Factors affecting the species composition of arable field boundary vegetation

Authors


* Present address and correspondence: David Kleijn, Département de Biologie/Ecologie, Université de Fribourg/Pérolles, CH-1700 Fribourg, Switzerland (fax + 41 26 3009740; e-mail david.kleijn@unifr.ch).

Summary

1.  In recent decades the botanical diversity of arable field boundaries has declined drastically. To determine the most important factors related to the species composition of arable field boundaries, the vegetation composition of 105 herbaceous boundaries, 1-m wide, in the central and eastern Netherlands was surveyed. Biomass samples of the boundary were taken at 0–33, 34–66 and 67–100 cm from the adjacent arable field.

2.  Farmers were interviewed with respect to boundary management and land use on the adjacent arable field. The two data sets containing the botanical data and the environmental variables were linked and analysed by multivariate analysis and analysis of variance.

3.  The nitrogen and phosphorus inputs to the crop were closely correlated to another environmental variable, crop rotation. Crop rotations dominated by maize cultivation received significantly higher nutrient inputs than the other rotations.

4.  Nitrogen, phosphorus and crop rotation were strongly correlated with the composition of the boundary vegetation. Species richness of the boundary vegetation was negatively related to nitrogen and phosphorus inputs to the crop, while total boundary biomass was negatively related to nitrogen inputs only. Furthermore, there was a significant relationship between the partitioning of the boundary vegetation between different functional groups and both nitrogen inputs and crop rotation on the adjacent arable field. No significant relationships were found between the vegetation composition and either herbicide use in the boundary or boundary management. This may have been due to the descriptive approach of the study: little variation in boundary management as well as herbicide use in the boundary was observed in the study area.

5.  The boundary vegetation was characterized by a peak in biomass production in the zone near the arable field. In this zone the perennial arable weed Elymus repens, as well as annual dicot species, were significantly more abundant, while perennial dicots and monocots were significantly less abundant compared with the zones further from the arable field.

6.  Efforts to protect field boundary vegetation need to focus on a reduction or cessation of fertilizer applications in the outer metre(s) of crops that require high nutrient inputs. The determination of the exact nature of the relationship between the vegetation composition of arable field boundaries and the intuitively important management and herbicide use in the boundary requires an experimental approach.

Introduction

In recent years scientific interest in the functional ecology of arable field boundaries has increased dramatically (van Emden 1990; Wiens 1992; Bennet, Henein & Merriam 1994; Boatman 1994; Halley, Thomas & Jepson 1996). Field boundary vegetation may benefit crop growth by serving as a windbreak (Forman & Baudry 1984), reducing soil erosion (Tim & Jolly 1994) and enhancing the abundance of both crop pollinators (Banaszak 1983, 1992) and natural enemies of crop pests (Sotherton 1984; Coombes & Sotherton 1986). However, its most pronounced impact lies in sustaining farmland wildlife. In the modern agricultural landscape, arable fields are virtual wildlife deserts and most animal and non-crop plant species depend on field boundaries for food, shelter, overwintering sites or simply a place to grow (Potts 1986; Kaule & Krebs 1989; Bunce et al. 1994; Tew, Todd & Macdonald 1994).

Maintaining a high level of botanical diversity in field boundaries is essential to many farmland animals. For instance, the abundance and diversity of arthropods, exemplified by butterflies, is higher in and along botanically rich field boundaries (Lagerlöf & Wallin 1993; Sparks & Parish 1995; Feber, Smith & Macdonald 1996). High plant species richness may even benefit crop production because parasitism of crop pests may be enhanced by the presence of nectar-producing plants outside the crop, which provide food for the nectar-feeding adults of many parasitoid species (Powell 1986; Hickman & Wratten 1996). Besides high species richness, the presence of a range of different vegetation types may be crucial to insects. Steffny, Kratochwil & Wolf (1984) concluded, for both butterflies and bumblebees, that the presence of a single species-rich vegetation type was not sufficient to meet their demands with respect to temperature requirements, food supply and resting and rendezvous sites. A mosaic of patches of different vegetation types was needed to meet all habitat demands of these insect groups.

