Introduction

A sustainable pathway is needed to increase agricultural production and achieve food security in the future, while simultaneously reducing the adverse environmental effects of agricultural production. The provision of ecosystem services from agricultural land, in particular, needs to be improved, and this has been increasingly highlighted by science and enacted in policy changes (Schaller et al. 2018). European agricultural landscapes have experienced diverging shifts towards intensification and specialization on the one hand, and marginalisation and abandonment on the other hand, and these dual trends are expected to continue into the future (Lambin et al. 2000; Monteleone et al. 2018; Stoate et al. 2009). Marginal agricultural landscapes, characterised by unfavourable biophysical conditions such as steep slopes, shallow and/or poor soils and inferior accessibility (Harvolk et al. 2014; Lüker-Jans et al. 2016), can increase biodiversity and habitat richness. This, however, requires low-input production, wide crop rotations, permanent grassland and small-parcelled mosaics. High Nature Value (HNV) farming systems, typical for such landscapes, are essential to biodiversity conservation and the provision of ecosystem services (Lomba et al. 2020; Strohbach et al. 2015). Intensive and traditional agricultural production, however, rests on homogenous landscapes, i.e. larger plots without landscape elements that could provide sufficient habitat structure or prevent soil erosion (Tscharntke et al. 2005).

Our research focuses on the eastern German Federal State of Brandenburg, where large-scale agricultural land use shapes the landscape. As in many other post-communist regions, large-scale agriculture persisted despite fragmented land ownership after restitution following German reunification (Hartvigsen 2014). Along with the large-scale farming structure, Brandenburg’s agricultural landscape is characterized by homogenisation and production intensification, both of which are associated with a decrease in biodiversity and adverse environmental effects, i.e. a decrease in soil and water quality (Thomson et al. 2019). These trends continue despite the EU’s efforts to increase financial support for sustainable land management practices within the Common Agricultural Policy (CAP). Concerns over whether these efforts have been able to impede adverse landscape structures lead to questions regarding how to quantitatively assess landscapes’ functioning so as to preserve and enhance biodiversity, habitats and thus, ecosystem service provision.

Quantitative landscape metrics characterise and allow comparison of agricultural landscapes across space and over time (Uuemaa et al. 2013). Typically, the number, size, shape and arrangement of patches of different land use/land cover types are used to describe a landscape’s structure, composition and dynamics (Lausch and Herzog 2002). Recent metrics also include the area under cultivation, mean patch size and Shannon’s Diversity Index as an agrobiodiversity indicator (Uthes et al. 2020) to characterize agricultural land use and management intensity (Schlesinger and Drescher 2018). Other measures of agricultural land management intensity rely on input use intensity, including labour or capital, management practices, output quantities such as per-hectare yields (Shriar, 2000), or the dependence on industrial goods such as machinery and fertiliser (Temme and Verburg 2011; Zasada et al. 2013). A conceptual framework to quantify and analyse land-use intensity proposed by Erb et al. (2013) integrated three dimensions: input intensity, output intensity and the associated system-level impacts of land-based production (e.g. changes in carbon storage or biodiversity). Estel et al. (2016) summarized that particularly in mapping indicators of cropland use intensity substantial progress has been made. Intensity indicators included yield gaps, fertilizer use, human appropriation of net primary production, field size or the extent of irrigated agriculture or tillage (Estel et al. 2016). The study by Estel et al. (2016) maps and characterizes cropping systems based on the MODIS Normalized Difference Vegetation Index (NDVI) time series and self-organizing maps. The results correspond well with indicators for agricultural intensity, such as nitrogen inputs or yields. Rega et al. (2020) combine the mapping of cropping systems with an indicator of management intensity to classify agricultural land across Europe. Many of these studies, however, have limited generalisability given restrictions on available data to small areas, regions or selected farms.

The Integrated Administration and Control System’s (IACS) dataset that is used to monitor and control the flow of payments for which farmers apply as part of the CAP offers promising applications to carry out plot-based characterisations of different types of agricultural landscapes. Several studies have successfully used this dataset to analyse agricultural land-use change (Lüker-Jans et al. 2016; Tomlinson et al. 2018) and to characterise farms based on crop choice and land use (Lomba et al. 2017; Uthes et al. 2020). On a broader scale, landscape archetypes and zones on a European or global level have also been identified and discussed in several studies (Eisenack et al. 2019; Levers et al. 2018; Oberlack et al. 2019; Václavík et al. 2013). This research, however, did thus far not take into account detailed regional specifications.

