Do the EU ’ s Common agricultural policy funds negatively affect the diversity of farmland birds? Evidence from Slovenia

(1)

Agriculture, Ecosystems and Environment 306 (2021) 107200

Available online 8 November 2020

(http://creativecommons.org/licenses/by-nc-nd/4.0/).

Do the EU ’ s Common agricultural policy funds negatively affect the diversity of farmland birds? Evidence from Slovenia

Tanja ˇ Sumrada

^a^,

*, Primo ˇ z Kmecl

^b

, Emil Erjavec

^a

aUniversity of Ljubljana, Biotechnical Faculty, Jamnikarjeva ulica 101, SI-1000 Ljubljana, Slovenia

bDOPPS – BirdLife Slovenia, Trˇzaˇska cesta 2, SI-1000 Ljubljana, Slovenia

A R T I C L E I N F O Keywords:

Farmland birds Land use change Common agricultural policy Agri-environment schemes Natura 2000

Boosted regression trees

A B S T R A C T

The paper investigates the relative influence of landscape characteristics, production intensity and the EU’s Common agricultural policy interventions on the diversity of farmland birds. For this purpose, data from the Farmland Bird Monitoring Scheme in Slovenia and high spatial resolution data from the national agricultural databases in the period 2008–2019 were analysed with the Boosted Regression Trees (BRT). The diversity of farmland birds was found to be highest in open, diversely cropped and extensively to moderately intensively managed landscapes in Natura 2000 sites where farm holdings were allocated a low average amount of both direct payments and payments for agri-environmental measures (AEM) and organic farming (OF). Furthermore, the highest diversity of the subgroup of grassland specialists was associated with very open and extensively managed grassland landscapes with low stocking density (<0.7 LU/ha). By contrast, the diversity of habitat generalists was highest in heterogeneous landscapes with a high diversity of land-use types, measured at the broader spatial scale. Areas with a higher allocation of direct payments and payments for AEM and OF were associated with lower farmland bird diversity, whereas high diversity was found in Natura 2000 sites and in some areas with natural constraints (LFA). Agri-environmental measures and the “Greening” measures had a negligible relative influence on bird diversity, possibly due to ineffective implementation and low uptake by beneficiaries.

The intensification of production, particularly in the beef and dairy sectors, which has been supported by the Common agricultural policy direct payments, and forest succession in marginal areas were identified as the potential key drivers of the recent farmland biodiversity loss in Slovenia. The future CAP income support schemes should be redesigned to ensure at least neutral if not positive overall effects on farmland biodiversity by gradual phasing-out of references to (historic) production levels, increased conditionality and more effective voluntary agri-environmental measures.

1. Introduction

As shown by the European Farmland Bird Indicator (FBI), populations of farmland bird species in Europe have undergone a widespread and rapid decline by as much as 57 % since 1980 (Gregory et al., 2019; PECBMS, 2019). Recent evidence attributes much of this biodiversity loss to agricultural intensification (Busch et al., 2020; Donald et al., 2001; Reif, 2013; Reif and Vermouzek, 2019), which has been related to higher pesticide and fertilizer use, changes in crop types and structure, altered mowing and grazing regimes and the introduction of new farming technologies (Buckwell and Armstrong-Brown, 2004).

Furthermore, land-use change has often led to the reduction of habitat heterogeneity at both landscape and field level, including the loss of

mosaic mixed farming and non-cropped (marginal) habitats such as hedgerows and field margins (Benton et al., 2003). All these factors can adversely affect the availability of food resources, nesting sites and other structures used by farmland birds as well as increase their direct mor- tality (Butler et al., 2010; Newton, 2004).

Another significant driver of farmland bird population declines has been the abandonment of agriculture and consequent overgrowth of areas with shrub and tree vegetation, which causes the gradual disap- pearance of species associated with open and ecotone habitats (Kmecl and Denac, 2018; Sirami et al., 2008; Zakkak et al., 2015). This development has been particularly notable in the marginal areas of Southern, Eastern and Northern Europe (Kuemmerle et al., 2016; Levers et al., 2018).

* Corresponding author.

E-mail address: tanja.sumrada@bf.uni-lj.si (T. ˇSumrada).

Contents lists available at ScienceDirect

Agriculture, Ecosystems and Environment

journal homepage: www.elsevier.com/locate/agee

https://doi.org/10.1016/j.agee.2020.107200

Received 4 November 2019; Received in revised form 21 August 2020; Accepted 26 September 2020

(2)

In addition to technological and socio-economic changes (van Vliet et al., 2015), agricultural intensification has been accelerated by the European Common Agricultural Policy (CAP) through income policy interventions in the form of market-price support, coupled and later decoupled direct payments, whereas it only partly succeeded in miti- gating land abandonment (Stoate et al., 2009). Despite much public debate, relatively few scientific studies have directly explored the links between CAP support and biodiversity decline (Hodge et al., 2015;

Uthes and Matzdorf, 2013). Budgetary transfers to farm holdings are rarely considered in farmland bird research, although they have a significant impact on farm management and such knowledge is essential for designing appropriate policy interventions (Mattison and Norris, 2005).

Since the Fischler reform in 2003, the majority of CAP income support is distributed in the form of area-based direct payments, which are said to be “decoupled” from production output and are conditional upon the implementation of minimal environmental standards enforced through the “Cross-compliance” system (Anania and Pupo D’Andrea, 2015). As opposed to market-price support and coupled direct payments, which increase effective market price and thus stimulate producers to invest more inputs (e.g. fertilizers and land) in order to gain higher outputs, the introduction of decoupled payments was expected to decrease overall production and stimulate less input-intensive agricultural practices (Matthews, 2013).

However, the connection between income support, land use practices and the consequent biodiversity response has proved to be complex and highly variable between regions (Brady et al., 2009; Doxa et al., 2012;

Stoate et al., 2009). In areas with intensive agriculture, decoupling seems to have often stimulated shifts in production systems towards more profitable crops with negligible overall input reduction and in some cases even increased production intensity (Vollaro, 2012). In other areas, transitions from extensive traditional farming to more specialized (livestock) systems have taken place with significant effects on landscape and agricultural practices (Ribeiro et al., 2018). On the other hand, decreased incentivisation of production has resulted in agricultural extensification (and to some extent land abandonment) in marginal regions (Acs et al., 2010). Here, landscape homogenization with increasing shares of grasslands is taking place due to the requirement of keeping at least minimum production levels in order for the land to be eligible for support (Brady et al., 2009).

There are several policy measures with the potential to mitigate the negative biodiversity effects of the above instruments, such as the

“Cross-compliance” system and other eligibility conditions, voluntary agri-environmental measures (AEM) and additional income support to marginal areas (LFA) (Matthews, 2013). Moreover, EU Member States have designated the Natura 2000 network of protected areas to conserve the most threatened species and habitat types (Orlikowska et al., 2016).

