Effects of agricultural intensification on plant diversity in Mediterranean dryland cereal fields


Correspondence author. E-mail: ljosemaria@ub.edu


1. Agricultural intensification, at field and landscape scales, has caused a decrease in weed richness and changes in species composition. In order to prevent this loss of diversity and develop efficient management regimes, it is important to understand how both aspects of intensification affect plant diversity and the way in which they interact.

2. This study assessed plant diversity at the centre, edges and boundaries of 29 organic and 29 conventional cereal fields distributed in 15 agrarian localities of the NE Iberian Peninsula. We focused on the composition of plant assemblages and on the specific richness at the field level, which was calculated for the whole set of species and for that of characteristic arable weeds. The percentage of arable land together with human settlements was used as a surrogate for landscape complexity and the amount of nitrogen inputs for land-use intensity.

3. Our results show that both scales of agricultural intensification have a similar negative effect on the total plant species and characteristic arable weed richness, and they also affect plant assemblages. Furthermore, we found no strong interaction between landscape and land-use intensity for explaining total plant richness.

4. The relative importance of farming intensity and landscape varies depending on the location within the field, which can be attributed to differences in the agricultural impact and limited seed dispersal from adjacent habitats. Management is the main factor in explaining differences among field centres, whereas changes at boundaries are mainly due to landscape characteristics, and at edges both factors are relevant.

5.Synthesis and applications. To preserve agricultural plant diversity it is equally important to prevent agricultural intensification at field and landscape scales. Policies enhancing low-intensity management techniques, such as organic farming, are the main way to promote diversity inside the fields and will be equally beneficial in simple and complex landscapes for total plant species richness. To maintain diversity within agricultural areas, it is also important to reduce farming intensity at the edges, which would favour characteristic arable weeds and margins’ overall plant diversity.


Agricultural intensification is a process occurring at field and landscape scales. The increase in external inputs (mainly nitrogen fertilization and pesticides), together with simplified crop-rotational schemes and improvement in seed-cleaning techniques have been identified as the major causes of land-use intensification at field scale (Sutherland 2002). Furthermore, intensification at the landscape scale has simplified the landscapes, through replacement of natural habitats with arable fields (Kleijn & Sutherland 2003). This leads to large, uniformly cropped areas with low spatial heterogeneity (Tscharntke et al. 2005; Gabriel et al. 2006).

In agroecosystems, arable weed communities have been especially harmed by this intensification process (Andreasen, Stryhn & Streibig 1996; Sutcliffe & Kay 2000; Hyvönen et al. 2003). Arable weed diversity not only has a conservational and aesthetic value but it also offers a variety of ecological and agronomic services, such as pest control, pollination and nutrient recycling (Altieri 1999; Marshall et al. 2003; Clergue et al. 2005). Moreover, the sensitivity of some weeds to agricultural intensification reinforces their suitability as indicator organisms to evaluate its effects on the conservation of agroecosystems (Albrecht 2003). However, arable weed communities include both ubiquitous generalists and characteristic species, which thrive almost exclusively in those habitats and are highly affected by intensification (Robinson & Sutherland 2002; Albrecht 2003; Romero, Chamorro & Sans 2008a). Therefore, the analysis of the response of characteristic arable species to intensification is necessary in order to formulate effective conservation strategies.

Several studies in central and northern Europe have focused on the effect of agricultural intensification on biodiversity at field and landscape scales. The comparison between contrasting farming systems (i.e., organic and conventional) has shown that organic farming enhances arable weed diversity (see reviews by Bengtsson, Ahnström & Weibull 2005; Hole et al. 2005). With the exception of Weibull, Östman & Granqvist (2003), research into the effect of landscape complexity on plant diversity has found that these aspects are positively correlated (Gabriel, Thies & Tscharntke 2005; Roschewitz et al. 2005a), because complex areas provide more alternative habitats and sources of recolonization from nearby habitats. Furthermore, landscape structure has been found to have a stronger influence on conventionally managed cereal fields than on those managed organically (Roschewitz et al. 2005a). Based on these results, Tscharntke et al. (2005) suggest that complex landscapes compensate for the negative effects of farming intensification on weed diversity at field scale (Compensation for landscape complexity hypothesis).

