Felipe-Lucia et al. 2014; Phalan et al. 2014; Tomscha and Gergel 2016). Thus, we urgently need to better understand how agricultural productivity and biodiversity conservation can be supported simultaneously in these complex landscapes. This challenge is particularly pressing in developing countries where food security and human well-being are of critical concern and coincide with threatened biodiversity hotspots (Millenium Ecosystem Assessment 2005).

Since the seminal paper of Green and collegues published in 2005, several studies have used the land sparing and land sharing framework to reconcile agricultural production and biodiversity conservation in rural landscapes (e.g., Kleijn et al. 2009; Clough et al. 2011; Phalan et al. 2011; Gabriel et al. 2013). These studies classically plot density-yield functions to identify species better adapted to land sparing (segregation of production and conservation) from those better adapted to land sharing (integration of production and conservation) by examining the shape of the curve. The former is characterized by a concave curve, the latter by a convex curve (Green et al. 2005). In either case, barring a few rare ‘winner’ species, most species are expected to be ‘losers’ which decline with increasing crop yields. Crop yield is often used as the primary agricultural indicator in studies employing such density-yield functions (Green et al. 2005; Kleijn et al. 2009; Clough et al. 2011; Phalan et al. 2011; Gabriel et al. 2013; Phalan et al. 2014; Law et al. 2015). However, as already pointed out by Baudron et al. (2017), the conclusions drawn by such studies would likely differ if agricultural productivity was evaluated in a more integrated fashion. For example, in addition to crop production, livestock and fuelwood comprise two very important products in many landscapes. Indeed, a quarter of the global terrestrial surface is used for grazing (Robinson et al. 2011), whilst fuelwood use remains important in many parts of the world lacking alternative energy sources (Bugaje 2006; Iiyama et al. 2014).

In this study, we simultaneously assessed multiple measures of agricultural productivity—crop, livestock feed, and fuel—as well as biodiversity in a rural landscape of Southern Ethiopia. To explore the potential co-benefits across landscape types, we compared four zones with contrasting land cover composition—cropland, grassland, and tree cover—to account for diverse resources provided by rural landscapes: cropland provides food (and secondarily crop residues used as feed and fuel); grassland provides primarily feed (and secondarily fuelwood); and tree cover provides primarily fuelwood (and secondarily feed). Our first objective was to assess how important components of biodiversity, namely bird and tree species composition and richness, changed with different land cover types associated with different levels of crop, feed, and fuel productivity. Our second objective was to uncover whether particular landscapes could simultaneously deliver high agricultural productivity (in multiple dimensions) and biodiversity conservation.

Materials and methods

Study area

We conducted our study in the woreda (district) of Arsi-Negele, located in the Oromia region of Ethiopia. The study area covers about 100 km2 between 38°42.14′–38°49.92′E and 7°15.05′–7°22.57′N. It borders the state forest of Munesa, and encompasses parts of the three kebeles (sub-districts) of Ashooka, Bombaso Regi and Gambelto, in which a total of six villages were studied. Altitudes here range between 1970 and 2200 m above sea level. The climate is sub-humid, characterized by a mean annual rainfall of 1075 mm per year (18-year average) and a mean annual temperature of 15 °C (16-year average). The study area is characterized by bimodal rainfalls, with a short rainy season from March to May, and a long rainy season from July to September. The natural vegetation is classified as dry afromontane forest (Tesfaye 2007). Wheat (Triticum sp. L.), maize (Zea mays L.), potato (Solanum tuberosum L.) and enset (Ensete ventricosum (Welw.) Cheesman) are the primary crops under cultivation. Most farmers keep livestock in the form of cattle, sheep, goats, and donkeys.

