The interspecific and intraspecific variation of functional traits in weeds: diversified ecological strategies within arable fields

Abstract

Arable weeds are a key component of the biodiversity of agroecosystems, but have faced a marked decline due to agricultural intensification. Recently, the crop edge has been considered as a potential refugia for many species. Indeed, weed species richness and abundance are higher in the crop edge than in the field margin and the field core. In this study we question whether weed functional diversity also varies among field elements and whether it is higher in the crop edge. We studied the interspecific and intraspecific variation of three functional traits (specific leaf area, canopy height and above-ground biomass) related to the response of weeds to competition and to agricultural practices, for seven weed species sampled in the crop edge, the field margin and the field core area in four winter-wheat fields. We show that trait values varied significantly with the species, the field element and their interaction. Within the field, all species had high specific leaf area, low canopy height and biomass, suggesting a shade-tolerance syndrome that could be a strategy in response to both competition with the crop and the disturbances induced by agricultural practices. In the crop edge, where the functional variation was the highest, two distinct functional strategies were observed, suggesting a resource partitioning under the predominance of weed–weed competition. In conclusion, the crop edge plays a key role in sustaining weed diversity, mostly because of its intermediate environmental properties that allow the coexistence of weeds with different ecological strategies.

Keywords:

• functional diversity
• specific leaf area
• canopy height
• above-ground biomass
• crop edge
• field margin
• field core
• weed ecological strategies

Introduction

Arable weeds are an important component of plant diversity in agroecosystems and play a key role in supporting biological diversity at all upper trophic levels. In particular, weeds are of primary importance as a food resource for many species of birds and insects (Robinson and Sutherland 2002; Marshall et al. 2003). Over the last decades, weed species diversity and abundance have drastically declined in several European countries due to post-war agricultural intensification and land-use change (Andreasen, Stryhn, and Streibig 1996; Sutcliffe and Kay 2000; Baessler and Klotz 2006; Storkey et al. 2012).

Weeds are not spatially limited within cultivated fields, but are present along field margins and within various semi-natural habitats embedded in the agricultural mosaics (Marshall and Arnold 1995). However, the management practices implemented by farmers to enhance crop production mostly impact weeds within arable fields and their surroundings. As a consequence, most studies focused on the spatial repartition of weeds in (i) the crop edge, which is the narrow linear area where the soil is tilled but uncultivated, and (ii) the field, which is generally subdivided into a field margin, i.e. the first few metres of the cultivated area, and a field core area (following the terminology of Cordeau et al. 2012). The biotic environment and the management intensity differ between these three elements. In the field core, the crop exerts an intense competition over many weeds because of early establishment, sowing at high density (Weiner et al. 2010) and important inputs of mineral fertilizers that confer to the crop individuals a long-term preferential access to light (Kleijn and van der Voort 1997). The intensity of weed control resulting from management practices (i.e. herbicide application, tillage and crop rotations) also varies among field elements. In general, the crop edge is characterized by a higher weed density (Marshall 1989), a higher weed cover (Kovács-Hostyánski et al. 2011) and denser seedbank (José-María and Sans 2011) than in the field, suggesting more favourable conditions for weed development in the crop edge. Moreover, the crop edge shows higher species richness and diversity than the field (Fried et al. 2009), and the field margin generally shows higher species richness than the field core (Romero, Chamorro and Sans 2008; Kovács-Hostyánski et al. 2011). Finally, decades of intense weed control in the field core have resulted in a depletion of soil seedbanks within cultivated fields (Robinson and Sutherland 2002), and crop edges may be considered as refugia for weed biodiversity (Fried et al. 2009).