Although the total area occupied by field boundaries may be substantial at a regional level, these landscape structures generally have a limited width. As a result of the high edge–area ratio, field boundaries may be more susceptible to disturbances than comparable vegetation that is concentrated in a single field. Disturbances may be caused by activities on the adjoining arable fields, such as close ploughing, misplacement of fertilizer or drift of herbicides, all of which may seriously suppress species richness in this habitat (Marshall 1987; Freemark & Boutin 1995; Kleijn & Snoeijing 1997). Furthermore, field boundaries are usually maintained by the farmers, and one of their primary maintenance objectives is weed control. As a result, herbicide use in the hedge bottom or ditch bank is rather common (Marshall & Smith 1987; Boatman 1992; de Snoo & Wegener Sleeswijk 1993), with serious consequences for the floristic diversity of boundaries. Other management activities, such as trimming hedges or different mowing regimes, may affect the vegetation composition as well. The general decline in botanical diversity of field boundaries in recent decades (Bunce et al. 1994) may be the result of some or all of these factors.

This study presents the results of a botanical survey in 105 arable field boundaries. The farmers cultivating the fields bordering these boundaries were interviewed with respect to their agricultural activities and boundary management. The two data sets were linked and analysed to determine what relationships exist between the boundary management, the land use on the adjacent arable field, and the composition of the boundary vegetation.

Methods

In June and July 1995, 105 field boundaries were surveyed on the sandy soils of the central and eastern Netherlands. To facilitate comparisons between sites, all of the boundaries selected had a relatively homogeneous grass-dominated vegetation that was not shaded by trees or shrubs. Relevés were made, following the ordinal scale of van der Maarel (1979), in 4-m long quadrats at 0–33, 34–66 and 67–100 cm from the crop–boundary transition. In the Netherlands, crops are generally planted right up to the permanent vegetation of the boundary, so the 0–33 cm relevé bordered the crop directly.

Biomass production in each of the 4-m quadrats was estimated by cutting all above-ground vegetation from 0·5 × 0·33-m subquadrats. The samples were separated into four functional groups: (i) annual dicotyledonous species (dicots), which were mostly arable weeds; (ii) Elymus repens (nomenclature following van der Meijden 1990), which was the most important perennial weed species associated with the boundary vegetation; (iii) perennial dicots, which were important for the aesthetic appearance of the boundary and for insect abundance and diversity because they flower abundantly (Weiss & Stettmer 1991; Frei & Manhart 1992); and (iv) monocotyledonous species (monocots) other than Elymus repens. Annual and perennial monocots were combined because of the very small contribution of the annual monocots (nine species with very low abundance). Dry weight of the biomass samples was determined after drying for 48 h at 80 °C.

From winter 1995 to summer 1996 the owners/tenants (hereafter farmers) of the arable fields adjacent to the surveyed field boundaries were identified and interviewed. The questions addressed (i) the preferred boundary type; (ii) the approximate age of the boundary; (iii) the type of boundary management; (iv) the use of herbicides in the boundary; (v) the crop rotation on the adjacent field; (vi) the mineral and organic fertilizer inputs per crop; (vii) the type of fertilizer spreader; (viii) the preventative measures taken to reduce fertilizer misplacement; (ix) the herbicide inputs per crop; and (x) the arable weed species (one or more) considered to be most problematic on their fields. The average nitrogen (N) and phosphorus (P) content per type of organic fertilizer was derived from Anonymous (1993). Subsequently, the N and P inputs per crop for each site were calculated by adding the inputs from the mineral fertilizer and the organic fertilizer; the field average was then determined by taking the 5-year mean of the crop rotation.

Analysis

Botanical data were collected from all 105 boundaries. However, the boundaries of nine sites were mown after taking relevés but before biomass samples had been taken. Furthermore, as the farmers of 30 of the 105 adjoining fields could not be found or declined to co-operate, a complete set of vegetation data and environmental variables was obtained for only 69 sites (for three sites no biomass and no interview data were obtained). Relationships between vegetation composition and environmental variables were therefore analysed for these 69 boundaries only.

The data were analysed by multivariate analysis and by analysis of variance. For the multivariate analysis the three relevés in each site were analysed as one 4 × 1-m relevé, to avoid pseudoreplication. The cover scores of each species were therefore averaged over the three relevés. Correlation between the relevés and the set of environmental variables was determined by canonical correspondence analysis (CCA) using canoco (Jongman, ter Braak & van Tongeren 1987) and depicted in ordination diagrams. Environmental variables used in the analysis were N inputs, P inputs (ordinal variables), type of boundary management, type of crop rotation and herbicide use in the boundary (nominal variables).