The Sida Malkatuka village and Dikitu Shirke village (in Ashooka kebele) border the state forest of Munesa and form a zone referred to as the ‘high tree cover’ zone in the rest of the paper (Fig. 1). Households in the high tree cover zone use the Munesa forest for fuel and livestock feed (Baudron et al. 2017). A second zone of medium tree cover encompasses Gogorri Lako Toko village (in Ashooka kebele) and Kararu Lakobsa Lama village (in Bombaso Regi kebele) and is located about 5.5 km away from Munesa forest (Fig. 1). Households from the medium tree cover zone make extensive use of a large communal grazing area for fuel and feed (Baudron et al. 2017). A third zone of low tree cover encompasses the villages of Shodna and Belamu (in Gambelto kebele) and is located about 11 km away from Munesa forest (Fig. 1). Households in the low tree cover zone lack access to common grazing or forest areas (Baudron et al. 2017).

Fig. 1
figure 1

Map of the study area illustrating the four zones (low tree cover, medium tree cover, high tree cover, and Munesa forest) where bird and tree biodiversity was assessed, and point counts (biodiversity points) in each zone. The land classification was done using RapidEye imagery from January 2015

Full size image

Land use classification and agricultural productivity

Contemporary land cover was determined using RapidEye imagery (5-meter resolution) from January 2015 and land was classified into five basic classes: cropland/bare soil, grassland, natural forest, plantations/woodlots, and enset homegardens, following the method described in Baudron et al. (2017). We defined a class ‘tree cover’ by merging the classes ‘natural forest’ and ‘plantations/woodlots’.

To relate our findings to different proxies of productivity, we interviewed the head of each household in the study area between December 2014 and February 2015. A total of 266 households were interviewed (88 in the high tree cover zone, 97 in the medium tree cover zone, and 81 in the low tree cover zone) using a standardized questionnaire addressing crop, livestock, and household fuel management. A farm typology was delineated based on self-categorization exercises conducted in each zone, and a stratified subsample of nine farms was selected in each zone (27 farms in total) for which resource flow maps (i.e., maps of each farm showing the flows of resources between components in the farm and to and from the farm) and resource use calendars were produced (Geifus 2008; Giller et al. 2011). In addition, the area of each field was measured using a hand-held global positioning system (GPS) Garmin Etrek 10. Empirical measurements of daily fuel consumption were conducted in nine of these 27 farms (one farm per type and per zone, selected randomly) once in March 2015 and once in August 2015).

Crop productivity per zone was calculated by dividing the total quantity of grain, tuber and fresh product harvested in the zone (from interview data) by the area of the zone, and multiplying this by the USDA’s specific standard value of dry matter content (https://ndb.nal.usda.gov/ndb/search). Feed productivity per zone was calculated by estimating the total biomass consumed by livestock in the zone and dividing it by the area. For each zone, the total biomass consumed by livestock was estimated by converting livestock numbers into Tropical Livestock Units (TLU), using a value of 250 kg live weight for one TLU (Houérou and Hoste 1977), and by assuming a daily feed intake of 5 kg DM TLU−1 (i.e., 2% of live weight). Oxen and bulls were assumed to be equivalent to 1.1 TLU, cows to 0.8 TLU, steers and heifers to 0.5 TLU, calves to 0.2 TLU, sheep and goats to 0.09 TLU, and donkeys to 0.36 TLU (Gryseels 1988). The total biomass consumed by livestock in a particular zone was then allocated between the zone itself (biomass consumed within the village), Munesa forest, and purchased feed, using resource use calendars. Fuel productivity per zone was calculated by estimating the total biomass used as household fuel in the zone and dividing it by the area. For each zone, the total biomass used as household fuel was estimated by multiplying the average daily consumption of each household type by the number of households from each type in the zone and by 365 days. The total biomass used as household fuel consumed in a particular zone was then allocated between the zone itself (fuelwood, crop residue and dung from the village), Munesa forest, and purchased fuel, using resource use calendars.

