To achieve greater environmental sustainability, the Common Agricultural Policy (CAP) of the European Union (EU) introduced a direct payment scheme (greening) in 2013, aimed at protecting and enhancing biodiversity, water, and soils in European farmland (European Union 2013; European Commission 2022). Mandatory greening practices include: (i) the maintenance of permanent grassland, (ii) the obligation of agricultural diversification of farmland (cultivation of at least two crops if the arable farmland exceeds 10 hectares), and (iii) the allocation of at least 5% of the arable area to ecological focus areas (EFAs) if the arable area of conventional farms exceeds 15 hectares (German Federal Environmental Agency 2014; European Commission 2017, 2022). Farmers must comply with these requirements to qualify for the greening payments, except farms already implementing sustainable farming practices such as organic farming (European Commission 2022).The ecological value of EFAs varies across EFA types (Cole et al. 2012; Cormont et al. 2016). While non-productive options such as buffer strips, fallow land, and landscape features are considered most beneficial for conserving biodiversity (Pe’er et al. 2017; Sutter et al. 2018; Tarjuelo et al. 2020a, 2020b), less conservation-effective (namely productive) options, such as planting nitrogen-fixing or catch crops, represent more than 70% of the registered EFAs (Zinngrebe et al. 2017; Nilsson et al. 2019; Pe’er et al. 2020). Results from interviews with stakeholders, farmers, and agricultural representatives have shown that farmers consider the EFA regulation complicated (Oppermann 2015; Bonke et al. 2021). Overall, economic and administrative motivations appear to be more important than ecological considerations in defining farm management practices concerning EFA implementation (Zinngrebe et al. 2017; Nitsch et al. 2018; Brown et al. 2021; Bonke et al. 2021). Nevertheless, farmers’ motivations for managing their land include the environmental context in which their farm is embedded (Brown et al. 2021; Santos et al. 2021). Farm management decisions may also be heavily constrained by the farm context (e.g. farm size), the field context (e.g. field size), or the field terrain and soil conditions (e.g. soil fertility) (Nitsch et al. 2018). Also, the success of particular land uses in enhancing biodiversity or ecosystem services may highly depend on their proximity to and connectivity with other semi-natural areas (Tscharntke et al. 2005, 2012). Hence, the consideration of a broader set of factors influencing farm management practices, including the ecological context of the farm, is urgently needed to broaden our understanding of how to manage agricultural landscapes in more sustainable ways(Tzilivakis et al. 2016; Singh and Leppanen 2020; Brown et al. 2021).

Farms across the EU differ in their environmental context according to their geographic location. Additionally, various policy tools exist to support the CAP greening policy and reach the EU environmental targets across member states. For instance, farmers can receive further economic support for areas considered challenging to cultivate (European Commission 2022). In our study region, locations near water bodies (SMEKUL 2021a) and within water protection (SMEKUL 2021b) or Natura 2000 (European Environment Agency 2022) areas have special restrictions regarding management practices or nutrient/pest control inputs. To account for regional differences, member states can select a subset of EFA options considered more suitable for their specific context. Furthermore, there is a weighting factor for each type of EFA to reflect their ecological benefits and implementation costs. Weighting factors can range from 0.3 (e.g. for catch crops) to two (e.g. for hedges) and are used to calculate the exact area per farm relevant for the policy subsidy payments. In Germany, farmers can select among 20 different land use options eligible as EFA (Zinngrebe et al. 2017; Lakes et al. 2020, see Table S1).