Management practices affect not only the taxonomic diversity of weed communities, but also their functional diversity, i.e. the variation in the functional traits of plant species. Several functional traits determine weed response to management practices, especially the amount of fertilizer applied and the herbicide pressures (Storkey, Moss and Cussans 2010). The usefulness of trait-based approaches in weed science has been emphasized in many recent papers (Storkey 2006; Fried et al. 2009; Gunton, Petit and Gaba 2011; Fried, Kazakou and Gaba 2012; Navas 2012), but all these studies considered functional variation across large scales (national or regional). There are currently no data about the spatial variation of traits at field scale. Furthermore, previous studies have explored the weed response to cropping systems by considering mean trait values per species. However, the intraspecific variability of trait values has recently been demonstrated to be important for species coexistence and the dynamics of communities (Jung et al. 2010; Albert et al. 2011; Bolnick et al. 2011). Indeed, exploring the intraspecific variation allows us to detect whether individuals within a local population exhibit different ecological strategies.

In the present study, we address the following questions: (i) does the crop edge show a higher range of response trait values compared to the field core and the field margin? (ii) if so, are there any differences in the range of ecological strategies of weeds among field elements? and (iii) does the intraspecific functional variability of weeds differ among field elements? We considered seven co-occurring common weed species and analysed their interspecific and intraspecific trait distributions across field elements in winter wheat. In the field margin and the field core, we assume a higher intensity of competition (mostly weed–crop) and of disturbances (i.e. agricultural practices) than in the crop edge. As a consequence, due to the higher environmental filtering (Cornwell and Ackerly 2009), the range of functional trait values is expected to be smaller in the field margin and the field core. In the crop edge, we assume that the lower disturbance and the absence of the crop competitor should promote weed–weed competition. Therefore, as predicted by the limiting similarity principle (MacArthur and Levins 1967), weeds should exploit different resource niches; hence the extent of trait differences among the species should be higher. We studied seven species that co-occur at different abundances among the three field elements and investigated whether these species show distinct sets of trait values and different ranges of trait variation depending on the field element where they grow.

Materials and methods

Field sites

The winter wheat fields sampled were located in Fénay, Côte d’Or (Burgundy, France 47°13′ N, 5°03′ E), within an intensive agricultural area mainly cultivated with winter crops (32% winter wheat, 20% oilseed rape, 14% mustard and 14% winter barley) in rotation with spring barley (9%). We chose four arable fields with a similar crop sequence (winter wheat – winter barley – winter oilseed rape) and a similar crop management (similar inter-row width and density, similar timing of treatment sequence). To avoid an effect of the surrounding landscape on the weed flora in field margins, the sampled areas within the four arable fields were always located along a stone path and away from forest edges and hedges.

Species selection

We studied seven common weed species known to occur in all field elements, though with different relative abundances in Fénay (Fried et al. 2007). All these species are dicotyledonous, have a C3 photosynthetic pathway, are therophytes, hermaphroditic and have partially overlapping flowering periods. They have different plant architectures. Veronica hederifolia L. and Veronica persica Poir. are creeping species. Fallopia convolvulus (L.) Á. Löve and Galium aparine L. are primarily climbing species while Viola arvensis Murray, Geranium dissectum L. and Papaver rhoeas L. are characterized by an erect habit.

Species traits

We use a few well-targeted functional traits that are ecologically meaningful (Bernhardt-Römermann et al. 2008), and chose three traits known to vary in response to disturbance and light competition: the specific leaf area (SLA), the plant canopy height and the plant above-ground biomass. SLA is a good indicator of weed–crop competition for light at the species-scale (Brainard, Bellinder, and DiTommaso 2005; Storkey 2005; Gaba et al. in press) and is also correlated with the relative growth rate of plants (Reich, Walters and Ellsworth 1997). A higher relative growth rate is usually linked with a higher return interval of disturbances (Gaba et al. in press). The canopy height can be considered as an indicator of light competition at the species-scale (Holt 1995) but also partly related to disturbance frequency (Gaba et al. in press). Finally, several studies have quantified the effect of competition among weeds and winter wheat based on the reduction of weeds above-ground biomass (Weiner et al. 2010). The above-ground biomass is also a surrogate of the plant performance and seed production (Lutman 2002; Brainard, Bellinder, and DiTommaso 2005).