Biodiversity assessment

A total of 96 point counts were selected in the study area: 24 in the Munesa forest and for each of the high, medium, and low tree cover zones. Point counts were selected randomly in GIS from a 150 m grid overlaid on a map of the area. Between May and September 2015, each tree within a radius of 50 m from the 96 point counts and with a diameter at breast height (~ 135 cm from ground level) greater than 10 cm was identified to species level by a trained local guide, and using relevant plant taxonomic literature (Hedberg and Edwards 1989; Edwards et al. 1995, 1997; Edwards and Hedberg 2000; Hedberg 2006). For each point, basal area was calculated by dividing the sum of the section area of all trees (at breast height) by the total surface area as an estimate of tree cover. In addition, each point count was visited three times between April and May (dry season) and three times between August and September (wet season) during one morning (between 6h00 and 10h00) and all birds within a 50 m radius and during a period of 10 min were identified visually and counted. A maximum of eight point counts were visited during a particular morning. Counting was avoided during rainy days or days characterized by heavy winds. Repeated bird counts were undertaken by a single experienced local bird guide.

Statistical analysis

Presence of individual tree species among the four zones were assessed by fitting the data, which was converted to a binary ‘presence’ or ‘absence’, into binomial generalized linear models. We assessed differences between zones in presence of tree species, and differences in these patterns between native versus introduced trees (classified as such using the online database from the Royal Botanic Garden, Kew www.plantsoftheworldonline.org). We used binomial generalized linear mixed effects models with count points as random variable and with zone and its interaction with ‘native/introduced’ status as fixed effects on the full dataset. Using a similar approach, we then assessed whether presence/absence among bird species differed between zones for different trophic guilds (following Wilman et al. 2014) as well as for different ranges (species with small versus large global ranges, following Phalan et al. 2011). Here, we used binomial generalized linear mixed effects models as for trees, but we added season (dry/wet) as well as point counts (N = 96) as random variables and also added zone and its two-way interaction with range and trophic guild as fixed effects. We summarised our results for trees and birds in graphs (Figs. 2, 3), in which we used species’ presences in the forest as a reference level and thus compared presences in the other zones against these values (see figure legends for details).

Fig. 2
figure 2

Presences of tree species a for all tree species and b for introduced versus native trees, in the four zones and in comparison to presences in the forest. To display results, we used the forest as the reference level. For all tree species combined, we calculated zone values as relative presences compared to forest. For introduced versus native trees, no introduced species were recorded in the forest, and we were therefore not able to calculate relative presences. To use forest as the reference level, we instead subtracted group-specific forest values from the values of all zones

Full size image
Fig. 3
figure 3

Presences of a four feeding guilds of birds (invertebrate eaters, fruit and nectar eaters, omnivores, and plant and seed eaters) and b small range versus large range species, in the four zones and relative to abundances in the forest

Full size image

Detrended correspondence analyses (DCA; Hill and Gauch Jr 1980) of the biodiversity data, illustrating changes in beta-diversity, were then used to assess how tree and bird communities changed along gradients of land use—i.e., gradients from low to high cover of cropland, grassland and trees. Environmental variables such as tree, cropland and grassland cover were plotted onto the DCA plots to reveal gradients along which bird and tree communities were organised. We removed 14 sampling points before running the tree DCA (13 and one from the low and the medium tree cover, respectively), because no trees were found at these sampling points. Using ANOVAs, we then tested whether tree and bird community turnover between sampling points—as indicated by the site scores of the first DCA axis (Hill and Gauch 1980)—was related to differences between zones.

Due to logistic reasons—including a lack of accessible and matching forest-agriculture gradients elsewhere in the study region—we were only able to assess productivity and biodiversity patterns along a single gradient. However, we found minimal autocorrelation for sampling points within the zones (see online Appendix 1), and thus believe our analysis to be robust. All analyses were carried out using R software (version 2.14.1, Foundation for statistical computing, 2011).

