By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
Invertebrates supporting natural pest control and pollination ecosystem services are crucial to world-wide crop production. Understanding national patterns in the spatial structure of natural pest control and pollination can be used to promote effective crop management and contribute to long-term food security.
We mapped the species richness and functional diversity of ground beetles and bees to provide surrogate measures of natural pest control and pollination for Great Britain. Functional diversity represents the value and range of morphological and behavioural traits that support ecosystem services. We modelled the rate at which functional diversity collapsed in response to species extinctions to provide an index of functional redundancy.
Deficits in functional diversity for both pest control and pollination were found in areas of high arable crop production. Ground beetle functional redundancy was positively correlated with the landscape cover of semi-natural habitats where extinctions were ordered by body size and dispersal ability. For bees, functional redundancy showed a weak positive correlation with semi-natural habitat cover where species extinctions were ordered by feeding specialization.
Synthesis and applications. Increasingly, evidence suggests that functionally diverse assemblages of ground beetles and bees may be a key element to strategies that aim to support pollination and natural pest control in crops. If deficits in both functional diversity and redundancy in areas of high crop production are to be reversed, then targeted implementation of agri-environment schemes that establish semi-natural habitat may provide a policy mechanism for supporting these ecosystem services.
By 2050, global population size is predicted to increase by 46% necessitating greater agricultural production to achieve food security (FAO 2006). Historically, increased yields have been achieved by improved agronomy, mechanized farming practices, chemical fertilizers, pesticides and new breeding techniques (Godfray et al. 2010). However, yield increases are frequently showing evidence of levelling off, so enhanced production must be achieved using new approaches (Godfray et al. 2010). While the development of new technology and crop varieties is crucial to improving yields, maximizing ecosystem services will also contribute to promoting agricultural productivity (Losey & Vaughan 2006; Gallai et al. 2009; Godfray et al. 2010). Natural pest control and pollination are ecosystem services that support agriculture and are delivered in part by invertebrates (Losey & Vaughan 2006; Gallai et al. 2009). Invertebrate pests damage 18% of world agricultural production, and while their control is achieved principally via chemical methods, the role of predatory and parasitic invertebrates is crucial (Symondson, Sunderland & Greenstone 2002; Losey & Vaughan 2006; Straub, Finke & Snyder 2008). In the USA, invertebrate natural pest control is worth $4·5 billion p.a., equivalent to 4·2% of US farm cash receipts (Losey & Vaughan 2006). Insect pollination is similarly important to agriculture and is estimated to support 9·5% of world food production (€153 billion) principally in the form of vegetables, fruits and oil producing crops (Gallai et al. 2009). While enhancing natural pest control and pollination could lead to increased crop yields, multiple threats to invertebrate populations are undermining the sustained delivery of these services (Kromp 1999; Straub, Finke & Snyder 2008; Potts et al. 2010). To properly manage ecosystem services in agricultural landscapes will require an improved understanding of both how they are distributed at policy-relevant (e.g. national) spatial scales and what their likely robustness to environmental change will be.
For both natural pest control and pollination, practical limitations mean that direct monitoring of ecosystem services at large spatial scales would be hard to implement. Surrogate metrics derived from invertebrate community structure may provide an alternative to mapping the delivery of ecosystem services. The abundance of invertebrates is one such metric and is known to be a key determinant of pollination and pest control (Kromp 1999; Symondson, Sunderland & Greenstone 2002; Potts et al. 2010). However, abundance is likely to be highly variable across landscapes as a response to local field or farm scale management (Bianchi, Booij & Tscharntke 2006; Straub, Finke & Snyder 2008; Woodcock et al. 2010). Best practices required to promote the abundance of invertebrates at farm scales are often well understood (e.g. Collins et al. 2002; Woodcock et al. 2010), with the limiting factor to their implementation depending on individual farmer management decisions or government policy requirements. However, where there is a limitation in the regional species pool, for example due to wide-scale species loss linked with agricultural intensification, this may place a more fundamental limit on the delivery of pollination and pest control (Straub, Finke & Snyder 2008; Stoate et al. 2009; Potts et al. 2010). For example, direct links between insect pollinator species richness and seed set have been found in many studies (Hoehn et al. 2008; Albrecht et al. 2012). While the effect of species richness on the delivery of natural pest control has been hard to predict in small scale mesocosm studies, there is evidence that species-rich assemblages are more likely to deliver improved pest control under real agricultural conditions (Straub, Finke & Snyder 2008). Species richness is a simple descriptor of community structure and takes no account of the range and value of behavioural or morphological species traits that contribute to ecosystems service delivery. Increased functional diversity of insect pollinators can promote the delivery of pollination services (Hoehn et al. 2008; Albrecht et al. 2012), while dissimilarity in functional traits among invertebrate predators may reduce negative competitive interactions, thereby promoting improved pest control (Schmitz 2007; Straub, Finke & Snyder 2008; Woodcock & Heard 2011).