Higher proportions of agricultural land enrolled into AEM or designated as Natura 2000 sites have both been associated with slower population decreases of some farmland birds (Gamero et al., 2017). However, most evaluations have revealed that these instruments have not been sufficient to halt the farmland biodiversity decline (e.g. European Commis- sion, 2019), including the recent “Greening” of the direct payments system (Pe’er et al., 2014). A better understanding of policy-induced and other drivers of farmland biodiversity loss is thus necessary (Ollerer, ¨ 2013; Stevens et al., 2007), especially in view of increasing Member States competences in the programming and implementation of the CAP (Henke et al., 2018).

Slovenia is a Central European country renowned for its high biodiversity (Perko et al., 2020). Over a third of the country’s territory is designated as Natura 2000, the highest proportion of all EU Member States (European Commission, 2018a). However, several indicators suggest a recent deterioration of agricultural ecosystems, including the Slovenian farmland bird index (FBI), which showed a significant overall decline (-21.9 %) in the period 2008–2018, with populations of grassland specialists declining even faster (-40.8 %) (Kmecl and ˇSumrada, 2018a).

The reasons for this development and the impact of policy support on farmland biodiversity in Slovenia have rarely been explored (Sla- be-Erker et al., 2019). Over the last decades, some regions seem to have experienced considerable land-use change in terms of forest overgrowth, altered agricultural practices and changes in landscape structure (Kaligariˇc and Ivajnˇsiˇc, 2014; Ogorevc and Slabe-Erker, 2018; Perko et al., 2020), possibly precipitating the decline of farmland bird populations (Ivajnˇsiˇc et al., 2020; Kmecl and Denac, 2018; Tome et al., 2020). On the other hand, the effectiveness of the CAP instruments aiming at biodiversity conservation, such as the AEM and “Greening”

measures, might have been limited by low budgetary allocation to targeted schemes, only partially effective design and insufficient spatial targeting (Kaligariˇc et al., 2019; ˇSumrada et al., 2020).

Our research aims to use high spatial resolution data to investigate the relative importance of landscape characteristics, production intensity and CAP interventions, as well as potential interactions between these factors, for the diversity of farmland birds. Furthermore, we aim to explore whether there are any differences in the diversity response along the landscape, production intensity and policy implementation gradi- ents. Data from the Slovenian farmland bird monitoring scheme in the period 2008–2019 were used to study the diversity of farmland specialists, whereas analysis of generalist species was conducted for com- parison purposes. To investigate the reasons for their particularly high population declines, a subgroup of farmland bird species that are strongly associated with grassland habitats was analysed as well (Kmecl and ˇSumrada, 2018). Given the complex nature of different drivers, an exploratory approach with boosted regression trees (BRT) was used, which combines the regression tree algorithm and machine learning (Elith et al., 2008).

2. Materials and methods 2.1. Study area

Slovenia is located at the junction of the Alps, the Dinaric Mountains, the Adriatic coast and the Pannonian Basin. Most of the territory is covered by forests (62 % in 2017), while agricultural landscape takes up 31 % (Fig. 1).

Grasslands are the predominant agricultural land-use type (19 %), with the largest shares in the Alps, Dinaric lowlands and eastern hilly areas. Grass- land communities are highly diverse (Fig. 2). They include alpine grasslands and pastures occurring on carbonate bedrock, distinct dry grasslands in the sub-Mediterranean region and several grassland types in continental Slovenia. The latter include wet grasslands on nutrient-poor soils, dry grasslands and mesic and manured grasslands (Perko et al., 2020).

Arable land (10 %) is concentrated in the lowlands, particularly in the Pannonian plain (Perko et al., 2020). Changes in the sowing structure were relatively small in the study period. In 2017, about 55 % of the arable land was used for cereals, with a predominant share of maize for grain and wheat, and 32 % for green fodder (predominantly silage maize and fodder crops such as grass, clover and alfalfa) (Bedraˇc et al., 2018).

Orchards and vineyards (3% of the territory in 2017) are relatively evenly distributed in most regions, following the once-widespread extensive mosaic landscape (Fig. 2). However, there is a somewhat higher con- centration of permanent crops in the low Pannonian and Mediterranean hills (Perko et al., 2020).

Areas with natural constraints (formerly known as less-favoured areas, LFA) were identified on 86 % of the country’s territory, mainly due to high altitudes and slopes that somewhat limit the potential for agricultural intensification (MAFF, 2015).

Unlike in most communist countries, but similarly to other ex- Yugoslav republics and Poland, only a small share of agricultural land (<15 %) in Slovenia was collectivised after 1945. Most private farms were until 1990 not allowed to manage more than 10 ha of cultivated land, which considerably slowed down agricultural development and largely maintained small-scale farming and the mosaic agricultural landscape (Jepsen et al., 2015, Suppl.).

(3)

After the break-up of Yugoslavia in the 1990s, Slovenia introduced a CAP-like protectionist agricultural policy. Relatively high market price support, increasing levels of production-coupled subsidies for some sectors, particularly beef and dairy, and investment measures created a favourable economic environment that sped up the intensification of production and changes in production technologies (Erjavec et al., 2003). Upon EU accession in 2004 and during the subsequent 2007 and 2013 CAP reforms, Slovenian decision-makers chose direct payment schemes that maximised product-related and historical payments (Erjavec et al., 2015; European Commission, 2015). To a large extent, this maintained the differences in the distribution of area-based payments between producers and potentially further influenced the choice of production orientation and intensity of use. The formerly highly uniform type of small-scale mixed farming has thus diversified into more specialised production types, whereas the traditional farming system has probably declined the most (Erjavec et al., 2006).

For many producers, however, the new economic situation was not stimulative enough to maintain marginal resources. Combined with other structural and demographic drivers, this has led to the development of extensive livestock production and in many cases, the gradual abandonment of farming (Jepsen et al., 2015, Suppl.). This happened despite considerable additional support to areas with natural and other constraints, which has been introduced already before the EU accession (Erjavec et al., 2003), and today still takes up about 10 % of the annual agricultural policy budget in Slovenia (ˇSumrada et al., 2020, Suppl.).

2.2. Bird data

Data for the calculation of species diversity were obtained from the farmland bird monitoring scheme in Slovenia (Kmecl and ˇSumrada, 2018), which is based on annual line transect surveys (Bibby et al., 2000). Experienced ornithologists walked predefined 2-km transects in the early morning twice a year. The first count took place from 1 April to 5 May and the second from 6 May to 30 June. In the subsequent

analyses, we used the maximum of the two counts. All bird species seen or heard were counted within two belts (0− 50 m and >50 m perpen- dicular to the route). The survey unit was a pair, which predominantly meant a singing or otherwise observed male, but also observed female, a group of fledglings or active nest (Koskimies and V¨ais¨anen, 1988). Based on a 2 ×2 km whole-country grid, transects were positioned only in squares with over 40 % of the area covered with agricultural land and selected to be approximately evenly distributed across Slovenian farmland (Fig. 1).

