Do the effects of crops on skylark (Alauda arvensis) differ between the field and landscape scales?

Bird data

We used data from the French Breeding Bird Survey (FBBS), a monitoring program in which volunteer skilled ornithologists count birds following a standardized protocol at the same plot, each year since 2001 (Jiguet et al., 2012). Each year, species abundances were recorded in each 2 km × 2 km squares whose centroids were located within a 10 km radius around a locality specified by the volunteer. On each plot, volunteers carried out ten point counts (5 min each, separated by at least 300 m) twice per spring within three weeks around the pivotal date of May 8th to ensure the detection of both early and late breeders. To be validated, counts must be repeated at approximately the same date between years (±7 days) and at dawn (within 1–4 h after sunrise) by a unique observer in the same order. The maximum count per point for the two spring sessions was retained as an indication of point-level species abundance. The counts obtained at the 10 points were summed to give the abundance for the entire square. The FBBS focuses on common birds that regularly breed in France, hence monitors the breeding skylark across the country.

Landscape data

For the identification of landscape factors affecting farmland birds, we carried out a literature review based on studies using data from French and UK breeding bird surveys (Chamberlain & Gregory, 1999; Devictor & Jiguet, 2007; Siriwardena, Cooke & Sutherland, 2012) or studies focusing on single factors, such as roads (Reijnen, Foppen & Meeuwsen, 1996). The variables used in this study are shown in Table 1. These variables were obtained from three national databases: the Land Parcel Identification System (LPIS) 2007–2010, used for the administration of the Common Agricultural Policy (CAP), the BD topo^® from Institut Géographique National, and Corine Land Cover 2006 (Table 2).

These geographic data were integrated into a single database, with priority given to the data with the best spatial resolution: the BD topo^®, followed by the LPIS and finally Corine Land Cover, mostly to cover the gaps in non-agricultural areas.

The French LPIS is not spatially explicit at crop level. It focuses on cropping blocks composed of one or several fields. Each block is a polygon, the attributes of which are the areas covered by the crops within it, with no specific information provided about the location of each crop within the block. It was not, therefore, possible to calculate indicators of crop configuration, and estimates of crop area were imprecise when the blocks intersected with FBBS squares. We resolved this problem by considering the area under a crop within such blocks to be equal to the area of the block within the square multiplied by the proportion of the crop in the block. The LPIS did not distinguish between spring and winter crops. Both winter and spring rapeseed crops were present, but this was of very little consequence because the spring type was largely underrepresented (0.2% of the area under rapeseed in France for 2007–2010, French Ministry of Agriculture, 2009). The “industrial set-aside” category of the LPIS was considered to correspond to rapeseed, based on cross-checking with data for the administrative area (French Ministry of Agriculture, 2009).

The relationships between birds and crops studied here may involve multiple ecological processes: the selection of the squares by skylarks in the year of observation, but also the demographic advantage or disadvantage conferred by the quality of the habitats within these squares. We tried to isolate this last term, to identify long-term effects on the carrying capacity of the landscape regardless of inter-annual crop variations. The four-year study period was too short to take large changes in cropping systems into account. We therefore pooled the data and used average values for both abundance and crop composition, for single squares followed for more than one year between 2007 and 2010.

Sample selection and landscape stratification

We initially selected the FBBS squares for 2007–2010, as LPIS data were available for the corresponding period. We then restricted the study to agricultural contexts, by selecting squares with more than 50% of their area under agriculture according to the LPIS.

According to Whittingham et al. (2007), habitat-density associations may be dependent on landscape type (e.g., openfield vs. grassland), bird density, and geographic context, with this last factor being the most important. We therefore stratified the FBBS squares as a function of landscape type and ecological region, as given by the digital map of European ecological regions (DMEER version 2003) from the European Environment Agency.

Arable crops, grass and trees in agricultural areas strongly influenced the abundance of skylarks (Chamberlain & Gregory, 1999; Robinson, Wilson & Crick, 2001). We therefore stratified the FBBS squares according to the grass and tree factors, with an indirect inclusion of arable crops, due to high correlation with grass (−0.87). FBBS squares were classified according to their position on either side of the curve defined by an equation, the parameters of which were selected so as to give equal weightings to both criteria according to their different ranges of variation, and to obtain two well-balanced groups: (1)${\displaystyle \sqrt{{\left(0.75*\text{grass area}\right)}^2+{\left(2*\text{tree area}\right)}^2}=100.}$

The group below the curve was called “open-field,” and the group above was referred to as “grassland” (Fig. 1).

The European ecological regions data incorporate information about climate, flora and topography. Some of these regions contain only marginal parts of France. We therefore retained only the “Southern temperate Atlantic” and “Western European broadleaf forest” regions, which together include 97% of the previously selected FBBS squares, and which split France into two roughly equal parts, corresponding to the West and the East (Fig. 2). The limit between ecological regions was approximated on the basis of administrative zones. Cross-referencing of the two stratifications yielded four groups: Openfield East; Openfield West; Grassland East; Grassland West.