Patterns of species richness and functional diversity may provide a surrogate measure of the current spatial distribution of ecosystem services. However, future land use and environmental change will have consequences for which, and how many, species persist over the long term (Kotze & O'Hara 2003; Potts et al. 2010; Williams et al. 2010). If species go locally extinct, then the unique traits that they contribute will be lost and overall functional diversity will decline, potentially impacting on ecosystem service delivery (Straub, Finke & Snyder 2008; Potts et al. 2010; Woodcock et al. 2010). The rate of decline in functional diversity with species loss provides an indication of the redundancy of a community in its capacity to deliver ecosystem services. Species are unlikely to go extinct at random, rather ordered patterns of extinctions reflecting sensitivities to environmental change will occur (Kotze & O'Hara 2003; Bommarco et al. 2010; Williams et al. 2010; Woodcock et al. 2012). For example, in Europe, large-bodied ground beetles are more prone to population decline than small species (Kotze & O'Hara 2003). Understanding what the potential consequences of ordered scenarios of species extinctions are on the robustness of pollination and pest control services is crucial to their long-term management.
We focus on UK arable farming systems that currently cover 4·4 million ha and have a net value of £ 3·1 billion p.a. (Defra 2010). We map the distribution of species richness and functional diversity for taxa important in the delivery of natural pest control (ground beetles: Coleoptera, Carabidae) and pollination (bees: Hymenoptera, Apidae) (Kromp 1999; Potts et al. 2010). We then model the consequences of ordered species extinctions from these communities to identify how robust their functional diversity will be in response to future environmental change. We predict that (i) once corrected for latitudinal differences in species pools, the distribution of species richness and functional diversity across Great Britain will reveal deficits in areas of high agricultural production (Stoate et al. 2009; Potts et al. 2010); (ii) the decline in functional diversity with species extinctions (hereafter referred to as functional redundancy) will be affected by the order with which species are lost; and (iii) increased availability of semi-natural habitat at a landscape scale will promote functional redundancy and so robustness to future environmental change. Note most UK habitats are modified and so are assumed to be at best semi-natural.
Materials and methods
Focal taxa for delivering natural pest control and pollination
Generalist predators are abundant and species rich in arable farmland and have been shown to reduce pest populations in 75% of field studies (Symondson, Sunderland & Greenstone 2002). Their spatial distribution is often well recorded nationally, particularly when compared to specialist pest control agents such as hymenopteran parasitoids. We used ground beetles as model taxa for assessing the distribution of these predators. Ground beetles have been used as indicators of anthropogenic disturbance and environmental change (Rainio & Niemela 2003) and are one of a suite of dominant generalist predators found in arable crops (Symondson, Sunderland & Greenstone 2002; Woodcock et al. 2010). They have been directly shown to reduce population sizes of economically significant agricultural pests, including aphids, slugs, root feeding flies and phytophagous beetles (Kromp 1999; Collins et al. 2002; Bommarco, Firle & Ekbom 2007; Zaller et al. 2009). Their abundance can also be actively encouraged through agri-environment schemes that provide financial incentives for farmer to modify land management (Woodcock et al. 2010). In the case of crop pollination, a variety of insect taxa have been linked with increasing seed set (Hoehn et al. 2008; Potts et al. 2010; Albrecht et al. 2012). However, bees (Apidae) are consistently identified as being primary pollinators for many crops (Potts et al. 2010) and are used here to assess the distribution of pollination services. Bee pollination has been shown to increase the yield and crop quality of oilseed rape, a principal UK arable crop (Bommarco, Marini & Vaissiére 2012).
