Present address: University of Plymouth, Seale-Hayne Faculty of Agriculture, Food and Land Use, Newton Abbott, Devon TQ12 6NQ, UK. †Bundesforschungsanstalt für Ernährung, Haid-und-Neustr. 9, 76131 Karlsruhe, Germany.
1 Understanding the spatial dynamics of insect distributions in farmland can provide insights into their ecological requirements and potential for management. Identifying the scale, location and persistence of species’ aggregations is an important step towards understanding the factors driving population distributions. This study examined how different carabid species were distributed in field and field margin habitats, and analysed their spatiotemporal dynamics.
2 We sampled 156 locations in a grid of pitfall traps over approximately 1 ha, comprising a hedgerow and adjacent parts of two arable fields. Traps were monitored between June and November 1994 to provide data with high spatial and temporal resolution of the two-dimensional distributions of six carabid beetle taxa.
3 The pattern of carabid activity-density over the experimental site was displayed in a series of contour plots. Spatial analyses by distance indices (SADIE) were used to provide aggregation indices (Ia) for the distributions of carabid species at different times.
4A second SADIE index (Im) of spatial association between two sets of counts was used to measure the spatial stability of species’ aggregation patterns through time. Im was also used to measure association or dissociation between the distributions of different species, and the spatial association between a species’ distribution and an environmental variable.
5 Each taxon was aggregated most of the time, but different taxa aggregated in patches within the field and/or hedgerow to different and varying extents. The field boundaries were important for some species, either as the only habitat in which they occurred (Amara spp.), as the major focus of more widespread distribution (Harpalus rufipes), or as a seasonal refuge (Nebria brevicollis). Within the crop habitat, different species also occupied different areas, providing some evidence to suggest species packing in space.
6 Spatial distributions between consecutive samples were strongly and positively associated, indicating stability of patch location over short time-scales. The strength of the association in turn was correlated with activity-density. When time lags were introduced between sample distributions, some negative association indices indicated displacement of patches. In most cases, however, positive association was maintained over long periods, demonstrating patch stability.
7 The spatial stability of patches suggests that future studies should focus on the detailed measurement of biotic and abiotic factors associated with patch location. Identifying the underlying causes of population distributions in farmland has important implications for managing habitat to enhance populations of beneficial predators, and targeting insecticide applications in precision agriculture to minimize their adverse side-effects.
It is increasingly recognized that a better understanding of farmland ecology is necessary to improve the targeting of environmental policies within the political spheres of European Union (EU) agricultural policy and World Trade Organization negotiations (Bignal 1998). Many agri-environment schemes, currently supported by governments within the EU, are aimed at reducing pesticide inputs and conserving biodiversity (Ovenden, Swash & Smallshire 1998). Farmland invertebrates play important roles as major contributors to total biodiversity on farmland and as food for other endangered groups, such as farmland birds (Fuller et al. 1995; McCracken & Bignal 1998). Furthermore, the invertebrate fauna includes many predatory groups, such as carabid beetles, which have an important role to play in natural pest control. Increasing natural predation of pests potentially reduces the need for pesticide inputs and decreases their undesirable side-effects on non-target species. Therefore understanding the ecology and managing populations of such organisms has been a major research focus in applied ecology in recent decades. However, despite significant progress, many aspects of insect ecology in farmland remain elusive or intractable. Difficulties arise because many important ecological and population processes operate at regional, farm or field scales, while experimental work is only feasible at the scale of the field or plots within fields.
Recent theoretical advances and simulation models have shown that the spatial dynamics of invertebrates in fragmented farmland ecosystems is a major factor in population and metapopulation processes (Sherratt & Jepson 1993; Halley, Thomas & Jepson 1996). With a more detailed understanding of insect distributions and their dynamics, populations might be more effectively managed for diversity and pest control within sustainable farming (Barker, Brown & Reynolds 1999). However, while spatial distributions are a characteristic ecological property of insects (Taylor 1984), it is difficult and time consuming to gather data with a suitably informative level of resolution.
