We used bird data collected during 2009–2011 in 47 survey areas situated within an area of 400 × 150 km across Southern Finland (Fig. 1; Supporting material S1). Finland is divided into three zones with different levels of agricultural subsidies, and our study areas were situated in zones A and B, containing 21% and 26% of Finland’s utilised agricultural area, respectively. In Finland, the greening measures are only applied to the three southern-most provinces while the rest are exempted due to a high forest cover. The nationally approved measures include diversification of cultivation with at least three crop plants in farms over 30 ha or two crop plants in farms with 10–30 ha in southern Finland (zones A and B, see Fig. 1) or two crop plants in farms over 10 ha further north (zone C), and an ecological focus area of 5% of field area in farms over 15 ha in zone A (Finlex 2017). The area of semi-natural grasslands or over five years old cultivated grasslands are to be retained within 95% of a national reference value based on the surface area in 2015 (132,000 ha). The requirement of 5% ecological focus areas can be achieved through permanent grassland, different kinds of fallows, short-term coppice and legume crops in farms larger than 15 ha.
Finnish farmland consists of mosaic landscapes where agricultural land is concentrated to patches of farmland surrounded by forest or other land use types. The size of these farmland patches vary from a few to several hundreds of hectares. Because of this mosaic structure, farmland patches contain well-delineated local communities of farmland birds. In this study, the 47 survey areas were delineated based on the extent of individual farmland patches, i.e. farmed areas surrounded by forested areas. Because the farmland patches varied in size, the survey areas also varied in size (mean ± SD = 234.2 ± 263.04, ha, min = 80, max = 1675; Supporting material S1). Survey areas either constituted an entire farmland patch or subsets of larger farmland patches, in which several survey areas were delineated to cover the farmland patches. As far as possible we used data collected in 2010, but when the 2010 data were not available we used data from 2009 or 2011. The total area surveyed was 12,300 ha, of which 10,572 ha represented cultivated land.
Field surveys and data preparation
We surveyed farmland birds using a territory mapping method with three survey rounds during early May to mid-June. The territory mapping was undertaken by a team of experienced field ornithologists, where each team member surveyed slightly over 100 ha farmland during one morning. Beginning at sunrise and ending roughly before noon, all farmland habitats within the survey area were thoroughly searched for farmland birds, which were marked on visit maps paying particular care to simultaneous observations on birds of neighbouring territories. Based on the three visit maps we interpreted the position of individual bird territories, which were subsequently represented as point objects in a GIS layer.
We used official digitized block maps (Integrated Administration and Control System database), supplemented with data on within-block boundaries of different crops based on field notes and aerial photographs. The maps were further supplemented with digitized rivers, major ditches, roads, forests and different open, bushy or wooded islets, as well as farmsteads and other built-up areas. These data were combined in a digitized vector map containing spatially explicit, georeferenced data on all crops cultivated in our study landscapes. During field surveys, we noted all spring-sown cereals on one hand and various types of sown grasslands on the other. Crop types included (i) non-permanent, sown leys for silage and (ii) pastures on arable land, often retained for some years as a part of crop rotation, (iii) spring-sown cereals, (iv) autumn-sown cereals, (v) spring-sown dicots (oilseed rapes, broad bean etc.), (vi) autumn-sown oilseed rape and caraway, (vii) fallows, and (viii) stubble fields (i.e. no-till spring-sown crops, including both cereals and oilseed rape), which together comprised all arable land within each sampling unit. Note that silage leys and pastures on arable land are different as habitats, the former being in intensive cultivation and the latter representing less intensive habitat for birds, existing for several years though not necessarily permanent in a strict sense. Using crop types rather than separating between all crops better reflects functional habitat types for farmland birds (Hiron et al. 2015).
We thereafter established circular sampling units with a radius of 200 m (12.56 ha) across all survey areas. The number of sampling units was maximized within the survey areas given three constraints; (i) at least 50% of the plot had to consist of open farmland, (ii) the sampling units were exclusive, i.e. no spatial overlap among the sampling units was allowed, and (iii) the centroid of a sampling unit had to be on an actively farmed field parcel. We discarded all sampling units containing abandoned farmland, which either consisted of grassy or bushy former fields. Thereafter we counted the number of territories of the 20 bird species breeding in open farmland in this study area (see below for details) for each sampling unit, along with information on land cover data. Following this procedure we obtained 657 sampling units, covering some 8200 ha of farmland across southern Finland. The selected sampling units included observations of 7624 individual bird territories of the selected 20 farmland bird species (Supporting material S2).
Calculation of response variables
In this study we selected all open-farmland bird species, belonging to two ecological groups: (i) 12 species breeding on arable land and along open field boundaries (hereafter termed field nesters), and (ii) 8 bird species breeding primarily amongst bushes and higher herb vegetation in non-crop habitats, such as field boundaries and other edge habitats (hereafter termed non-crop nesters; see Supporting material S2). Thus we explicitly focused on bird species breeding in open farmland and not on farmland species breeding in forest edges or farmsteads (Josefsson et al. 2017). We classified farmland birds into these two groups based on earlier published classifications developed for Finnish conditions (Tiainen and Pakkala 2001). We used total species richness, the species richness of field nesters and non-crop nesters, as well as the diversity of farmland birds [using the inverse Simpson’s index (1/D)], as community response variables. In addition, we analysed abundances of four most common individual species: the field breeders skylark (3936 territories) and the meadow pipit (686 territories), and the non-crop nesters common whitethroat (924 territories) and winchat (443 territories).