The aim of this study is to analyse the effects of both aspects of agricultural intensification (landscape simplification and land-use intensity) on plant diversity and assemblages in dryland Mediterranean cereal fields. As species are not homogeneously distributed over the entire field (Marshall 1989; Wilson & Aebischer 1995; Clough et al. 2007; Romero, Chamorro & Sans 2008a), we focused our study on three different areas of the field (the centre, the cultivated edge and the adjacent uncultivated boundary), in order to evaluate the importance of farming, landscape simplification and their interaction in the aforementioned habitats. We hypothesized that the effect of management intensity should be higher in the centre and decrease towards the outer part of the field, whereas the importance of the surrounding landscape decreases inversely. These trends would be caused by spatial differences in the impact of farming practices within a field, which is higher in the centre, and the limited plant seed dispersal from adjacent habitats. Nevertheless, it must be taken into account that landscape complexity and land-use intensity are often closely related (Roschewitz, Thies & Tscharntke 2005b; Hendrickx et al. 2007), and this may hinder interpretations of the results based on both aspects. Therefore, we selected landscape and farming descriptors that are not correlated, in order to understand how both factors and their interaction affect agricultural biodiversity as well as their relative importance, and thus help to develop agri-environment schemes to counteract the current tendency of agricultural biodiversity loss.

To our knowledge this is the first study that analyses the relative effect of farming and landscape simplification in dryland Mediterranean areas at field centres, edges and boundaries. Similar studies have already been carried out in more temperate areas (Roschewitz et al. 2005a; Gabriel et al. 2009), but the particularities of our study area make it relevant to investigate whether the trends are similar to that of central and northern Europe. In contrast to these areas, the Mediterranean agricultural landscapes of the region studied are characterized by an uneven topography and scarcity of broad plains where intense cultivation is concentrated (Folch et al. 1997). In addition, the Mediterranean climate, characterized by low rainfall and high year-to-year variation in water availability, affects cereal crop yields and competitive interactions among plants (Liancourt, Callaway & Michalet 2005). Moreover, the higher diversity of arable weed communities (Holzner & Immonen 1982) and that of natural habitats surrounding and intermingled in agrarian Mediterranean areas may modify responses to landscape simplification.

Materials and methods

Study area

The study was conducted in 2008 in a dryland cereal region situated in Central Catalonia, the NE Iberian Peninsula (41°24′–42°05′N; 1°05′–2°05′E). Fifteen localities defined as sectors of 2·5 km radius were selected so as to vary as much as possible in the degree of landscape complexity, ranging from structurally simple landscapes, where arable land was the main habitat (96%), to structurally rich landscapes characterized by a low percentage of arable land (19%). All localities are in an area that covers c. 85 × 75 km (Fig. 1). The climate is Mediterranean, with mean annual precipitation and temperatures ranging from 400 to 850 mm, and from 11 to 14 °C respectively (Ninyerola, Pons & Roure 2005). The elevation of study sites ranges from 240 to 850 m a.s.l., and the soils are loamy and clayish. The natural habitats in the study area include woodlands, mainly pines (Pinus halepensis Mill. and Pinus nigra Arnold) as well as evergreen oaks Quercus ilex L. and deciduous oaks Quercus faginea Lam.; shrublands, including stands of resprouting oaks and young pines; perennial-dominated grasslands and riverine vegetation. Arable weed communities in the study area are dominated by ruderal and generalist annual species and characteristic arable weeds. The latter are mainly Mediterranean and submediterranean species and have shown a clear decline in abundance over the last century (Chamorro et al. 2007).

Figure 1.

 Situation of the study area within Catalonia (above) and of the 15 localities within the study area (below). The grey scale of 1 × 1 km pixels refers to the percentage cover of intensive land-use (PIL).

In each locality an organic and a conventional farm was selected that differed in farming intensity. Agricultural practices in the conventional farms comprised an annual full-field application of herbicide and no insecticide use. This level of management was representative of the study area. At each farm, we surveyed two winter cereal fields (wheat or barley), except in one locality where only one field per farm was sampled due to early harvesting. The mean area of the fields per farm did not differ between management practices (organic = 1·79 ± 0·13 ha, mean ± SE; conventional = 1·82 ± 0·09 ha; Wilcoxon’s paired tests within locality: = 0·525), and neither did the perimeter (organic = 647 ± 27 m; conventional = 624 ± 21 m, = 0·359), nor the area/perimeter ratio (organic = 27·10 ± 1·22 m, conventional = 28·99 ± 0·99 m, =0·208). We did not find any significant correlation between field descriptors and landscape complexity assessed as the percentage of natural habitats, (area: Pearson’s correlation coefficient ρ = 0·03, = 0·796; perimeter: ρ = 0·07, = 0·597, area/perimeter ratio: ρ = −0·03, = 0·839).