For both ground beetles and bees, a limited set of species are found in association with arable crops, and these are considered to be the key providers of ecosystem services in these systems (see Appendices S1 & S2 in Supporting Information). The subset of ground beetles found in arable crops was determined using large-scale data sets of ground beetles recorded from 250 arable fields and four break crops (Firbank et al. 2003). This subset was confirmed by comparing it to other published data sets (see Appendix S1, Supporting Information). Only predatory (zoophagous) ground beetles were included, limiting the pool to 60 species from 25 genera. As cereal crops do not rely on insect pollination, we consider here bees known to pollinate oilseed rape Brasica napus L. (Brasicaeae), which by area is the dominant UK insect-pollinated crop (Defra 2010). Forty-five species of bee from seven genera were determined to be oilseed rape pollinators based on both published (Woodcock et al. 2013) and unpublished non-quantitative surveys (18 UK farms surveyed in 2011; S. Faulk, P. Harvey and D. Sheppard, pers. comm.).
Distribution maps for ground beetles and bees were derived from records stored in the National Biodiversity Network of the UK Biological Records Centre (BRC). National biodiversity recording is typically carried out by volunteers, and so non-standardized recorder effort is a common problem (Hill 2012). To correct for variable recorder effort, we used the ‘Frescalo’ algorithm to determine the probability of individual species occurrence in 10-km grid squares (Hill 2012). This method uses a Poisson modelling process incorporating information on benchmark species to correct for sampling effort (Hill 2012). For each 10-km grid square in Great Britain (2824 squares total), the probability of ground beetle and bee species occurrence was determined. From this, the species richness of ground beetles and bees involved in natural pest control and pollination was determined for each grid square. These data were used in all subsequent calculations of functional diversity and redundancy. As semi-natural habitats provide important resources for both ground beetles and bees (Thiele 1977; Bianchi, Booij & Tscharntke 2006; Potts et al. 2010; Woodcock et al. 2010), we determined the percentage cover of this resource in each 10-km grid square based on the UK Land Cover Map (Morton et al. 2011). Semi-natural habitat combined the cover of grasslands (rough, acid, neutral and calcareous, but not improved with NPK fertilizer), wetlands (bogs, fen, and marshland), heathland (heather grassland and dwarf shrub heath), woodland (broadleaf and coniferous) and montane habitat.
Traits are defined as physical or behavioural characteristics that evolve in response to competitive interactions and abiotic conditions. They influence survival, fitness and rates of resource processing, and so their diversity is linked with ecosystem service delivery (Schmitz 2007; Hoehn et al. 2008; Straub, Finke & Snyder 2008; Woodcock & Heard 2011; Albrecht et al. 2012). We identified traits based on three broad categories: (i) pollination/hunting efficiency, (ii) foraging range/dispersal and (iii) key aspects of species ecology and behaviour (Forsythe 1983; Juliano 1986; Ribera et al. 1999; Kotze & O'Hara 2003; Bommarco et al. 2010; Williams et al. 2010; Woodcock et al. 2010; Wamser et al. 2011). A full description of the traits and their relevance for the delivery of ecosystem services are given in Table 1. For each 10-km grid square, the functional diversity of ground beetles and bees was determined using the ‘Functional Dispersion’ index (FDis) using the traits described in Table 1 (Laliberté & Legendre 2010). The FDis index represents the average distance of species in multidimensional space from a centroid defined by a distance matrix weighted by the probability of individual species occurrence. Species encountered more frequently will have a greater effect on the value of FDis. All traits in the analysis were given equal weighting. As the traits for both bees and ground beetles (Table 1) were represented by a mixture of variable types (both continuous and categorical), the Gower method was used to calculate the distance matrix and all traits scores standardized to have a range of 0–1 (Gower 1971; Laliberté & Legendre 2010). This index is not affected by species richness (Laliberté & Legendre 2010).