Plant sampling and traits measurement

In each field, we delimited one area in each of the three field elements considered (the crop edge, the field margin and the field core). Up to ten individuals per species per element were collected from each of the four cultivated fields sampled. In the crop edge and the field margin, samplings were carried out along a 50 m transect. In the field core, a random sampling was performed. To avoid sampling genetically related individuals, the minimum distance between samples was always 50 cm. Measurements were conducted in early June 2012 within a single week when plants had reached the end of their vegetative growth period (numerous nodes or side shoots visible) or when they were at the beginning of their flowering period. The canopy height was measured in situ as the distance (in cm) between the top of the main photosynthetic organs and the soil. Two leaves (fully formed with no signs of damage or senescence) were then sealed in plastic bags containing moist paper towels to prevent dehydration. The above-ground biomass was collected and stored in paper bags. Immediately upon returning from the field, individual leaf area was measured using a leaf area meter (LI-3100 Area Meter, LI-COR inc, Lincoln, NE, USA). For small leaves, the surface area was determined using a software image analysis: Visilog 6.7 (Noesis, France). Then, leaves and above-ground biomass were dried at 80°C for 2 days and weighed (in g). The SLA (in m² g^–1) was calculated as the ratio of the leaf surface to the leaf dry mass.

Data analyses

All data analyses were carried out using the R software version 2.15.1 (R Development Core Team, 2012). Variation in functional trait values among individuals was explored with a standardized principal component analysis. For each trait, a linear mixed model was implemented, that included the field element, the species and the “field element × species” interaction as fixed effects and a random intercept field effect. To meet the assumptions of homogeneity of variance and normality of the residuals, the SLA and the height were square-root transformed and the biomass was log₁₀-transformed. Bartlett test and Shapiro–Wilk test were performed on the standardized residuals to check the need for transformation. Mixed effects models analyses were performed using the package lme4. A likelihood-ratio test based on the χ² distribution was used to evaluate the significance of the additional effect of each variable in the model after first including all other variables (i.e. type III analysis of deviance). We applied sum to zero contrasts before fitting models. We used Tukey contrasts for multiple comparisons of means to compare significant differences between field elements.

Interspecific variation in functional traits was quantified using a measure of overlap between the two distributions of trait values obtained for each pair of species. To avoid the effect of a too-small sample size, we only considered species with a total of more than 10 individuals measured per field element. First, a measure of overlap was estimated for each trait based on non-parametric Gaussian kernel density estimation using the R functions implemented by Geange et al. (2011) but with slight modifications to follow recommendations from Sheather and Jones (1991) for the choice of bandwidth. Second, a multi-trait measure of overlap was estimated based on standardized trait values using the metric F_β (Villéger, Novack-Gottshall, and Mouillot 2011). F_β is a measure of functional dissimilarity between two convex hulls in a multidimensional space; it is maximal when two species have no common range of traits and minimal when they are identical. We therefore used 1 – F_β as a measure of functional overlap.

We used a randomization method to test whether the functional overlap between species within each field element was significantly lower than under the null hypothesis of no functional divergence among species. If individuals from two species share the same distribution of trait values, their actual taxonomic identity is irrelevant; the randomization procedure is therefore a permutation of individuals, independently of their species name. We generated 10,000 null distributions and used a one-tailed direct test with significant threshold of p < 0.05 to test for a non-random structure (see Supplementary Figure 1 for details). Multiple comparisons were made; therefore a sequential Bonferroni adjustment was applied (Holm 1979).

Intraspecific variation in functional traits within each species was quantified using a measure of overlap between the two distributions of trait values obtained for each pair of field elements. The same overlap metrics and randomization method as described above were used to test whether the functional overlap between field elements for each species was significantly lower than under the null hypothesis of no intraspecific functional divergence between field elements. Individuals were randomized independently of the field element from which they were sampled.

Results

Over the four cultivated fields, a total of 386 plants were sampled and measured, 185 of which were located in the crop edge, 108 in the field margin and 93 in the field core area. The average number of plants sampled per species was 54 ± 24.