Insects can be highly aggregated even in apparently uniform cultivated habitats (Hengeveld 1979), but this potentially informative detail of population distributions is not necessarily apparent from transect sampling. By identifying the location of patches and correlating these with coincident biotic and abiotic factors, it may be possible to identify the principle driving factors that determine population distributions. This knowledge may facilitate the manipulation or management of habitats for biodiversity, conservation or pest control. Similarly, the stability of patches can indicate how populations move between habitats in response to seasonal factors, natural or anthropogenic disturbances, or how species share resources in the same space. This information can also enhance our understanding of the interactions between species, or between species and environmental factors. Such information is necessary in order to minimize the detrimental effects of agricultural interventions on beneficial insects, and to optimize the management of non-crop habitats to enhance population size and survival.
Sampling designs that are spatially structured, for example on a grid of pitfall traps, provide a simple but labour-intensive way of obtaining two-dimensional information on carabid distributions. Pitfall trap catches depend on both population density and insect activity, to provide a quantity usually referred to as activity-density (Thomas, Parkinson & Marshall 1998). Thus, data from a grid of traps are easily presented as contour plots of activity-density that can be assessed visually. Statistical analyses of aggregated populations are generally restricted to descriptions of frequency distributions of counts, for example the negative binomial, or relationships between means and variances, such as Taylor’s power law (Southwood 1978a; Taylor 1984), to give a measure of the degree of aggregation. A major constraint of these methods is that they generally make little or no use of spatial information (Perry & Hewitt 1991). Some spatial information can be used in autocorrelation methods (Cliff & Ord 1981) to test for independence of samples as a function of distance between sample units (Midgarden, Youngman & Fleischer 1993). Statistical tests for differences in the degree of aggregation between two populations generally depend on testing for significant differences between parameters of their two distributions. However, because these methods do not explicitly use any spatial information, it is not possible to differentiate between two population distributions with the same level of aggregation but which occupy different space because of different behavioural or other ecological traits. If significance tests indicate differences or similarities in the level of aggregation between two populations, the best options to pursue have until recently still relied on textual or graphical reference to the original data (Syrjala 1996).
Spatial Analysis by Distance IndiciEs (SADIE) (Perry et al. 1996) is a new class of analytical techniques, applicable to experimental designs where data, in the form of counts, are available together with the spatial co-ordinates of each sample unit. The SADIE method provides an index of aggregation (Ia) for spatial data and a probability (Pa) that the data are not distributed randomly. Perry (1995) gives a clear and concise summary of the method. The SADIE method has now been developed and extended to enable the derivation of indices of spatial association or dissociation between two data sets (Perry 1998; Korie et al. 2000). The two data sets may be counts of the same species taken at different times; two species sharing habitat space; or a species and an environmental variable that is quantifiable in counts. Perry (1997, 1998) gives details of the methods. Briefly, an index of association (Im) is formed by combining two intermediate indices, If, which is formed explicitly from non-spatial information in coincident sampling units (analogous to a correlation coefficient), and Iz, which is formed explicitly from directional information in coincident and nearby sampling units.
We examined extensive data describing the spatial distribution of six taxa of carabid beetle (five species and one identified to genus). Populations were sampled at frequent intervals in the summer and autumn from a grid of pitfall traps spanning parts of two adjacent cereal fields and the hedgerow field boundary between them. We addressed the following questions. (i) How are different species distributed in field and field margin habitats? (ii) What are the temporal dynamics of these distributions?
Materials and methods
The field site at IACR – Long Ashton Research Station (OS grid reference ST537693) comprised an area encompassing parts of two adjacent fields of winter barley, separated by a 5·3-m wide hedgerow orientated north–south. Both fields had identical histories in continuous cereal production for the previous 3 years. Before that, the west field had been grass pasture for 4 years following orchard; the east field had been in continuous wheat for 6 years following orchard. Both fields had identical soil of heavy clay. The ground flora of the hedgerow was dominated by ivy Hedera helix L. on a slightly raised bank, with a shrub layer and a number of mature trees along the 120-m length. On the east side of the hedgerow, a 4-m wide experimental field margin had been established in the previous year. This comprised three replicate blocks of four vegetation types in 10-m plots: (i) Lolium perenne L.; (ii) mixed grasses and wild flowers; (iii) natural regeneration; and (iv) winter barley, arranged randomly within the blocks. The effects of these treatments are reported elsewhere (Thomas & Marshall 1999).