Plant species surveys

A total of 58 fields were surveyed in spring 2008 before harvest, between May and the beginning of July. In each field, we delimited three areas (hereafter positions): the field centre, 20 m away from the edge; the field edge adjacent to the margin, understood as the first cultivated metre; and the field boundary adjacent to the crop, namely the first metre of non-cultivated habitat surrounding the field. Four blocks were randomly established along the field perimeter, which consisted of three 1 × 10 m plots parallel to the margin and placed in each position (3 positions × 4 blocks = 12 plots per field). The species composition of each plot was recorded as the list of all species present in 20 quadrats of 25 × 25 cm evenly distributed in each plot. Those species considered characteristic of phytosociological order Secalietalia cerealis Br.-Bl., 1936, were identified as characteristic arable weeds in the sense of Albrecht (2003) and Romero, Chamorro & Sans (2008a). Nomenclature of plant species and phytosociological adscription follows that of de Bolòs et al. (2005).

Landscape complexity analysis

The landscape around each field was characterized within a circular sector of 1 km radius using Catalan Habitats Cartography (Departament de Medi Ambient 2004) produced at a 1 : 25 000 scale. Proportion of arable land has been a widely used estimator of agricultural landscape simplification as it correlates strongly with other landscape structure descriptors (Roschewitz et al. 2005a; Gabriel et al. 2006; Rundlöf & Smith 2006). However, in our Mediterranean landscape, arable land is usually interspersed with other minor crops and human settlements, and therefore we use the percentage of arable land, mainly arable fields but also including vineyards, almond groves and olive trees, together with associated human settlements, as an estimator of landscape simplification, hereafter percentage cover of intensive land-use (PIL). PIL is the inverse of the percentage of natural habitats, and previous studies have shown its negative correlation with diversity of habitats, as well as a significant positive correlation with arable patch aggregation (Romero et al. 2008b).

Land-use intensity analysis

Farmers were interviewed about their farming practices. From the information available, only those variables differing among farmers which affect biodiversity were analysed (McLaughlin & Mineau 1995; Hole et al. 2005; Gabriel et al. 2006). These variables were: mean annual inputs of exogenous nitrogen, use of herbicide, weed harrowing, seed origin (purchase of industrially selected seeds or re-use of own seeds), cultivated diversity (number of different plant-families cropped in the last 5 years) and cereal ratio (proportion of cereal crops in a rotational scheme). Preliminary analysis had shown a high variability of management within organic and conventional farms (Table 1), especially among organic ones, which led us to evaluate the intensity of management in a more complex way than the traditional organic/conventional dichotomy. Therefore, we analysed the individual correlations among the management variables, their relationship with PIL and with plant diversity, as well as the relationship between combinations of them. Finally, we selected the amount of nitrogen inputs (hereafter N) as a land-use intensity indicator, which has already been used in other studies (Kleijn et al. 2009). A comparison of N between organic and conventional fields with a Student’s t-test showed that it was significantly higher in conventional fields (mean = 150·24) than in organic ones (mean = 30·78; t = 5·669, < 0·0001). It was also higher at farms where industrially selected seeds were sown (mean = 133·05) vs. those who re-used their own seeds (mean = 62·15, t = −2·36, P = 0·028) and it correlated significantly with cereal ratio (Pearson’s correlation coefficient ρ = 0·58, P < 0·0001) and cultivated diversity (ρ = 0·49, < 0·0001). Moreover, it was also independent of PIL both when organic and conventional farms were analysed together (ρ = 0·00, = 0·977) or separately (organic: ρ = −0·09, = 0·633; conventional: ρ = −0·12, = 0·535). These results showed that N is suitable for testing the relative effect of landscape simplification and farming intensity on plant diversity and assemblages.