The average biomass variables (mean of three positions) and total species number per boundary site were used to determine correlations between the above-mentioned environmental and vegetation variables. Where explanatory variables were qualitative, one-way anova with unequal replications was used to analyse their effects (GENSTAT 5 Committee of the Statistics Department 1993). In case of rejection of the null hypothesis (no significant effect of the explanatory variable), differences between different levels of these variables were tested for significance using least significant difference (LSD) tests. The quantitative variables N inputs and P inputs were analysed by means of linear regression analysis followed by t-tests.

The effect of position in the boundary (distance from the field) on species number was analysed by means of anova using data from all 105 sites, while its effect on the biomass production of different functional groups was analysed for the 96 sites for which biomass data were obtained. In these analyses, sites were used as replicates.

Prior to all analyses, residuals were plotted vs. fitted values to test for constancy of variance of the errors. If variance increased with increasing values of species numbers or biomass production, ln-transformed data were used in the analyses. All percentage data were angular transformed prior to analysis.

Results

Field boundary and crop cultivation characteristics

The surveyed field boundaries were managed in three qualitatively different ways. Cutting and removing the boundary vegetation was rather uncommon in the study area: less than 10% of the farmers removed the cuttings after mowing. Cutting the boundary vegetation without removing the mown vegetation was by far the most popular management type, while one-third of the boundaries was not cut at all (Table 1). Approximately half of the field boundaries were not managed by the farmers but by municipalities or other governmental institutions, due to the fact that many of the surveyed boundaries bordered public roads or major watercourses. Herbicide use in the field boundary itself was not very common (Table 1), and most of the farmers that used herbicides in the boundary used them spot-wise against species such as Cirsium arvense, Rubus fruticosus and Urtica dioica.

Table 1.  Characteristics of 75 field boundaries and their adjoining arable fields, based upon a questionnaire survey carried out in 1995–96. Characters in bold are abbreviations used in Tables 2 and 3 and Figs 1 and 2
  • *

    Rijkswaterstaat and Waterschappen.

  • including the seven pneumatic spreaders.

  • herbicide frequency data are based upon the crops of 1995 only.

Type of boundary managementCutting + removing (cr)Cutting  – removing (cn)No management (nm)     22 
54525    22 
Management performed byFarmerMunicipalityGovernmental services*   22 
37308   22 
Herbicide use in boundary?No (nh)Yes (h)     22 
678       
Crop rotation of the arable fieldMostly/only maize (ma)Alternating maize/grass (mg)Potato, cereals, sugar beet, maize (po)Other (ot)      
3712215    
Type of fertilizer spreaderNo spreaderSingle discTwin discOscillating spoutPneumatic spreader     
30137187     
Preventative measures takenNoneReduced speedHeadland deflectorBorder discTilting spreaderOther    
2346633    
Fertilizer inputs on field (organic + mineral)MinimumMeanMaximum       
Nitrogen (kg ha−1 year−1)36222397      
Phosphate (kg ha−1 year−1)13  94178      
Frequency of herbicide application on field
8

64

2

1
      
Ten worst crop weeds indicated by farmers1. Chenopodium album (35)2. Echinochloa crus-galli (31)3. Elymus repens (25)4. Solanum nigrum (24)5. Stellaria media (10)6. Cirsium arvense (8)7. Capsella bursa-pastoris (7)8. Polygonum convolvulus (6)Polygonum persicaria (6)Viola arvensis (6)

Land use on the arable field could be categorized into four groups, of which the (more or less) continuous growing of silage maize and the rotation of cereals, potato, silage maize and sugar beet (not necessarily in that order) were the most common ones (Table 1). The ‘other’ category included continuous low-input cereal production as well as a fallow–crop rotation in which fallow dominated. As most maize fields were only fertilized organically, 30 farmers did not have fertilizer spreaders. Single and twin disc spreaders and oscillating spout spreaders were common, while the fairly high numbers of the very accurate (and expensive) pneumatic fertilizer spreaders may be the result of the inclusion of three fields of agricultural research stations near Wageningen. Half of the farmers that possessed fertilizer spreaders did not take any preventative measures to reduce fertilizer misplacement outside the field, while the other half took a variety of measures (Table 1), including fertilizing the edge by hand. Average N and P inputs were, respectively, 222 and 94 kg ha−1 year−1. Herbicides were applied predominantly once a year; the only farmer that applied herbicides three times a year did so as part of a low-dosage weed control strategy in sugar beets. Farmers considered four weed species to be by far the most troublesome: Chenopodium album, Echinochloa crus-galli, Elymus repens and Solanum nigrum, respectively.