The carabids examined in this study were adult Pterostichus melanarius (Illiger), P. cupreus (Linnaeus), Nebria brevicollis (Fabricius), Agonum dorsale (Pontoppidan), Harpalus rufipes (Degeer) and Amara spp. Bonelli. Members of the genus Amara were not identifi-able to species in the field, but were shown in a parallel study (Thomas & Marshall 1999) to be predominantly Amaraplebeja Gyllenhal and Amaraovata Fab. All data refer to combined counts of males and females unless indicated. All species in this study are polyphagous (Hengeveld 1980) and likely to be found in the diet of farmland birds. They are known to feed to some extent on pests including slugs (Symondson et al. 1996), aphids (Sunderland & Vickerman 1980; Wallin et al. 1992) and weed seeds (Briggs 1965; Jorgensen & Toft 1997). Some ecological characteristics of these species are given in Table 1.
D, diurnal; DN, diurnal and nocturnal; N, nocturnal
BS: breeding season
S, spring and summer; A, autumn
EM: main period of emergence
SS, spring and summer; S, summer
AC: main period of adult activity
S, summer; A, spring and autumn after aestivation
The carabids were sampled with pitfall traps (plastic cups, diameter 60 mm, depth 70 mm, with snap-on lids) set in plastic sleeves such that the rim of the trap was flush with the soil surface. The traps were arranged in a 12 × 13 grid at 10-m intervals along the north–south axis, and at 5·3-m intervals along the east–west axis (scaled to the width of the hedgerow for parallel mark–recapture dispersal studies) (Thomas, Parkinson & Marshall 1998). Six columns of traps were to the west of the hedgerow, one column was between the hedgerow and the field margin plots on the east, and a further six columns of traps were in the field to the east of the field boundary. At each of the 156 grid intersections, three pitfall traps were set in a triangular arrangement, approximately 0·5 m apart, giving a total of 468 traps. Data from each set of three pitfall traps were pooled for analysis. The pitfall traps were operated dry, with a little soil and a few stones in the bottom to provide shelter, and were protected from the rain and birds by inverted plant pot saucers supported a few centimetres above the traps on wire legs.
Trapping commenced on 5 June 1994 by removal of the trap lids. Beetles captured overnight were identified, recorded and immediately released at the site of capture after the lids had been replaced on the traps. This eliminated potential bias from individuals falling in traps for a second time immediately after release. The trap lids were removed again on the following day and the procedure repeated. Beetles were thus able to move freely over the experimental site for at least 24 h before being exposed to the possibility of recapture. Trapping continued on alternate days until 17 August, when the experiment was terminated to allow the crop to be harvested. Crop density (number of tillers m−2) was measured on 10 August at two random points near to each set of pitfall traps at the grid intersections.
Traps were reset and run for 1 day on 29 September and then from 4 October to 4 November to provide data on the autumn-active species N. brevicollis. Over this period limited access to fields meant the traps were opened on Mondays then run continuously with data collected daily until Fridays, when the traps were closed. Captured beetles were released a metre or two away, to the north or south of the open traps. Released beetles were seen to conceal themselves immediately under stones or in crevices within a few centimetres of their release point. Any bias from beetles falling into traps immediately after release was considered negligible. Activity-density data are presented as the pitfall catches summed over each 4-night period. The cereal crop in the experimental arena had been harvested and left in set-aside. However, the extreme eastern line of traps had been lost under the plough, leaving a 12 × 12 grid of traps.
Contour plots of activity-density were drawn using Surfer version 6·04 (Golden Software Inc. 1997, Golden, CO, USA) using the Kriging option to interpolate contours. SADIE analyses (200 simulations per test) were performed on each data set to derive the index of aggregation Ia (Perry 1995). Data are aggregated if the index Ia is greater than unity, and are considered significantly different from random if the associated probability Pa < 0·025. Further analyses were performed on pairs of data sets to derive the index of association Im (Perry 1997, 1998; Korie et al. 2000). Im was used to assess the stability of spatial distributions between sampling periods, and association or dissociation between the spatial distributions of different species. Two sets of counts are considered spatially associated (i.e. the positions of patches remain spatially stable) if the index Im is greater than zero, or disassociated if Im is negative. The probability that two distributions are associated or dissociated is considered significant if Pm < 0·025 or Pm > 0·975, respectively.