Table 1.   Characterization of farming practices of organic and conventional fields (60 fields evenly distributed in 15 organic and 15 conventional farms)
  1. Median and range (in brackets) of nitrogen inputs (mean annual inputs of exogenous nitrogen); cultivated diversity (number of different plant-families sown in 5 years); cereal ratio (proportion of cereal in a rotational scheme; 1 accounts for cereal monoculture for more than 10 years). Seed origin and herbicide application is assessed by the proportion of farmers using the stated practices.

Nitrogen inputs (kg ha−1)142·8 [58·2, 325]18·8 [0, 108·7]
Cultivated diversity1 [1, 3]2·5 [1, 5]
Cereal ratio1 [2/3, 1]2/3 [1/5, 9/10]
Seed origin
 Purchase of seeds10/152/15
 Re-use of own seeds5/1513/15
Herbicide use15/150/15
Weed harrowing0/155/15

Plant diversity analysis

Species diversity within each field was estimated by means of total species richness. Species richness was also calculated separately for the set of characteristic arable weeds. We analysed its variability using mixed-effects models, which account for non-independent errors that may occur due to hierarchically nested designs. We tested the effect of N, PIL and position [centre (C), edge (E) and boundary (B)] and their interactions as fixed factors, and locality and field, nested within locality, as random factors. As the absolute values and the ranges of PIL and N differed strongly, they were standardized to have a mean of zero and a standard deviation of one, which facilitated comparison of their effects based on regression coefficients (Hendrickx et al. 2007). Orthogonal contrasts to compare the different levels of the factor position were fixed a priori, in order to compare the field centre with the outer areas of the field (B&E vs. C), and also to check for differences between the edge and the boundary (E vs. B). Adequacy of the models was checked by confirming the normality of the residuals within the different groups, the homoscedasticity of the residuals and the correlations between the observed and fitted values.

To evaluate the possible models explaining our data we used the methods described by Burnham & Anderson (2002). For each species set (total and characteristic species), 19 models with all possible combinations of the explanatory variables were compared by Akaike Information Criterion (AIC), which allows direct comparison of models with different combinations of parameters. We calculated the AICc, which includes a correction for small sample sizes. Afterwards we estimated the size of information loss for the various models compared with the estimated best model (Δi = AICci − AICcmin), and we calculated an Akaike weight (wi) for each model, which is the probability that a certain model would be selected as the best fit model if the data were collected again under identical circumstances. We calculated confidence sets of models fitted to each data set, which are the smallest subset of candidate models for which the sum of wi reaches a given value, in this case 0·90.

In relation to the analysis carried out on the effect of each variable, we base our inferences on the entire set of selected models, using multi-model inference, which averages the estimates and standard errors of the parameters in the different models weighted by the Akaike weights. To evaluate the relative importance of the different factors we summed the Akaike weights of all the models containing a certain variable, which accounts for the probability that this variable would be in the best approximating model if we collect the data again under identical circumstances. Because poor predictors are not expected to have selection probabilities close to zero, we also computed the 95% confidence intervals of these variables to evaluate the significance of their contributions. Statistical analysis was carried out using r 2.8.1 (R Developement Core Team 2008) with package lme4 (Bates, Maechler & Dai 2008) for mixed models.

Plant composition analysis

The species composition at different positions within each field was analysed using multivariate analysis of floristic surveys, based on presence/absence data. Only 14 localities were considered to achieve a balanced design, and species present in just one field were removed (113 out of 432 species found). The Jaccard dissimilarity indexes were computed between all the lists of plant species of each field-position combination (56 fields × 3 positions = 168 sites) and a Non-metric Multidimensional Scaling (NMDS) analysis was performed, with k = 2 (number of dimensions). NMDS is commonly regarded as the most robust unconstrained ordination method in community ecology (Minchin 1987), and restricting the number of dimensions to two facilitates its graphical representation. Finally, the position as a factor was fitted onto this ordination and its significance was tested with random permutations of the data restricted within each locality.

In addition, we conducted a permutational multivariate analysis of variance, separately for each position, using distance matrices, to analyse how our explanatory variables (N and PIL) affect species composition. This analysis allows partitioning distance matrices among sources of variation and fitting linear models to distance matrices. Again we used the Jaccard dissimilarity index and, for each position, species present in just one field were removed. The significance of the explanatory variables was obtained by means of F-tests based on sequential sums of squares from permutations of the raw data, restricting permutations within each locality so as to take into account hierarchal sampling. This method is a sound alternative to both parametric manova and to ordination methods for describing how variation is attributed to different experimental treatments or uncontrolled covariates and it is also analogous to redundancy analysis (Legendre & Anderson 1999). Analyses were carried out using vegan package for r (Oksanen et al. 2009).