Trait variability among field elements and species

The first axis of the principal component analysis on individual traits’ values explained more than 60% of the variation. Individuals were classified along this axis according to their height and SLA (Figure 1A). The variation in biomass was correlated with the variation in height, but was relatively independent from the variation in SLA. The ordination plan showed a clear difference between the crop edge and the field. In the crop edge, a wider range of trait values was observed, especially for the biomass (Figure 1B). In comparison, the field core and the field margin were both characterized by a similar trend of variation of SLA values (Figure 1B).

Figure 1. Standardized principal component analysis of the variation of three functional traits in seven weed species, measured at the individual level within three field elements. (A) Contribution of traits to the first and second principal components. (B) Location of plants sampled from different field elements on the ordination plan; 90% confidence ellipses around field element centroids are drawn. (C) Location of plants belonging to different species on the ordination plan. EPPO codes refer to abbreviations as presented in Table 2.

Functional traits values significantly varied with the field element, the species and the “field element × species” interaction (Table 1). Important differences in trait values were observed between the weed individuals sampled in the crop edge and those sampled in the field margin. On average, individuals sampled in the crop edge were significantly taller, with higher biomass and smaller SLA values than in the field margin (Table 1 and Figure 1).

Table 1. Linear mixed model analyses on functional trait variation for seven weed species in three field elements (crop edge, field margin and field core). Tukey’s post hoc tests investigate the pairwise differences between field elements. Square-root (SLA and canopy height) or log₁₀-transformations (above-ground biomass) were applied to meet the assumptions of homogeneity of variance (Bartlett test) and normality of the residuals (Shapiro–Wilk test). *p < 0.05, **p < 0.01, ***p < 0.001, ns, not significant.

		Specific leaf area		Canopy height		Above-ground biomass
Fixed factors (Type III)	df	χ²	p-value	χ²	p-value	χ²	p-value
field elements	2	118.058	***	54.250	***	190.350	***
species	6	68.378	***	170.002	***	68.629	***
field elements × species	12	34.493	***	97.916	***	86.184	***
Tukey contrasts		Estimate	p-value	Estimate	p-value	Estimate	p-value
field margin – crop edge		0.036081	***	– 1.4636	***	– 1.8693	***
field core – crop edge		0.039621	***	– 0.4669	0.0969^ns	– 1.2520	***
field core – field margin		0.003540	0.749^ns	0.9967	***	0.6174	**

Ecological strategies of weed species

Based on functional overlap, two groups of species could be distinguished within the crop edge, respectively composed of G. aparine, G. dissectum and P. rhoeas (hereafter “group 1”) and F. convolvulus, V. persica and V. arvensis (“group 2”). Veronica hederifolia could not be included in the analyses because the sample size was too small; however, Figure 2 suggests that it would belong to group 2. The multi-trait analysis showed a significant functional dissimilarity between these two groups, without any significant differences within a group (Table 2A). The single-trait analysis showed an overlap significantly lower than expected under random assumption (Table 2A), particularly with regard to height and biomass, which are significantly higher in group 1 (Figure 2). Although SLA provided a less clear among-group distinction, species constituting group 1 showed lower values of this trait (Figure 2). By contrast to the crop edge, only a small amount of trait divergence among species was observed in the field margin, except for V. hederifolia, a creeping species that differed from all other species by its height (Table 2B). Finally, as previously illustrated in the principal component analysis graphs (Figure 1), a similar pattern was observed in the field core and the field margin with no significant functional divergence among species. But, as several species had fewer than 10 individuals in this field element, overlap estimation and randomization testing could not be implemented for the field core.