The extensive data set permitted 37 contour plots to be drawn for each of the six taxa. However, for conciseness, contour plots are presented as the accumulated catch over six or seven consecutive samples (Figs 1–3). Agonum dorsale and N. brevicollis were only captured during the first half of the summer sampling period and are therefore represented in only three contour plots each (Fig. 3a–f). Autumn samples of N. brevicollis were of 1 day, 29 September (Fig. 3g), and accumulated 4-day catches over 5 weeks during October and November (Fig. 3h–l). In each contour plot, north lies at the top of the page. The hedgerow, running from north to south, is marked with a hatched bar, and the field margin plots lie to the east of the hedgerow.
Figures 1–3 show each species to have contrasting distribution patterns. The most striking differences were between field and hedgerow species. Pterostichus cupreus and P.melanarius (Fig. 1) occurred in large patches within the cropped area of the fields, although each species dominated fields on different sides of the hedgerow. In contrast, Amara spp. and H. rufipes (Fig. 2) were found in small patches in association with the hedgerow, while Agonum dorsale (Fig. 3) occurred in small patches in both the field and hedgerow. During the summer, N. brevicollis (Fig. 3) occurred in the hedgerow, with activity declining to zero by 10 July as they entered aestivation. By late September, N. brevicollis had re-emerged from the hedgerow to occupy the fields during the autumn.
Contour plots provide a convenient visual image of species distributions. However, they can conceal subtle differences because of the somewhat arbitrary grouping of data into convenient classes defining contour levels. Figure 4 shows examples of three plots with the original data posted over the sampling points. Figure 4a shows the aggregation ‘hot spots’ of Amara spp. within the field boundary, with catches of one or two individuals at other positions close to the boundary, but a virtual absence from the cropped areas of the fields. Figure 4b shows similar aggregation hot spots of H. rufipes in the field boundary but, in contrast to Amara spp., H. rufipes was also widely dispersed at low densities in the cropped areas of the adjacent fields. Similarly, although P. cupreus dominated the field to the east of the field boundary (Fig. 4c), this species was virtually absent from the field boundary itself but was present at low densities in the field to the west of the field boundary. These latter individuals are likely to have given rise to the emergence of the second generation in the west field (Fig. 1f).
The changes in activity-density (total number of beetles caught in pitfall traps per day, summed over the whole sampling grid) are given for each species and total carabids in Fig. 5. As with the spatial distributions, each species showed a different temporal pattern of activity-density throughout the summer. Day-to-day fluctuations were probably due to random sampling errors and changes in activity in response to, for example, weather factors, although these were not measured specifically in the present study. Pterostichus cupreus activity-density fluctuated around 250 between 6 June and 18 July (Fig. 5a), after which activity-density dropped to a low of about 50 by 3 August. This was followed by a period of wide fluctuations, peaking at 500 as the following generation of teneral adults emerged, identified by their soft elytra.
Pterostichus melanarius activity-density (Fig. 5b) started low but gradually built up during the summer in a series of four increasing peaks. These peaks are known to be activity-dependent because mark–recapture analysis on this species showed the absolute population density to be relatively stable following a period of emergence (Thomas, Parkinson & Marshall 1998).
In contrast, activity-density of Amara spp. (Fig. 5c) started low, increased to a peak and then decreased, while H. rufipes (Fig. 5d) started low and continued to increase throughout the sampling period. Agonum dorsale and N. brevicollis (Fig. 5e,f) both started high and decreased to zero before the end of the sampling period.
Aggregation (Ia) was measured for each taxon on each sample date by SADIE analysis (Fig. 6). The distributions of the two Pterostichus species (Fig. 6a,b) and N. brevicollis in the autumn (Fig. 6f) gave Ia values significantly (Pa < 0·025) greater than unity on nearly all dates, indicating a high level of aggregation. The aggregation indices, however, showed a periodic trend of increasing and decreasing Ia values, suggesting the contraction and expansion of aggregation foci. For the other species, and N. brevicollis in the summer, distributions were aggregated (Ia > 1) on only approximately 60% of the sample dates (Fig. 6c–f), despite the obvious patchiness of their distribution patterns in the contour plots (Figs 2 and 3). Corresponding Pa values did not generally indicate significant differences from random distributions (Pa > 0·025).