We divided bird species into three groups. Farmland specialists (29 species) are the indicator species used in the Slovenian Farmland Bird Index (FBI) (Table 1). The selection of these species was based on expert opinion, supported by the literature review (Kmecl and ˇSumrada, 2018).

Grassland specialists (9 species) were identified among the farmland bird species according to their habitat preference by using redundancy analysis (RDA) (for details see Kmecl and Sumrada, 2018). Generalists ˇ were defined as the 10 most abundant species according to the farmland bird monitoring surveys, which are not farmland specialists and also not predominately connected to a single habitat according to the national habitat selection studies (Miheliˇc et al., 2019).

In the period 2009–2018 altogether 145 transects were censused. For each transect, the Shannon diversity index (H’) of each group of species was calculated for each year when the transect route was surveyed.

Since some transects were not censused each year in the study period, a total of 885 diversity indices for each group of species were thus used as statistical units in further regression analysis.

2.3. Predictor variables

Based on literature review and available spatial databases, potential predictor variables were defined and divided into four groups: (1) land use and landscape diversity, (2) intensity of agricultural production, (3) extent of the area where environmental measures were implemented and (4) average amount of payments for crucial groups of CAP Fig. 1. Land use in Slovenia in 2014 with the location of the transects used in the Slovenian farmland bird monitoring scheme (land use data source: National database of agricultural and forest land use, Ministry of Agriculture, Forestry and Food of the Republic of Slovenia).

(4)

interventions.

Data originated from three sources: the Integrated administration and control system (IACS) database of Slovenia, the National database of agricultural and forest land use (AFLU) and spatial layers of Natura 2000 sites. IACS and AFLU databases were set up to facilitate the distribution of CAP funds and are managed by the Ministry of Agriculture, Forestry and Food. The IACS dataset includes annual spatially referenced data on agricultural land units belonging to farm holdings that have applied for CAP funds, their enrollment in various agricultural measures and crops cultivated in spring of the respective year. For each farm holding, annual data on livestock and agricultural subsidies were obtained as well. In the AFLU database, land-use types of land units are identified with the computer-aided interpretation of orthophotos and other spatial data.

Each year about a third of the country’s territory is reinterpreted (Lisec et al., 2013). As the boundaries of the Natura 2000 sites were amended twice during the study period, we prepared annual datasets following the Natura 2000 regulation in force in the previous year.

The values of the potential predictor variables were calculated using ArcGIS (ESRI, version 10.2.2) for transect buffer zones (<200 m of the transect paths) for each year (2008–2017). BRT models’ explaining and prediction power are not affected by multicollinearity (Elith et al., 2008). However, redundant (strongly correlated) predictors do make the interpretation of predictors’ importance more difficult since both predictors in a correlated pair can appear in the list of important predictors.

That is why we chose to exclude one predictor in highly correlated pairs (r>0.72, Spearman rho correlation coefficient), which was based on ecological reasoning (Appendix A) (Dormann et al., 2013). The final set for further analyses contained 18 variables (Table 2).

Landscape diversity was described with two variables, which were calculated as the Shannon diversity index (H’). Land-use diversity (HUSE) measured the overall heterogeneity of the landscape and was derived from 25 land-use types used in the AFLU database (Appendix A).

The heterogeneity of agricultural land-use was measured by the diversity of crop types (including grasslands) (HMOS), which are moni- tored in the IACS database (Appendix A). Among the individual land-use types, the extent of buffer area covered with permanent grasslands (GRA), maize (MAI) and woody vegetation (FOR) (AFLU database) were singled out for their potential importance as indicators of bird-habitat association and as structural elements in the agricultural landscape (Hinsley and Bellamy, 2000; Reif and Hanzelka, 2016). The extent of buffer area cultivated by all cereals (CER) was also considered but was later excluded due to high correlation with other variables (see Ap- pendix A).

Existing spatial databases in Slovenia do not include indicators that would allow for direct monitoring of production intensity. However, the intensity of animal production can be approximated with stocking density (LU), which was calculated as the weighted average of livestock units per transect buffer area (IACS database). Since 59 % of farm holdings in Slovenia were livestock-breeding or mixed farms in 2016 (SURS, 2019), we concluded that this could be a sufficient overall proxy.

The variables measuring the implementation of environmental measures were calculated as the extent of buffer area under the key agri- environmental and conservation policy instruments. Agri- environmental schemes were grouped according to the targeted production systems (i.e. grasslands and landscape features AEMLA, arable land AEMAR and permanent crops AEMPC), while 4 schemes targeting Fig. 2.Typical examples of agricultural landscapes in Slovenia: a) extensively managed mosaic farmland in hilly areas, b) extensively grazed and open sub- Mediterranean dry grassland areas, c) lowland areas and karst floodplains with wet grasslands, d) intensively managed lowlands. Credits: Tomaˇz Janˇcar (a) and Primoˇz Kmecl (b, c, d).

(5)

Table 1

Classification of bird species into groups, their frequency and average abundance on the surveyed transects (F - farmland birds; G - grassland species, R - generalists).

