Weed Species Trait Selection as Shaped by Region and Crop Diversity in Organically Managed Spring Cereals

: Weeds remain a challenge in organic arable farming, as well as supply ecosystem services. The aim is to control weed densities while hosting a diverse and manageable weed community, preventing domination of few deleterious species. Therefore, we want to understand how speciﬁc species are stimulated, and which traits are selected for. This study focuses on crop diversity hypothesizing that (1) regions and (2) crop diversity function as ﬁlters for speciﬁc weed species traits. We conducted a weed monitoring in spring cereals over 2 years on organic farms in ﬁve northern European regions. Management and weed trait variables collected for the occurring species allowed an RLQ fourth-corner analysis. The weed communities were regionally speciﬁc, but trait selection was not observed, except in Latvia. Hence, the regional species pool provided different species with similar traits. Crop diversity within the management of spring cereals, such as undersowing and cereal frequency in the rotation, affected weed traits. The number of years under organic production selected no traits, although species numbers are known to increase. Hence, general weed species diversity increased, irrespective of traits. We conclude that organic management may support the agility within the weed community against selection of species and act as a buffer rather than as ﬁlter.


Introduction
During the last few decades, the floral diversity in arable fields has declined severely, driven among others by the use of fertilizers and herbicides [1,2]. The presence of weeds provides, however, a plethora of beneficial ecosystem services in arable fields [3]. Therefore, a more sustainable weed management is an important step toward ecological intensification [4]. On the other hand, weed pressure still remains the main production-limiting factor in agricultural systems, especially in those systems forgoing the use of herbicides [5][6][7]. Thus, the aim is to continue controlling weeds and, within the remaining weed community, host weakly competitive and manageable species in the absence of herbicides. There have been arguments made for balanced weed communities, in order to mitigate weed problems [8,9]. Furthermore, there are signs indicating that weed diversity and evenness are capable of decreasing total weed biomass, as well as mitigating crop yield losses [10].

Vegetation Data
A weed survey was conducted on 58 organic farms in the northern European countries of Denmark, Finland, Germany, Latvia, and Sweden ( Figure A1, Appendix A) [18]. The survey was conducted in spring cereals during 2015 and 2016. In total, 207 fields were monitored at the crop flowering stage (stages 61-69 of the BBCH scale) [31], after all physical weed control measures were completed. During the survey, weed densities and the number of weed species were documented for each field. Within these arable fields, three plots of 100 m 2 were surveyed for the density of all individual weed species. To estimate the densities, a classification scale was used, which included 10 density classes, exponentially increasing from fewer than 0.2 individuals per m 2 to more than 200 individuals per m 2 . To avoid edge effects, the plots were randomly located in the field, at least 10 m away from the nearest boundary. Several weed species were impossible to identify at time of the survey and, thus, only classified and recorded at the genus level, such as Vicia spp.
To allow for analysis, the classification scale was converted into density values using a logarithmic mean. The Latin names were sourced from the Flora Europaea [32], and species are displayed in the ordination graph with EPPO codes [33].

Crop Management and Environmental Data
The farmers whose fields were surveyed completed a questionnaire about their farm and field management. From this, information was documented about the site, the current cereal species, the crop sequence for the previous 5 years, primary tillage, weed management, and yield. From the farm and field information, five classes of environmental or management data were selected for the research objective: (1) 'crop' (if the cereal crop present is sown on its own, intercropped, or undersown), (2) 'pre-crop' (if the preceding crop in rotation was a spring cereal, winter cereal, grass clover ley, row crop, or other spring sown crop), (3) 'crop diversity frequency' (the number of uses of crop mixtures, undersown crop, or winter catch crop used in the last 5 years of rotation), (4) 'rotation' (the number of cultivations of cereals, grass clover ley, or other crops in the last 5 years of rotation, and (5) 'harrowing' (the use of physical weed control in the surveyed spring cereal). Primary tillage was not included as a variable, because inversion tillage was the common primary tillage practice on all the farms involved in the survey.

Trait Data
A trait database was compiled for the observed species. In total, 149 species were included (Table A1, Appendix A). Exceptions were made for species identified at the genus level for reasons of unidentifiable trait variance within the genus, as well as voluntary crops. Both these groups were excluded from the species list. The trait database was based on the database of Bàrberi et al. (2018) [34] and expanded further to include species found during the monitoring, but not previously listed (Table A2, Appendix A; includes sources).