SADIE failed to give Ia values > 1 on several dates, especially for the hedgerow species data sets where low or zero counts occurred in much of the grid. The SADIE analyses were therefore repeated using the same accumulated data as had been used to draw the contour plots (Figs 1–3) to reduce the number of zero counts. Results are given above the contour plots (Figs 1–3). Pooling data over several sample dates generally increased the Ia values and indicated aggregation for all four hedgerow species over all periods, except Amara spp. between 12 and 22 July (Fig. 2d) and N. brevicollis from 18 to 28 June (Fig. 3e).
Because most of these hedgerow species were largely confined to the field boundary, there were still a large number of zero counts in the rest of the grid, even when data were pooled over six sample dates. Aggregation may then be indicated simply from the resolution of the hedgerow habitat as a single patch within the experimental area. To compensate for these effects and to test for aggregation into patches within the hedgerow habitat, a second analysis was performed using the accumulated data restricted to the four columns of traps most closely associated with the field boundary. However, no further resolution of aggregation within the field boundary was gained; some Ia values increased slightly, while others decreased. The average Ia values for Amara spp., H. rufipes and N. brevicollis were 1·33, 1·26 and 1·16, respectively, when data from the whole grid were analysed. These changed to 1·34, 1·08 and 1·12, respectively, when analyses were restricted to data from the four columns of traps in the proximity of the hedgerow.
If species packing in space occurs, the distribution of total carabids might be expected to be more random or regular than individual species. To test this, SADIE analyses were performed on the distributions of total carabids on each sampling date. This analysis also indicated high levels of aggregation even though the individual species occupied different areas. The differences among their respective activity-densities was too great to create a more uniform overall distribution when summed, with the two abundant Pterostichus species exerting proportionally more influence on the value of Ia. Similarly, if the location of aggregation hot-spots of individual species moved around the experimental site throughout the sampling period, the accumulated distributions of all taxa, pooled over all sample dates, might be expected to be random or regular. The accumulated catch of all species over the whole experimental period was also analysed and showed total carabids to be aggregated at this spatiotemporal scale as well (Ia = 1·5, Pa = 0·02). This gives some indication of persistent, localized, aggregations of beetles within the field site, throughout the entire sampling season.
Association between successive samples
Im values for each pair-wise comparison between successive data sets are given in Fig. 5. Im values greater than two are considered large, indicating a high degree of association (J. Perry, personal communication). Figure 5 shows that most comparisons between successive distributions resulted in positive values of Im and that many of these were well in excess of two. Thus there is an exceptional level of association present in the data. Another interesting observation is the apparent high correlation between activity-density and Im. This is difficult to explain because density is explicitly removed from the measurement of association and there should be no correlation. However, when activity-density was low, for example 6 June, 20 June and 26 July (Fig. 5b), there were large numbers of zero counts in at least one of the data sets. Under these circumstances, the power of the test to discriminate between two data sets is reduced and the test returns correspondingly low values for Im. The large number of exceptionally high Im values may result from the short time intervals between samples (48 h). This limits the potential for the populations to disperse between successive samples and the measured association is therefore correspondingly large.
Association between time-lagged samples
The high level of association between carabid distributions in successive samples suggested that measuring association between distributions with time lags between sampling occasions should give a better indication of their long-term temporal stability. Samples from 12 June, 24 June, 12 July and 9 August were selected for testing, as these dates coincided with peaks of activity-density in P. cupreus and P. melanarius (Fig. 5a,b), thus reducing any potential zero count problems. The 29 September sample was also selected, because these data were collected 16 weeks after the first sample and 6 weeks after a break from data collection, during which time the field was harvested. All possible pairwise comparisons between these samples were tested for all species where data were available, but Amara spp. were absent in September and H. rufipes and Agonum dorsale were absent from the late summer and September samples. Associations between the accumulated weekly autumn distributions of N. brevicollis were also analysed and Im values are given in Table 2.