Species English name Groups of species Frequency (%) No. of pairs

Acrocephalus palustris Marsh Warbler F 34.2 1.2

Alauda arvensis Skylark F, G 37.5 2.3

Anthus trivialis Tree Pipit F, G 24.7 1.2

Carduelis chloris European Greenfinch R 77.1 2.4

Carduelis cannabina Eurasian Linnet F, G 23.1 0.5

Carduelis carduelis European Goldfinch F 69.3 1.8

Columba oenas Stock Dove F 18.8 0.5

Columba palumbus Woodpigeon F 67.2 2.0

Corvus cornix Hooded Crow R 90.7 7.1

Cyanistes caeruleus Eurasian Blue Tit R 67.8 1.7

Dendrocopos major Great Spotted Woodpecker R 71.1 1.4

Emberiza calandra Corn Bunting F, G 22.6 0.9

Emberiza cirlus Cirl Bunting F 18.8 0.8

Emberiza citrinella Yellowhammer F 59.8 2.6

Falco tinnunculus Common Kestrel F 57.4 0.9

Fringilla coelebs Common Chaffinch R 91.3 6.1

Galerida cristata Crested Lark F 17.7 0.4

Hirundo rustica Barn Swallow F 86.2 6.2

Jynx torquilla Eurasian Wryneck F 49.9 1.0

Lanius collurio Red-backed Shrike F, G 72.3 2.0

Lullula arborea Woodlark F, G 20.2 0.8

Luscinia megarhynchos Common Nightingale F 33.7 2.0

Motacilla alba White Wagtail R 81.9 2.0

Motacilla flava Western Yellow Wagtail F 12.7 0.6

Parus major Great Tit R 93.4 6.1

Passer montanus Eurasian Tree Sparrow F 77.9 5.8

Phoenicurus phoenicurus Common Redstart F 18.4 0.4

Pica pica Eurasian Magpie R 61.2 1.4

Picus viridis European Green Woodpecker F 45.6 0.8

Saxicola rubetra Whinchat F, G 11.9 0.6

Saxicola torquatus European Stonechat F 58.6 1.4

Serinus serinus European Serin F 69.9 2.2

Streptopelia turtur European Turtle Dove F 18.1 0.3

Sturnus vulgaris Common Starling F 91.5 10.1

Sylvia atricapilla Eurasian Blackcap R 99.2 10.6

Sylvia communis Common Whitethroat F, G 42.9 1.3

Turdus merula Eurasian Blackbird R 91.1 6.6

Upupa epops Eurasian Hoopoe F, G 14.5 0.2

Vanellus vanellus Northern Lapwing F 14.7 0.6

Table 2

Predictor and response variables used in the BRT models, calculated for buffer areas around transect routes, which were surveyed in 2008-2018 (% - percentage of buffer area; H - Shannon diversity index; EUR/ha and LU/ha - weighted average amount of payments or livestock units, which was calculated as P =(1/A)*(a1*x1 + a2*x2 +…), where A is buffer zone area and ai is area with xi value).

Abbrev. Description Unit Average SD Min Max

Land use and landscape diversity

FOR Area covered by woody vegetation % 13.6 11.46 0.0 60.8

GRA Area covered by permanent grassland % 35.1 24.59 0.0 97.8

MAI Area sown with maize % 17.4 16.74 0.0 94.4

HUSE Land use diversity H 1.3 0.46 0.1 2.4

HMOS Crop diversity H 1.3 0.66 0.0 2.5

Intensity of production

LU Stocking density LU/ha 0.71 0.42 0.0 2.9

Extent of the area with environmental measures

ESPG “Greening” measure Permanent grasslands % 3.4 12.57 0.0 76.6

GRE “Greening” measures Crop diversification and Ecological focus area % 9.7 22.57 0.0 99.7

OF Organic farming % 4.4 11.60 0.0 88.9

AEMNA AEM for Natura 2000 grassland species and habitat types % 1.8 7.27 0.0 60.4

AEMLA AEM for landscape conservation and extensive grasslands % 4.9 8.43 0.0 44.6

AEMAR AEM on arable land % 13.4 21.42 0.0 90.1

AEMPC AEM in permanent crops % 2.4 6.75 0.0 52.1

N2000 Natura 2000 sites % 34.1 42.65 0.0 100.0

Payments for key groups of agricultural policy measures

DP Direct payments EUR/ha 290.2 103.97 70.1 522.7

PLFA Payments to areas facing natural or other constraints EUR/ha 68.9 53.60 0.0 221.6

PENV Agri-environmental and Organic farming payments EUR/ha 104.5 88.92 0.0 403.2

Other predictor variables

YEAR Survey year 2008 2018

Response variables

SDIF Farmland bird species diversity H 2.081 0.300 0.645 2.699

SDIG Grassland bird species diversity H 0.713 0.537 0.000 1.872

SDIR Generalist bird species diversity H 1.793 0.283 0.349 2.239

(6)

priority species and habitat types in Natura 2000 sites were defined as a separate group (AECMNA) (Appendix C). Organic farming (OF) was identified as a separate measure due to its specific production system (MAFF, 2015). The elements of the new “Greening” direct payment scheme were divided into two groups. Environmentally sensitive permanent grasslands (ESPG) limits ploughing of grasslands in Natura 2000 sites, whereas measures “Crop diversification” and “Ecological focus areas” (GRE) apply to arable land on farms that manage >10 ha and >15 ha of cultivated land, respectively (Official Gazette of RS, 2019).

Because “Greening” measures were first introduced in 2015, zero values were applied to buffer areas in the preceding years. The extent of the Natura 2000 area was narrowed down to sites with at least one priority farmland species or habitat type (N2000), as this is the basis for the targeted management actions.

The weighted average amount of payments received by farm holdings in the buffer areas was calculated for three groups of agricultural policy instruments. Farm income support from the first pillar of the CAP was aggregated in the Direct payments (DP) variable, which included the basic payment scheme, coupled production support schemes and since 2015 also payments for practices beneficial for the climate and the environment (“Greening”), payments for areas with natural constraints and payments for young and small farmers (Official Gazette of RS, 2019). Payments to areas facing natural or other specific constraints (PLFA) are additional income support in the framework of the rural development policy, which is mainly directed to farms in hilly and mountain areas and imposes no specific management requirements (MAFF, 2015). On the other hand, Agri-environment and Organic farming payments (PENV) include various voluntary measures for sup- porting the utilisation of above-standard environmental practices.

In the regression analysis, one-year time lag was assumed in population response. Therefore, we took predictor variables from year N (2008–2017) and response variables (diversity indices) from year N +1 (2009–2018). To test for the robustness of results, the data were also analysed without the time lag. Additionally, survey year (YEAR) was included in all models to control for temporal trends of variation.

2.4. Boosted regression trees (BRT)

Boosted regression trees (BRT) are an advanced form of regression analysis that is based on machine learning. They optimise predictive performance by adaptively combining large numbers of simple tree models, i.e. the algorithm iteratively adds trees into the model and new trees model the residuals of the previous run (Elith et al., 2008). The method has many advantages: there is no need for prior data trans- formation or elimination of outliers, it can describe complex nonlinear relationships, automatically models interactions, reliably identifies relevant variables and has a strong predictive performance (Elith et al., 2008). Compared to the more established mixed model analysis, BRT have also been shown to better identify subtle, intermediate effects of predictors, as well as the complex relationships of multiple correlated variables (Buston and Elith, 2011). BRT have therefore been increas- ingly used in ecological research over the past decade, albeit relatively rarely in farmland bird studies (e.g. Kmecl and Denac, 2018; Meffert and Dziock, 2012). Since the data were quite heterogeneous (payments, indices, land use data, areas under different payments) and we expected complex responses, we chose BRT over more conventional methods (such as GLM/GAM).

We carried out all statistical calculations using the program R (R Core Team, 2018). To perform BRT analysis, we used the R packages gbm (Hijmans et al., 2018) and dismo (Hijmans et al., 2018) and additional code written by J. Elith and J. Leathwick for the calculation of confidence intervals.

The analysis parameters were set as follows: small learning rate at 0.005, which results in a more accurate model, tree complexity at 5, which enables the modelling of complex relationships, and bag fraction 0.5, for all runs. Cross-validation (CV) was used for model development

and evaluation with 10 randomly chosen subsets, which were used for testing in runs. The minimum of predictive deviance determined the final number of trees. For the importance of variables, the number of times a variable is selected for splitting of the trees was used, weighted by the squared improvement to the model as a result of each split, and averaged over all trees (Elith et al., 2008). The relative influence of each predictor variable on the respective bird diversity was expressed in percentages summing up to 100 % (Table 4) (for explanation see Elith et al., 2008). As a measure of the predictive performance of the model, the predictive deviance was calculated and expressed as the percentage of the null deviance (100*(null deviance – CV deviance/CV deviance), with the help of test (held-out) data during the cross-validation procedure.