Data Analysis
A multivariate analysis of the weed species composition was performed to study the dispersal of species and sites. The weed species records were used as presence/absence data for the multivariate analysis, as the large variation of species density data caused extreme and uninterpretable ordination patterns. A correspondence analysis (CA) [35] was performed on the whole dataset, and the resulting ordination plot displayed sites and species dispersal.
The relationships between region and traits and between crop management and traits were studied by means of a RLQ method, which addresses the fourth-corner problem [36,37]. This analysis was performed on the basis of three datasets. The first was the so-called R-table, which consists of 'environmental' or, in our case, management data per field. The second was the L-table, which contains the density data of each species for each field. The third was the Q-table, which contains the trait data for each species. The analysis combines several multivariate techniques in order to relate the species traits to the management or environment data.
The three datasets were first analyzed through the use of ordination methods. This meant a CA for the L table and a Hill-Smith ordination for the Q and R tables [38]. The results from these ordinations were used for the RLQ analysis. The relationships between the management variables and the traits were then tested by means of a fourth-corner analysis [36]. To test the crop management variables more closely, a backward selection was made, excluding the traits which had no significant correlation with management variables. Qualitative traits were tested on all levels but showed no significant relationships. Therefore, the following traits were entered as qualitative, and interpreted from the functional group analysis: RLF, GTF, GLS, SSG, and SNC. Associations between two categorical variables were tested with Pearson chi-square statistics (X 2 ), associations between a categorical variable and a continuous variable were tested using a pseudo-F and Pearson's r correlation ratio, and associations between two continuous variables were tested using a Pearson correlation coefficient.
Following the fourth-corner analysis, biplots for the management and trait data were compiled, which, after the species distribution on the basis of management variables and traits, were plotted in the ordination space. The functional groups were identified using the hierarchical cluster analysis, using Ward's method based on the Euclidean distances, for more detailed interpretation of the underlying dynamics. The functional groups were observed to further understand the interactive mechanisms of the management and trait variables.
All the statistical analyses were performed with R software version 3.5.1 [39], utilizing the ade4 package [40].

Region Interactions with Traits in the Weed Community
The ordination plot of the CA performed on the weed communities demonstrated a clear geographical clustering (Figure 1). The full species list is available in Table A1 (Appendix A). According to their individual weed communities, fields within a region were visibly similar and clustered in a north-to-south orientation. In contrast, Latvia deviated from this picture by settling into its own cluster. This clustering reflected the regionally diverse environmental factors, which in turn created a weed community that was regionally distinct.
The only weed trait regions selected for was Raunkiaer life form (RLF), and it specifically discriminated Latvia (Table 1). Latvia as a region selects for perennial species, specifically for geophytes (plants with underground storage organs) and chamaephytes (dwarf shrubs). When the data from Latvia were tested on their own (data not shown), we found tendencies that perennial species positively correlated with crop and pre-crop (X 2 = 0.366 and X 2 = 0.322, respectively) and negatively correlated with the use of winter catch crops and other crops (other than cereals, grass, or clover ley) in the rotation (X 2 = 0.242 and X 2 = 0.245, respectively). The other traits and regions had no relationships.