Results

Characteristics of the four zones

Area, population, percentages of land covers, and productivities of the four zones (forest, high tree cover, medium tree cover, and low tree cover zones) are displayed in Table 1. The zones were of comparable area, except for the medium tree cover zone, which was markedly larger. The number of inhabitants was the largest in the high tree cover zone and the lowest in the low tree zone (excluding the forest zone). The four zones displayed clear differences in terms of landscape composition, and particularly in terms of tree cover (see % tree cover, No of trees and basal area in Table 1). In addition, the medium tree cover zone had a greater grassland cover, and the low tree cover zone a greater cropland cover. When considering total productivity (crop, feed and fuel), the high tree cover zone was the most productive of the three agricultural zones. The medium tree cover zone had the lowest crop productivity of the three agricultural zones while the high tree cover zone had a crop productivity similar to the low tree cover zone. The high tree cover zone also had the highest feed productivity, followed by the medium tree cover zone, the low tree cover zone, and Munesa forest. Munesa forest had the highest fuel productivity of the four zones, followed by the medium tree cover zone, the low tree cover zone and the high tree cover zone. People living in the high tree cover zone were using fuel and feed from both the high tree cover zone itself as well as the Munesa forest.

Table 1 Characteristics of the four zones investigated in the study
Full size table

Tree and bird biodiversity along the forest-agriculture gradient

In total, 4132 individual birds, 96 bird species, 4473 individual trees, and 52 tree species were recorded. Details of these numbers for the four zones are given in Table 1. Tree species presences differed between zones (zone; GLMER, Wald χ2 = 17593150, P < 0.001), where overall species presences were highest in the forest and decreased along the forest-agriculture gradient (Fig. 2a). However, patterns differed between introduced versus native trees (zone x status; GLMER, Wald χ2 = 2174048749, P < 0.001). Native tree species decreased continuously along the gradient, while introduced tree presences, which were absent in the forest, peaked in agricultural zone with medium tree cover. Introduced tree species presences were much higher in the high tree cover zone and the medium tree cover zone than in the low tree cover zone. This integration of planted tree species and crop production in these former two zones could be considered as examples of agroforestry systems (Fig. 2b).

Presence of species representing different bird guilds showed distinct patterns in the different zones (zone × guild; GLMER, Wald χ2 = 209.5, P < 0.001). Plant and seed eaters and omnivores tended to increase along the forest-agriculture gradient while fruit and nectar eaters tended to decrease (Fig. 3a). Invertebrate eaters and omnivores showed highest presences towards the forest, but not in the forest itself (Fig. 3a). Presences of small versus large range species differed across zones (zone × range; GLMER, Wald χ2 = 151.2, P < 0.001), with large range species tending to increase in agricultural land while small range species showed highest presences in the high tree cover zone (Fig. 3b).

Community composition along gradients of land use change

The tree DCA revealed distinct communities along tree, cropland and grassland cover gradients i.e., differed in the forest, high tree cover, medium tree cover, and low tree cover zones (Fig. 4a–c). The scores of first DCA axis differed among zones indicating significant turnover in the tree community (ANOVA, F1,84 = 44.4, P < 0.001). Similarly, the bird DCA showed that the bird communities differed along gradients of tree, cropland and grassland cover i.e., differed in the forest, high tree cover, medium tree cover and low tree cover zones (Fig. 5a–c). As for trees, species turnover in the bird community was related to differences among zones along the forest-agriculture gradient (ANOVA, F1,94 = 499.5, P < 0.001).