Partial dependence (marginal) plots were used for the visualisation of models, where loess smoothing was applied with the R package ggplot2 (Wickham, 2016). Marginal plots indicate the projected response of bird diversity along the gradient of each predictor’s marginal effect while accounting for the average effects of all other variables in the model (Elith et al., 2008). The effects can be interpreted visually as a positive/negative or a uni-/multimodal associations. These plots also enabeld us to determine cut-off/threshold values of predictors, where applicable. In addition, we used 3D plots to explore six most important pairwise interactions in each model.

The data recorded on the same transect (but in different years) were potentially not fully independent. We checked whether there is a need to include the transect as random effect or continue with pooled data of all 145 transects. We tested the independence of the data with the method described in Wood (2006, sec. 6.5) and Read et al. (2008). We modelled the residuals of the original BRT models as a linear null model, based on an intercept term and error terms with no covariates, and as a linear null mixed model with transect as a random effect, but no fixed effects. We then compared both models with ANOVA – package nlme (Pinheiro et al., 2018), to show whether modelling the nestedness (random effect of transects) has an important effect on variance. ANOVA was not significant for all of the models calculated.

We checked whether the spatial autocorrelation of model residuals was significant with a spline correlogram with 95 % pointwise bootstrap confidence intervals and maximum lag distance of 20 km (Zuur et al., 2009). We plotted the correlograms with the R package ncf (Bjørnstad, 2018).

3. Results

All three BRT models (for all farmland species, a subgroup of grassland species and generalists) had high explanatory and predictive ability (Table 3). None of the models showed spatial autocorrelation or lack of independence.

3.1. Farmland bird species

The extent of woody vegetation, landscape diversity and payments for agricultural policy instruments were found to be the most important predictors of farmland bird diversity (Table 4).

The marginal effect of individual predictors (Fig. 3) indicate that the diversity of farmland birds was likely to be high in areas with up to about Table 3

BRT model properties for three groups of bird species.

Group of species Number of

trees Deviance explained

(%) CV evaluation

(%)

Farmland species 4800 81.3 46.5

Grassland

species 4050 82.6 56.7

Generalist

species 4700 88.1 61.7

(7)

25 % of woody vegetation (FOR), but rapidly decreased with increasing overgrowth. Farmland bird diversity was likely to increase with higher crop diversity (HMOS), which indicates the importance of mosaic farmland and is also supported by a negative association with the extent of grasslands (GRA) when GRA exceeded about 50 % of the area. The influence of the diversity of land-use types (HUSE) reflects a lower bird diversity in landscapes which were either highly fragmented or relatively simple.

Farmland bird diversity was higher in areas with natural constraints, especially if they received average payments (PLFA) of about 75 EUR/

ha, as well as in Natura 2000 sites (N2000). On the other hand, bird diversity was negatively associated with areas that receive a higher average amount of direct payments (DP). A similar, though less pronounced, pattern was found for areas where payments for AEM and organic farming (PENV) exceeded about 300 EUR/ha. A negative association also appeared in landscapes with high stocking density (LU), i.e.

above about 1.5 LU/ha.

3.2. Grassland species

The highest diversity of grassland specialists (Fig. 4) was found in open landscapes (<10 % of woody vegetation, FOR) with a high percentage of grassland area (>50 %) and very low stocking densities (<0.7 LU/ha), which were located in Natura 2000 areas (N2000). Such landscapes were also characterised by either very low or relatively high crop diversity (HMOS), whereas the association with the diversity of land-use types seems to be the opposite (HUSE).

Similarly to all farmland bird species, the diversity of grassland birds was likely to be higher in some areas with natural constraints; i.e. those receiving average payments (PLFA) of about 75 EUR/ha. Furthermore, a negative association between bird diversity and the average amount of direct payments (DP) was found, which was particularly strong in areas with more than 150 EUR/ha and areas exceeding 400 EUR/ha. On the other hand, no visible association was identified for the amount of AEM and OF payments (PENV).

3.3. Generalist species

Diversity of land-use types was by far the most important predictor variable for the diversity of generalist bird species, explaining more than a third (35 %) of the data deviance (Fig. 5). The diversity of generalists was found to be low in relatively homogeneous landscapes characterised by low diversity of both land-use types (Shannon index H’ <0.8, HUSE) and crops (Shannon index H’ <0.2, HMOS). In contrast to farmland specialists, the diversity of habitat generalists was likely to be lower in very open (<10 % of woody vegetation, FOR) and extensively managed (<1.0 LU/ha) landscapes. However, the diversity of generalists was also likely to be somewhat lower in the most intensively managed landscapes (>1.5 LU/ha). On the other hand, increasing overgrowth (>25 % FOR) did not appear to be associated with further increases in generalist diversity.

The average amount of payments to areas facing natural or other constraints (PLFA) and direct payments (DP) were moderately influential predictors (Table 4), although no particular association was found.

Interestingly, the diversity of generalists was somewhat lower when a large share of landscapes was included in the AEM on arable land and in the Organic farming measure (OF). However, these associations do not seem to be very strong.

3.4. Interactions between predictor variables

Analysis of the important pairwise interactions for each of the three models shows that most of the predictor variables also enable combined interpretation (Table 5, for three-dimensional plots see appendix D).

In the case of the farmland bird species, the six most important pairwise interactions included either the extent of woody vegetation (FOR) or the average amount of direct payments per hectare (DP) (Table 5). The diversity of farmland birds was highest in open land- spaces that are located in Natura 2000 areas (N2000) and in some areas with natural constraints (PLFA). Furthermore, the Shannon diversity index of land use types (HUSE) in such landscapes can be expected to Table 4

Heat map of the relative influence (%) of predictor variables in the three BRT (boosted regression trees) models. Values above 5.56 % (i.e. the average in case all 18 variables had equal influence) are shaded in intervals 5.57–10.00 % (grey), 10.01–15.00 % (light orange) and above 15.01 % (dark orange). (For interpretation of the references to colour in this Table legend, the reader is referred to the web version of this article).

(8)

extend between 0.5 to 2.0. Interestingly, the highest diversity of farmland birds was also fund in landscapes that receive the lowest amounts of both direct payments (DP) and agri-environmental and organic farming payments (PENV), whereas landscapes that supported the lowest bird diversity seemed to have received the highest income support. On the other hand, a positive interaction was found between DP and implementation of the agri-environmental schemes on arable land (Figure D1).

A model for grassland bird diversity showed a negative interaction between the extent of woody vegetation (FOR) and permanent grasslands (GRA). Further important pairwise interactions reinforce the interpretation of the influence of individual predictor variables (see chapter 3.2) that the highest diversity of grassland specialists can be found in very extensively managed and open landscapes, which were designated as Natura 2000 sites and as some areas with natural constraints (Table 5, Figure D2).