The aim of this paper is to close this gap by first identifying and characterising different types of agricultural landscapes based on an integrative approach that jointly acknowledges landscape structure, diversity and management derived from plot-based information, and second, by depicting their spatial patterns. We illustrate our approach for the Federal State of Brandenburg and pose the following research questions:

RQ1: How can agricultural landscapes be characterized in terms of landscape structure, diversity and management on a small scale. What spatial patterns can be detected?

RQ2: How can regionally specific agricultural landscape types be identified? Which spatial concentration of those types exist?

Our analysis relies on IACS data for Brandenburg and uses metrics built from a combination of agricultural landscape structure, diversity and management indicators rather than single-variable metrics. We thereby include detailed regional specifications in contrast to existing classifications for Brandenburg, i.e. agro-ecological zones (Landbaugebiete), defined by site conditions and the resulting productivity (Landesamt für Ländliche Entwicklung, Landwirtschaft und Flurneuordnung, 2016) or single-variable crop and livestock information provided by the Thünen Atlas (Thünen Institut 2014). We are thus able to capture different dimensions of agricultural landscapes which may indicate or be used as a proxy for selected ecosystem services (ESS), e.g. habitat richness or biodiversity. While landscape metrics are most frequently applied to grids and administrative areas, we use hexagons, which have been shown to better capture spatially continuous phenomena, such as agricultural landscapes, due to their spatial smoothing effect towards the edges (Birch et al. 2007; Schindler et al. 2008).

The outcomes of this study provide important insights. First, integrated agricultural landscape characteristics can be used to develop environmental and agricultural policies that are better tailored to local and regional characteristics. Second, the results of this study can subsequently be used to prioritize areas and set the scope for measures regarding agricultural land use, particularly enforcing multifunctional agricultural landscapes. Last, our methodological approach allows transferability to other EU regions, where identification of different types of landscapes offers locally- and regionally-adapted designs of environmental and agricultural policy measures, environmental policy impact evaluation and landscape planning.

Material and Methods

Study Region

The state of Brandenburg is located in north-eastern Germany, covers 29,640 km2 and is a heavily agricultural state, with approximately 45% of its area comprised of agricultural land (Amt für Statistik Berlin-Brandenburg 2016)—making it an ideal setting to study landscape composition. 12% of Brandenburg’s agricultural area is dedicated to organic agriculture, which is relatively high compared to other German states, and has been steadily increasing (MLUK 2019). Nevertheless, the utilized agricultural area has remained constant, with about 77% cropland and 23% permanent grassland (Troegel and Schulz 2018). Brandenburg completely surrounds Germany’s capital city of Berlin (Fig. 1), where land use and its composition are heavily influenced by strong urbanization trends such as demand for residential land in the suburban areas. Demand for regional food production in the neighbouring state has been growing, as has the use of cropland for renewable energy production (Gutzler et al. (2015), leading to considerable increases in maize production for subsidized biogas fermentation in Brandenburg (Federal Environmental Ministry 2000).

Fig. 1
figure 1

Input data source samples and hexagonal grid in the study area Brandenburg, Germany

Full size image

Brandenburg’s agricultural land exhibits a high share of low-quality soils; almost two-thirds are sandy and sandy-loamy soils. According to Gutzler et al. (2015), this situation, paired with low rainfall (on average, less than 600 mm/year), makes agricultural production challenging. This is one reason why Brandenburg farmers either produce in the organic niche, benefiting from the high prices paid in Berlin for regional, organic food, or apply a high level of technology, including heavy-duty machinery and intensive use of fertilizers and agrochemicals (Gutzler et al. 2015). Maize replaced rye as the main crop in 2013, followed by wheat and rapeseed (Troegel and Schulz 2018). As in all eastern German states, agricultural in Brandenburg is dominated by large farm enterprises with an average size of approximately 250 hectares, four times the German average (Gutzler et al. 2015; Troegel and Schulz 2018). Livestock production has been in continuous decline in Brandenburg; according to the most recent available agricultural census, its livestock density in 2010 was a low 0.4 livestock units (LU) per hectare in comparison to other federal states, such as Lower Saxony’s 1.1 LU per hectare (Statistisches Bundesamt 2019). We, therefore, focus on cropland and grassland in our analysis. Furthermore, in contrast to Uthes et al. (2020), we propose an areal characterisation of landscapes instead of farming systems where livestock numbers are more relevant.