Fig. 4
figure 4

Tree communities along gradients of a cropland cover, b grassland cover, and c tree cover, plotted on a plane defined by the two first vectors of the detrended correspondence analysis (DCA). The contour lines indicate the degree to which the individual land cover type contributes to total cover (i.e., cover values ranging from 0 to 1). Data points represent the sampling sites in the four different zones (i.e., forest, high tree cover, medium tree cover, and low tree cover zones). Individual land cover types were measured in a 50 m radius around the sampling sites

Full size image
Fig. 5
figure 5

Bird communities along gradients of a cropland cover, b grassland cover, and c tree cover, plotted on a plane defined by the two first vectors of the detrended correspondence analysis (DCA). The contour lines indicate the degree to which the individual land cover type contributes to total cover (i.e., cover values ranging from 0 to 1). Data points represent the sampling sites in the four different zones (i.e., forest, high tree cover, medium tree cover, and low tree cover zones). Individual land cover types were measured in a 50 m radius around the sampling sites

Full size image

Discussion

Multifunctional agricultural landscapes can help maintain species of high conservation value

Tree and bird communities differed along gradients of tree, cropland, and grassland cover (Figs. 4, 5). Cropland, grassland, and tree cover provide different bundles of services: cropland provides food primarily (and secondarily crop residues used as feed and fuel); grassland provides feed primarily (and secondarily fuelwood); and tree cover provides primarily fuelwood (and secondarily feed). Thus, increasing production of these products in different mixtures is likely to change the tree and bird communities in different ways, as demonstrated by our analysis. We found that individual species that may be negatively affected in landscapes that focus on crop production may benefit in landscapes that focus on fuelwood or feed. By focusing solely on crop productivity, many past studies (e.g., Kleijn et al. 2009; Clough et al. 2011; Phalan et al. 2011; Gabriel et al. 2013) have potentially placed too much emphasis on trade-offs while missing possible synergies (see below).

We found that bird species benefitting from crop production tended to include plant and seed eaters, omnivores, as well as large range species (Fig. 3a). These species may not be considered of high conservation value (and may even cause crop losses for farmers in the case of plant and seed eaters, see below). However, we also found that small range species, that is species for which local conservation tends to be more critical, were more abundant in the agricultural zone with high tree cover compared to the other zones, surpassing even the forest zone (Fig. 3b). Thus, agricultural landscapes appear to host species of high conservation value which were seldom encountered even in the forest (e.g., the wattled ibis, Bostrychia carunculata, and the brown-rumped seedeater, Crithagra tristriatus, both of which are highland species endemic to the horn of Africa).

Conservation theories and planning have long been criticized for oversimplifying the role of the landscape matrix for many species (Franklin and Lindenmayer 2009). In part, this stems from an over-emphasis on the stark contrasts portrayed in Island Biogeography perspectives which assume landscape mosaics composed of biodiversity-rich patches of natural vegetation (e.g., forest) embedded in an otherwise hostile agricultural (biodiversity-depleted) matrix. The importance of the matrix as habitat is becoming increasingly appreciated (Vandermeer et al. 1998; Dauber et al. 2003; Swift et al. 2004; Vandermeer and Perfecto 2005; Donald and Evans 2006; Fischer et al. 2006; Chazdon et al. 2009; Reed et al. 2016). Our results also challenge this simplified view of the matrix, demonstrating that the matrix is not a ‘green desert’ but is rather part of the habitat for several species, and even the main part of the habitat for some small range species when managed as a complex mosaic of open vegetation patches (e.g., cropland, grassland) and patches covered by trees (such as the high tree cover zone). Such mosaics may support a wider range of species (including in particular open-habitat species) than continuous forest (Pasari et al. 2013; Kremen and Merenlender 2018). In many natural ecosystems, large herbivores play—or used to play—a major role in creating and maintaining patches of open vegetation, on which several species depend, either totally or partially (Owen-Smith 1988; van der Waal et al. 2011). In human-dominated landscapes, where large wild herbivores are rare or have been extirpated, agriculture is often the dominant force maintaining open patches and creating heterogeneity. Domestic herbivores may mimic the functions once provided by wild herbivores (Benton et al. 2003; Vera 2009; Wright et al. 2012) and contribute to the maintenance of biodiversity, e.g., open-habitat bird species (Wright et al. 2012). In tropical forests, traditional shifting cultivation practices create patches of open grassy fallows in an otherwise homogeneous forest cover. The resulting landscape mosaic may provide diverse habitats and food sources to several species e.g., populations of endangered Asian elephant in Sri Lanka (Wikramanayake et al. 2004).