Table 1. Traits used to define functional diversity and redundancy of natural pest control and pollination services delivered by ground beetles and bees
Ground beetles (pest control)
Efficiency in delivering pest control/pollination
Diet specialization (Collembola specialists, obligate predators and omnivores): The range of potential pest species eaten will affect natural pest control
Diet specialization (polylectic vs. oligolectic): The range of plants foraged upon will affect specificity to the crop and ability to persist on secondary resources across complex landscapes (Williams et al. 2010)
Visual and sensory acuity (ratios of Eye: head width and Antennae: body length): Defines the relative size of key sensory organs used during hunting and foraging (Bauer et al. 1998; Ribera et al. 1999; Woodcock et al. 2010)
Temporal range of pollinating activity: These are defined by i) the start month of the flight period and ii) the total duration of flight period (months). This will influence the likelihood on congruence of bees with flowering crops
Feeding rate (body mass): Body size (mg) affects partitioning of prey types between species (Radloff & DuToit 2004) and is positively related to feeding rates (Juliano 1986) and negatively related to reproductive output (Kotze & O'Hara 2003)
Mobility and utilization of complex landscape structure
Foraging range (Femora width: length ratio): Used as an index of walking speed and so potential area covered foraging (Forsythe 1983; Ribera et al. 1999)
Foraging range (Intertegular distance categorized as 1–3 mm, 3–4, 4–6 mm and >6 mm): Intertegular distance is correlated with bee foraging ranges (Greenleaf et al. 2007) and so affect resource utilization across complex landscapes (Bommarco et al. 2010). As intraspecific range in ITD can be large, it was treated as categorical
Flight (Wings full, absent or dimorphic): Presence of wings affect dispersal ability and utilization of fragmented landscapes (Kotze & O'Hara 2003)
Biology and behaviour
Diurnal activity (nocturnal, diel or both): Activity period will influence what pests are likely to be encountered, their activity rates on an off plants and so interspecific resource partitioning (Luff 1978)
Social behaviour (social or solitary): As social bees are more sensitive to pesticides, increased diversity in this trait will promote pollination under typical agricultural management (Williams et al. 2010).
Breeding period (autumn/winter or spring/summer): Breeding periods affect activity rates and so encounter with prey throughout the year and can influence rates of colonization of arable fields after winter (Wamser et al. 2011)
Nesting behaviour (mining, cavity nesting or other): Affects sensitivity to tillage regimes and so persistence under different agricultural management (Williams et al. 2010)
Brood number (single, double or continuous): May influence population recovery rates after agricultural management
As the pool of species found in northern latitudes is limited by fundamental climate requirements (e.g. Thiele 1977), both the species richness (SR) and functional diversity (FD) of ground beetles and bees were characterized by a negative latitudinal cline. Without correcting for latitudinal gradients in species richness, any management intended to support ecosystem service providing taxa (e.g. agri-environment schemes) might be biased to northern clines based on the misconception that there was a local ecosystem service deficit. To account for this, we calculated a derived index of species richness (SRLat) and functional diversity (FDLat) represented by the residuals from a linear regression of species richness or functional diversity with latitude (ground beetles: SR = 63·52–3·91 × 10−5 × latitude (m); FD = 0·24–1·217 × 10−8 × latitude; bees: SR = 44·25–3·80 × 10−5 × latitude). For bee functional diversity, FDLat was based on the residuals from a third-order polynomial response to latitude (FD = 0·24 + 1·13 × 10−7 × latitude – 3·36 × 10−14×latitude 2 – 2·50 × 10−19 × latitude 3).
