Agricultural landscapes provide the essential ecosystem service of food to growing human populations; at the same time, agricultural expansion to increase crop production results in forest fragmentation, degrading many other forest-dependent ecosystem services. However, surprisingly little is known about the role that forest fragments play in the provision of ecosystem services and how fragmentation affects landscape multifunctionality at scales relevant to land management decisions.
We measured the provision of six ecosystem services (crop production, pest regulation, decomposition, carbon storage, soil fertility and water quality regulation) in soya bean fields at different distances from adjacent forest fragments that differed in isolation and size across an agricultural landscape in Quebec, Canada.
We observed significant effects of distance-from-forest, fragment isolation and fragment size on crop production, insect pest regulation, and decomposition.
Distance-from-forest and fragment isolation had unique influences on service provision for each of the ecosystem services we measured. For example, pest regulation was maximized adjacent to forest fragments, while crop production was maximized at intermediate distances-from-forest. As a consequence, landscape multifunctionality depended on landscape heterogeneity: the range of field and forest fragment types present.
We also observed strong negative and positive relationships between ecosystem services that were more prevalent at greater distances-from-forest.
Synthesis and applications. Our study is one of the first to empirically measure and model the effects of forest fragments on the simultaneous provision of multiple ecosystem services in an agro-ecosystem at the landscape and field scales relevant to landowners and managers. Our results demonstrate that forest fragments, irrespective of their size, can affect the provision of multiple ecosystem services in surrounding fields, but that this effect is mediated by fragment isolation across the landscape. Our results also suggest that managing habitat fragmentation and landscape structure will improve our ability to optimize ecosystem service provision and create multifunctional agricultural landscapes.
Maintaining the provision of multiple ecosystem services in agricultural landscapes while increasing crop production is a critical global challenge (Jordan & Warner 2010; Sachs et al. 2010). Agricultural landscapes provide numerous goods and services important for human well-being, including food, fibre, water quality regulation, soil formation, flood regulation and recreation (Zhang et al. 2007; Power 2010). Yet management of agricultural landscapes generally focuses on crop production and the expansion of agricultural lands (Saunders, Hobbs & Margules 1991; Robinson & Sutherland 2002), typically leading to the loss of other ecosystems and their associated biodiversity (Green et al. 2005; Phalan et al. 2011), ecosystem services and, ultimately, loss of landscape multifunctionality (i.e. the ability to provide multiple ecosystem services; Robertson & Swinton 2005; Tscharntke et al. 2012a).
There is considerable indirect evidence that forest fragments within agricultural landscapes affect many ecosystem services: they are habitat for insect species that provide pollination and pest regulation services in adjacent fields (Tscharntke et al. 2005; Bianchi, Booij & Tscharntke 2006; Ricketts et al. 2008; Holzschuh, Steffan-Dewenter & Tscharntke 2010), and plant species that provide climate regulation and water purification services (Foley et al. 2005); they can alter microclimate conditions that affect crop production (Kort 1988); change dispersal patterns for fungi and soil organisms that affect decomposition (Plantegenest, Le May & Fabre 2007); and alter water and nutrient flow through landscapes (Brauman et al. 2007). However, we have little direct empirical data that demonstrate how forest fragments influence sets of ecosystem services simultaneously, the distances over which this influence might occur, or how fragment size and connectivity can alter these patterns and the relationships between different ecosystem services.
One result of the expansion of agricultural lands is forest loss and fragmentation (Saunders, Hobbs & Margules 1991; Tscharntke et al. 2005). This, in turn, causes loss of connectivity and biodiversity, changes to ecosystem function and may alter the supply and distribution of ecosystem services (Tscharntke et al. 2012b). Theory predicts that changes to landscape structure and forest connectivity should affect ecosystem service provision through two main mechanisms (Mitchell, Bennett & Gonzalez 2013). First, forest fragmentation can directly influence the movement of organisms and matter important for service provision. For example, fragmentation can alter the ability of pollinators, crop pests, pest predators (Tscharntke, Rand & Bianchi 2005; Kremen et al. 2007), water and nutrients (Brauman et al. 2007; Power 2010) to move across a landscape, which might affect pollination, pest regulation or nutrient cycling, among other services. Secondly, habitat loss and disruption of connectivity between fragments affects demography, leading to changes in biodiversity, and ecosystem functions that contribute to service provision (Loreau, Mouquet & Gonzalez 2003; Leibold et al. 2004). However, while current theory predicts that forest fragment size and connectivity should influence ecosystem service provision, the actual effects of forest fragments on multiple ecosystem services have rarely been measured or modelled (Mitchell, Bennett & Gonzalez 2013), especially at the scale of fields and forests relevant to land managers. Effective management of agricultural landscapes for multiple ecosystem services requires measurements of changes in service provision as forest cover varies, and an understanding of the processes that underlie these patterns.