Figure 2. Distribution of trait values ​per species and field element. The horizontal line on each beanplot is the mean value of the distribution. The graphical representation is based on a non-parametric Gaussian kernel density estimation. The canopy height and the specific leaf area were square-root transformed while the above-ground biomass was log₁₀-transformed. CE, crop edge; FM, field margin; FC, field core. EPPO codes refer to abbreviations as presented in Table

Table 2. Interspecific comparisons of trait distribution within each field element. Overlap in functional trait values between weed species sampled within (A) the crop edge and (B) the field margin. Non-parametric Gaussian kernel density estimation was used to measure functional overlap for each trait (top). 1 – F_β is a measure of functional overlap in the multi-trait space (bottom). Values in bold are significantly lower than under the randomization null model after applying sequential Bonferroni–Holm adjustment (P < 0.05). Abbreviations used to represent species refer to EPPO codes as follow: Galium aparine (GALAP), Geranium dissectum (GERDI), Papaver rhoeas (PAPRH), Fallopia convolvulus (POLCO), Veronica hederifolia (VERHE), Veronica persica (VERPE) and Viola arvensis (VIOAR).

A	Crop edge
Biomass	GERDI	PAPRH	POLCO	VERPE	VIOAR
GALAP	0.803	0.858	0.203	0.413	0.331
GERDI		0.725	0.195	0.412	0.323
PAPRH			0.195	0.404	0.321
POLCO				0.515	0.665
VERPE					0.771
SLA	GERDI	PAPRH	POLCO	VERPE	VIOAR
GALAP	0.746	0.758	0.558	0.338	0.497
GERDI		0.734	0.517	0.267	0.516
PAPRH			0.725	0.514	0.732
POLCO				0.735	0.737
VERPE					0.651
Canopy height	GERDI	PAPRH	POLCO	VERPE	VIOAR
GALAP	0.715	0.744	0.423	0.343	0.509
GERDI		0.921	0.454	0.386	0.614
PAPRH			0.467	0.403	0.623
POLCO				0.851	0.689
VERPE					0.659
1-Fβ	GERDI	PAPRH	POLCO	VERPE	VIOAR
GALAP	0.157	0.219	0.032	0.022	0.032
GERDI		0.221	0.019	0.012	0.014
PAPRH			0.019	0.058	0.033
POLCO				0.074	0.279
VERPE					0.153
B	Field margin
Biomass	GERDI	PAPRH	POLCO	VERHE	VIOAR
GALAP	0.516	0.573	0.576	0.340	0.709
GERDI		0.207	0.531	0.627	0.380
PAPRH			0.413	0.099	0.663
POLCO				0.321	0.672
VERHE					0.213
SLA	GERDI	PAPRH	POLCO	VERHE	VIOAR
GALAP	0.686	0.499	0.355	0.286	0.546
GERDI		0.671	0.581	0.455	0.815
PAPRH			0.478	0.363	0.737
POLCO				0.521	0.622
VERHE					0.505
Canopy height	GERDI	PAPRH	POLCO	VERHE	VIOAR
GALAP	0.633	0.726	0.801	0.106	0.661
GERDI		0.427	0.633	0.201	0.617
PAPRH			0.642	0.049	0.580
POLCO				0.101	0.647
VERHE					0.419
1 – Fβ	GERDI	PAPRH	POLCO	VERHE	VIOAR
GALAP	0.303	0.003	0.031	0	0.026
GERDI		0.002	0.024	0	0.015
PAPRH			0.026	0	0.073
POLCO				0	0.210
VERHE					0