Functional redundancy has been defined in many different ways, but is considered here to be a measure of the rate of decline in functional diversity with species extinctions. This is defined by the slope parameter (β) of a linear regression between the number of species that have gone extinct and the change in functional diversity (FDis) after each species is lost. High rates of decline in functional diversity in response to species loss indicate a community with low functional redundancy. Such a community would be limited in its capacity to maintain ecosystem services where environmental change resulted in local species extinctions. While biologically unlikely, a null model of random species extinction was used to assess the relative rates of decline in functional diversity compared with species extinctions ordered in a biologically realistic manner (see below) (Kotze & O'Hara 2003; Bommarco et al. 2010; Williams et al. 2010). For each 10-km grid square, species were deleted until only one remained. Following each species deletion, the functional diversity of the remaining assemblage was calculated based on their combined traits (Table 1). The deletion process was repeated 500 times, and a mean slope parameter (βRandom) defining the null model of functional redundancy was calculated for each 10-km grid square.
The slope parameters for this null model were compared to an equivalent slope (βOrdered) defined by species extinctions ordered by traits known to affect population sizes and local extinction rates in both ground beetles and bees. These were the following:
Body size: For ground beetles, species loss was ordered so that the largest species (body mass) went extinct first, reflecting observed declines in European ground beetles linked to their reduced dispersal and lower reproductive rates (Kotze & O'Hara 2003). For the bees, smaller species (based on intertegular distance) were assumed to go extinct first. Although it has been suggested that smaller bees may be better suited to surviving in small habitat patches (Williams et al. 2010), larger bees have greater foraging ranges and so are better able to utilize widely distributed resources in fragmented agricultural landscapes (Greenleaf et al. 2007; Bommarco et al. 2010).
Diet specialization: Species with specialist niches, such as a limited diet breadth, are more likely to undergo population declines in both ground beetles (Kotze & O'Hara 2003) and bees (Bommarco et al. 2010). For ground beetles, extinctions were in order of collembola specialist, obligate predators and then omnivores. For bees, oligophagous flower foraging species were deleted before polyphagous species.
3a. Ground beetle flight: Ground beetles with wing dimorphism can colonize fragmented and isolated habitat, then once established flightless morphs of the same species are superior competitors (Kotze & O'Hara 2003). In contrast, obligate flightless species are ill suited to persist in highly fragmented landscapes, while obligate fully winged species tend to be comparatively poor competitors once colonized (Kotze & O'Hara 2003). We modelled species extinctions in order of flightless, full-winged and then wing-dimorphic species.
3b. Sociality: Social bees are more sensitive to pesticides and isolation from semi-natural habitats than solitary species (Bommarco et al. 2010; Williams et al. 2010) and so were modelled to go extinct first.
Following the same procedure as described above, the slope parameter (βOrdered) was calculated following sequential extinctions from each 10-km grid square. Where traits used to describe the order of species loss were categorical (e.g. solitary vs. social bees), species were deleted at random within a particular trait level before moving onto the next. As for the null model, this process was repeated 500 times and an average slope parameter calculated. A relative index of functional redundancy (FRRelative) was then calculated as the percentage difference between these decline slopes for random and ordered species extinctions (FRRelative index = (β Random - β Ordered)/β Random × 100). Positive values of FRRelative indicate a rate of decline in functional diversity that is lower than would be expected if species extinctions had been entirely at random, with the converse of this being true for negative values.
The response of the latitude-corrected species richness (SRlat), latitude-corrected functional diversity (FDlat) and all functional redundancy indexes (FRRelative) to the percentage cover of semi-natural habitat in 10-km grid squares was assessed using general linear models in SAS v9.1 (SAS Institute Inc., Cary, NC, USA). Following Borcard & Legendre (2002), we used principal coordinates of neighbour matrices (PCNM) to account for spatial autocorrelation in these models. Geographical distances among sampling points (taken to be the south-east corner of each 10-km grid square) were used to obtain eigenvectors that describe the spatial structure of the data at a wide variety of scales. These eigenvectors were subsequently included as covariates in GLM models. As the PCNM method calculates a large number of eigenvectors describing a complex range of spatial structures underpinning the data (equivalent to c. 50% of all the 2824 sampling points), we tested the first 200 of these as univariate correlations against each response variable. Only those shown to be significantly (P <0·05) correlated with a response variable were included in final models with semi-natural habitat cover. Note that as the PCNM eigenvectors have only been included as covariates to account for underlying spatial structure, they are not be reported in the results section. While the percentage cover of arable crop in each 10-km grid square was considered as a potential covariate describing land use intensity, its strong negative correlation with the percentage cover of semi-natural habitat (F1,2562= 280·3, P <0·001, β= −0·95) and resulting lack of independence made its inclusion inappropriate. Paired t-tests were also used to determine whether there was an overall difference in the slope of decline in functional diversity resulting from random (β Random) or ordered (β Ordered) species extinctions.