Once the squares had been assigned to these four groups, we eliminated those considered potentially unsuitable for the crop of interest, by retaining the squares in which its area was non-zero. All squares were considered potentially suitable for skylark according to the large range of this species and the presence of favorable agricultural habitats. We eventually obtained eight samples, corresponding to four groups ∗ two factors of interest (the rapeseed and maize areas; Table 3).

Statistical analysis

Crop compositions are constrained by agronomic rules. For example, rapeseed is systematically grown in rotations with cereals, to the benefit of both species, as this approach improves weed and pest management. Successive rapeseed crops are usually separated by at least three years in the rotation (e.g., rapeseed followed by wheat and barley before a return to rapeseed). Even in landscapes dominated by such a short rotation, the area under rapeseed therefore cannot exceed one third of the total area under annual crops. By contrast, maize can be cultivated either in rotations or as a monoculture; its area is therefore not limited. These structural relationships may make it difficult to establish isolated responses to individual crops. Confounding effects may occur between crops, or between crops and the total area under annual crops or grass. Before investigating responses, we checked the correlations between these variables for each square sample.

We estimated the relationships between skylark abundance and rapeseed of maize areas for the various squares according to an information theoretic approach. We first built three nested general linear models, where abundance depended on: (1) an autocovariate to minimize the effects of the spatial autocorrelation of abundances (Augustin, Mugglestone & Buckland, 1996), (2) the autocovariate, and the set of fixed elements listed in Table 1, but without the forest area (i.e., used of the whole set would generate collinearity due to the sum of areas being equal to 400 ha), (3) the autocovariate, the set of fixed elements, and the tested factor, i.e., rapeseed or maize area. We used negative binomial regressions due to overdispersion of the count data. Then we considered for each model and all the possible combinations of predictors. The resulting models were compared with Akaike information criterion (AICc with small sample size correction), and we used model-averaging to calculate parameter estimates and 95% confidence intervals for the top models (ΔAICc < 2). The influence of sampling on the results was assessed by repeating the analysis 100 times on two third of the data.

Implementation

Data were input and managed with the PostgreSQL 9.2.4 relational database server and its spatial extension PostGIS. The choice of this software was based on its ability to handle entire national databases. The statistical analyses were performed with R version 3.0.1, and the ‘spdep’ ‘MASS’ and ‘MuMin’ packages.

Origins of the discrepancies between the field and landscape scales

Our results suggest that field-scale studies do not take agronomic and ecological mechanisms operating at larger spatial and temporal scales into account, which is consistent with ecological theory: upscaling involves moving to higher levels of biological organization and larger spatiotemporal extents. This increases complexity and tends to decrease the generality of ecological findings (Lawton, 1999). Diverse biotic and abiotic interactions within the landscape may exacerbate or mitigate impacts. For example, the benefit of organic farming is smaller at farm level than at field level according to Bengtsson, Ahnström & Weibull (2005).

In our case, the discrepancies between field and landscape may be accounted for by mitigation due to the diluted impact of the crop in landscapes to which other habitats make a major contribution. For rapeseed, the constraints on crop rotation have a strong mitigating effect. Rapeseed cannot account for more than one third of the total area under annual crops, and is associated with other more favorable crops, such as cereals. This threshold probably mitigates all the potential unfavorable effects observed at the field scale. Maize crops are not subject to such constraints and can dominate the landscape, leading to an absence of such mitigating effects. Furthermore, fixed landscape elements have a greater weighting than crops.

Mitigating effects, such as those described above, are consistent with the hypothesis of simple additive effects of crop areas in the landscape during the breeding season. They may account for absence of expected effects, but not opposite effects, such as that of rapeseed in one context. We can explain this last case only by abandoning the hypothesis of simple additive effects, and considering more complex processes. This reasoning is more speculative and we suggest here three hypothetical processes compatible with our results:

1. “Remote” effects of the crop extending beyond the crop: Rapeseed crops may interact with neighboring habitats because this crop provides more insects than other crops, as it is more attractive to herbivorous insects and pollinators (Hebinger, 2013). However, the scenario in which rapeseed acts as a source of food spilling over into neighboring fields remains theoretical. Studies of the food resources for birds associated with crops (Stoate, Moreby & Szczur, 1998; Cléré & Bretagnolle, 2001; Moreby & Southway, 2002) are scarce and seldom comparable, due to methodological differences.

2. Delayed effects from winter to the breeding season. Rapeseed is a favorable crop for skylark in winter. Powolny (2012) observed it was the most selected crop with alfafa, as its leaves provided a useful source of food during this critical period. For resident populations, this process may have visible effects during the breeding season. A beneficial association between set-aside in winter and bird density in the same area in spring was observed by Whittingham et al. (2005) for a resident passerine, the yellowhammer (Emberiza citrinella). In line with this hypothesis, the positive response to rapeseed may be accounted for by cumulative effects throughout the year, whereas field-scale studies generally focus on partial effects during the breeding season. This mechanism depends on the migratory behavior of the skylarks. A resident population would benefit from rapeseed all over the year, whereas a migratory would not. Resident and migrant populations are poorly delimited in continental Western Europe and a mixture of resident and migratory behavior was observed in one population from the Netherlands (Hegemann et al., 2010).