Calculation of landscape variables
We calculated the following predictors corresponding to the current greening measures: (i) the proportion of fallows within the sampling units as a proxy for ecological focus areas; (ii) crop type diversity (see below for details) within the sampling units, and (iii) the proportion of grasslands (all types of grasslands except for rotational silage leys, i.e. crop type (i) listed in the section Field surveys and data preparation above) within the sampling units as a proxy for permanent and longer-term grasslands. Fallows are comprised of environmental fallows (i.e., a non-productive field set aside for at least 2 years under an agri-environment scheme, Toivonen et al. 2013), other long-term fallows (combined on 421.0 ha in total, present in 41% of all sampling units), and rotational fallows with stubble or bare-ground fallows (in 8% of all sampling units totalling 100.2 ha). The two fallow types were combined because of the relatively low sample size of rotational fallows.
We included two groups of adjusting landscape variables in this study (Table 1). First, in order to describe the landscape structure in terms of non-crop landscape characteristics, we measured the following predictors: (i) distance to the nearest forest from the centroid of each sampling unit (following Piha et al. (2007); (ii) the proportion of built-up habitats (following Devictor and Jiguet 2007), including all built-up areas and human settlements, but in addition also small islets with trees or bushes found primarily close to roads, settlement and barns; and (iii) an index for field edge density describing the relative amount of non-crop field boundaries in the sampling units. We chose to measure distance to forests instead of proportion forests within the buffers because the former will more accurately describe the landscape context for all sampling units, including those which had no forests within 200 metres from the centroid (40% of all 657 sampling units). We combined islets with trees or bushes with built-up areas because accounting for islets by themselves would not have been statistically feasible, as they constituted a tiny fraction of total land cover. We calculated the relative amount of field edge density by dividing the number of blocks intersecting with the sampling units with the area of arable land within respective sampling units. Thus, low values indicated a low relative density of field boundaries and high values a high relative density of field boundaries.
We constructed our measure of crop diversity using major crop types listed above instead of all available crop classes listed in the block database or stipulated in the national regulation on greening. A classification into major crop types by their sowing timing and taxonomy has been shown to be ecologically more relevant for farmland birds as compared to a more detailed distinction between crops as it is implemented in greening (Hiron et al. 2015). We defined crop type diversity as the inverse Simpson’s index following Palmu et al. (2014), calculated on the proportions of seven groups of crop types recorded spatially explicitly for each sampling unit. Defined in this way our measure of crop diversity was highly correlated with crop type richness (rS = 0.74, P < 0.0001), and equated to crop type richness only if crops had equal proportions within each sampling unit.
Our grassland variable included a variety of grasslands, including truly permanent grasslands and sown but grazed grasslands on arable land but not short-rotational silage leys, which were not grazed and typically kept for one or 2 years. Rotational grazed grasslands are typically kept for several years, and they covered 173.3 ha in total (present in 16% of all sampling units). In Finland, only 1.4% of the field area has been classified as permanent grassland for the whole country (Bascou 2012). Permanent grassland consisted of three different land-use types as specified in the Integrated Administration and Control System database: semi-natural permanent grazed pastures; semi-natural permanent grazed pastures on wetlands, and semi-natural grasslands characterised by herb- and grass-dominated vegetation, but not currently grazed or mown. Together, these permanent grasslands were found in 15% of the 659 sampling units, covering a total area of 110.6 ha.
We found a significant moderate correlation between edge density and crop type diversity (rP = 0.40). To account for collinearity between these predictors we regressed crop type diversity against edge density and used the residuals of this regression (Graham 2003) to measure crop type diversity, while accounting for field size. Thus, all predictor variables showed sufficiently low correlations with each other (rP ≤ 0.30; Graham 2003; Zuur et al. 2009). We log-transformed all predictors to improve linearity and thereafter scaled the predictors to zero mean and unit variance. Scaling the predictors allowed us to assess the individual and joint effects of predictors given the average of other included predictors. We thereafter constructed individual statistical models for total species richness, species diversity (inverse Simpson’s index) and abundance of the most common species by first including the three adjusting landscape variables (distance to forests, proportion of built-up areas within the circular sampling units, and farmland edge density) and the three variables corresponding to greening measures (crop diversity and the proportions of fallows and grasslands within the sampling units). We first defined full models considering all two-way interactions between the three adjusting landscape variables and the three variables corresponding to greening measures, and thereafter we removed all non-significant interactions one by one to simplify the models. All models included the survey area identity nested within a variable describing regional identity (Supporting Material S1), to control for non-independence between sampling units within the same survey area, and for larger-scale autocorrelation between regionally clustered survey areas (Zuur et al. 2009).
Total species richness of farmland birds was analysed using generalised linear mixed models with Poisson error distributions as implemented in the function glmer() available in the library lme4 (Bates et al. 2015), whereas species diversity was analysed using linear mixed models using the function lme() in the library nlme (Pinheiro et al. 2015). Finally, while analysing the abundances of the most common farmland birds we evaluated four alternative error structures (Poisson, zero-inflated Poisson, negative binomial and zero-inflated negative binomial) by comparing model AIC:s using the library glmmADMB (Fournier et al. 2012). Following this procedure, the abundance of skylarks, meadow pipits, and whinchats was modelled using a negative binomial distribution, whereas the common whitethroat was modelled using zero-inflated Poisson error distributions. We verified model assumptions by visual examinations of model residuals, and by confirming that model residuals were not spatially autocorrelated using correlograms as implemented in the library ncf (Bjornstad 2016).