To determine correlations between the environmental variables, the nutrient inputs for each type of management, crop rotation and herbicide use were analysed (Table 2). While there was no obvious relationship between the type of boundary management or herbicide use in the boundary and nutrient inputs, the nutrient input levels differed significantly between different types of crop rotation. Crop rotations that were dominated by maize cultivation had significantly higher N and P inputs than the other two crop rotations (Table 2).

Table 2.  The mean (+ SE) nutrient levels (kg ha−1 year−1) applied to crops in fields with different crop rotations and adjacent to boundaries with different types of management or herbicide use. Data were analysed by anova, followed by LSD tests (n = 69). Mean nutrient levels with different superscripts are significantly different at P < 0·05. Nutrient levels without superscripts are not significantly different
ManagementRotationHerbicide
cr (5)cn (40)nm (24)ma (34)mg (12)po (18)ot (5)No (61)Yes (8)
Nitrogen170222232259a271a163b61c221230
(32·4)(14·4)(20·0)(14·0)(21·0)(10·9)(8·72)(11·4)(44·2)
Phosphorus8810190115a115a69b22c9697
(18·2)(6·9)(9·7)(6·7)(10·7)(6·4)(2·9)(5·6)(19·7)

Effects of land use and management on the boundary vegetation

For the species ordination, 2·9% of the variation in the data was explained by axis 1, which had a 0·824 correlation between the species and land-use variables. This accounted for (only) 22·0% of the explained species–environment relationships. Axis 2 explained a further 2·4% of the species variation and had a species–environment correlation of 0·857. This accounted for a further 18·1% of the species–environment relationship. From Fig. 1a,b, it can be seen that the environmental variables that were most strongly correlated with the first axis were the other (ot) crop rotation (0·50), unmanaged boundaries (nm) (0·47) and P level (−0·45). The second axis was primarily correlated with the potato–cereal–sugar beet–maize (po) (0·57) and maize only (ma) (−0·42) crop rotations. Figure 1a shows that species-poor plant communities were predominantly found in boundaries along arable fields with continuous maize cultivation that received high N loads. More species-rich communities were correlated with moderate to low nutrient inputs in boundaries that were not cut. Most of the plant species found in the field boundaries were grassland species, annual and perennial ruderals and species adapted to treading. Figure 1b shows that perennial species, such as Cerastium arvense, Hypochaeris radicata and Rumex acetosella, from nutrient-poor habitats were found in boundaries in which no herbicides were applied along fields with moderate to low nutrient levels. Surprisingly, Calluna vulgaris, a species of extremely poor soils, was most strongly correlated with high N inputs on the field (Fig. 1b).

Figure 1.

Figure 1.