Relative effect of farming and landscape complexity on plant diversity

In total, 432 plant species were recorded, of which 140 were found in the centre, 285 at the edge and 398 on the boundary of the field. From all these species, only 35 were characteristic arable weed species (see Table S1, Supporting information); and 26, 29 and 28 species were recorded at the centre, edge and boundary respectively. The confidence sets of models fitted to total species richness and that of characteristic arable species are summarized in Table 2.

Table 2.   Confidence sets of models for which Σwi > 0·90, for total species richness and characteristic arable weed richness
PILNposPIL × NPIL × posN × posAICcΔiwi
  1. Results from information-theoretic-based model selection. x indicates variable inclusion in each individual model; PIL, percentage cover of intensive land-use; N, nitrogen inputs; pos, field-position (boundary, edge and centre). AICc, Akaike’s information criterion corrected for small sample size; Δi, the AICc differences compared with the most parsimonious model; wi, Akaike weights (more details in text).

Total species
xxx x 1251·3400·62
xxxxx 1253·261·920·24
xxx xx1255·103·760·10
Characteristic arable weeds
xxxx  741·8200·40
xxxxx 741·840·020·39
xxxx x745·944·120·05
xxx x 746·154·330·05

Overall species richness was influenced by the gradient of land-use intensity, landscape complexity and distance to the margin (Table 3), as we observed strong support of PIL, N and position based on the estimated magnitude of the effects, the relative importance (based on Akaike’s weights) and the confidence intervals, which do not include 0. Thus, species richness was higher in complex landscapes than in simple ones, and it was also higher in low-intensity farming systems compared with high-intensity ones. Both aspects of the intensification had a similar size effect (similar regression coefficients) and similar relative importance on total species richness (Fig. S1, Supporting information). Moreover, we found strong interactions between PIL and position, which indicate that the surrounding landscape had a greater influence on the boundary and its effect decreased progressively through the inner positions of the field (Fig. 2). Although the PIL × N and N × position interactions appeared as explanatory variables in the confidence sets of models (Table 2), they did not receive strong support from their relative importance values and confidence intervals (Table 3), and therefore we avoid making inference based on them.

Table 3.   Model-averaged estimate, unconditional standard error (UnSE), relative importance (RI) and 95% confidence interval (CI) of predictor variables and their interactions for total species richness and for characteristic arable weeds
  1. PIL, percentage cover of intensive land-use; N, nitrogen inputs; pos B&E vs. C, edge and boundary against centre; pos E vs. B: edge against boundary.

Total species
(Intercept)37·031·601(33·88, 40·17)
PIL−5·371·401(−8·11, −2·63)
N−4·800·901(−6·56, −3·05)
pos B&E vs. C10·030·361(9·32, 10·73)
pos E vs. B−7·870·621(−9·09, −6·65)
PIL × pos B&E vs. C−2·250·361(−2·96, −1·54)
PIL × pos E vs. B2·690·631(1·46, 3·91)
PIL × N0·530·890·28(−1·20, 2·27)
N × pos B&E vs. C0·190·360·14(−0·51, 0·90)
N × pos E vs. B−0·460·620·14(−1·68, 0·77)
Characteristic arable weeds
(Intercept)3·570·321(2·94, 4·19)
PIL−0·920·301(−1·51, −0·33)
N−0·800·221(−1·23, −0·37)
pos B&E vs. C0·630·081(0·47, 0·79)
pos E vs. B0·400·141(0·12, 0·68)
PIL × pos B&E vs. C−0·170·080·50(−0·33, −0·01)
PIL × pos E vs. B0·060·140·50(−0·22, 0·34)
PIL × N0·570·210·90(0·15, 0·99)
N × pos B&E vs. C0·020·080·12(−0·14, 0·18)
N × pos E vs. B−0·090·140·12(−0·37, 0·20)
Figure 2.

 Total number of species at the centre (empty circles), edges (black circles) and boundaries (squares) of each field in relation to the percentage cover of intensive land-use (PIL). Regression lines from mixed effect models.