Five out of six most important pairwise interactions in the model for generalist species include the diversity of land use types (HUSE), which

also had the highest relative influence among predictor variables (Table 5). Furthermore, the interaction between the extent of woody vegetation and stocking density was found (Figure D3).

4. Discussion and conclusions

In this study, we investigated the relative influence of landscape characteristics, intensity of animal production and agricultural and nature conservation policy interventions on diversities of farmland birds, grassland specialists and generalist species.

4.1. Landscape structure, production intensity and bird diversity

The diversities of farmland and grassland species were found to be higher in landscapes where woody vegetation (FOR) takes up to about 25 % and 10 %, respectively. This finding supports the often emphasised importance of conserving woody landscape features in agricultural ecosystems, which can be important for some farmland species, e.g. as Fig. 3. Marginal plots for the nine most influential predictor variables in the BRT model for the Shannon diversity index of farmland bird species in the Slovenian agricultural landscape. Percentages (%) indicate the relative influence of individual variables. Gray strips represent confidence intervals calculated with 500-fold bootstrapping. The top line of each plot shows data distribution in deciles. Legend: AECMAR – Agri-environmental schemes on arable land, DP – Direct payments, FOR – Area covered by woody vegetation, HMOS – Crop diversity, HUSE – Land use diversity, LU – Stocking density, N2000 – Natura 2000 sites, PENV – Agri- environmental and Organic farming payments, PLFA – Payments to areas facing natural or other constraints.

(9)

nesting sites and perches (Hinsley and Bellamy, 2000). However, many grassland species and other field nesters tend to avoid woody vegetation, possibly due to higher predation risk and social interaction, and thus respond negatively to habitat fragmentation (Besnard et al., 2016;

Bonthoux et al., 2017). Our results support these findings, since the negative effects of the increasing extent of woody vegetation were found to be more pronounced for grassland specialists. However, due to data limitations in this study, we were not able to distinguish between the effects of woodlands and small landscape features, such as hedgerows, solitary trees and shrubs. Other research has shown that woodlands often support different bird communities, especially if they are larger in size (Bonthoux et al., 2017).

Both farmland bird and grassland species diversity rapidly decreased with progressing overgrowth, which seems to be particularly problem- atic for grassland specialists. Smaller percentage of forest (up to approx.

25 %) does seem to sustain high farmland species diversity, while for grassland specialists the decline is more rapid. In contrast, the diversity of generalists remained relatively stable regardless of the percentage of

woody vegetation. Similarly as in other Southern and Eastern European countries (Sirami et al., 2008; Zakkak et al., 2015), forest succession may thus be one of the critical drivers of the recent farmland bird declines in Slovenia (Kmecl and Denac, 2018), especially given the speed of this process in some areas important for grassland conservation (Kaligariˇc and Ivajnˇsiˇc, 2014).

In our study, land use diversity (HUSE) was used to measure heterogeneity at a large spatial scale; i.e. of broader land-use types such as arable land, woodland and urban areas, whereas crop diversity measured heterogeneity of agricultural land at the field level. Although there is generally a positive relationship between habitat heterogeneity and biodiversity (Stein et al., 2014), habitat specialists, such as farmland species, tend to negatively respond to high heterogeneity of land use (Assandri et al., 2019a; Pickett and Siriwardena, 2011), since it can increase habitat fragmentation and disturbance. On the other hand, such landscapes are usually more readily occupied by generalist-dominated bird communities (Devictor et al., 2008). Our study partly confirms these findings, since the diversity of habitat generalists was found to be Fig. 4. Marginal plots for the nine most influential predictor variables in the BRT model for the Shannon diversity index of grassland bird species in the Slovenian agricultural landscape. Percentages (%) indicate the relative influence of individual variables. Gray strips represent confidence intervals calculated with 500-fold bootstrapping. The top line of each plot shows data distribution in deciles. Legend: DP – Direct payments, FOR – Area covered by woody vegetation, GRA – Area covered by permanent grassland, HMOS – Crop diversity, HUSE – Land use diversity, LU – Stocking density, N2000 – Natura 2000 sites, PENV – Agri-environmental and Organic farming payments, PLFA – Payments to areas facing natural or other constraints.

(10)

highest in landscapes with a high diversity of land-use types, when measured at large scale, although this influence was less evident for farmland bird diversity.

On the other hand, heterogeneity at the field level is essential for many farmland specialists, because it determines e.g. the availability and quality of foraging and nesting sites (Benton et al., 2003; Hinsley and Bellamy, 2000). In our study, the diversity of farmland birds was found to be highest in open, diversely cropped landscapes (HMOS) and in extensively to moderately intensively managed landscapes (LU) where grasslands take up to about 50 % of the area. The decline of farmland birds has been associated with both the loss of mosaic landscape (Doxa et al., 2012; Pickett and Siriwardena, 2011) and increased productivity (Reif and Vermouzek, 2019). Our results indicate the importance of extensive mosaic landscapes with small-scaled mixed farming in Slovenia, where traditionally an important extent of farmland was kept as grasslands due to steep slopes (Appendix B; Ivajnˇsiˇc et al., 2020; Miheliˇc et al., 2019; Perko et al., 2020).

The highest diversity of grassland specialists was found in open landscapes with a high share of grasslands (>50 %) and very low

stocking density (<0.7 LU/ha). In Slovenia, such landscapes are char- acteristic of dry sub-Mediterranean and Dinaric grasslands with extensive grazing production systems as well as of late-mown wet grasslands in karst floodplains (Appendix B; Miheliˇc et al., 2019; Perko et al., 2020). This is consistent with findings of other ecological studies, which indicate that these species tend to prefer large patches of unfragmented grasslands (Besnard et al., 2016; Bonthoux et al., 2017) and often respond negatively to intensification of grassland use (Assandri et al., 2019b; Broyer et al., 2017; Kmecl and Denac, 2018; Tome et al., 2020).

4.2. Impacts of agricultural policy

The second most influential group of predictor variables were CAP policy instruments, expressed as budgetary transfers per unit area. The diversity response of farmland specialists was negative in areas with an increasing amount of direct payments (DP), whereas adverse effects on generalist diversity were not observed. In Slovenia, the differences in these allocations stem from the original model of direct payments, which was based on coupled production support. In this system, the highest Fig. 5. Marginal plots for the nine most influential predictor variables in the BRT model for the Shannon diversity index of generalist bird species in the Slovenian agricultural landscape. Percentages (%) indicate the relative influence of individual variables. Gray strips represent confidence intervals calculated with 500-fold bootstrapping. The top line of each plot shows data distribution in deciles. Legend: DP – Direct payments, FOR – Area covered by woody vegetation, GRA – Area covered by permanent grassland, HMOS – Crop diversity, HUSE – Land use diversity, PLFA – Payments to areas facing natural or other constraints.