Table 2. Indices of spatial association (Im) from SADIE analyses with time lags between different carabid distributions. The data for N. brevicollis (autumn only) are accumulated over 4 days from the date shown. Figures in bold indicate Pm < 0·025
P. melanarius (males)
P. melanarius (females)
P. melanarius (total)
N. brevicollis (autumn)
Pterostichus cupreus showed strong positive associations between the 12 June distribution and those of 24 June and 12 July. By 9 August there was only weak positive association or dissociation with earlier samples. This dissociation strengthened by September, indicating an inversion of the spatial distribution when compared with all samples taken before August. This ties in with the observation of teneral P. cupreus with soft elytra emerging in August from the west field. Teneral P. cupreus are known to emerge at this time (Table 1) (Lindroth 1992). However, it is interesting to note that this generation emerged in a different area from where adults of the previous generation had been most active. In contrast, P. melanarius distributions showed strong positive associations between all samples, indicating highly stable distribution patterns. However, separate analyses of male and female distributions showed some differences. Distributions of males between 12 June and 24 June were strongly associated, but this positive association gradually diminished through July and August before strengthening again between early June and September and between July, August and September. Thus there was an initially stable distribution that shifted during late June and July, before returning to the original distribution. Pterostichus melanarius females showed only a very weak association between distributions of 12 June and 24 June, stronger positive association between June and July distributions, but dissociation between early June and both August and September distributions. All other associations were strongly positive. In other words, there appeared to be an early dispersal phase followed by a long-term stable distribution.
Amara spp. and Agonum dorsale showed very strong positive associations between 12 June and 12 July. For Amara spp., Im values declined or became negative by August. Agonum dorsale activity-density declined to zero after mid-July, so tests over long time intervals were not possible. Harpalus rufipes appeared to show dissociation between distributions over shorter time lags and stronger association over longer time lags. Nebria brevicollis , which aestivated during the summer, showed strong dissociation between its summer distribution in the hedgerow and its early autumn distribution in the field (Im = −1·84; Pm = 0·958). During the autumn, N. brevicollis distributions between all samples were highly associated, especially after the second week.
Associations between the distributions of male and female P. melanarius were also tested on the sample dates used above. Early in the season (12 June) there was no significant association (Im = 0·11; Pm = 0·44). By 24 June there was a weak association (Im = 1·17; Pm = 0·21). Later in the season, the association between the distributions of males and females had become strong and highly significant, Pm < 0·0025 (12 July: Im = 7·32; 9 August: Im = 6·58; 29 September: Im = 6·84). Thus, as the season progressed, the distributions of males and females coincided more closely.
The contour plots (Fig. 1) show a clear dissociation between the distributions of P. melanarius and P. cupreus. SADIE analysis on these data was hardly necessary, other than to demonstrate a strong negative value for the association index (Im = −2·94; Pm > 0·99).
Analysis of the environmental variable, crop density (Fig. 7), was performed omitting the line of traps where no crop was sown between the hedgerow and the field margin. Crop density was aggregated (Ia = 1·63; Pa < 0·005) and strongly associated with P. cupreus (Im = 4·73; Pm < 0·0025) and P. melanarius (Im = 2·31; Pm = 0·05) total accumulated distributions. This appears a counter-intuitive result, as two dissociated distributions were also both positively associated with a third distribution common to both.
A classic study by Greenslade, reviewed by Southwood (1978b), which included most of the species investigated in the present work, demonstrated species packing among carabids in terms of their seasonal and circadian activity periods. Our data also provide evidence of species-specific patterns of seasonal activity-density. Furthermore, we provide evidence that suggests species packing in space, with different species occupying different parts of the experimental site. The Amara spp. were confined to a few small aggregations within the field boundary, with little evidence of movement into the cropped areas of the field. Harpalus rufipes was also distributed in field boundary aggregations but present at low densities within the cropped areas of the field. Agonum dorsale was found in small aggregations both in the field and the field boundary. The field aggregation appeared to be associated with a weedy area at the base of a telegraph pole. Nebria brevicollis moved into the field boundary as an aestivation site in the early summer and recolonized the fields in the autumn. Thus, the field boundary habitats were important for some species, either as the only habitat in which they occurred (Amara spp.), as the major focus of their more widespread distribution (H. rufipes), or as a seasonal refuge (N. brevicollis). There was no evidence of any relation between species distributions and the field margin plots to the east (but see Thomas & Marshall 1999 for a more extensive analysis). However, a dense aggregation of Amara spp. persisted in a permanent grassy patch situated at the north end of the field margin plots.