Species richness and functional diversity
The spatial distribution of ground beetle and bee species richness (SRLat) showed deficits in both the south-west and north-west regions of Great Britain (Fig. 1). In contrast, central and eastern parts of England associated with high levels of arable crop production supported high levels of SRLat for both ground beetles and bees, with this trend extending to eastern parts of Scotland for the bees. However, this was somewhat reversed for the companion surrogate measure of ecosystem service delivery, functional diversity (FDLat). In contrast to SRLat, central and eastern England had deficits in FDLat for both the ground beetles and bees (Fig. 1). For the ground beetles, functional diversity was highest in the West of the UK, although this distribution was somewhat patchy. For the bees, FDLat was highest in Scotland, Wales, northern and south-west England.
For both the ground beetles (F1,2699= 64·9, P <0·001) and bees (F1,2691= 237·7, P <0·001), SRLat was negatively correlated with the percentage cover of semi-natural habitat in 10-km grid squares (Fig. 2). In contrast, FDLat was positively correlated with the availability of semi-natural habitat for both the ground beetles (F1,2693= 133·3, P <0·001) and bees (F1,2695= 79·9, P <0·001), although the slope was greater for the ground beetles (Fig. 2).
The rate of decline in ground beetle functional diversity in response to species extinctions (βOrdered) was found to be significantly different from that predicted by the null model of random species extinction (βRandom). However, the direction of this difference varied with species trait. Where beetle species extinctions were ordered by diet specialization (t2823 = 55·98, P <0·001), the rate of decline in functional diversity with species loss was lower than predicted by the null model. Whereas the rate of decline in functional diversity was higher than predicted by the null model when beetle extinctions were ordered by body size (t2823 = −52·5, P <0·001) and ability to fly (t2823 = −117·6, P <0·001), for bees, species extinctions ordered by body size led to greater rates of decline in functional diversity with species loss compared with the null model (t2823 = −60·0, P <0·001). Where social bees were modelled as becoming extinct before solitary bees, there was conversely an increase in the rate of decline in functional diversity with species loss (t2823 = −57·6, P <0·001). However, for bees, the loss of dietary specialists before generalists resulted in a lower rate of decline in functional diversity, compared with the null model (t2823 = 54·6, P <0·001).
Functional redundancy (FRRelative), describing the percentage difference in the decline slopes for random and ordered species extinctions, was correlated with the percentage cover of semi-natural habitat for both the ground beetles and bees. For the ground beetles, FR Relative was positively correlated with semi-natural habitat cover where species extinctions were ordered by both body size (F1,2708= 30·8, P <0·001; Fig. 3a) and ability to fly (F1,2697= 21·2, P <0·001, Fig. 3c), although not by diet specialization (F1,2703= 0·18, P >0·05). For the bees, FRRelative was positively correlated with the percentage cover of semi-natural habitat where species extinctions were ordered by diet specialization (F1,2648= 7·09, P <0·01, Fig. 3b), although this correlation was not significant where extinctions were ordered by social structure (F1,2665= 2·30, P >0·05) or body size (F1,2662= 0·98, P >0·05). The slope coefficients for the response of FRRelative to the cover of semi-natural habitat resulting from bee extinctions ordered by diet specialization were small (β = 0·02) compared with those reported for the ground beetles (body size: β = 0·75; ability to fly: β = 0·51). This suggests that over the range of semi-natural habitat covers encountered, the change in FRRelative for the bees would be largely inconsequential (Fig. 3).