As a base for our indicator and cluster calculation, we created a hexagonal grid with a cell size of 10 km2 (N = 2 836, Fig. 1). The size of the cells captured the landscape level and the spatial configuration of plots within each cell (mean plot size = 7.9 ha). Since administrative areas vary in size and form, the hexagonal grid provides a smoother surface for analysis (Birch et al. 2007; Schindler et al. 2008) and has been applied in studies using landscape metrics for characterizing agricultural landscapes (Griffith et al. 2000). We selected only those cells that are located entirely within the Brandenburg state, including overlaps with Berlin administrative areas.

Data

We used plot-based information on the cultivation of agriculture in Brandenburg in 2018 from the IACS to identify agricultural characteristics. Farms apply for area-based payments to ensure income support according to EU CAP regulations, managed and controlled in a standardised way in all EU member states through IACS. In Brandenburg, the baseline map for the registration is a digital cadastre of field blocks established in 2015. The field block cadastre covers the agricultural area in Brandenburg that is eligible for EU subsidies and is updated based on orthophotos. A field block is a coherent agricultural area surrounded by permanent borders (e.g. roads, paths, trees) with a predominantly uniform primary land use. One or more farmers can use a field block, meaning that the area of one field block may be split between each farmer who applies for subsidies. As a result, the georeferenced agricultural land use data covers only those plots for which farmers applied for subsidies in 2018. The outlines of the plots are generally aligned with the underlying field blocks, but they may have been edited by the farmer due to the specific land use in a specific year. Hence, the size and outlines of plots registered for subsidies can change over time. In addition to agricultural use at the plot level, landscape elements located in a field block, such as hedges, rows of trees and single trees, are also registered. In Brandenburg, landscape elements were registered and located with a single point until 2016, but now they are digitised with spatial outlines (e.g. groups of trees). We, therefore, focused on the categories of grassland, cropland and landscape element, which were assigned based on cultivated crops (Kulturart) for 2018. These landscape elements include ecological priority areas for which farms can get extra support within the EU CAP. However, we did not include landscape elements in the final cluster analysis. All subcategories were then aggregated to the categories: cropland, grassland and landscape elements (Fig. 1).

To account for specific types of arable land use, we identified plots that were likely to have been cultivated without crop rotation and used maize as a specific crop type. We also included information about whether a plot is under organic or conventional management, both of which are indicated in the IACS data.

In addition to IACS data, we used the Open Street Map (OSM) data and regional planning data (settlement locations) and soil quality (Fig. 1). We used the OSM data for all building footprints in Brandenburg from September 2019 to assess the degree of urbanisation in each hexagon. OSM is an open-source, crowd-sourced mapping platform that has high coverage and good quality in countries such as Germany (Fan et al. 2014; Jokar Arsanjani et al. 2015). We used April 2019 settlement data from the Landesentwicklungsplan Hauptstadtregion Berlin Brandenburg for calculating the mean Euclidean distance to settlements for each cell. The Bundesanstalt für Geowissenschaften und Rohstoffe (2014) provides a soil quality rating (SQR) on a 0–100 point scale (Mueller et al. 2007), which indicates a rough estimate for crop yield potential. Soil quality points suggest the potential productivity and are an official measure in Germany that was constructed to combine pedologic, scientific and agronomic considerations within one measure. A low (high) number represents very low (high) productivity (BMJV 2007; Scheffer et al. 2010).

Indicator Calculation, Metrics and Spatial Patterns

To answer RQ1, we selected a set of landscape metrics to characterise agricultural landscapes based on a literature review according to three categories (Table 1).

Table 1 Metrics to describe landscape structure, diversity and management with description of indicators, calculation of metrics and data sources
Full size table

Landscape structure: median plot size (ha), edge density (calculated as a share of the total hexagon area in km/10 km2), number of buildings (N) and mean distance to settlements (km).

Landscape diversity: agriculture share of total hexagon area (%), Shannon Diversity Index (SDI), share of landscape elements in a total agricultural area (%).

Management: share of organic of total agricultural area (%), share of cropland of total agricultural area (%), share of maize of total agricultural area (%), soil quality (values from 0–100).