The hedgerow also appeared to be important as a feature imposing spatial structure. Considering the similar nature of the two fields either side of the hedgerow, in terms of soil type and cropping history, it is difficult to explain why the two Pterostichus species dominated in different fields for most of the time. This distribution suggests that hedgerows might act as a barrier to movement, dividing potentially continuous populations into local populations at the field scale. Populations at the farm-scale may then be defined as metapopulations and modelled in this context (Sherratt & Jepson 1993). Mark–recapture studies on the same site showed P. melanarius (Thomas, Parkinson & Marshall 1998) and N. brevicollis (Fernández García, Griffiths & Thomas 2000) moving between fields to a limited extent, while P. cupreus showed no movement between fields (C.F.G. Thomas, unpublished data). Other workers have shown carabids moving through barriered sections of hedgerows (Mauremooto et al. 1995), but rates of movement through the hedgerow were generally much lower than over similar distances in the open field.
Fournier & Loreau (1999) have also demonstrated different spatial distributions of P. melanarius and P. cupreus. However, rather than dominating different fields, their study showed these two species’ distributions to be related to the proximity of a hedgerow. Pterostichus cupreus dominated their catches close to a recently planted hedge, with fewer found at increasing distances up to 110 m into the cropped area of their study field. The inverse was the case with P. melanarius; these differences were attributed to microhabitat preferences. Our study found no such pattern, although sampling only extended approximately 30 m into the crop. However, the distributions of the two Pterostichus species and N. brevicollis (Figs 1 and 3) suggested that the edges of larger spatial distributions extending into the field were being observed. The size and scale of the trapping grid is an important factor to consider in interpreting distributions. Recent whole-field studies conducted over pairs of adjacent fields (grid size > 8 ha) have shown that P. melanarius and P. cupreus occur in large patches both in the centres and at the edges of fields (N.J. Brown and C.F.G. Thomas, unpublished data). Factors that more or less suit different species may be distributed in patterns unrelated to the location of field boundaries. Edaphic and physical factors (such as soil type and texture, topography, drainage pattern and aspect), together with anthropogenic factors (such as habitat distribution, crop history and management) and ecological factors (Table 1), are likely to interact in complex ways to drive the observed patterns of carabid distributions in farmland. Future work therefore should include measurements of a wider range of parameters than is currently undertaken. SADIE could then be used to determine which physical, ecological and anthropogenic parameters are associated with carabid distributions, and the reasons why different studies sometimes give rise to inconsistent observations may be resolved.
Very little of the ecological information given in Table 1 was, on its own, easily related to the observed carabid distributions. Factors such as the periods of emergence, activity and breeding are likely to expose different species to some crop management practices, but these effects are beyond the scope of this paper. The predatory beetles were found mainly in the field and, as expected, those with mixed or mainly plant diets were largely confined to the florally diverse hedgerow. The only other observation of interest was the broad correlation between size of beetle, the patch size in their distributions, and the magnitude of the aggregation index. The larger beetles tended to exhibit distributions with larger patch sizes. The average value of Ia given for each taxon in Figs 1–3, in ascending order of magnitude, were Agonum dorsale 1·22, H. rufipes 1·26, Amara spp. 1·33, N. brevicollis (autumn) 1·61, P. melanarius 2·06 and P. cupreus 2·09. With the exception of H. rufipes, this sequence closely follows the increasing size of the different taxa.
A surprising finding was the spatial stability of the patch structure of all species through time, as evidenced by the large number of positive associations by SADIE analysis between successive and time-lagged distributions. Although this spatial stability may be expected among the seed-feeding species, through their association with resources in the hedgerow habitats, P. melanarius and P. cupreus also had highly stable field distributions for significant periods. Both these latter species are highly mobile (Wallin & Ekbom 1988, 1994; Thomas, Parkinson & Marshall 1998). It might be expected, therefore, that any observed patchiness in beetle distributions would be ephemeral, but our evidence contradicts this assumption. Even if the location of high-density patches shifted only slowly across the site throughout the course of the experiment, the distributions of the accumulated totals would be expected to be random or regular. However, the accumulated totals also showed significant aggregation. Clearly, there must be strong attraction to certain areas, or repulsion from others, for the patch structure to persist in the way we have observed.
The apparent periodic increase and decrease in aggregation indices for P. melanarius and P. cupreus (Fig. 6a,b) suggested a sequential contraction and expansion of distributions from aggregation foci. During the study period, rainfall occurred in four discrete episodes (Thomas, Parkinson & Marshall 1998) that appeared to be related to the peaks of activity-density observed in P. melanarius and, to a lesser extent, other species. Between episodes of rain, gradual drying of the soil into contracting foci of moist habitat may drive this observed pattern. Manipulation of soil moisture in fields by sheltering areas from rain has been shown to have an effect on some groups of invertebrates, although not beneficial taxa (Frampton, van den Brink & Gould 2000). Alternatively, such a pattern could be explained by the periodic emergence of cohorts of teneral beetles from favoured oviposition sites. It should be emphasized that this is conjecture and further detailed studies are required at the appropriate scale, including quantification of the dynamics of putative driving factors, to confirm or refute these speculations.