It could be argued that our study lacks a natural/baseline habitat as defined by Phalan et al. (2014). Indeed, Munesa forest was not devoid of human influence, and was intensively used for feed (grazing) and fuel: an estimated 80 GJ ha−1 of biomass was extracted every year. This could be seen as a limitation of our effort to understand the link between land management and biodiversity. However, we argue that very few forests are truly unexploited by people: many of them are used for grazing and fuelwood collection (as in this study), or for harvesting of various food products (Hladik et al. 1990; Fa et al. 2003; Vinceti et al. 2008; Nasi et al. 2011; Ickowitz et al. 2014; http://www1.cifor.org/pen). Even rainforests thought to be primary or old growth in nature have been found to have sustained periods of extensive human use (Willis et al. 2004; Levis et al. 2017; Maezumi et al. 2018). Only ~18% of the land globally (excluding Antarctica) is actually without measurable human impact (Venter et al. 2016). Thus, our experimental setup is likely to reflect the rule rather than the exception.

Multifunctional agricultural landscapes can support high agricultural productivity

The high tree cover zone had crop productivity similar to the low tree cover zone (29 and 30 GJ ha−1, respectively), despite farmers in the latter having greater market access (in Arsi Negele) and having transformed their landscape for crop production more intensely than in the other zones (Table 1). Crop productivity was similarly dominated by staples in the three zones (in average, ~93 to 95% of the total crop productivity—in GJ—was represented by wheat, maize, and potato for farms of the three zones). Total productivity (crop, feed and fuel) declined with decreasing tree cover totalling 160, 141, and 129 GJ ha−1 in the high, medium, and low tree cover zones, respectively (Table 1). These results challenge the notion that strong trade-offs exist between biodiversity conservation, including the maintenance of trees, and agricultural productivity. Our results point to the existence of landscape mosaics delivering high agricultural productivity in multiple dimensions while simultaneously playing a key role in biodiversity conservation. Interactions between forests/trees and agriculture may explain the high productivity observed in the high tree cover zone (Foli et al. 2014; Reed et al. 2017). In agroforestry systems, trees may benefit crop yield through erosion control (Young 1989), the capture of nutrients below the root zone of annual crops and recycling in the topsoil through litter decomposition (Chikowo et al. 2003) and regulation of the local climate (Ong et al. 2000; Shiferaw Sida et al. 2018). Some tree species may also fix atmospheric nitrogen (Ajayi et al. 2011), or mobilize phosphorus through root exudation or mycorrhiza (Watt and Evans 1999; Smith and Read 2008), potentially benefiting associated crops. Finally, trees may provide habitats to beneficial species, such as pollinators (Blanche et al. 2006; Ricketts et al. 2008; Garibaldi et al. 2013), predatory and parasitoid arthropods controlling crop pest populations (Dix et al. 1995; Bianchi et al. 2006) or insectivorous birds (Koh 2008; see section below). The diversity of productions in complex mosaics may also lead to greater resilience to shocks e.g., climate, or input availability (Duriaux Chavarría et al. 2018).

Trophic composition of the bird communities differed between the three agricultural zones and the forest, with important implications for smallholder farmers. Compared to the forest, agricultural zones hosted less fruit and nectar eaters but more plant and seed eaters. In the high tree cover zone, plant and seed eaters (potential ‘pests’ which may consume crops) were almost as abundant as invertebrate eaters (potential ‘natural enemies’ which may provide a service to farmers by consuming insect pests; Fig. 3a). In contrast, in the low tree cover zone, plant and seed eaters were far more abundant than invertebrate eaters. These differences may explain the fact that crop productivity in the high tree cover zone—despite a high density of trees that may compete with crops for radiation, water and nutrients—was similar to crop productivity in the low tree cover zone. As such, farmers in the high tree cover zone may benefit from ecosystem services provided by birds (such as predation on insect pests) and other wild species, whilst farmers in the low tree cover zone may sustain ecosystem disservices from a community of species adapted to feed on the dominant resource in the landscapes: crops.