We calculated the respective indicator values for the year 2018 at the aggregated level of the hexagons. We focused on measures to describe agricultural land use, management, agricultural intensity and diversity and spatial configuration.

Plot size captures the spatial configuration of plots and is frequently used to characterise agricultural landscapes (Dengler 2009; van der Zanden et al. 2016). We calculated median plot size within hexagons from the reported management units in the IACS data by using the centroid of the plots, considering each plot only once even though it might have overlapped between two cells.

The ecological role of habitat diversity and plot edges for farmland biodiversity (including functional biodiversity) has been demonstrated by several authors (Benton et al. 2003; Burel and Baudry 2005; Weissteiner et al. 2016). We, therefore, calculated edge densities and the SDI. Edge density characterises the fragmentation of the agricultural landscape, i.e. with increasing edge density, the number of farmland patches increase and their patch size decreases (Su et al. 2014).

Organic agriculture is a production type in which mineral fertiliser and synthetic pesticide usage are subject to stricter regulations than in conventional agriculture (Gabriel et al. 2010). Because organic production is considered less harmful to the environment and key for more sustainable agricultural production, it has been included as a share of organic agriculture as an indicator.

To differentiate between cropland and grassland, we included the share of cropland of the total agricultural area, following the argument that most grasslands in eastern Germany are managed rather extensively (Matzdorf et al. 2008). Though grasslands can also be managed intensively, particularly in regions with high livestock densities, Brandenburg is characterised by few ruminant livestock and rather extensively used grasslands under agro-environmental measures, whereby farmers receive additional compensation payments through the EU CAP for extensively-managed grasslands (Matzdorf et al. 2008).

We measured cropland intensity by the share of maize that is likely to be used for biogas and cultivated as a long-term, self-following crop, i.e. without crop rotation (Gutzler et al. 2015; Lüker-Jans et al. 2016). We included all maize types (i.e. silage maize and corn maize) in our analysis. According to the German expert group for renewable energy (FNR 2013), the expansion of maize monocultures (no mixed cultivation on a plot) is expected to be on par with the intensification of crop production (Vergara and Lakes 2019). Areas with a high share of maize may indicate intensive production of crops for biogas, which often comes at the expense of food production areas (Grundmann and Klauss 2014; Lüker-Jans et al. 2016).

The SDI, as a measure of agrobiodiversity, is widely used (Uthes et al. 2020; Vaz et al. 2014). It considers the abundance of different crop types. We calculated the SDI for all listed cultivated plants within the IACS data (N = 158) according to the following formula:

$${\text{SDI}} = - \mathop \sum \limits_{i = 1}^{n} p_{i } lnp_{i}$$

where

\(p_{i}\) = share (%) of crop/crop and usage i in a total agricultural area

\(lnp_{i}\) = natural logarithm of pi

The diversity measure equals minus the sum, across all crop types, of the proportional abundance of each crop type, multiplied by that proportion (Griffith et al. 2000).

According to Uthes et al. (2020), landscape elements such as hedge or tree rows are important features for a diverse landscape structure. We thus calculated the share of landscape elements in the total agricultural area within each hexagon.

We used the SQR as a measure for yield potential, which has often been used in land market analyses, such as those of Hüttel et al. (2016) and Ritter et al. (2015).

To assess the degree of urbanisation, we calculated the number of buildings in each hexagon and the mean distance to settlements. According to Su et al. (2011), proximity to urban centres parallels the intensity of urbanisation and the decrease in human influences on the environment. Additionally, Piorr et al. (2018) emphasise that agricultural landscapes ‘differ in the way they are influenced by the proximity to urban areas, being part of functional urban–rural linkages, urban pressures and opportunities’, e.g. regarding the farming systems and the involvement of (urban) communities.

For visualization of the results, we classified the metrics share of agriculture, cropland, maize and organic agriculture by equal intervals in 20% steps. For the indicators related to the number of buildings, distance to settlements, soil quality, median plot size, edge density and the SDI, we used natural breaks (jenks) for classification.

To identify spatial patterns, we calculated the spatial autocorrelation values for all single metrics with continuous values. We used Global Moran’s I statistics, which characterise the spatial dependency of values between the hexagons (Moran 1950). We used all six neighbours (Queen’s contiguity) of each hexagon. The value of Moran’s I ranges from − 1 (perfect negative autocorrelation) to 1 (perfect positive autocorrelation), with 0 indicating spatial randomness (Moran 1950).