Crop Management Interactions with Traits in the Weed Community
The subsequent analysis excluded the data from Latvia, as this region expressed a distinct trait selection on local level. The first two axes of the RLQ accounted for 46.8% and 21% of the total inertia ( Figure 2). The Monte Carlo test indicated the relationships between traits and environment which were generally significant (p < 0.05, based on 9999 permutations). The first two axes of the RLQ showed relatively low correlations (23% and 20%, respectively), but the variance of the environmental scores was well preserved on the first two axes with 76%. The variance of the traits scores on the first two axes was up to 84%. According to the fourth-corner analysis ( Table 2 and Figure 2), elements of the rotation in which spring cereals were cropped had the strongest interactions with weed traits. Undersowing was positively associated with Grime's life strategy, while the years of cereal in the rotation correlated negatively with duration of flowering. These significant results were followed by visible trends in the results (Table 2). Harrowing negatively impacted the plant height and selected for rosette forming species, reducing creeping or ascending species. The years of cereals or grass clover ley selected differently for growth form, with cereals selecting for ascending and creeping species, while grass clover ley selected for rosette species. Grass clover ley was also associated with longer flowering periods. A pre-crop of spring cereals was positively associated with nitrophile species. When the present crop was pure cereals, these tended to select for species with a competitive life strategy. A present crop with undersown crops was associated positively with species with a more stress tolerant or diverse life strategy. Specific leaf area was not affected by any of the tested crop management variables; this is echoed in Figure 2, where this trait is positioned unattached of the other interactions. The number of years a field was under organic cropping was tested, but no trait selection was observed.  The cluster analysis identified seven functional groups ( Figure 3). The distribution values of the nine traits of each of these groups are presented in Figure 4A-I. Group 1 was characterized by competitive perennials, group 2 was characterized by large annuals, group 3 consisted mostly of annual grasses, group 4 was composed of rosette forming perennials, which could be characterized as grassland species, group 5 included rosetteforming autumnal annuals, group 6 specifically involved big-seeded annuals, and group 7 consisted of perennial grasses. When studying Figures 3 and 4 while observing Table 2, patterns emerge. Group 4, consisting of grassland species and characterized by a diversified life strategy, was correlated with the use of grass clover ley and undersown crops. Groups 2 (larger annuals) and 6 (big seeded annuals) were positioned in the direction of growth form, seed weight, and plant height; however, it is less clear how these groups load in the absence of harrowing. Both groups had the same growth form composition and could both be influenced by the frequency of cereal and grass clover ley in the rotation. The duration of flowering positively correlated with the use of grass clover ley, and the group occurring on basis of this trait was group 5 (rosette-forming annuals with long flowering periods). Group 2 with large annuals occurred opposite and toward the cultivation of cereals in the rotation and displayed shorter flowering periods. Group 3, broadly consisting of annual grasses with indifferent or complex seasonal germination, positioned itself along the SSG axes and opposite the use of winter catch crops. Winter catch crops posed a selection pressure on species more clearly autumn-or spring-germinating. Group 1 and 7, large perennials and perennial grasses, showed no clear association according to their positioning. From the lack of interaction of crop management and the Raunkiaer trait, it is clear that annuals and perennials are not directly affected as traits by the crop management.