2. Groups 1 and 2 refer to species grouping based on functional overlap (see the main text for details).

Intraspecific trait variability

Different patterns of intraspecific variation were observed among the seven species. Intraspecific variation of F. convolvulus, V. hederifolia, V. persica and V. arvensis among field elements was mainly supported by a single functional trait, i.e. SLA, while for G. aparine, G. dissectum and P. rhoeas, intraspecific variation was important for both SLA and biomass (Figure 1C). At the intraspecific level, the multi-trait functional overlap measured by 1 – F_β was always close to 0, indicating a significant difference in the multidimensional functional space between individuals sampled in adjacent field elements of the same field (Table 3). Although not detected on all pairs of field elements and for all species, the importance of intraspecific variability was also broadly highlighted by the single-trait approach. The intraspecific divergence was most frequently significant for the biomass and height between the crop edge and the field margin. Within all species, plants growing in the crop edge had a higher biomass than in the field margin. Except for V. persica (a creeping species), plants growing in the crop edge were also significantly taller than their conspecifics from the field. The SLA of individuals growing in the crop edge was significantly lower than in the field margin, except for two species, G. aparine and G. dissectum. These two species are however characterized by lower SLA values in crop edge than in field margin (see Supplementary Figure 2 for details). Less consistent patterns were observed when comparing the field core with the other two elements. SLA was however generally higher for individuals sampled in the field core. Among the species studied, some are known to be more abundant in the crop edge (G. aparine, F. convolvulus and V. persica), others are known to be more abundant in the field core (V. arvensis), whereas G. dissectum and P. rhoeas are represented in all field elements (Fried et al. 2007). The general trends observed (Table 3) do not suggest that patterns of intraspecific variability are associated with the ecological preferences of the species: intraspecific variability was not higher nor lower in the preferred habitat of each species.

Table 3. Intraspecific comparisons of trait distribution between field elements. Overlap in functional trait values between plants sampled from different field elements within each weed species. Non-parametric Gaussian kernel density estimation was used to measure functional overlap for each trait (top). 1 – F_β is a measure of functional overlap in a multi-trait space (bottom). Values in bold are significantly lower than under the randomization null model assumption after applying sequential Bonferroni–Holm adjustment (p < 0.05). CE, crop edge; FM, field margin; FC, field core; NA, data not available due to too small sample sizes. EPPO codes refer to abbreviations as presented in Table 2.

	Field elements compared
Biomass	CE – FM	CE – FC	FM – FC
GALAP	0.217	NA	NA
GERDI	0.264	0.181	0.546
PAPRH	0.051	0.424	NA
POLCO	0.314	0.673	0.285
VERPE	0.029	0.054	0.761
VIOAR	0.199	0.298	NA
SLA	CE – FM	CE – FC	FM – FC
GALAP	0.803	NA	NA
GERDI	0.635	0.062	0.382
PAPRH	0.379	0.711	NA
POLCO	0.218	0.422	0.301
VERPE	0.122	0.214	0.558
VIOAR	0.358	0.444	NA
Canopy height	CE – FM	CE – FC	FM – FC
GALAP	0.407	NA	NA
GERDI	0.307	0.639	0.424
PAPRH	0.199	0.742	NA
POLCO	0.434	0.551	0.234
VERPE	0.621	0.500	0.550
VIOAR	0.286	0.073	NA
1 – Fβ	CE – FM	CE – FC	FM – FC
GALAP	0.008	NA	NA
GERDI	0	0	0
PAPRH	0.001	0	NA
POLCO	0.001	0.003	0
VERPE	0	0	0.001
VIOAR	0.014	0	NA

Discussion

In this study we demonstrate the existence of different functional strategies of weeds among field elements based on in situ individual measurement of three key functional traits. In the field margin and the field core, we observed higher values of SLA, lower values ​of above-ground biomass and, although less clearly, smaller canopy height than in the crop edge. SLA is generally higher under canopy cover (Brainard, Bellinder, and DiTommaso 2005; Storkey 2005), which, in association with a smaller plant size, defines a shade-tolerance syndrome (Storkey 2005; Valladares and Niinemets 2008). This suggests that the main factor limiting the growth of individuals within the field is the lower light availability under crop canopy (Holt 1995). A high SLA is also known to be correlated with a high relative growth rate, an important ability in frequently disturbed environments (Gaba et al. in press). The relatively low canopy height and above-ground biomass observed within fields could also be associated with a predominance of late emerging plants that escaped important abiotic filters, such as herbicide applications or late freezing, but had a reduced growth and biomass accumulation because of an intense competition for light with the crop late in the season (Baumann, Bastiaans, and Kropff 2001; Chauvel, Guillemin, and Letouze 2005; Torra and Recasens 2008). Hence, within the field, disturbances and competition for light could lead to a predominance of late emerging plants with a single fast-growing shade-tolerant strategy that are able to produce seeds under canopy cover.