In the present study, we used field-level information on land use and management from the Integrated Administration and Control System (IACS) (SMEKUL 2019), as well as geospatial information on small woody features and nature protection areas, to assess which structural, terrain and soil conditions influence the probability of a field being assigned to a specific EFA. Since EFA policies focus on the farm level, our predictors included variables reflecting the farm context in terms of the diversity of additional habitats within the farm boundaries, the relative usability of a specific field for production, and the terrain and soil conditions of the field compared to the other fields of the same farm. Overall, we expected that fields whose structural (size, compactness, proximity to other landscape elements) or terrain and soil conditions (slope, soil fertility) are below the farm’s average are more likely to be assigned to an EFA. Additionally, we expected that large farms with a smaller proportion of protected areas or small woody features, whose fields are predominantly suitable for production, would most likely meet the policy requirements with productive EFA options. Since landscape elements are not easily movable and can be used to receive EFA policy payments, we anticipated that farms with landscape features within their boundaries are likely to register them under the policy. Conversely, farms without landscape features within their boundaries are more likely to allocate productive EFA options than other EFA types.

Materials and methods

Study region

The study area is the Mulde River Basin, located in the western part of the federal state of Saxony, Germany (Fig. 1). It covers an area of 5814 km²; the terrain varies from flat to hilly, and the elevation ranges between 24 and 1214 m.a.s.l. (Sachsen Staatsbetrieb Geobasisinformation und Vermessung 2016). The climate is predominantly continental, with total annual precipitation between 570 and 1260 mm and mean annual temperatures between 7.4 and 14.1 °C (DWD 2020). 38% of the study area is covered by arable land, with winter wheat, oilseed rape, winter barley, and maize as predominant crops, followed by permanent grassland (mowing pastures and meadows).

Fig. 1
figure 1

Depiction of the geographical location of the study area and the distribution of productive, non-productive, and landscape features as EFAs used in our analysis. The square in the top right represents an inset of the data

Data preparation

Our data included field-level information on land use and management of 879 farms (41,936 fields) that complied with EFA policy in 2019 (SMEKUL 2019). The median number of fields per farm was 25. The median farm size was 100 hectares, and the median field size was 6 hectares. Farmers could select among 20 land use types to be registered as EFA (Table S1). In our study, a farm consisted of a group of fields registered in the IACS database (SMEKUL 2019) under the same anonymized farm identification number. A field consisted of a parcel with uniform land use (a certain type of crop or EFA). We grouped our predictor variables into three categories; predictors related to the farm context, the field context, or the terrain and soil conditions (Table 1).

Table 1 Overview of the explanatory variables used in the regression models grouped by category

Farm context predictors

This category included: the farm area (FARM AREA) calculated as the area of all fields registered under the same anonymized farm identification number, the percentage of the farm covered by Natura 2000 areas (NATURA), the percentage of the farm covered by landscape features registered as EFA (LAND FEAT), the percentage of the farm area covered by fields near small woody features (i.e. in a buffer of 50 m, SWF), the percentage of the farm area designated as naturally constrained (ANC), and the percentage of the farm area designated as areas of importance for drinking water and medicinal spring protection areas (WATERPROT) (Table 1).

Field context predictors

We used z transformation to standardize the values of each field concerning the farm context. For instance, a positive value of the field area means that the respective field is larger than the average field of the farm to which it belongs. The field context predictors category included: the standard z-score of the area of the field (FIELDAREA), the standard z-score of the compactness index of the field (COMPACTNESS), the standard z-score of the distance of the field to the farm centroid (DISTANCE), a binary predictor of the presence of a small woody feature in the proximity of 50-meter buffer (SWF), information on whether the field belonged to an area designated as naturally constrained (ANC), and the presence of a water body in the proximity of 20-meter buffer (WATERBODY) (Table 1). We selected the 20-meter buffer distance of water bodies used by Paulus et al. (2022), who studied the spatial allocation of agri-environmental schemes in our study region. The five meters buffer represents an intermediate distance that considers the effects of field edges on pollinators (Reynolds et al. 2022) or natural enemies (Büchi 2002).