Discussion
Evident from the clear clustering ( Figure 1) the weed communities found in this study were regionally specific and influenced by both the local climate and the soil conditions, as well as the local management. The finding that weed species compose locally unique weed communities is coherent with other studies on the effects of local soil type, altitude, and climatic conditions on weed species throughout Europe [23,29,[41][42][43]. However, when analyzing, if region selected for certain traits, the results indicate limited influence (Table 1). With the Latvia region as the exception, we found our first hypothesis not supported. An accumulative effect of 'region' as a filter for trait selection has not been often investigated, although soil type, temperature, and precipitation have been studied by others [24,42]. Fried et al. (2008) [20] stressed the weak effects of location, as subordinate to the stronger selection of crop choice and sowing date.
Although we conducted the study within spring cereals solely, meaning one crop type and spring sowing, the filter of region was still minor. The Raunkiaer trait selection in Latvia toward hemicryptophytes and geophytes weed species, locally presented in high densities of Elytrigia repens (L.) Nevski, Equisetum arvense L., and Taraxacum officinale F. H. Wigg., could be explained by the relative high use of undersowing, ley farming, and fodder crops. These different choices in management reduce the incidences of (soil) disturbance, favoring perennial species [25,44]. The ordination plot based on the weed species of all sites (Figure 1) revealed that the species composition was very different and specific for each region. Together with the weak selection for traits, this suggests that, although the weed communities were different in species composition, they served the same set of traits. Hence, the regional species pool providing different species with similar traits allows other species in each region to use the opportunities provided by spring cereal fields. Another explanation for the minor differences in traits could be the limited traits geared toward geography, such as latitude and temperature, which were not included in the analysis [41,42].
The crop management variables on undersowing and cereal frequency in the rotation were found to affect weed traits (Table 2), confirming our second hypothesis. The undersowing frequency connected with Grime's life strategies selected for weed species with a more diverse strategy. As shown in Figure 4, group 4 consists mostly of species which combined all three strategies (competitive, stress-tolerant, and ruderal). The repetitive undersowing of legumes and grasses selected for species which can deal with the additional competition, not through 'competitiveness' per se, but by being adaptable. Gunton et al. (2011) [17] found an interaction of the Grime's life strategy trait with the crop architecture, where single-stemmed or open-rosette crops selected for more competitive species, showing the opposite movement. The dominance of cereals in the previous years of the rotation selected for species with shorter periods of flowering. However, general cereal cropping in the agronomic context of this study was based on spring sowing varieties (4:1 ratio for spring-autumn in our data), thus pushing for weeds with a shorter flowering period by spring tillage [16] and sowing date [17,24].
The functional groups found in Figures 3 and 4 reflected the minor crop management influences and, thus, formed recognizable functional weed groups primarily on the basis of their inherent traits. These groups reflected a rather ordinary organic spring cereal community behavior. The argument has been made before that organic management recreates the arable conditions for this diversity of weed species, allowing them to resettle in the specific niches for the specialist flora selected for by the agronomic constrains [45].
No trait selection was found by the number of years under organic production. This is remarkable as the literature has discussed the selection for perennial species under organic management, especially in the Nordic regions [6,[46][47][48]. However, most of these studies were in comparison to conventional management, and it appears from our results that this selection might be initial and does not persist over time. The selection for perennial traits observed by previous studies could have taken place within the first few years of organic cropping. Melander et al. (2016) [49] observed that it took 4-5 years of organic farming to build up a perennial weed problem, hinting at early establishment, although it could take up to 9-10 years in some sites. Hofmeijer et al. (2021) [18] found an increase in species numbers and diversity under organic management over time. Hence, the allocation of weed species reflects the regional species pool but is not trait-related.
The low trait selection in this study might be due to the organic management in general. Needless to say, species numbers in organic arable farming are consistently higher than under conventional management [50][51][52][53]. Richness in the species pool increases the functional redundancy of the weed community [54], i.e., multiple species are able to play equivalent roles, presenting similar functional and physical traits. Multiple species filling functional groups can be assumed to react to agronomical management or other environmental filters. The lack of regional selection for traits supports that even though communities consist of different species, the same traits are fully covered in the agronomical niches.
We see our hypothesis further explored in the recent literature, although for the inverse effect. Studies looking into the intensification of agricultural practices have demonstrated a decrease in weed diversity [55], number of species [56], and, ultimately, functional redundancy [57,58]. Together, these declines could lead to loss of resilience. Organic agriculture is arguably less intensive than conventional, lacking the addition of herbicides and inorganic fertilizers, while also being strong weed species filters. In our study, we observed the reverse effect on the clearly observed increase in species diversity [18]. This could mean a higher level of functional redundancy. Hence, organic management could support a certain agility within the weed community against selection of species and could be additionally considered to buffer rather than filter weed traits.

Conclusions
The lack of regional trait selection provides insight in the behavior of different weed species within similar agricultural niches, where the exception of Latvia diverts back to the influence of management rather than location. The observed trait selection by crop diversity and the general effects of organic management found in this study are promising for the approach that weed diversity is able to mitigate the dominance of deleterious species [8,9,59]. Generally, smart and diverse implementation of crop management could stimulate a diverse weed flora, which in turn can form a manageable arable weed community. Additionally, the discussed inverse effect of the implementation of less intense or organic management indicates potential for the prevention of species losses, ecosystem service provision, and mitigation of systemic disturbances, such as changes in agricultural management or environmental factors. Hence, this potential buffering effect requires further exploration. Acknowledgments: The authors would like to thank all the participating farmers, stakeholders, and students for their facilitating role in this study. The authors also thank Friederike de Mol for her statistical advice.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.       Seed germination period determines the match between a weed species life cycle and the growing cycle of a crop and, hence, its ability to escape disturbance posed by farming practices Duration of the flowering period indicates the length of the reproduction phase. Mechanical removal of weed seeds before shedding is an excellent strategy preventing weed seeds from entering the seed bank. Duration of the flowering period also informs on the provision of floral resources for higher trophic levels.