Here, we describe how forest fragments in an agricultural landscape affect the patterns of six above- and below-ground ecosystem services: crop production, pest regulation, decomposition, carbon storage, soil fertility and water quality regulation (Table 1). We hypothesized that forest fragments would influence the provision of multiple ecosystem services, that these effects would decay with distance-from-forest and that this influence would depend on fragment size and isolation, as well as the service itself. We also hypothesized that because the overall influence of fragments would covary with forest fragment size and isolation, the presence of fragments would affect the relationships (synergies and trade-offs) between ecosystem services.
Table 1. Ecosystem services and indicators analysed
Relationship between indicator and service provision
Soya bean yield
Kilograms of soya beans hectare−1
Soya bean aphids plant−1
Proportion of leaves damaged by insects
Proportion of buried cotton fabric decomposed
Proportion of buried litter decomposed
Soil organic matter
Per cent carbon in soil by weight
Per cent nitrogen in soil by weight
Water quality regulation
Soil phosphorus saturation
Phosphorus-sorption saturation (per cent P binding sites occupied)
Materials and methods
Study Site and Sampling Design
We measured ecosystem service indicators in soya bean fields of the Richelieu River watershed east of Montreal, Quebec (Fig. S1, Supporting information) in 2010 (n =15 fields) and 2011 (n =19 fields). New fields were chosen each year because soya bean is grown in rotation with corn in this region. Soya bean fields were selected adjacent to forest fragments that spanned the gradients of fragment size (range 0·5 to 4880 ha; mean 530 ha) and fragment isolation present in this landscape. Using the Quebec provincial Système d'information écoforestière data set, we quantified forest fragment size and calculated fragment isolation using proximity index for each fragment in FRAGSTATS 3·3 (University of Massachusetts, Amherst MA, USA). Proximity index (PI) is the sum of fragment areas divided by the nearest edge-to-edge distance squared between each focal fragment and neighbouring fragments within a specified distance. We used 2000 m, as it represents an intermediate scale to which arthropod groups respond (Chaplin-Kramer et al. 2011) and is larger than most fields in our system. Higher PI values indicate lower fragment isolation (i.e. a more connected forest landscape).
Within each soya bean field, a transect perpendicular to the adjacent forest fragment edge was established. Agricultural fields in Quebec follow the seigneurial system of land distribution, where fields are arranged in long narrow strips. This meant transects followed a gradient of distance-from-forest, while other field and management variables were kept consistent. We expected that (i) forest fragments would affect below-ground services at smaller distances than above-ground services due to differences in dispersal distances for above- and below-ground organisms (van der Putten et al. 2001), and that (ii) the effects of forest fragments on ecosystem services would decay with distance (i.e. the majority of variation in service provision would occur at small distances-to-forest). Therefore, below-ground services were measured at 0, 5, 10 and 25 m from forest in 2010 and 0, 10, 25, 50 and 100 m in 2011. Above-ground services were measured at 0, 50, 100, 200 and 500 m from forest in both years. Additionally, the above-ground service of crop production was measured at 10 and 25 m in both years.