Cluster Analysis to Identify Agricultural Landscape Types and Spatial Concentrations

To answer RQ2, we applied a two-step cluster analysis using selected metrics to identify different types of agricultural landscapes in Brandenburg.

Lausch and Herzog (2002) emphasise that when working with landscape metrics, one is confronted with the question of which indicators are relevant for the area and the problem under investigation. We, therefore, determined Spearman’s correlation coefficients to reduce redundancies (Lausch and Herzog 2002). After the Spearman correlation analysis, eight selected indicators showed values < 0.4 (Fig. 9 in Appendix). However, we relied on seven input variables for the cluster analysis, having excluded the share of landscape elements. This indicator was not included because the values are generally very low in the hexagonal grids, with low variance except for a few outliers (65% of all hexagons have a < 1% share), and if included in the cluster analysis, the results showed no variance within clusters. The final cluster analysis input indicators included soil quality, number of buildings, edge density, shares of organic agriculture, cropland and maize, and median plot size for each hexagon in 2018 (Fig. 2).

Fig. 2
figure 2

Workflow including data and processing steps from indicator calculation to cluster analysis for identifying agricultural landscape types

Full size image

We followed the approach of Lüker-Jans et al. (2016), that characterised agricultural land use patterns using k-means clustering. Here, we applied a two-step cluster analysis because of its ability to deal with large datasets, including variables that are not normally distributed, and the possibility of automatically determining the optimum number of clusters (Chiu et al. 2001). In the first pre-clustering step, the Bayesian information criterion (BIC) was calculated for each cluster, which was then used to generate an initial estimate of the number of clusters. The second step refined the initial estimate by determining the greatest change in distance between the two closest clusters in each hierarchical clustering stage (Chiu et al. 2001). We note that 178 hexagons could not be clustered due to missing soil quality data in those cells; consequently, no type could be assigned.

For goodness assessment of the cluster number, the model fit was evaluated using the silhouette coefficient, which is a measure of the cohesion and separation of clusters. A value above 0.2 indicates a fair cluster quality (Tkaczynski 2017).

Since the cluster values are categorical, we calculated the join count to determine the degree of spatial concentration or dispersion among a set of spatially adjacent polygons (Plant 2012). To calculate the join count for each cluster value, we set the reference cluster value to 1 and all other cluster values to 0.

Results and Discussion

Characteristics and Spatial Patterns of the Agricultural Landscape in Brandenburg

With respect to RQ1 and regard to the categories landscape structure, landscape diversity and management, we found the following results. For brevity, total values (min, max, median and standard deviation) for all hexagons are provided in Table 3 in the Appendix.

Landscape structure (Fig. 3)

Fig. 3
figure 3

Maps for landscape structure indicators including median plot size (ha), edge density (m/10km2), number of buildings (#) and mean distance to settlements (m)

Full size image

Most hexagonal cells show a rather low median plot size between 0.1 ha and 7.4 ha (N = 2347). However, median plot sizes can reach up to 27.0 ha to 46.9 ha (N = 9) which include both, cropland and grassland plots. The ecological value of certain plot sizes depends on the agricultural use; according to Crist and Peters (2014), larger and older plots of grassland might support greater biodiversity of insect species than do smaller plots. Other studies suggest that agricultural diversification—the compositional heterogeneity of crops within a landscape—supports an increase in both biodiversity and yields (Burchfield et al. 2019; Thomson et al. 2019).

Edge density represents the composition of plots. Hexagons that did not have strictly rectangular plot shapes showed higher edge density values. From this, we infer that such shapes might increase agricultural landscape diversity, in line with Uthes et al. (2020). Edges might operate as zones for ecologically valuable elements, such as hedges or tree rows.

Brandenburg contained many rural hexagons with a low number (0–421) of buildings (N = 2238). The highest settlement densities (number of buildings > 6041) were in cells adjacent to Berlin and to regional centres such as Neuruppin, Schwedt/Oder, Fürstenwalde, Cottbus and Jüterbog (N = 17). At the same time, most of the hexagons were characterised by short mean distances to the nearest settlement associated with high spatial autocorrelation (Moran’s I = 0.51). Approximately 66% of cells showed a mean distance below 2 km.