Canonical correspondence analysis (ordination diagram of 69 field boundaries with environmental variables represented by lines. (a) Only sites are shown and for each site the species richness is plotted. (b) Only species are shown. For the abbreviations of the environmental variables see Table 1. Aar, Anthoxanthum aristatum; Ael, Arrhenatherum elatius; Apa, Atriplex patula; Asa, Avena sativa; Avu, Artemisia vulgare; Bho, Bromus hordeaceus; Bst, Bromus sterilis; Car, Cerastium arvense; Cav, Cirsium arvense; Cca, Crepis capillaris; Ccr, Carduus crispus; Cfo, Cerastium fontanum ssp. vulgare; Cse, Calystegia sepium; Cvu, Calluna vulgaris; Ear, Equisetum arvense; Ecr, Echinochloa crus-galli; Ere, Elymus repens; Ghe, Glechoma hederacea; Gte, Galeopsis tetrahit; Hmo, Holcus mollis; Hmu, Hordeum murinum; Hra, Hypochaeris radicata; Hsp, Heracleum sphondylium; Lco, Lapsana communis; Lpu, Lamium purpureum; Lse, Lactuca serriola; Lvu, Linaria vulgaris; Mar, Myosotis arvensis; Mma, Matricaria maritima; Mre, Matricaria recutita; Ope, Ornithopus perpusillus; Par, Phalaris arundinacea; Pav, Polygonum aviculare; Pdu, Papaver dubium; Ppr, Phleum pratense; Pre, Potentilla repens; Pse, Prunus serotina; Psy, Pinus sylvestris; Ptr, Poa trivialis; Qro, Quercus robur; Qru, Quercus rubra; Rar, Ranunculus acris; Rat, Rumex acetosella; Rob, Rumex obtusifolius ssp. obtusifolius; Rps, Robinia pseudo-acacia; San, Scleranthus annuus; Sas, Sonchus asper; Sav, Spergula arvensis; Sce, Secale cereale; Soe, Symphytum officinale; Sof, Sisymbrium officinale; Sol, Sonchus oleraceus; Svi, Senecio viscosus; Svr, Setaria viridis; Tpr, Trifolium pratense; Tvu, Tanacetum vulgare; Udi, Urtica dioica; Var, Veronica arvensis; Vcr, Vicia cracca; Vsn, Vicia sativa ssp. nigra.

Figure 1.

Figure 1.

Canonical correspondence analysis (ordination diagram of 69 field boundaries with environmental variables represented by lines. (a) Only sites are shown and for each site the species richness is plotted. (b) Only species are shown. For the abbreviations of the environmental variables see Table 1. Aar, Anthoxanthum aristatum; Ael, Arrhenatherum elatius; Apa, Atriplex patula; Asa, Avena sativa; Avu, Artemisia vulgare; Bho, Bromus hordeaceus; Bst, Bromus sterilis; Car, Cerastium arvense; Cav, Cirsium arvense; Cca, Crepis capillaris; Ccr, Carduus crispus; Cfo, Cerastium fontanum ssp. vulgare; Cse, Calystegia sepium; Cvu, Calluna vulgaris; Ear, Equisetum arvense; Ecr, Echinochloa crus-galli; Ere, Elymus repens; Ghe, Glechoma hederacea; Gte, Galeopsis tetrahit; Hmo, Holcus mollis; Hmu, Hordeum murinum; Hra, Hypochaeris radicata; Hsp, Heracleum sphondylium; Lco, Lapsana communis; Lpu, Lamium purpureum; Lse, Lactuca serriola; Lvu, Linaria vulgaris; Mar, Myosotis arvensis; Mma, Matricaria maritima; Mre, Matricaria recutita; Ope, Ornithopus perpusillus; Par, Phalaris arundinacea; Pav, Polygonum aviculare; Pdu, Papaver dubium; Ppr, Phleum pratense; Pre, Potentilla repens; Pse, Prunus serotina; Psy, Pinus sylvestris; Ptr, Poa trivialis; Qro, Quercus robur; Qru, Quercus rubra; Rar, Ranunculus acris; Rat, Rumex acetosella; Rob, Rumex obtusifolius ssp. obtusifolius; Rps, Robinia pseudo-acacia; San, Scleranthus annuus; Sas, Sonchus asper; Sav, Spergula arvensis; Sce, Secale cereale; Soe, Symphytum officinale; Sof, Sisymbrium officinale; Sol, Sonchus oleraceus; Svi, Senecio viscosus; Svr, Setaria viridis; Tpr, Trifolium pratense; Tvu, Tanacetum vulgare; Udi, Urtica dioica; Var, Veronica arvensis; Vcr, Vicia cracca; Vsn, Vicia sativa ssp. nigra.

Interestingly, the results of the regression analyses indicated that three variables that may have affected the composition of the boundary vegetation indirectly (N and P inputs and crop rotation) were significantly related to vegetation composition, while management and herbicide use, variables that affected the boundary vegetation directly, were not (Table 3). The amount of N applied was negatively related to species richness, total boundary biomass and the proportion of perennial dicots in the vegetation. P input was negatively related to plant species richness only (Table 3). No significant (negative) relationship was found between total vegetation biomass productivity and species richness, although the F-value approached significance (F1,67 = 3·60, P = 0·062).