In contrast to the field, in the crop edge – characterized by an absence of direct crop competition and less frequent disturbances – two different functional strategies seem to coexist. The first strategy (previously defined as group 1) is characterized by a stem elongation and larger investment into supporting tissues but partially lower SLA, which can be regarded as a shade-avoidance syndrome (Franklin and Whitelam 2005). The second strategy (previously defined as group 2) is characterized by higher SLA values but smaller plants, which is more consistent with shade tolerance (Storkey 2005; Valladares and Niinemets 2008). The coexistence of these two light acquisition strategies could result from the predominance of weed–weed competition and a partial reduction of trait similarity among species. Studying larger sets of fields in diverse localizations will be necessary to validate these trends and a better characterization of the ecological filters at play, especially the ‘weed–crop’ and ‘weed–weed’ competitive interactions and disturbances, should be performed to fully understand how weeds respond to environmental conditions.

One salient result of our study is the substantial amount of intraspecific variation of weeds among field elements. Significant changes in trait values in response to varying environmental conditions were observed for all species. Furthermore, the large intraspecific variation observed in some species fully accounts for the presence of additional functional syndromes in the crop edge as compared to the field. Intraspecific trait variation may have two sources, i.e. genetic variability or phenotypic plasticity. Our study scheme does not allow distinguishing among the possible sources of intraspecific variation; however, we think that phenotypic plasticity is more likely. First, a high phenotypic plasticity in response to competition and light environment was previously described in weeds able to grow in a wide array of different habitats (e.g. Griffith and Sultan 2012). Second, as the spatial scales considered here are very small (i.e. about 1 m between the crop edge and the field margin, up to 20 m compared with the field core), genetic differentiation among field elements is unlikely. Moreover, most weeds are typically considered barochorous (Benvenuti 2007), constraining dispersal to a few metres away from the mother plant. However, seeds can travel much farther via secondary dispersal by humans (i.e. anthropochory), especially by soil-working machinery, reducing the likelihood of genetic differentiation. At the field scale, both the biotic environment and the management intensity differ between the field elements, which may select for genotypes adapted to the field elements’ environmental conditions. This would result in fine-scale genetic differentiation, as observed previously in some weeds (Hettwer and Gerowitt 2004). However, the three adjacent field elements considered in this study are all tilled, which probably results in a genetic homogenization of the seed bank over time due to secondary seed dispersal. Hence, the intraspecific variability observed in this study would be mostly due to phenotypic plasticity but further studies would be necessary to characterize the extent and role of plasticity and dispersion in the weed species studied here.

From a practical point of view, the occurrence of large intraspecific variation implies that the use of fixed mean field approach based on average trait values for analysing the functional diversity of weed communities is likely to be misleading. Whereas such an approach could be relevant when comparing the trait distribution over large environmental gradients (e.g. across a country), it should be avoided when analysing species response to environmental factors that vary at a finer scale, especially for plastic traits such as SLA and canopy height. In situ measurement of individual traits seems necessary to better understand how weed trait distribution is structured across various habitats within agroecosystems.

Using a functional approach, our study confirms that crop edges are favourable habitats for the development of many weeds. Several development strategies unexpressed in the cultivated field itself are observed within crop edges. In addition, the higher biomass production, regardless of the species, suggest a higher potential for the replenishment of soil seed banks within crop edges as compared to fields (José-María and Sans 2011). This is in agreement with previous results from Kohler et al. (2011) showing that, for threatened arable weeds, seed production is maximized in the presence of ploughing but without crop competition, two conditions that characterize the crop edge area.

Acknowledgements

This work was funded by a grant from INRA, Research Division “Plant Health and the Environment”. The authors thank Audrey Alignier, Rémy Destrebecq, Camille Drouhin, Gilles Louviot and Quentin Martinez for their assistance in the field.