Soil and terrain conditions

This category included values of soil fertility (SOIL FERTILITY), soil moisture (SOIL MOISTURE), wind (WINDEROSION), and water erosion (EROSION(KLSR) extracted from the digital soil map at a scale of 1:50,000 from the Saxon State Ministry for Energy, Climate Protection, Environment, and Agriculture (SMEKUL 2021c) (Table 1). The soil fertility index considered the capillarity, risk of waterlogging, cation exchange capacity, stone content, and plant-available water in the root space of the soil (Table S2). The soil moisture index was calculated based on parameters such as the soil pore system, climate, and terrain (Table S3). The wind erosion indicator described the erosion hazard depending on the soil type and long-term average wind speed (Table S4). The water erosion risk considered the soil type, slope length, slope inclination, and regenerative erosivity of the soil (Table S5). All predictors were tested for collinearity using pairwise correlation analysis. Since no correlations of Pearson’s |r| > 0.7 occurred (Dormann et al. 2013), no variables were excluded (Fig. S1–3).

Statistical analysis

We used multinomial regression models to evaluate the influence of variables related to the farm and field context and the field terrain and soil conditions on EFA allocation in farmland. Multinomial regression is a logistic regression employed to predict the probability that a case is in a particular category when the dependent outcome includes more than two categories. Multinomial models break the outcome variable into a series of comparisons between pairs of categories, on which the probability (odds) of an observation being a member of a specific group rather than the reference group is calculated. In multinomial models, a logistic transformation of the odds (Log (OR)) is the dependent variable. The odds value can range from 0 to infinity and express the likelihood that an observation is a member of the reference group rather than a member of the group used for comparison. It holds that odds = p/(1 − p). The odds ratio (OR) estimates the change in the odds of membership in the target group for a one-unit increase in the predictor and it is calculated by using the regression coefficient of the predictor as the exponent:

$${\text{log}}\left( {{\text{odds}}} \right) = {\text{logit}}\left( {\text{p}} \right) = {\text{log}}\left( {{\text{p}}/\left( {{\text{1}} - {\text{p}}} \right)} \right){\text{ }} = {\text{ }}\beta _{0} {\text{ }} + {\text{ }}\beta _{1} {\text{ x}}_{{\text{1}}} {\text{ }} + \ldots$$

Where p = the probability that a case is in a particular category, β0 is the intercept, and β1, \(\ldots\) are the regression coefficients of x1.

In order to identify specific responses according to the EFA group (e.g. productive, non-productive, landscape features) or EFA type (e.g. catch crops, nitrogen-fixing crops, etc.), we ran four multinomial models. All models included the same predictor variables (described in Table 1) and had fields dedicated to agricultural production (defined as NOEFA) as a reference value. We used Model 1 to identify differences in drivers of EFA allocation according to their productive value. Besides the reference value (NOEFA), Model 1 included three additional response categories: (i) productive EFA, (ii) non-productive EFA, and (iii) landscape features registered as EFA. We used Model 2, Model 3, and Model 4 to identify important drivers for specific EFAs (See Figs. S4, S5, S6). Model 2 incorporated only productive EFAs (c = 6) as the response categories. Model 3 included response non-productive EFAs (c = 6), while Model 4 had EFA landscape features (c = 8) as response categories. Given that we only considered farms with at least one field dedicated to one of our focal EFAs (defined by the model), all models differed in the number of farms and fields used as input (Table 2). All data analysis and data processing was conducted in R version 4.0.5 (R Core Team 2021) using the packages nnet (Fox and Weisberg 2019), car (Fox and Weisberg 2019), caret (Kuhn 2008), dplyr (Wickham et al. 2019), raster (Hijmans and van Etten 2012), sf (Pebesma 2018), and terra (Hijmans 2023). Details of the code can be found in the GitHub repository related to this manuscript (

Table 2 Description of the model ID, input data (number of fields and number of farms), and response categories of the four models used in our analysis. Response categories referred to the discrete outcomes used in each model. Class frequency refers to the number of fields used as data input for the discrete outcomes on each model. NOEFA refers to fields used for agricultural production