Table 3.  The relationships between environmental factors and species richness, biomass of the vegetation and the partitioning of biomass between different functional groups in 69 field boundaries in the central and eastern Netherlands. The effects of nitrogen and phosphorus were analysed by linear regression analysis, followed by t-tests. –/+: negative/positive relationship. The other variables were analysed by anova, followed by LSD tests. The nature of the relationship of the qualitative variables ‘field boundary management’, ‘crop rotation’ and ‘herbicide use in the boundary’ is given in the text and in Fig. 2. For definition of qualitative variables see Table 1
Number of speciesBiomassMonocots (%)Perennial dicots (%)Annual dicots (%)Elymus repens (%)
d.f.   −   / + FP   –   / + FP   –   / + FP   −   / + FP   –   / + FP   –   / + FP
Nitrogen18·040·0065·820·019 + 1·020·3166·770·011 + 0·140·7130·060·808
Phosphorus18·740·0040·000·9941·200·2770·220·6380·540·465 + 3·680·059
Management2 0·530·588 3·020·055 0·370·693 0·790·459 0·480·619 0·960·388
Rotation3 4·670·005 0·110·956 4·200·009 1·960·128 2·600·059 3·120·032
Herbicide1 0·050·832 1·790·185 0·490·488 0·310·579 0·510·479 0·420·518

The relationship between management and boundary productivity also approached significance (Table 3). In this case, boundaries from which the cuttings had been removed tended to have a higher biomass (735 gm−2) compared with boundaries that were cut but from which cuttings were not removed (532 gm−2) and boundaries that were not cut at all (476 gm−2).

Species numbers were distinctly different in boundaries along fields with different crop rotations (Table 3). Boundaries next to fields with the continuous maize production (ma) rotation or alternating maize and grass (mg) rotation, were significantly species poorer (9·0 and 11·0 species 4 m−2, respectively) than those next to the ot rotation fields (15·0). Boundaries next to po rotation fields had intermediate numbers of species (12·4). Field boundary productivity was not significantly different between fields with different crop rotations but the partitioning among the functional groups was (Fig. 2). Field boundary vegetation next to fields with mg rotation contained significantly more Elymus repens compared with boundary vegetation next to ma and ot crop rotations, with intermediate levels along fields with the po rotation. Furthermore, field boundaries next to fields with ma rotation had a significantly higher percentage of monocots compared with boundaries next to fields with mg rotations. Boundaries next to fields with po and ot rotations had intermediate percentages of monocots. The percentage of annual and perennial dicot biomass in the boundary vegetation did not differ significantly between fields with different crop rotations

Figure 2.

Mean biomass production (g dry weight m−2) and partitioning among different functional groups in the first metre of the field boundary vegetation next to arable fields with different crop rotations. Diagonally hatched bars, monocots; cross hatched bars, perennial dicots; filled bars, annual dicots; transparent bars, Elymus repens. Different characters indicate significant differences between bars with the same patterns only. Absence of characters indicates absence of significant rotation effects.

The vegetation composition at different positions in the field boundary showed some characteristic and statistically significant differences, although the species richness of the 105 arable field boundaries did not change significantly with increasing distance from the arable field. However, there was a significant effect of position on the number of perennial species (F2,208 = 7·42, P < 0·001), the number of perennial species being significantly lower in the first 33 cm (5·5 species 1·33 m−2) compared with the positions further from the arable field (6·1 at 34–66 m and 6·0 at 67–100 cm). Annual species showed the opposite pattern (F2,208 = 12·14, P < 0·001), 2·4 species 1·33 m−2 in the first 33 cm of the boundary vs. 1·8 and 1·6 at 34–66 and 67–100 cm, respectively. Biomass production in the first 33 cm was significantly higher than in the sample quadrats further from the arable field (Fig. 3).The two functional groups containing arable weeds, the annual dicots and Elymus repens, occupied significantly decreasing proportions of the boundary vegetation biomass with increasing distance from the field. In contrast, monocots other than Elymus repens and perennial dicots increased their relative proportions in the boundary vegetation (Fig. 3).

Figure 3.

Biomass production (g dry weight m−2) of the boundary vegetation with distance from the arable field and the relative contribution (% of biomass production) of monocots (diamonds), perennial dicots (circles), annual dicots (squares) and Elymus repens (triangles). Different letters indicate significant differences within lines with the same symbols only. For clarity, characters indicating significant differences in biomass production have been omitted, however, 0–33 cm differs significantly (P < 0·01) from 34–66 and 67–100 cm.