Similarly to total species richness, characteristic arable species were also negatively affected by landscape simplification and land-use intensification (Table 3) and there was also a strong position effect. The results for the interaction between the landscape and position indicated that the landscape effect was lower in the centre of the field, and that there was no strong difference between the edge and the boundary. The major effect of the interaction between PIL and N, which was mainly related to the low characteristic species pool, revealed that they occurred under low PIL and N values by preference (Fig. S2, Supporting information). Furthermore, the interaction N × position received weak support, as shown by its relative importance and confidence intervals.

Plant assemblages

The NMDS analysis (k = 2, non-metric fit: r2 = 0·947) showed a clear spread of the sites according to their positions (Fig. 3; fit of the factor position: r2 = 0·458, < 0·0001). In the centre of the fields, we found mainly common ruderal and arable weeds (i.e., Lolium rigidum Gaudin, Polygonum aviculare L., Papaver rhoeas L. and Convolvulus arvensis L.) and also some characteristic arable weeds (Table S1, Supporting information); in the boundaries we recorded typical species from the natural communities surrounding the field (dry grasslands, shrublands, woodlands and ruderal communities), whereas in the edges we found plants of both types of environments.

Figure 3.

 Site ordination (NMDS) based on floristic similarities of centre (empty circles), edge (black circles) and boundary (squares) of 56 fields (k = 2, non-metric fit: r2 = 0·947). The labels of each position correspond to the averages obtained after fitting the factor onto the ordination (< 0·0001).

Landscape simplification and farming intensity also affected species composition, and their effects varied in relation to the different positions within the field (Table 4). The effect of management increased from the boundary to the centre of the field (boundary: 1·9%; edge: 2·7%; centre: 6%), whereas the percentage explained by surrounding landscape decreased (14·5%, 13·6%, 10·2%). In all the positions, a significant interaction of both factors was found, explaining around 3% of the variability.

Table 4.   Results from permutational analysis of variance in species composition of 56 different fields (d.f. = 55), conducted separately for each position
  1. Sums of squares (SS), partial R-squared (r2) and levels of significance (***< 0·001; **< 0·01; *< 0·05; •< 0·1) of the different sources of variation considered (N, nitrogen inputs; PIL, percentage cover of intensive land-use, and their interaction), based on 9999 permutations.

N0·2200·019 ns0·2900·027**0·7160·060***
PIL × N0·3330·029**0·2780·026*0·3120·026•
Total11·386 10·583 11·967 


Distribution pattern of plant species

Plant species richness and assemblages in dryland arable fields are highly influenced by position in the field. The NMDS analysis revealed that the similarities between plant communities respond to the different locations within the field (Fig. 3), following the same pattern as described by Dutoit et al. (2007). In addition, our results reveal that there is a gradient in total species richness from the boundary to the centre of the field (Fig. 2), in concordance with other studies (Marshall 1989; Wilson & Aebischer 1995; Clough et al. 2007; Romero, Chamorro & Sans 2008a). This pattern reflects the different impact of farming practices and landscape in these different areas. The importance of management increases towards the field centre, limiting species occurrence, whereas the importance of adjacent habitats, which are a source of species seed pool, decreases with increasing distance to them. Nevertheless, the distribution pattern of characteristic flora is different because, although it concentrates in the outer areas of the field (edge and boundary), it reaches its maximum at the edge (Table 3). This pattern can be attributed to the lower efficacy of farming practices (crop sowing, fertilization and weed control) at the edges compared with the centre (Kleijn & van der Voort 1997; Romero, Chamorro & Sans 2008a). Furthermore, the lower presence of characteristic species in the boundaries compared with the edges is related to their dependence on regular cultivation and their inability to thrive in competitive habitats (Marshall 2009).

Effect of agricultural intensification on plant diversity and assemblages

Our results show that plant diversity is affected similarly by both aspects of agricultural intensification. The analyses used the amount of nitrogen inputs (N), which reflects the differences in management practices between and within organic and conventional farms, as a proxy for land-use intensity, and the PIL for landscape simplification. Similarly to most European studies we found that species richness is reduced under high-intensity farming practices (Bengtsson, Ahnström & Weibull 2005; Hole et al. 2005; Kleijn et al. 2009), and that it decreases with landscape simplification (Gabriel, Thies & Tscharntke 2005; Roschewitz et al. 2005a), which highlights the importance of neighbouring habitats. Moreover, our results strongly support the view that the relative importance of both scales of agricultural intensification depends on spatial position within a field.