7% of all fields were allocated with a productive EFA, 6% with a non-productive EFA, and 5% to landscape features registered as EFA. Among productive EFAs, catch crops, and nitrogen-fixing were the most commonly adopted options. Fallow land and buffer strips were the most frequently allocated non-productive EFAs, and hedges and rows of trees were often registered as landscape features. Farms covered with a low percentage of Natura 2000 (NATURA), landscape features registered as EFA (LAND FEAT), small woody features (SWF), or water protection areas (WATER PROT) mostly allocated productive EFAs (Fig. 2). Conversely, farms with higher values on the same variables were more prone to allocate non-productive EFAs or landscape features.

Small and linear fields (FIELD AREA, COMPACTNESS) had higher probabilities of being allocated with non-productive EFAs or landscape features (Fig. 2). Non-productive EFAs and landscape features were allocated further away from small woody features (SWF) or water bodies (WATER BODY), but they were frequently present in fields designated as naturally constrained for agricultural production (ANC). Conversely, large and rounded fields were most prone to be assigned with productive EFAs. Contrary to our expectations, EFA fields were not allocated in fields with low soil fertility (SOIL FERTILITY). Nevertheless, both productive and non-productive EFAs were allocated in fields with lower risks of water erosion (EROSION (KLSR)).

Among productive options, catch crops and nitrogen-fixing crops responded differently to the same variables (Table 3, Fig. S4). While nitrogen crops were frequently allocated on small farms (FARM AREA) with a lower percentage of landscape features (LAND FEAT), small woody features (SWF) and water protection areas (WATER PROT), catch crops were allocated in farms with higher values of Natura 2000 (NATURA), landscape features (LAND FEAT (%)), and water protection areas (WATER PROT) (Table 3, Fig. S4). Catch crops were frequently assigned to large and rounded fields (FIELD AREA, COMPACTNESS) in naturally constrained (ANC) areas. Fallow land was more likely to be allocated in farms with lower values of NATURA, LAND FEAT, and WATER PROT), and in small and linearly shaped fields (Table 3, Fig. S5). Conversely, buffer strips were most often established on farms with higher values of NATURA, LAND FEAT, SWF or WATER PROT. Hedges and rows of trees were most abundant in farms with low percentages of SWF and ANC, but within these farms, to fields located in ANC areas (Table 3, Fig. S6). Most variables related to terrain and soil conditions had non-significant effects for the majority of EFA types, with the exception of slope and erosion KLSR.

Fig. 2
figure 2

Model coefficients (logarithm of odd ratio) of the multinomial regression used to evaluate the drivers of EFA allocation according to the EFA group (model 1). Explanatory variables are grouped into three categories (farm context, field context, soil and terrain conditions). Colored circles indicate p-values lower than or equal to 0.05. Farm context indicators (except farm area) reflected the percentage of farm area covered by: Natura 2000 designated area (NATURA (%)), landscape features registered as EFA (LAND FEAT (%)), small woody features (SWF (%)), naturally constrained areas (ANC (%)) or water protection areas (WATER PROT (%)). SWF indicates the presence (1) or absence (0) of a small woody feature in a buffer of 50 m around the field. ANC indicates whether the field is in a naturally constrained area. WATER BODY indicates the presence (1) or absence (0) of a water body in a buffer of 50 m around the field. EROSION (KLSR) refers to the water erosion risk

Table 3 Effect direction of the two most frequently adopted EFA of each category. The plus symbol (+) indicates significant positive effects, the minus symbol (−) indicates negative significant effects and ns indicates non-significant effects at the significance level of ≤ 0.05