Discussion

Farmers mentioned a variety of preferred field boundary types, varying from bare soil to ‘as colourful as possible’. However, the 85% of the responding farmers who mentioned weeds in relation to the preferred boundary vegetation pointed out that farmers consider weed abundance to be a very important aspect of arable field boundaries. Most of the weed species considered problematic by farmers, such as Chenopodium album or Solanum nigrum, are annual dicots and this group was only marginally present in the surveyed field boundaries (species from this group contributed only 1·7% to the total boundary biomass). In contrast, the perennial grass species Elymus repens contributed on average some 20% to the biomass of the boundary vegetation, and may be considered the only weed species that can infest crops from the boundary in the study area. In this respect it is interesting that only 12% of the farmers had used herbicides in their boundaries. Studies in other areas revealed that more than half of the farmers applied herbicides to the boundary vegetation (Marshall & Smith 1987; Boatman 1992; de Snoo & Wegener Sleeswijk 1993). The lack of any relationship between herbicide use in the boundary and vegetation composition may therefore be due to the low number of boundaries that were subjected to (spot-wise) treatment with these chemicals.

The CCA demonstrated that the vegetation in arable field boundaries is highly disturbed; individual field boundaries may consist of species from a wide range of ecosystems. For example, Calluna vulgaris was highly correlated with arable fields with continuous maize cultivation and high N inputs. Other species with similar distributions to Calluna vulgaris were Pinus sylvestris, Lamium purpureum, Artemisia vulgaris and Sisymbrium officinale. While Calluna vulgaris and Pinus sylvestris are species of undisturbed, nutrient poor soils, Lamium purpureum, Artemisia vulgaris and Sisymbrium officinale are characteristic of disturbed and nutrient rich soils. This suggests that the vegetation in field boundaries consists of a set of species that are remnants of the (pre-agricultural) natural vegetation, and a set of very common species that are part of the present agricultural system. Kleijn et al. (1998) found that field boundaries in different areas and even countries are dominated by a small number of very common species that are well adapted to the conditions in the arable landscape. The omnipresence of these species results in a reduced diversity in boundary vegetation types. Efforts to conserve or restore species-rich and diverse field boundaries should therefore focus on promoting the set of remnant species at the expense of the set of ‘agricultural’ species.

Both the CCA and the regression analyses revealed that the activities of farmers on their fields are related to the botanical composition of the vegetation next to those fields in a number of ways. Increased nutrient application levels to the crop were strongly related to reduced species richness in the boundary vegetation. Furthermore, regression analysis revealed that boundary biomass was negatively related to the N levels applied to the bordering arable field. Interpretations of these results should be made with care as nutrient inputs are strongly correlated with crop rotation (Table 2). The descriptive approach used in this study does not allow a separation of the direct effects of high nutrient levels on the boundary vegetation and indirect effects caused by different cultivation activities or even crops. For instance, higher nutrient application levels on the field generally increases boundary biomass (Kleijn 1996). The significant adverse relationship found in this study between N inputs and boundary biomass may have been caused by the fact that fields with high N levels were predominantly planted with maize. Maize is considerably taller than any other crop and the effects of shading may therefore have been considerably more severe in boundaries along maize crops, resulting in lower biomass production in these boundaries. In contrast, the relative proportion of the different functional groups responded in a similar way to that found in other studies, as perennial dicots decreased significantly with increasing nutrient levels (Berendse 1983; Kleijn & Snoeijing 1997). P was not significantly related to any of the biomass variables which may be caused by the P saturation of soils in many sandy regions of the Netherlands (Breeuwsma, Reyerink & Schouwmans 1989; Oenema & van Dijk 1994). N may therefore be limiting to crop and boundary vegetation while P is not.

The boundary vegetation along fields with different crop rotations contrasted remarkably. Along fields with continuous maize cultivation, field boundaries were least species rich and almost completely dominated by grasses, while fields with alternating maize and grass production were bordered by boundaries with a significantly higher proportion of Elymus repens. This high proportion of Elymus repens may be caused by the fact that this is probably the only species that can dominate in both grasslands and arable fields and, in contrast to clonal dicots such as Cirsium arvense, is difficult to control in both crops. The observed differences may in part have been caused by the different nutrient input levels associated with the crop rotations. Nevertheless, crop rotation itself has an important impact on boundary composition, as exemplified by the significant difference in the abundance of Elymus repens in boundaries next to ma and mg rotations, two rotations with comparable nutrient input levels.