Ecosystem Service Indicators
We used eight indicators to quantify our six ecosystem services (Table 1). Each indicator was chosen because it (i) was expected to vary over the spatial scale of our transects, (ii) was feasible to quantify during a single growing season, (iii) matched metrics already used in the region (CRAAQ 2010; Raudsepp-Hearne, Peterson & Bennett 2010) and (iv) was relevant to local landowners. We focused on using indicators that reflect some of the multiple processes underlying each ecosystem service; therefore, some services had multiple indicators. For example, we measured both litter and cotton decomposition to quantify total decomposition from both soil macro- and micro-organisms. In addition, we chose to interpret each indicator only for the service each most strongly determined, although some indicators are involved in multiple services. For example, soil phosphorus saturation and nitrogen both contribute to water quality and soil fertility. However, in our system, water quality corresponds most strongly with soil P (Bochove et al. 2007), and soil fertility with soil N. Finally, some of our indicators correspond positively with service provision (i.e. soya bean yield), others negatively (i.e. aphid numbers, herbivory). We present our results in terms of service provision; therefore, in some cases, lower values of an indicator signal higher service provision.
Above-Ground Ecosystem Service Indicators
Soya bean yield was measured as soya bean dry weight. At each distance-from-forest, we collected soya bean plants from two crop rows along a distance of 0·5 m just before harvest (22–24 September 2010; 27–30 September 2011). Plants were dried at 50 °C for 48 h, mechanically threshed, and the separated soya beans were then weighed. Yield (kg ha−1) was calculated based on the area of field sampled.
We measured pest regulation through aphid abundance and herbivory levels. This definition incorporates both ecosystem services (i.e. predation or parasitism) and disservices (i.e. pest pressure or colonization), that may respond to forest fragments differently and at different scales, and measures net effects, the variable of greatest interest to farmers. Soya bean aphids are an important pest of soya bean crops in Quebec, and economically harmful levels (>250 aphids plant−1) have increased pesticide use (Ragsdale et al. 2011). Aphid abundances were measured biweekly by counting all live individuals on the above-ground parts of five soya bean plants at each distance-from-forest (Gardiner et al. 2009), twice in 2010 (27–30 July and 9–13 August) and three times in 2011 (19–22 July, 1–5 August and 17–20 August). These sampling periods were timed to coincide with regional aphid population peaks (CRAAQ 2014). Because we used commercial soya bean fields, destructive plant sampling was not possible. Instead, we estimated plant size by recording total leaf number (Sivakumar 1978) and insect herbivory by counting the number of leaves with obvious insect damage (i.e. holes).
Below-ground Ecosystem Service Indicators
Water quality regulation, carbon storage and soil fertility were estimated using soil properties. At each distance-from-forest, a composite sample of five soil cores, 2 cm each in diameter and 0–15 cm deep, was collected in the middle of each growing season (7–12 July 2010; 4–16 July 2011). Each sample was air-dried for 1 week, ground to pass through a 1-mm mesh and dried at 50 °C for 48 h.
Water Quality Regulation
We quantified soil phosphorus saturation index (i.e. the ratio of soil P to Al), which provides a measure of the ability of a soil to both release P and bind additional P (Kleinman & Sharpley 2002). As the index increases, the potential for excess P to enter run-off and contribute to eutrophication rises; for our region, values >12% indicate excess P (Beauchemin & Simard 2000). We calculated P saturation using Mehlich-3 extractions: 2·5 g of soil and 25 mL of Mehlich-3 solution were shaken for 5 min, filtered and then analysed colourimetrically for P and spectrophotometrically for Al.
Carbon Storage & Soil Fertility
Agricultural soils as carbon stores are important regulators of climate change, and soil carbon is vital to soil structure and fertility and is therefore often used as a proxy for soil ecosystem services (Bommarco, Kleijn & Potts 2013). Soil N is also critical for crop growth especially for corn (CPQ 2000), which in our region is grown in an annual rotation with soya bean and requires high N input. Soil C and soil N were measured as percentage by weight using an elemental auto-analyser on 60 μg of soil.