(11)

amounts of payments per ha were received by intensive holdings engaged in cattle rearing for meat and dairy (Erjavec et al., 2006).

Slovenia introduced this model of support after 1998 and consolidated it upon accession to the EU. Subsequently, significant differences between farm holdings were maintained, although most payments were con- verted to decoupled (regional) payments during the 2007 CAP reform (Erjavec et al., 2015) and partial convergence of payments began after 2014 (European Commission, 2015). This variable may thus be used as a proxy for production intensity, particularly in the beef and dairy sector, which is also indicated by the relatively strong correlation between DP and stocking density (LU) (Table A1, Appendix A).

Since 2000, the intensification and restructuring of livestock production in Slovenia have sped up. A significant share of farms has abandoned livestock breeding or has been abandoned altogether, whereas some farms have further increased in size and intensity of production (Erjavec et al., 2018). With the increased demand for fodder, which in Slovenia is typically produced on arable land and grasslands managed by the farm holdings themselves, local production has inten- sified, likely causing at least some of the recently observed deterioration and loss of farmland habitats (Trˇcak et al., 2012, Trˇcak et al., 2010).

According to the results of our study, this process may have been particularly detrimental to a subgroup of grassland specialists, since the effect of the amount of DP/ha on their diversity was found to be more pronounced than on that of all farmland bird species (Bas et al., 2009).

The intensification of production in the beef and dairy sectors, supported by CAP direct payments, is thus conceivably another critical driver of the recent farmland biodiversity loss in Slovenia.

The diversity of farmland birds and grassland specialists was found to be high in some areas with natural and other constraints, which are mostly located in mountainous, hilly and karst regions (MAFF, 2015).

Income support in the form of aids to areas with natural constraints (PLFA) was designed to maintain farming in regions at risk of abandonment, and is often highlighted as (indirectly) beneficial for biodiversity (Matthews, 2013). However, a positive association between PLFA/ha and farmland bird diversity might simply be a result of the fact that such areas have often experienced smaller or delayed production changes and have therefore maintained higher biodiversity (Frenzel et al., 2016). Additional economic and ecological research would, therefore, be needed to prove their effectiveness, especially since this measure imposes no specific management requirements (MAFF, 2015).

A similar caveat might apply to the Natura 2000 network. Listing species under Annex I of the EU Birds Directive was found to positively affect their population trends in both old and new EU Member States

(Donald et al., 2007; Koschov´a et al., 2018). Furthermore, the management of the Natura 2000 network seems to have some positive effects on common farmland bird populations as well (Gamero et al., 2017).

However, most of the qualifying bird species with steep or moderate population declines on Slovenian Natura 2000 sites are grassland or farmland specialists (Denac et al., 2018). We therefore argue that the Natura 2000 network in Slovenia successfully identified areas with the highest diversity of farmland birds, while any conjectures regarding the effectiveness of implementation of the Birds and Habitats Directives are yet to be substantiated.

Finally, the diversity of farmland bird species responded negatively in areas with the highest payments for agri-environmental measures (AEM) and organic farming (OF) (PENV/ha). In the 2008–2014 and 2015–2021 programming periods, the highest payments per hectare in Slovenia were allocated to OF and AE schemes in horticulture and arable farming, and the predominant share of enrolled land (and consequently of disbursements) was included in the arable schemes (Appendix C). The majority of CAP funding for environmental measures was thus directed towards sectors with more intensive production, i.e. mainly to arable land in the Pannonian lowlands and some regions with a higher con- centration of intensively managed permanent crops (Appendix B). This probably explains our finding that areas with very high average PENV/

ha demonstrated a lower diversity of farmland species.

In principle, environmental measures can effectively promote biodiversity in an intensively managed landscape if they create a large ecological contrast and if at least some semi-natural habitats are still preserved to serve as a source of colonisers (Kleijn et al., 2011). How- ever, our results indicate that the extent of land under AE contracts had a negligible relative influence on farmland bird diversity (Table 4).

Furthermore, among different types of AES, targeted conservation schemes on Natura 2000 sites were found to have the lowest relative influence on both farmland and grassland species. These findings confirm the conclusions of previous evaluations in Slovenia, which revealed a limited effectiveness of the design and implementation of AES, as well as low enrolment in many areas (Appendix B & C; Kaligariˇc et al., 2019; ˇSumrada et al., 2020).

A low level of implementation and unclear ecological effects might also explain the small relative influence of the “Greening” direct payment schemes (GRE), which have been implemented on arable land since 2015. Already prior the last CAP reform, Pe’er et al. (2014) noted that these measures would affect only a small proportion of farms in the EU since their implementation would be limited to medium and large farms. This is particularly relevant for Slovenia, where the average farm managed only 6.9 ha of agricultural land (of which 2.5 ha arable) in 2016 (Erjavec et al., 2018). By contrast, the conservation of environmentally sensitive permanent grasslands (ESPG) should have a more substantial influence, especially since our results indicate higher farmland bird diversity in Natura 2000 sites. However, ESPG is limited to only some designated areas in Slovenia, where it restricts ploughing, whereas no precise management requirements are put in place (Official Gazette of RS, 2019).

4.3. Data limitations and future research

The effects of production systems and their consequent land use on bird diversity could probably be more thoroughly investigated if more direct measures of habitat quality on particular fields could be included in the study, such as grazing period, mowing dates and fertilization of meadows, crop yields and presence of landscape features (e.g. Broyer et al., 2017; Reino et al., 2010). Since such indicators are not included in the existing national databases, proxies were used to measure the intensity of livestock production, whereas similar indicators for arable, wine and fruit growing farms were not available. The impacts of these sectors on bird diversity may thus be underestimated in this study (cf.

Dross et al., 2018).

Despite these data limitations, the high explanatory and predictive Table 5

Two-way interactions between predictor variables in the three BRT (boosted regression trees) models. Smaller values indicate weaker interactions. For the explanation of variables’ abbreviations and their units see Table 2.

Model Predictor 1 Predictor 2 Interaction size

Farmland species

FOR HUSE 0,71

FOR N2000 0,26

FOR PLFA 0,16

DP HUSE 0,14

DP AEMAR 0,13

DP PENV 0,12

Grassland species

LU FOR 0,65

LU PLFA 0,6

LU GRA 0,58

FOR GRA 0,31

HMOS PLFA 0,22

FOR N2000 0,21

Generalist species

HUSE DP 0,86

HUSE HMOS 0,49

HUSE AEMAR 0,42

HUSE PLFA 0,31

FOR LU 0,15

HUSE LU 0,07

(12)

power of the models indicates that the selected approach and set of variables can predict bird diversity in Slovenian agricultural landscapes reasonably well. Furthermore, it enabled the identification of the potential key drivers of farmland bird declines. However, since the biodiversity response was measured with diversity indices only, further research is needed to investigate temporal correlations and causality between the probable drivers and bird population changes.