The current study may have failed to demonstrate any effect of management on species richness, perhaps because no boundaries were included in the study that had a management regime really favourable to the establishment of species-rich plant communities. The five boundaries undergoing the most favourable management regime with respect to plant conservation, cutting and removing the vegetation, were also the boundaries with the highest biomass production and can therefore not be expected to sustain species-rich plant communities. This trend in biomass production between boundaries with different types of management again illustrates the limitations of the current descriptive approach. Generally, productivity declines as cuttings are being removed annually (Berendse et al. 1992), while in this study the highest productivity levels were found in boundaries from which cuttings had been removed. Therefore, the productivity of the boundary probably determines the type of management a farmer chooses and it is not the management that determines the productivity of the boundary. If productivity is very low, a farmer may decide to do nothing because this saves time and money. To determine the causal relationship between type of management and the botanical boundary composition, an experimental approach is more suitable.

Vegetation composition near the arable field and further from the field differed considerably. Generally, the first 33 cm next to the arable field was more productive, had a higher percentage of weedy species and a lower percentage of monocots and perennial dicots. The increased productivity near the field was probably related to the capture of arable nutrient resources by boundary plants (Kleijn 1996; Kleijn, Joenje & Kropff 1997). The abundance of arable weeds may be caused by the higher level of disturbances this zone experiences. In an experimental study, the exact position of the field–boundary transition fluctuated from year to year by 0·23 m (SE = 0·16) due to inaccuracies of cultivation activities (D. Kleijn, unpublished results). This high frequency of disturbance, in combination with the high fertility of the habitat, enhances growth of annuals and perennial ruderal species such as Elymus repens (Tilman 1987; Wilson & Tilman 1991).

Implications for field boundary management

A strong point of this study is that it is, to the knowledge of the authors, the first one that attempts to link the activities of the farmers to the vegetation composition in the field boundary. An important finding is that fields with crop rotations that have a high percentage of maize cultivation, which are also characterized by high nutrient input levels, tend to be surrounded by field boundaries that are generally species poor and monotonously dominated by grass species. Initiatives to conserve or restore the boundary vegetation along fields with crops that need high nutrient inputs need to focus on a reduction or cessation of fertilizer application in the outer metre(s) of the crop. Such measures may also have an environmental spin-off, as arable fields often border ditches or streams and reduced fertilizer application in the crop edge bordering these surface waters will reduce leaching into water resources.

A drawback of this study is that causal relationships cannot be made. An obviously important outcome of the present study was the fact that only five out of 75 farmers removed the cuttings after mowing the boundary vegetation. This may explain why species numbers were low even in the most species-rich boundaries and where biomass levels of the boundary vegetation were not extremely high. Experimental studies (Parr & Way 1988; van Schaik & van den Hengel 1994) clearly demonstrate that mowing herbaceous vegetation without removing the cuttings, or not mowing at all, results in more species-poor ruderal vegetation. Furthermore, perennial arable weed species such as Cirsium arvense and Elymus repens are generally promoted by this type of management. Mowing and removing the cuttings of field boundary vegetation at least once a year may prove to be the most effective way to stimulate the botanical diversity in field boundaries. Furthermore, from a weed control point of view, mowing and removing the cuttings may reduce weed problems because it promotes grassland species over arable weeds. Therefore farmers may be stimulated to adopt this type of mowing regime on both agronomic and wildlife arguments.

Acknowledgements

The authors are grateful to Malou Dekker and Ineke Snoeijing for their assistance in collecting and processing the data. The critical comments of Jon Marshall, Ivo Raemakers, Karlé Sykora and two anonymous referees on earlier drafts of the manuscript, and the advice of Jacques Withagen on matters of statistics, were greatly appreciated. This study was made within the framework of the EC-funded project (AIR3-CT920476) ‘Field boundary habitats for wildlife, crop and environmental protection’.

Received 21 May 1998; revision received 19 November 1999

Ancillary