Slight differences in the spatial stability of male and female P. melanarius distributions suggest behavioural differences that could be related to mate finding and oviposition behaviour, but this would require further work to elucidate. A lack of association between male and female distributions early in the season may reflect behavioural differences or it may be an artefact of the small sample size at that time. Similarly, the P. cupreus distribution, although stable for most of the time, shifted markedly at the end of the summer when the following generation emerged. A possible explanation might be differential larval survival in the two fields but, again, further detailed study would be required to determine the causal factors.
It was not the purpose of our study to identify causal factors of aggregation. However, P. cupreus has been shown to respond to high prey (cereal aphid) density with a systematic search pattern, which may increase their residence time in areas of high prey density (Wallin & Ekbom 1994). There is also some evidence to suggest that P. melanarius aggregates in areas of high prey density (Bohan et al. 1998). And there is evidence for carabid aggregation in response to soil moisture (Hengeveld 1979) and vegetation (crop) density (Honek 1988). In our study, crop density was aggregated and associated with the distributions of P. cupreus and P. melanarius, but this does not rule out the possibility of similar responses of crop and beetle to a common abiotic factor. Bohan et al. (2000) found no association between distributions of either a polyphagous predatory beetle or slugs (a known prey item) with soil moisture content or dry grain weight. It was concluded that because no associations were found between these factors, changes in slug and beetle distributions were best explained by predation. However, many interacting factors are likely to be involved in determining species’ distributions and a degree of circumspection is necessary when interpreting data. Associations or dissociations between species’ distributions and putative biotic and abiotic driving factors should have biological meaning and relevance before conclusions are drawn concerning causation or higher order trophic interactions are invoked as the only remaining explanation.
Other possible causes of aggregation might include competitive exclusion, affinity to oviposition sites or limited dispersal from areas with low larval mortality. Whatever the underlying causal factors might be, the largely non-overlapping distributions of the different species indicate that the factors are species-specific and require further elucidation.
We have demonstrated how insect aggregations can persist at the same location for considerable periods, even for a highly mobile species. Future work should test for stable and persistent patch structure between years at the whole-field scale and should measure a range of other putative biotic and abiotic parameters, in order to develop hypotheses concerning the causal factors underlying these distributions. If the locations of carabid population distributions are predictable and stable in the long term, the potential for reducing agrochemical input to farmland will be increased, through the wider use of precision application methods that avoid non-target effects of insecticides and the enhancement of biological pest control. If the most important factors governing these distributions are persistent, large-scale properties of fields, for example soil type and moisture retention, may be identifiable by remote sensing. Such information could then be easily incorporated into precision farming systems. Similarly, identifying the factors governing the distribution of carabids in relation to specific qualities of different field boundary types will enable the size and survival of beneficial carabid populations to be enhanced through suitable management of these important non-crop habitats.
SADIE does seem to be sensitive to small sample size. Accumulating successive data sets could partially solve the problem, but there may still be times when visual analysis of contour plots is the only solution. Nevertheless, SADIE is a powerful tool for analysing population spatial distributions. Since the data reported in the present paper were collected and analysed, SADIE has been further developed (Perry et al. 1999) to include ‘red-blue plots’ to identify the specific locations of clusters and gaps. This has been used to examine possible associations between distributions of predators, prey and vegetation cover (Holland, Perry & Winder 1999). We anticipate further developments of SADIE and a rapid and widespread adoption of these sampling and analytical techniques to address current concerns in many areas of applied ecology.
This work was partly funded under the EC AIR3 programme, grant no. AIR3-CT 920476/920477, ‘Field boundary habitats for wildlife, crop and environmental protection’. IACR – Long Ashton receives grant-aided support from the Biotechnology and Biological Sciences Research Council of the United Kingdom. Thanks are due to Joe Perry for help and advice with the SADIE analyses, and to two anonymous referees for their constructive comments on an earlier draft.
Received 2 December 1999; revision received 6 June 2000