We quantified decomposition by measuring leaf litter decomposition and soil microbial and fungal activity, both important for decomposing organic matter in soils (Barrios 2007). For litter decomposition, 5 g of air-dried Acer saccharum Marsh. (sugar maple) litter was placed in 1-mm mesh 10 × 10 cm nylon bags (Harmon, Nadelhoffer & Blair 1999). Each bag had eight 5-mm holes to allow entry of soil macrofauna (Smith et al. 2009). Three litterbags were buried at a 45° angle and 15 cm depth at each distance-from-forest for c. 3 months (24 June–23 September 2010; 22 June–28 September 2011). Prior to soya bean harvest and field tillage, litterbags were collected and frozen until processed. To process, remnant litter was removed from each bag, dried overnight at 55 °C and then ashed at 360 °C to obtain the ash-free dry mass and correct for soil contamination (Smith et al. 2009). The percentage mass lost was calculated for each bag, representing total decomposition (physical breakdown and mineralization). Soil microbial activity was estimated using cotton fabric squares (Tiegs et al. 2007). Four 5 × 5 cm squares of unbleached, undyed cotton fabric were weighed and then buried vertically in the soil at 10 cm depth at each distance-from-forest. The fabric remained in the soil for c. 3 weeks (23 June–16 July 2010; 27 June–22 July 2011). Each square was then removed and frozen until processing. Squares were hand-washed of all soil, air-dried and weighed to determine mass loss from microbial and fungal decomposition.
We used generalized additive mixed models (GAMMs; Wood 2006) to model each ecosystem service indicator as a function of distance-from-forest, forest fragment isolation (PI) and fragment size. We used GAMMs as opposed to generalized linear models as we had no a priori expectation that the relationships would be linear. GAMMs were fit using the ‘gamm’ function in the ‘mgcv’ package of R version 3.0.2. We used cubic regression spline smoothers with ‘shrinkage’ for each explanatory variable in the GAMMs, with field as a random factor. ‘Shrinkage’ is a method to minimize the degree of smoothing in the model for each explanatory variable, reducing each relationship to a linear function where possible (Wood 2006). Forest fragment isolation and size were log10-transformed in every model. We analysed data from each year separately as attempts to include year as a random or repeated measure prevented model convergence. For aphid abundance and herbivory, where multiple censuses were performed each year, we fit repeated-measures GAMMs for each year separately. For these models, each census was nested within field as a random factor, and we modelled compound symmetrical correlation between the censuses nested within distance-from-forest within field; other correlation types failed to improve model fit as evaluated using AICc values. For analyses of aphid abundance, we also included plant size (i.e. number of leaves) as a covariate. For each ecosystem service indicator, the appropriate distribution type (i.e. Gaussian, negative binomial or binomial) and link function were used (see Table S1 in Supporting Information). Standard diagnostic plots were inspected to evaluate model fit.
To evaluate relationships between ecosystem service indicators at distances both relatively near- and far-from-forest fragments, we calculated Spearman rank correlations between our indicators at each distance-from-forest. We split transects in half to define ‘near’ and ‘far’ locations: above-ground, values from 0, 50 and 100 m were considered near-to-forest, while 200 and 500 m were far; below-ground, 0, 10 and 25 m were defined as near, and 50 and 100 m as far. Each indicator was transformed so that higher values corresponded to higher values of service provision (e.g. decreased herbivory or decreased soil phosphorus saturation equalled increased provision of pest regulation or water quality regulation, respectively); these same data were also used to evaluate landscape multifunctionality. We pooled data from 2010 and 2011 for the Spearman rank and multifunctionality analyses, except for aphid abundance and herbivory as these showed distinct patterns between years. Despite this, relationships between these two indicators and all other service indicators were generally consistent between years. To evaluate correlations between below-ground and above-ground service indicators, we used data from 2011 where measurements of both sets of services overlapped (i.e. 0 m for near- and 100 m for far-from-forest). Correlation values were tested as being different than zero (Spearman's rho α < 0·05).
As a measure of landscape multifunctionality, we calculated the inverse Simpson index in R using the ‘vegan’ package at each distance-from-forest for above- and below-ground service indicators separately. Using the ‘nlme’ package in R, polynomial relationships for distance-from-forest and field as a random factor, we fit nonlinear mixed-effects models to determine the effect of distance-from-forest, forest fragment isolation category [i.e. log10(PI) < 1·16; 1·16 < log10(PI) < 2·31; log10(PI) > 2·31] and their interaction on the inverse Simpson index values. The three isolation categories equally divided the range of PI values present and were used to simplify the presentation of the multifunctionality results. The appropriate nth-order distance-from-forest polynomial term in each model was determined using AICc values.
