Bocage landscapes are characterized by a network of hedgerows that delimits arable fields. Such landscapes provide many ecosystem services, including biodiversity conservation, but their effects on weed communities remain largely unknown. Bocage landscapes could affect weed communities through two main processes: plant spillover from hedgerows and increased environmental heterogeneity in arable fields. These bocage effects are also likely to vary between farming systems (conventional vs organic), due to differences in management practices.

We sampled weed communities more than 20 m from field margins in 74 arable fields (37 per farming system). Fields were located along two independent landscape gradients of total length of hedgerows (with or without a shrub layer) and organic farming cover, in Brittany (France). We analysed the effect of ‘bocage’ (i.e., the density and complexity of hedgerow networks) and farming systems at field and landscape scales on species and functional diversity of weed communities. Further, we used fidelity to non-crop habitats and Ellenberg indicator values to assess the ‘plant spillover’ and ‘environmental heterogeneity’ hypotheses, respectively.

Weed communities were more diverse and more abundant in organic farming systems. In addition, weed communities were more diverse, but not more abundant, in denser and more complex bocage landscapes. ‘Bocage’ increased species diversity of weeds, but also community-weighted variance of specific leaf area, plant height and seed mass. Positive effects of ‘bocage’ on weed diversity were driven by increased environmental heterogeneity rather than spillover of transient species from hedgerows. ‘Bocage’ effects were independent of farming systems at field and landscape scales.

Synthesis and applications. Maintaining diverse weed communities is key to agroecological weed management and biodiversity conservation in agricultural landscapes. Farmers are often concerned that hedgerows harbour competitive plants spreading into field edges, thereby increasing weed pressure. However, our study shows that dense and complex bocage landscapes promote weed diversity in field cores, most likely by increasing environmental heterogeneity. Thus, bocage landscapes could actually enhance ecosystem services provided by weed communities and reduce weed-crop competition.

This study was conducted in the southern part of the Zone Atelier Armorique, a Long-Term Socio-Ecological Research (LTSER) site in Brittany, northwestern France (47°59'35N, 1°45'12W). This region is characterized by dense hedgerow networks and crop-livestock farming systems. Agricultural lands are dominated by wheat and maize fields and temporary grasslands, and dominant soil types are Brunisols and Luvisols. Hedgerows are mostly old (i.e., planted at least before World War II), and generally composed of oak (Quercus robur) or chestnut (Castanea sativa) trees planted on a bank and pruned for firewood every 9–12 years. When present, the shrub layer is generally dominated by hazel (Corylus avellana), hawthorn (Crataegus monogyna), blackthorn (Prunus spinosa), spindle (Euonymus europaeus), broom (Cytisus scoparius) or gorse (Ulex europaeus). A more detailed description of hedgerow features in the study area, based on a sample of 40 hedgerows, is provided by Boinot and Alignier (2022). Observation sites (n = 37) are located along two independent landscape gradients (r = 0.15, p-value = 0.20): total length of hedgerows (ranging from 6.3km to 18km within a buffer radius of 1000m around focal fields) and organic farming cover (ranging from 0% to 37% of total area within a buffer radius of 1000m). Each site contains a pair of fields less than one kilometre apart and managed under conventional vs organic farming (CF vs OF). Among these fields, 62 were grown with winter cereals, whereas 12 OF fields were grown with winter cereal-legume mixtures (we resorted to mixtures when we could not find any pure crop in the area). Management practices and landscape gradients were similar between pure crops and mixtures under OF.

For each field, weeds were sampled on ten 1m² located in the field core, at least 20 m from the field margin. We sampled 20 pairs of fields in 2019 and 34 pairs of fields in 2020, in June and July before crop harvest. Pairs of fields sampled in 2019 were different from those sampled in 2020, but evenly located along the same range of landscape gradients. We visually estimated the cover of each species found in quadrats, with an accuracy of ± 5%. Ten quadrats per field were necessary to obtain a fairly accurate estimate of weed diversity, as indicated by species accumulation curves. Thus, we calculated the mean cover (%) of each species over the 10 quadrats for each field. Then, we measured total weed cover (i.e., the cumulative cover of all species), and species diversity (Hill-Shannon index) for each field. Hill-Shannon diversity considers both the number of species within a community and their relative cover to calculate the mean rarity of species within a community. There was no weed in two CF fields, which were therefore assigned the lowest possible diversity value (zero). We described the functional structure of weed communities using the community-weighted mean (CWM) and community-weighted variance (CWV) of the three traits included in the leaf-height-seed scheme: specific leaf area (SLA), plant height and seed mass. Mean trait values for each species were collected in Ecoflora (Fitter and Peat, 1994) and LEDA (Kleyer et al., 2008) databases. Species with known trait values always represented more than 90% of the total weed cover for each field, thereby providing sufficient representativeness.

To estimate plant spillover, we used the phi coefficient of association, which distinguishes ‘transient’ species that rely on regular re‐colonisation from neighbouring non-crop habitats, from ‘resident’ species typical of arable habitats and buffered by persistent seedbanks (Metcalfe et al., 2019). To avoid circular analysis, we used data from vegetation surveys at national scale. Weed frequency in arable fields was obtained from the ‘Biovigilance Flore’ network 2002–2012 (Fried et al., 2008), whereas plant frequency in uncropped field margins came from the ‘500 ENI’ network (Fried et al., 2018). For each field, we computed the CWM and CWV of fidelity to estimate the contribution of spillover to plant diversity in arable fields. An increase in CWM Fidelity indicates greater spillover of transient species from neighbouring non-crop habitats. Further, an increase or a decrease in CWV Fidelity indicates co-existence or selection of fidelity values, respectively. Species with known fidelity values always represented more than 90% of the total weed cover, except for one field that was therefore excluded from analyses.

To estimate environmental heterogeneity within arable fields, we used CWM and CWV of Ellenberg indicator values for temperature (EIV-T), soil moisture (EIV-F), light (EIV-L), and nutrients (EIV-N). An increase in CWV values indicates higher environmental heterogeneity. Further, a shift in CWM values suggest a change in mean environmental conditions. Ellenberg values were collected in the Baseflor database (Julve, 1998). Species with known Ellenberg values always represented more than 90% of the total weed cover for each field.

Kermap (https://kermap.com/en/) generated hedgerow mapping, using Computer Assisted Photo-interpretation based on the IGN orthophotograph of 2017. We rasterized vector maps with a resolution of one pixel for 5 m × 5 m to compute the ‘bocage’ metric, i.e., the density and complexity of hedgerow networks, within circular buffer radii of 250m, 500m, 750m and 1000m around each field, using Chloe software (Boussard and Baudry, 2017).

We assessed the effects of local farming systems (CF vs OF). OF systems were characterized by the absence of herbicide treatment, lower N-fertilization but higher soil disturbances in our study site. We considered both organic and mineral fertilization, converting organic fertilization into nitrogen amounts per hectare. Maize - winter wheat is by far the most common crop rotation in CF systems, whereas OF systems generally have more complex crop rotations, including polycultures and temporary grasslands. On average, surveyed fields have been under OF for 23 ± 6 years. Finally, we computed the total cover of OF systems (%OF) within circular buffer radii of 250m, 500m, 750m, and 1000m around each field, based on geographical data of CF vs OF systems in the Department of Ille-et-Vilaine in 2019 (provided by the CartoBio team of the French Agency for the Development and Promotion of Organic Agriculture).

Microsoft Excel and Microsoft Word (or Open Office)