4.4. Policy implications and recommendations

The recent dramatic loss of biodiversity in agricultural ecosystems in Europe and other continents (IPBES, 2018; Rosenberg et al., 2019) re- quires urgent changes in the way natural resources are used. It has been claimed that it is insufficiently explored whether existing conservation measures are successful and how strong their effects are in relation to other socio-economic factors (Kleijn et al., 2011). Our study contributes to filling this gap by evaluating the impacts of policy interventions on biodiversity in the context of landscape and production system characteristics.

On the case of Slovenia, we show that distribution of the CAP income support, most notably direct payments, might strongly influence the diversity of farmland birds, which were used as an indicator of wider farmland biodiversity (Gregory and van Strien, 2010). The effects of the current direct payment schemes were generally found to be negative, with higher public funding per hectare negatively associated with pre- dicted biodiversity. Due to the common policy framework, we argue that something similar might apply to the other EU Member States, although some variability can be expected (Henke et al., 2018). In designing the future CAP, decision-makers should thus pay particular attention to this link and design income support schemes that are at least neutral, if not actively promoting farmland biodiversity. The gradual phasing-out of references to historic production levels in the payment schemes, which is foreseen by the CAP post-2020 reform (European Commission, 2018b), may reduce some of the adverse impacts (Matthews, 2013).

However, the question is whether this will suffice or a more targeted and result-oriented approach is needed (Brady et al., 2009).

One critical approach might be to increase conditionality, which restricts eligibility for income support to farms that implement suitable management practices, to ensure the long-term viability of common farmland bird populations (Matthews, 2013). Our results indicate that both the “Cross-compliance” system and “Greening” instruments have so far probably failed to ensure sufficient minimal standards. The new system of Conditionality, which is expected to combine both systems (European Commission, 2018b), should therefore be substantially strengthened in order to reach biodiversity objectives.

In our study, negative or insignificant effects were also evident in the case of voluntary agri-environmental and organic farming measures.

Here, the bulk of support was directed towards areas that harbour a relatively low diversity of farmland birds, while the ecological impacts of these measures remain unclear. By contrast, traditional small-scale farmland seems to be facing low uptake and payment amounts for environmental measures, whose effectiveness is further watered down by their inadequate design (Kaligariˇc et al., 2019). One should be careful when attempting to generalise these findings at the EU level as the implementation of AES greatly varies between EU Member States (Bat´ary et al., 2015). Nevertheless, biodiversity objectives in this part of Europe do not yet seem to be sufficiently integrated into agricultural policy (ˇSumrada et al., 2020). Future CAP strategic planning should ensure that this gap is closed by defining measurable targets consistent with EU nature conservation policy, designing result-oriented measures with sufficient funding and setting up effective monitoring of results (ECA, 2020; ECA, 2011). The proposed “Eco-schemes”, which will be voluntary for farmers and designed at the Member State level, represent a new additional opportunity in this regard. However, similarly to existing AEM, the effectiveness of these schemes will depend on the quality of their design and significant support by budgetary funds

(Meredith and Hart, 2019).

It is still largely unknown what kind of contemporary production systems would be needed to maintain traditional farming systems in Europe and how they could be effectively promoted to mitigate the negative impacts of structural changes in agriculture (Keenleyside et al., 2014). Most of the research on farmland biodiversity and policy instruments was conducted in the countries of Western and Northern Europe, which are dominated by intensively farmed landscapes (Uthes and Matzdorf, 2013). The suitability of applying the same approaches to marginal regions in the new Member States, many of which have maintained small-scale farming and are subject to broader negative socio-economic trends, has thus frequently been called into question (Bat´ary et al., 2015; Sutcliffe et al., 2015). Further research on the economics of such systems and the ecological effectiveness of different policy instruments in extensively managed landscapes are therefore necessary for improving the design of conservation measures.

Funding

This work was partly supported by the Ministry of Agriculture, Forestry and Food of the Republic of Slovenia[public procurement number 2330-16-310015] and the Slovenian Research Agency [research program P4-0022 (B)].

Declaration of Competing Interest None.

Acknowledgements

We would like to thank all dedicated ornithologists who enable the yearly collection of data in the Slovenian Farmland Bird Monitoring Scheme (SIPKK), which is organised by DOPPS-BirdLife Slovenia. We are grateful to Jane Elith for clarifications regarding the BRT analysis and to anonymous reviewers for their valuable comments on this paper.

Appendix A. Supplementary data

Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.agee.2020.107200.

References

Acs, S., Hanley, N., Dallimer, M., Gaston, K.J., Robertson, P., Wilson, P., Armsworth, P.

R., 2010. The effect of decoupling on marginal agricultural systems: implications for farm incomes, land use and upland ecology. Land Use Policy 27, 550–563. https://

doi.org/10.1016/j.landusepol.2009.07.009.

Anania, G., Pupo D’Andrea, M.R., 2015. The 2013 reform of the Common agricultural policy. In: Swinnen, J. (Ed.), The Political Economy of the 2014-2020 Common Agricultural Policy: An Imperfect Storm. Centre for European policy studies (CEPS);

Rowman and Littlefield International, Brussels; London, pp. 33–86.

Assandri, G., Bogliani, G., Pedrini, P., Brambilla, M., 2019a. Toward the next Common Agricultural Policy reform: determinants of avian communities in hay meadows reveal current policy’s inadequacy for biodiversity conservation in grassland ecosystems. J. Appl. Ecol. 56, 604–617. https://doi.org/10.1111/1365-2664.13332.

Assandri, G., Bogliani, G., Pedrini, P., Brambilla, M., 2019b. Species-specific responses to habitat and livestock management call for carefully targeted conservation strategies for declining meadow birds. J. Nat. Conserv., 125757 https://doi.org/10.1016/j.

jnc.2019.125757.

Bas, Y., Renard, M., Jiguet, F., 2009. Nesting strategy predicts farmland bird response to agricultural intensity. Agric. Ecosyst. Environ. 134, 143–147. https://doi.org/

10.1016/j.agee.2009.06.006.

Bat´ary, P., Dicks, L.V., Kleijn, D., Sutherland, W.J., 2015. The role of agri-environment schemes in conservation and environmental management. Conserv. Biol. 29, 1006–1016. https://doi.org/10.1111/cobi.12536.

Bedraˇc, M., Bele, S., Koˇzar, M., Moljk, B., Breˇcko, J., Pintar, M., Zagorc, B., 2018.

Poroˇcilo o stanju kmetijstva, ˇzivilstva, gozdarstva in ribiˇstva 2017 [Report on the state of agriculture, food, forestry and fisheries in 2017]. Agricultural Institute of Slovenia; Ministry of Agriculture, Forestry and Food of the Republic of Slovenia, Ljubljana.

Benton, T.G., Vickery, J.A., Wilson, J.D., 2003. Farmland biodiversity: is habitat heterogeneity the key? Trends Ecol. Evol. (Amst.) 18, 182–188. https://doi.org/

10.1016/S0169-5347(03)00011-9.