For the total number of plant species, as well as for the species composition, the landscape effect has its maximum on the boundary and decreases at the edge, until it has no effect in the centre (Fig. 2). In complex areas, immigrating species from non-crop areas surrounding the field lead to higher levels of diversity along the boundaries. The effect of landscape on plant diversity at the edges is mediated through the boundaries because of the neighbourhood effect (Gabriel et al. 2006). The fact that no landscape effect was found in the centre of the field, which has also been reported by Marshall (2009), may be a response to the high-intensity of management, which inhibits the expression of plant diversity. It could also be a consequence of short-range dispersal of seeds from adjacent habitats, which would limit the neighbourhood effect (Devlaeminck, Bossuyt & Hermy 2005). More research is needed on seed dispersal from neighbouring habitats to different positions of the field to gain further insight into the mechanisms explaining the spatial species diversity distribution pattern.


Our results show that the importance of farming related to the landscape effect decreases from the field centre towards the boundary, both for the species assemblages and plant diversity analysis. We only found a weak interaction between land-use intensity and the spatial position within a field for species richness, which showed that the absolute species reduction related to the increase of farming intensity was similar in each position. Nevertheless, it should be taken into account that the species pool differed in each position (being greater in the boundary), and therefore the relative contribution of management to the number of species was minimal in the boundaries, and reached a maximum in the centre of the fields. These findings, which follow the pattern described by Gabriel et al. (2009), can be explained by a greater impact of agricultural practices in the centre of the fields.

Interaction between landscape and farming

Although recent studies (Tscharntke et al. 2005; Concepción, Díaz & Baquero 2008) have pointed out the importance of the interaction between landscape and farming, we did not find any strong interaction between these factors for total species richness, and it only became relatively important for the small subset of characteristic arable species. Hence, our results support the idea that the species reduction related to the increase of the intensity of management is similar in complex and simple landscapes for total plant species (Fig. S1) but not for characteristic arable species (Fig. S2). Rare species, defined as those with a relative cover <1% (Kleijn et al. 2009), and characteristic arable species are highly sensitive to agricultural intensification. Therefore, their conservation is especially threatened where agricultural intensification takes place: in simple landscapes or on conventional farms.

Implications for conservation

Our study supports the view that to conserve biodiversity in farmed landscapes, it is equally important to prevent agricultural intensification at field and landscape scales. At landscape scale, policies should limit boundary reduction and encourage the recuperation of natural habitats, whereas at the field scale policies should encourage low-intensity farming practices. In our Mediterranean context, these measures would be beneficial for total plant diversity in both simple and complex landscapes. Moreover, it seems to be the only factor explaining total diversity in the centre of the fields in Mediterranean areas. Therefore, due to the relative simplicity of adjusting management at a local scale, compared with the landscape scale, it is crucial, regardless of the landscape, to focus on the promotion of low-intensive farming practices.

As our results point out, edges are a refuge for characteristic arable flora, which are the weeds that have been mostly affected by agricultural intensification and may persist in the seedbank at the field edge (Marshall 1989; Wilson & Aebischer 1995). Accordingly, agri-environment schemes limiting the intensity of management (prohibiting the use of pesticides and/or fertilizers) at the edges become crucial to preserving them. Such measures will be more cost-effective in terms of enhancement of characteristic arable weed richness in complex areas, where natural habitats cover an important part of agricultural landscapes.

To our knowledge, our study is the first to show that farming also influences field margin plant diversity in Mediterranean agricultural landscapes, at least in the area adjacent to the crop edge, and this is in accordance with results from temperate areas (Marshall & Moonen 2002; Boutin, Baril & Martin 2008). Therefore, to preserve the diversity of field margins it is important to limit negative management actions, such as direct removal of boundaries or herbicide spraying, as well as to prevent indirect effects of farming in the cropped field. In this context, the promotion of low-intensity farming practices or appropriate agri-environment schemes would be effective measures in Mediterranean agroecosystems.


We thank the farmers for their collaboration and the members of the Department of Plant Biology of the University of Barcelona for field and office assistance. We particularly thank Albert Romero and Lourdes Chamorro for their help on the experimental design and for providing useful discussions. This research was funded by the Spanish Ministry of Education and Science with a fellowship to the first author and the projects CGL2006-13190-C03-01 and CGL2009-13497-C02-01.