Species richness and functional diversity
By mapping national-scale patterns of species richness and functional diversity, we provide crucial information for the development of targeted mitigation measures intended to support ecosystem services (Bianchi, Booij & Tscharntke 2006; Woodcock et al. 2010). Contrary to our prediction, low levels of species richness (once corrected for latitude) were not spatially linked with regions of high crop production, in particular, the intensively managed arable landscapes of central and eastern England (Defra 2010). Similarly species richness was negatively correlated with the cover of semi-natural habitats. This may on the surface appear to contradict evidence that habitat loss and degradation driven by intensive agriculture has led to declining ground beetle and bee species richness (e.g. Kromp 1999; Kotze & O'Hara 2003; Bommarco et al. 2010; Potts et al. 2010; Williams et al. 2010). However, it is important to take into account that we focused not on overall species richness, but instead on a subset of species known to be linked with arable agriculture and so likely to deliver ecosystem services. Species most likely to suffer from the effects of intensive agriculture are likely to be non-crop habitat specialists (Kotze & O'Hara 2003; Bommarco et al. 2010; Potts et al. 2010); however, such species were excluded from our analysis. Species found in arable crops are likely to possess adaptations that predispose them to colonization and survival in agricultural habitats (Thiele 1977). Thus, it is not unexpected that such species would at least be associated with areas of agricultural production, although their densities may well be relatively low in many such areas (Kotze & O'Hara 2003; Potts et al. 2010; Williams et al. 2010). This highlights a failing of using species richness as an indicator of ecosystem services. Specifically, it is an unweighted measure of invertebrate community structure that makes no distinction between rare and ubiquitous species; consequently, it may lack the resolution of information on rarity to be an inadequate indicator of ecosystem service provision.
In contrast, functional diversity, while dictated by species composition, has the advantage of being weighted by the probability of species occurrence. As species become rarer in landscapes denuded of semi-natural habitat, their contribution to overall functional diversity and so ecosystem service provision is reduced (Laliberté & Legendre 2010; Woodcock et al. 2010). This in part explains why species richness and functional diversity are, respectively, negatively and positively correlated with the percentage cover of semi-natural habitat. Enhancement of semi-natural habitat cover will promote functional diversity of ecosystem service providers in crops by increasing the probability of species occurrence. The implementation of agri-environment schemes may therefore be more valuable in diversifying the trait structure of ground beetles and bees than necessarily promoting increased species richness (Woodcock et al. 2010).
Functional redundancy and the order of species extinctions
Functional redundancy was typically lowest where extinctions were ordered according to traits known to affect species sensitivity to environmental change (Kotze & O'Hara 2003; Bommarco et al. 2010; Williams et al. 2010). Where species extinctions were ordered by body size (bees and ground beetles), flight ability (beetles) and sociality (bees), the decline in functional diversity with species loss was higher than occur under random extinction scenarios. Body size is strongly intercorrelated with a wide variety of traits, including dispersal, reproductive capacity and diet breadth (Kotze & O'Hara 2003; Greenleaf et al. 2007; Bommarco et al. 2010). These other trait characteristics will be systematically lost from the community with size-dependent extinctions, leading to an increased rate of collapse in functional diversity with species loss for both ground beetles and bees. Other species characteristics, not considered here, may also exacerbate the consequences of collapse in functional diversity with ordered species loss. For example, social bees have been found to be responsible for four times as many visitations to flowers as solitary bees (Albrecht et al. 2012). An increased likelihood of their local extinction may therefore have greater than predicted consequences for the delivery of pollination services (Williams et al. 2010). Land management could be adapted to preferentially support populations of species at the sensitive ends of a particular trait spectrum. For example, reducing levels of pesticide application or isolation from semi-natural habitat will benefit population stability of social bees, thus reducing the rate at which they go extinct (Williams et al. 2010). Such targeted management could therefore be used to promote functional redundancy in arable systems.