Effects of Distance-From-Forest
Distance-from-forest had strong effects on soya bean yield, soya bean aphids and insect herbivory in adjacent soya bean fields (see Table S1 and Fig. S2, Supporting information). Soya bean yield peaked c. 100 m from forest in both years (Fig. 1a,b) and was on average 117% and 55% greater at 100 m than directly adjacent to forest in 2010 (Fig. 2a) and 2011, respectively. Aphid patterns were opposite those of soya bean yield in 2010, with 75% and 117% less regulation at 100 m than at 0 and 500 m, respectively, although aphid numbers in 2010 were very low and patchy, which could have affected our analysis (Fig. 1c and 2c). Aphid numbers were on average 12 times higher in 2011, and the relationship between aphid numbers and distance-from-forest disappeared (Fig. 1d); however, a pattern similar to 2010 was present at the end of 2011 (see Fig. S3, Supporting information). Herbivory regulation was higher close to forest in both years (Fig. 1e,f), although herbivory was on average 38% higher in 2011. Herbivory regulation was 77% and 19% lower at 500 m than at 0 m in 2010 (Fig. 2e) and 2011, respectively. Below-ground, distance-from-forest only affected decomposition in 2011, although the below-ground transect in 2010 only spanned 25 m. In 2011, there was 20% less cotton decomposition at 100 m than at 0 m from forest (Fig. 1h). Relationships between distance-from-forest and other below-ground service indicators were not statistically significant (see Table S1, Supporting information).
Effects of Forest Fragment Isolation
Fragment isolation affected aphids, herbivory and cotton decomposition (see Fig. S4, Supporting information). Average soya bean yield was greater next to connected compared to isolated fragments by 15% in both 2010 and 2011 (Figs 1a,b and 2b), but these relationships were not statistically significant (see Table S1, Supporting information). In 2010, aphid regulation increased with fragment isolation and was 86% higher in fields adjacent to the most isolated fragments (Figs 1c and 2d). A different pattern was observed in 2011, with aphid regulation on average 4·2 times lower in fields next to isolated fragments (Fig. 1d). Herbivory regulation increased with fragment isolation (Figs 1e,f and 2f); in 2010, it was 58% greater in fields next to isolated fragments and 16% greater in 2011. Cotton decomposition was greatest at intermediate levels of forest fragment isolation, but the position of this peak shifted between years (Figs 1g,h and 2h). In both years, cotton decomposition was lowest in fields adjacent to the most isolated fragments (81% less in 2010 and 72% in 2011). Other relationships between ecosystem service indicators and forest fragment isolation were not statistically significant (see Table S1, Supporting information).
Effects of Forest Fragment Size
Only three indicators showed a relationship with forest fragment size: a negative relationship with aphid regulation in 2011, a concave-up relationship with cotton decomposition in 2010 and a concave-down relationship with soil nitrogen in 2010 (see Table S1; Fig. S5, Supporting information).
Effects of Forest Fragments on Ecosystem Service Relationships
We found both positive and negative correlations between service indicators, and typically, the strength of these relationships varied with distance-from-forest (Figs 3 and S6, Supporting information). Strong positive correlations were limited to below-ground service indicators and varied in fields with distance-from-forest. Only the relationship between soil carbon and nitrogen was consistently positive. Negative correlations between indicators were more frequent far from the forest; only the trade-offs between aphid regulation–soya bean yield and 2011 herbivory regulation–soya bean yield were statistically significant near-to-forest. While the strength of service indicator relationships varied with distance-from-forest, we did not observe any synergies changing to trade-offs or vice versa as distance-from-forest varied.
Effects of Forest Fragments on Landscape Multifunctionality
There was no single combination of forest fragment isolation and distance-from-forest where all above- or below-ground service indicators were maximized as measured using the inverse Simpson index (Fig. 4). Above-ground, multifunctionality was greatest directly adjacent to forest fragments and decreased significantly over 100 m, except for isolated forest fragments (see Fig. S7, Supporting information). There was little variation in below-ground multifunctionality with distance-from-forest or fragment isolation, although connected fragments at 25 and 50 m did have slightly higher values.
We provide clear empirical evidence that forest fragments influence the provision of multiple ecosystem service indicators in adjacent agricultural fields. While some of these effects have been observed or modelled for a few services (Ricketts 2004; Bodin et al. 2006; Farwig et al. 2009), our results expand on these studies by simultaneously considering the effects of distance-from-forest, fragment isolation and fragment size on multiple above- and below-ground services.