In this study, we used a spatially-explicit approach to analyze how the farm context, the field context, and the terrain and soil conditions influenced the allocation of EFAs in farmland. We demonstrated that the farm and the field contexts, rather than the terrain and soil conditions, are decisive for the assignment of a particular EFA type to a particular field. In our study region, EFA types known to benefit biodiversity, such as fallow land, buffer strips, and landscape features (Martin et al. 2019; Pe’er et al. 2022), were used on farms already embedded in contexts with a higher cover of nature conservation and non-crop habitats (Fig. 2). In contrast, productive EFAs such as catch crops or nitrogen-fixing crops were often found on farms where cropland dominated (Fig. 2). These results have important implications for conserving farmland biodiversity. Considering that the effectiveness of conservation measures in agricultural areas is highest in landscapes with intermediate levels of land use heterogeneity (“intermediate landscape-complexity hypothesis”, Tscharntke et al. 2012), the potential of EFAs to contribute to biodiversity conservation in farmland can be strongly moderated by their spatial distribution (Sutter et al. 2018; Concepción and Díaz 2019; Concepción et al. 2020).

EFAs on farms embedded in complex landscapes may contribute little to restoring existing biodiversity on this land. On the other hand, EFAs on farms embedded in simplified landscapes may also contribute little to restoring the biodiversity that has already been lost in these areas. The allocation of EFAs in an intensively used area, however, has the potential to increase landscape complexity, for instance by increasing crop diversity or the number of non-crop elements within farm boundaries (Concepción et al. 2012). Nevertheless, EFAs should not be used as the only tool to conserve biodiversity in farmlands, as their benefits for threatened or specialist species depend highly on their ecological quality (Herzog et al. 2017; Pfiffner et al. 2018). Further, many EFAs only benefit farmland generalist species (Kleijn et al. 2006; Aviron et al. 2007, 2009) and need to be complemented with additional nature protection efforts or improved and tailored agri-environmental schemes in order to protect and enhance a wider range of biodiversity (Aviron et al. 2009).

Farm area in our study region was a determining factor only for a subset of the EFA options (Table 3). Fallow land and nitrogen-fixing crops were most often adopted by small farms, while hedges were most frequently used by larger farms (Table 3). Large farms with a higher cover of landscape features in their boundaries used these elements to fulfill the policy requirements. Interestingly, farms that did not fulfill their EFA quota with landscape elements tended to allocate productive instead of non-productive EFA (Fig. 2). These results support previous findings on the prevalence of economic rather than ecological considerations on farmers’ decisions on greening policy allocation (Zinngrebe et al. 2017). Productive EFAs (e.g. catch crops and nitrogen-fixing crops) are mainly targeted to reduce soil erosion and facilitate nutrient uptake (Cerdà et al. 2022; Quintarelli et al. 2022) and may help to facilitate crop production in the same field on the following year (Brown et al. 2021; Wittstock et al. 2022). On the other hand, the placement of non-productive EFAs and landscape features is not spatially flexible and may appear disadvantageous from the farmer’s point of view, given that working with large machinery may be more difficult if individual trees or hedges are present at the field boundaries. While productive EFA options were commonly assigned to large, rounded fields with lower soil erosion hazard, non-productive EFAs were allocated in small, linear fields in areas considered disadvantaged for crop production (Fig. 2). A similar pattern was found by Paulus et al. (2022) for Agri-environmental schemes allocation in our study region. Paulus et al. (2022) demonstrated that Agri-environmental schemes are more often allocated in areas with low potential for agricultural intensification, supporting the notion that the lack of careful spatial targeting of agricultural policies may diminish their ecological contribution.


Farmers’ economic considerations play an essential role in deciding which type of EFA to choose (e.g. productive versus non-productive) and where to place it (e.g. large and rounded fields versus small, linear, and less fertile fields). Our results demonstrate that the farm and the field context have a significant impact on the spatial distribution of greening measures within the farm boundaries. These findings are especially important for possible future spatial targeting of conservation and nature protection measures (including eco-schemes) on farmland. This relates to several measures that are spatially flexible, such as fallow land and buffer strips—options that are widely recognized in expert-based assessments as highly valuable for biodiversity and ecosystem services (Pe’er et al. 2017; Traba and Morales 2019).