3. Effect of the crop as a function of its area, with positive effects in small areas becoming negative with increasing area size. Quadratic responses of this type may be accounted for by ecological processes, such as ‘landscape complementation’ (Dunning, Danielson & Pulliam, 1992), in small areas, followed by a detrimental loss of appropriate habitats when the crop area exceeds a given threshold. This scenario may be rendered more complex by adding a temporal dimension, as complementation between crops may occur during the breeding season. The growth of rapeseed makes the vegetation structure unsuitable for nesting (Wilson et al., 1997), but some studies suggest rapeseed is more suitable in the early season than later. Eraud & Boutin (2002) found that skylark density in rapeseed decreased throughout the breeding season, and a positive association between rapeseed and skylark was observed in some cases in early breeding season (Chamberlain & Gregory, 1999) or with underdeveloped rapeseed (Whittingham, Wilson & Donald, 2003). This could cause skylark to shift to other more favorable crops, as observed in the case of winter wheat (Chamberlain et al., 1999; Donald et al., 2002; Hiron, Berg & Pärt, 2012). According to this hypothesis, the area of rapeseed is less important than crop diversity allowing the succession of suitable crop mosaics on the landscape during the breeding season. Habitat diversity on 1 km² lanscapes, however, was found to have a negative effect on skylark abundance in UK lowlands (Chamberlain et al., 1999; Pickett & Siriwardena, 2011). Facing this, Chamberlain et al. (1999) questioned the equal weight given to each crop in their diversity index. The solution probably lies in the development of crop diversity indices taking into account the growth dynamics of the crops and not their simple nature.

In conclusion, the effects of crops were not simply additive when switching from field to landscape, but the underlying causal mechanisms remain unclear. If we are to understand such processes, we must take into account subtle interactions between crops, and between crops and fixed elements, and further investigations of the shape of the responses are required.

Importance of context

We observed contrasts between ecological regions (for rapeseed) and between openfield and grassland contexts (for maize). These sources of variability were expected, but their true origins remain unclear. Possible underlying mechanisms were discussed by Whittingham et al. (2007) and Schaub et al. (2011). The reflections of these authors call into question the tendency to oversimplify objects for conceptual reasons (lack of prior knowledge of their variability) or practical reasons (data availability). Indeed, regional differences may indicate that the bird populations evolved differently, with different habitat preferences (unlikely according to Whittingham et al. (2007)), or types of migratory behavior, with consequences mentioned here above. Regional differences may also result from ecological or agronomic gradients that are unknown or cannot be described at the required resolution, e.g., agricultural practices (pesticide use, previous crop, soil tillage) resulting in differences in a given crop between regions, from the bird’s point of view. For example, Shrubb (1988) showed that, in winter, lapwings (Vanellus vanellus) could differentiate between wheat following rapeseed and wheat following wheat, due to the stimulatory effect on the soil fauna of the organic manure applied after rapeseed in the cropping systems of West Sussex. This example highlights the complexity of the agronomic processes potentially affecting crops and subsequent species-habitat associations. The spatial variation of the responses raises a practical problem. In a perspective of applied research, the question is not so much determining whether or not there is an effect, as identifying the conditions and locations in which this effect is expressed. However, it was not our goal. We aimed instead simply to highlight differences, revealing gradients operating at large scales and the influence of some key elements of the landscape.

Consequences for management

Our results concerning field/landscape inconsistencies and variations with local and regional context may reasonably be assumed to apply to situations other than that of the effect of spatiotemporal crop allocation on skylark. We consider here implications for future studies on both sides of decision-making and local management. Our findings call into question the analytical approaches aggregating the effects of individual habitats in methods for assessing and planning land use over large scales (e.g., life cycle assessment (Geyer et al., 2010), land use scenarios (Brandt & Glemnitz, 2014)). We need to refine the models to catch possible interactions and non-linear responses. For this purpose, field- and landscape-scale studies are complementary and can be put together in both top-down and bottom-up directions, by constructing a hypothesis at one scale and verifying it at the other. We also need to accept that the explanation “the effects are context-dependent” is unsatisfactory in a perspective of applied research for rural extension. The adaptation of management measures advocated by some authors (Whittingham et al., 2007; Schaub et al., 2011) implies an ability to define the boundaries of contexts precisely. It is easy to recycle existing administrative entities, but this may be difficult to justify if we are focusing on the bird’s viewpoint. We still need to open the fuzzy box of “context,” with empirical (mapping the responses) or mechanistic (identifying the underlying causes) methods.

These are programmatic rather than practical considerations. We should stress that our study provides no evidence directly useful for advice and rural extension. Skylark abundance was used as a biological indicator, not as an indicator for management regardless of geographic context. Moreover, if we consider crop allocation as a means of improving the status of farmland birds, the effect sizes obtained were so small that the gain would be minimal for a large range of possible losses (crop allocation suboptimal for gross margin, work organization, agronomy, etc.). By contrast, responses were general and so imprecise that local improvements based on local diagnosis could not be excluded. Is it better to prescribe the same remedy for all patients, on the basis of imprecise models, or to take time the time to examine each case separately? This debate is beyond the scope of agronomy and ecology.