Wild birds control crop and livestock pests in many agroecosystems. For example, wild insectivorous birds have been found to regulate the populations of insect pests attacking apple (Mols and Visser 2007) and oil palm (Koh and Wilcove 2008). Managers may enhance this natural pest control by erecting nest boxes in the first case (Mols and Visser 2007) and by maintaining ground and epiphytic ferns that serve as nesting sites in the second (Koh 2008). Similarly oxpeckers—birds native to Africa and feeding specifically on ectoparasites of large herbivores—are known to reduce the tick burden of domestic cattle (Samish 2000). Populations of oxpeckers may be maintained by banning the use of certain acaricides and conserving wild ungulates in the landscape. Raptors may also prey on rodents (a major pest in cereal-based farming systems) and their presence in the field may be stimulated by providing natural or artificial perches (Kay et al. 1994). Thus, the specific services and disservices provided by wild communities in diverse landscape mosaics may positively impact agricultural productivity.

Conclusions

This study suggests that focusing solely on crop productivity when examining the interactions between agricultural productivity and biodiversity—as is the case for most studies using the land sparing and land sharing approaches—risks missing possible synergies. To the best of our knowledge, this is the first study linking important elements of biodiversity to the three main products valued by rural communities in most landscape mosaics: crops, livestock (or feed), and fuel. Areas used for livestock grazing and fuel collection tend to be relatively undisturbed land units with higher tree cover (e.g., rangelands, forests) compared to cultivated croplands, and such areas may host higher levels of biodiversity as a consequence. Therefore, considering livestock and fuel production in addition to crop production could provide a very different view on the relationship between agriculture and biodiversity; one less dominated by trade-offs and one where coexistence and synergies are more prevalent. This, in turn, would lead to very different policy recommendations.

Although this study only focusses on a few provisioning ecosystem services, it highlights the importance of considering bundles of ecosystem services when evaluating the performance of landscape mosaics (Bennett et al. 2009). The land sparing and land sharing approaches have been developed and chiefly used by conservation ecologists, and as a result convey a strong ecological focus. In these approaches, agricultural productivity has also commonly been assessed through a narrow lens (often exclusively crop yields), disregarding local perspectives of what is actually important to people in terms of ecosystem services (Pascual et al. 2017). In addition, most biodiversity of interest to conservation ecologists tends to be species found in closed forest, rather than species found in more disturbed landscapes and providing ecosystem services to people (Gibson et al. 2011). Dimensions of ecosystem services and human wellbeing, as encapsulated in landscape sustainability science (Wu 2013), need to be incorporated to design interventions that truly balance the need of people and nature.

Our results also suggest that biodiversity may play a critical role in supporting agriculture (e.g., ecosystem services provided by trees and invertebrate-eating birds). We found agricultural productivity to be highest in the landscape with the highest tree cover, and lowest in the landscape most transformed for agricultural production. This challenges the conventional paradigm of agricultural intensification based on landscape simplification and dependence on external inputs. Livelihoods may be better supported by landscape approaches based on multi-sector interconnections between different land uses and ecosystem services (Milder et al. 2012). Productive landscapes should be designed to make the most of the services provided by biodiversity, and replace or complement mechanical and chemical inputs where possible (Bommarco et al. 2013). Such farming systems are likely to: (1) be more benign to the environment than modern conventional farming, (2) depend on self-organization rather than (largely fossil-fuel based) energy, and (3) use biophysical resources more efficiently (Doré et al. 2011; Tittonell 2014).