Effects of Distance-from-Forest and Fragment Isolation on Ecosystem Services
Each ecosystem service was maximized at a unique distance-from-forest and level of fragment isolation (Figs 1 and 2). Key mechanisms for these patterns likely involve the effects of fragments on the movement of different service-providing organisms, matter or energy (Mitchell, Bennett & Gonzalez 2013). For example, decreased crop production within 50 m of forest fragments was likely to be a consequence of competition for water and light from forest plants (Kort 1988), although soil compaction from farm machinery, which is often concentrated at field margins (Hamza & Anderson 2005), might have also contributed. The peak in soya bean yield we observed 75–200 m from forest could have resulted from decreased competition in combination with increased pollination near fragments (Ricketts et al. 2008); soya bean is primarily self-pollinated, but insect pollination can increase yield (Chiari et al. 2005).
While crop production was affected primarily by distance-from-forest, aphid regulation appeared more tightly tied to forest fragment isolation, with specific effects dependent on overall aphid numbers. Aphid populations in North America undergo regular outbreak cycles, with long-distance dispersal occurring primarily via atmospheric movements (Ragsdale et al. 2011). Forest fragments can interrupt this long-distance dispersal, and aphid landing is often concentrated next to movement barriers such as forest fragments (Irwin, Kampmeier & Weisser 2007). At the same time, aphid populations are affected by increased predator pressure near forest fragments (Chaplin-Kramer et al. 2011) and by resource levels (i.e. soya bean growth). In 2010, aphid numbers were relatively low, and the peak at 100 m from forest was likely to be due to a combination of increased aphid dispersal, increased soya bean growth and relief from increased aphid predation closer to forest fragments. In 2011, aphid numbers were high enough that we suspect their short-distance dispersal within fields overwhelmed any effects of forest fragments on aphid landing patterns or soya bean growth. Instead, we saw lower aphid numbers in more connected landscapes, suggesting that landscape connectivity benefited aphid predators (Thies, Roschewitz & Tscharntke 2005; Tscharntke et al. 2005).
Interestingly, herbivory regulation was greatest close to isolated forest fragments. While forest fragments in our system most likely provided habitat for insect predators that control soya bean pests, forests and increased forest connectivity could have facilitated herbivore dispersal (Bianchi, Booij & Tscharntke 2006), especially far-from-forest where predator pressure is often reduced. Finally, for cotton decomposition, we observed the highest rates of decomposition near forest fragments with intermediate isolation. Forest fragments can act both as sources (Edman et al. 2004) and landscape barriers (Plantegenest, Le May & Fabre 2007) to the airborne dispersal of fungi and micro-organisms, leading to increased decomposition near forest fragments. Why fragments with intermediate isolation might show increased decomposition remains unclear.
Ecosystem Service Trade-Offs and Synergies
Our results show that changes in field size and distance-from-forest could affect the presence and strength of trade-offs between services in this system. While ecosystem service trade-offs and synergies have been quantified elsewhere (Raudsepp-Hearne, Peterson & Bennett 2010; Gamfeldt et al. 2013), we currently know little about what drives these relationships (Bommarco, Kleijn & Potts 2013) or how they change across scales relevant to management. This lack of knowledge limits our ability to manage landscapes for multiple services, especially if trade-offs identified at one scale (i.e. the watershed) are not present at the scale that the landscape is managed (i.e. the field).
We observed both positive (indicative of synergy) and negative (indicative of trade-off) relationships between service indicators whose strength changed with distance-from-forest (Fig. 3; Bennett, Peterson & Gordon 2009). Strong positive or negative relationships were more common far-from-forest, suggesting that forest fragments can moderate ecosystem service relationships. This could occur through changes in environmental conditions, or if fragments affect the spatial subsidy of organisms to fields. For example, strong trade-offs between soil carbon and herbivory and between soil nitrogen and herbivory may not be present near forests because increased predator diversity and abundance decrease insect pest populations (Tscharntke, Rand & Bianchi 2005; Bianchi, Booij & Tscharntke 2006), weakening the link between plant quality and pest performance. Similarly, strong synergies between decomposition and soil carbon or soil nitrogen may be weaker near forest fragments because other drivers of decomposition, including temperature and moisture (Prescott 2010), may vary to a greater degree and obscure relationships between decomposition and soil nutrients.