Where species extinctions were ordered by diet specialization, functional redundancy was consistently higher than predicted by the random model for both the ground beetles and bees. For the bees, it may be the case that while diet specialization is a predictor of responses to environmental change (e.g. sensitivity to habitat fragmentation), its consequences on ordered extinction rates do not occur independently of interactions with other traits. For instance, Bommarco et al. (2010) demonstrated that body size can be important in predicting the response of bees to habitat loss, but only when considered in the context of the dietary specialization of individual species. For dietary generalists, species of small size were more affected by habitat loss than larger-bodied species, with the reverse true for dietary specialists. It should be noted that Bommarco et al. (2010) considered this finding to be a potential artefact resulting from the possibility that the majority of small diet specialist bees had already gone extinct from the landscapes investigated. It is quite possible, however, that a similar mechanism is in operation with ground beetles, where the importance of diet specialization as a predictor of extinction rates is moderated by other as yet unconsidered species traits.
Semi-natural habitats to promote functional redundancy
For the ground beetles, correlative relationships suggested that their functional redundancy could be promoted by increasing the availability of semi-natural habitat at landscape scales, but only where extinctions are ordered by body size and flight ability. While there was some evidence that bee functional redundancy also increased with semi-natural habitat cover, the strength of this trend was too weak to make inferences that would be biologically relevant to applied management. For the bees, the spatial structure of semi-natural habitat may be more important in predicting the occurrence of individual species and their associated traits than simply its overall percentage cover in a 10-km grid square (Bommarco et al. 2010; Potts et al. 2010). Bees may also be more specific in what elements of semi-natural habitats represent viable alternative resources in an agricultural landscape (Potts et al. 2010) (e.g. those rich in flowers), particularly when contrasted with ground beetles that may be more plastic in their habitat associations (Thiele 1977). For this reason, the importance of semi-natural habitat as a key landscape element supporting robustness to environmental change may have been underestimated for the bees due to a limited capacity to define exactly which habitats were important. The existence of positive, albeit sometimes weak, correlations between functional redundancy and semi-natural habitat does emphasize the role that landscape scale conservation could play in supporting ecosystem service robustness by creating new semi-natural habitat (Bianchi, Booij & Tscharntke 2006; Lawton et al. 2010; Potts et al. 2010; Woodcock et al. 2012). As agri-environment schemes are implemented in association with arable agriculture, they represent a policy mechanism that can be used in promoting robustness of pest control and pollination by establishing new semi-natural habitat (Bianchi, Booij & Tscharntke 2006; Woodcock et al. 2010; Pywell et al. 2011). However, as the utility of different agri-environment scheme for pollinators and natural pest control agents differs, research effort is still required to assess best management practices to support these taxa (Woodcock et al. 2010; Pywell et al. 2011).
For invertebrates, our ability to predict large-scale patterns in ecosystem service provision has been limited by our understanding of the mechanistic relationship between community composition, functional diversity and ecosystem service provision rates. Although not considered in the current study, management at local scales that promotes abundances of these taxa will also be important in the delivery of ecosystems services. While research is increasingly focusing on interactions that underpin these relationships for both pest control and pollination, current predictions of service delivery must be based on assumptions that would be likely to be refined in time (Hoehn et al. 2008; Straub, Finke & Snyder 2008; Woodcock & Heard 2011; Albrecht et al. 2012). Independent of this, there remains a pressing need to develop new approaches to determine the distribution of ecosystem services, particularly where this allows responses to future environmental change to be predicted. Without such methodologies, we will be unable to manage agricultural landscapes in a pre-emptive manner and be limited to reactionary approaches that attempt to prop up failing levels of pollination and pest control in response to falling crop yields. This is clearly a serious long-term issue, as while there are many methods to establish semi-natural habitats to benefit pest control and pollination supporting invertebrates, they all take time to implement (Lawton et al. 2010; Woodcock et al. 2010; Pywell et al. 2011). The identification of landscapes that may be vulnerable to deficits in ecosystem services delivery, now or in the future, allows interventions to be implemented in a timely fashion that will secure their value and function in the long term (Lawton et al. 2010).
This work was partly supported by the PRESS2 project of the PEER research consortium. Thanks to James Bullock for his valuable comments on the article and the Biological Records Centre and all individuals involved in the recording of these taxa for access to long-term data sets. Species distribution data used in this article are accessible via the National Biodiversity Network's Gateway (http://data.nbn.org.uk/).