Effects of distance-from-forest and fragment isolation on multiple ecosystem services
We observed that different ecosystem service indicators were maximized at different distances-from-forest fragments and found no evidence that multifunctionality is maximized at a single distance-from-forest, degree of fragment isolation or combination of the two (Fig. 4). While the importance of species diversity for multiple ecosystem functions and services is now the focus of investigation (Isbell et al. 2011; Gamfeldt et al. 2013), landscape structure and heterogeneity have not been explored to the same extent (Symstad et al. 2003; Fahrig et al. 2011). Our results show that both distance-from-forest and forest fragment isolation are likely to be important for service provision. Ecological theory predicts this, because the key species and processes that mediate each service are expected to vary at different scales (Kremen 2005). Landscape heterogeneity may therefore be necessary to maintain biodiversity (Fahrig et al. 2011) and create multifunctional agricultural landscapes, but the combination and amount of different landscape elements is likely to depend on the agro-ecosystem in question, the biodiversity and ecosystem processes that it encompasses, and the needs and preferences of the landowners and policymakers that manage it.
The distinct patterns of ecosystem service indicators we observed within fields, across forest fragments, and between years, highlight the difficulties of designing agricultural landscapes to optimize ecosystem service provision. Any change in landscape structure or heterogeneity is likely to have varying or even opposing effects on different services (de Groot et al. 2010). For example, clearing forests for fields may increase crop production but also jeopardize other functions upon which crop production relies, such as pest regulation or pollination (Power 2010). The effects of changing landscape structure or heterogeneity are also likely to depend on the spatial scale considered. For instance, forest fragments might act as movement barriers and increase aphid numbers in nearby fields, but simultaneously sustain aphid predators and aphid control across the wider landscape. Effective management for pest regulation may therefore require cross-scale cooperation between landowners and an understanding of the patterns of service provision at different scales (Bommarco, Kleijn & Potts 2013). Our results suggest that distance-from-forest (i.e. field size) and the configuration of forest fragments may be managed to influence landscape multifunctionality; however, improving the provision of all services is likely to be difficult. There is a pressing need to strongly link landscape structure at different scales with the provision of sets of ecosystem services and understand how landscape structure affects the species and processes underlying these patterns.
Agricultural expansion often leaves landscapes composed of small, isolated forest fragments while fields grow in size and connectivity (Robinson & Sutherland 2002). As global food demand increases, pressure to clear natural habitat and increase agricultural area will grow. Our results demonstrate that landscape structure, and specifically the characteristics of forest fragments, can affect multiple ecosystem services, and that these effects are mediated both by distance-from-forest in adjacent fields and forest fragment isolation across the landscape. Managing landscape structure to control forest fragment connectivity may therefore be a more effective tool to manage agricultural landscapes for multiple ecosystem services than simply limiting further forest loss. At the same time, the effects of distance-from-forest and fragment isolation on different ecosystem services vary widely. This emphasizes the importance of incorporating a variety of forest fragment types across agricultural landscapes to maximize multifunctionality. Knowledge of the effects of landscape structure and forest fragmentation on agricultural ecosystem services, and an enhanced understanding of how the movement of key organisms and matter affect service provision, has the potential to enhance the design and management of multifunctional agricultural landscapes in the future.
We thank the farmers of the Montérégie for allowing us to use their fields; M. Luke, E. Hartley and E. Pickering-Pedersen for field assistance; H. Lalonde for soil analysis; and D. Maneli of the Gault Nature Reserve for logistical support. Comments from three anonymous reviewers greatly improved the quality of the manuscript. This work was supported by an NSERC PGS-D scholarship to MGEM, an NSERC Strategic Projects Grant to EMB and AG, a grant from the Ouranos Consortium to AG and EMB, and funding from the Quebec Centre for Biodiversity Science to MGEM, EMB and AG; AG is supported by the Canada Research Chair Program.
Ecosystem service indicator data, R scripts and plot locations: DRYAD entry doi:10.5061/dryad.41r51 (Mitchell, Bennett & Gonzalez 2014).