Landscape diversity (Fig. 4)

Fig. 4
figure 4

Maps for landscape diversity indicators including share of agriculture (% of total area), Shannon Diversity Index and Share of Landscape Elements (% of total UAA)

Full size image

The share of agricultural area per cell was evenly distributed between the 20% step-classes, with the exception of those with a very high agricultural share (> 80%, N = 349 of 2761). The highest agricultural shares were found in cells with the highest values for mean soil quality (> 62). Only 75 hexagons contained no agricultural land at all and were instead covered by forest, urban centres or water surface. The spatial autocorrelation analysis of agriculture share returned a Moran’s I value of 0.59 indicating relatively high spatial concentration of hexagons.

The SDI, calculated as a proxy indicator for agrobiodiversity, showed low positive spatial autocorrelation (Moran’s I = 0.37). Areas with a high share of organic agriculture, however, showed no explicitly higher values for the SDI.

The share of landscape elements was generally low in relation to the total agricultural area (between 0–1% for 73% of the hexagons) due to the chosen landscape scale. However, they perform a number of functions, such as serving as windbreaks, modifying the microclimate and assisting in soil retention and water purification (Stoate et al. 2009). They also enhance landscape diversity and connectivity, are explicitly acknowledged as important cultural features and have recreational, aesthetic and heritage value (van der Zanden et al. 2016).

Management (Fig. 5)

Fig. 5
figure 5

Maps for Management indicators inlcuding share of organic, share of maize, share of cropland (% of total UAA) and mean soil quality (points)

Full size image

About 8% of hexagons showed values of more than a 60% share of organic agriculture with high spatial autocorrelation (Moran’s I = 0.56). Best (2006) states that farmers’ decision to switch to organic management is dependent on multiple factors, but might include socialisation factors, such as neighbours’ perceptions and social connectivity. The decision to switch is also influenced by higher uncertainties in yields leading to a fear of lower income and dependence on subsidies (Best 2006).

Most of the agriculturally used areas were characterised by a high share of cropland (53% of hexagons show values > 80% cropland share), and cells with a low share of cropland had low soil quality. In contrast, the maize share was below 20% in 1,955 of 2,761 cells (71%), with low spatial autocorrelation (Moran’s I = 0.30).

Brandenburg’s mean SQR ranges between 37 and 79 which indicates generally rather low soil quality and thus yield potential. Mean soil quality shows low values of spatial autocorrelation (Moran’s I = 0.33).

To identify types of agricultural landscapes and cover systematic patterns, reduction of dimensionality was necessary and subsequently implemented as the two-step cluster analysis.

Types of Agricultural Landscapes and Spatial Patterns

In this section we answer RQ2, identifying types of agricultural landscapes and their spatial concentrations. The two-step clustering analysis returned the most optimal results with a cluster number of six, with a relatively low BIC value of 7 894.076 and the highest distance measure of 1.546 (Table 4 Appendix). The silhouette measure of cluster cohesion and separation indicates a fair quality (0.3) of the resulting number of six clusters. Based on the cluster analysis, we identified six different types of agricultural landscapes in Brandenburg: peri-urban, high fragmentation, low fragmentation, high intensity, low intensity and organic production described in more detail in Fig. 6.

Fig. 6
figure 6

Description of agricultural landscape types derived from two-step cluster analysis

Full size image

Median values of clusters are summarized in Table 2 and Fig. 7:

Table 2 Centroid of clusters with the lowest (italic) and highest (bold) values
Full size table
Fig. 7
figure 7

Comparison of input variables’ value of importance and statistic values for clusters; colorbars represent clusters (1–6) with median and 25% and 75% quantile, white boxes represent the combined values showing median, 25% and 75% quantile

Full size image

The map in Fig. 8 shows the spatial distribution of clusters in Brandenburg. For the join count, results with significance level p > 0.01, and thus cells with only one or two total neighbouring hexagons, were excluded. We identified a high positive spatial autocorrelation for the high intensity (N = 98) and organic production (N = 95) clusters. That is, one agricultural landscape type was likely located next to another agricultural landscape of the same type. The spatial clustering of high-intensity agriculture that we found in our results may be attributed to the underlying spatial clustering of high soil quality.

Fig. 8
figure 8

Map of agricultural landscape types in Brandenburg, Germany in 2018 with exemplary satellite imagery for each type (Google)

Full size image

We found that hexagons with a high percentage of organic farming had lower median plot sizes (cluster 6, Fig. 7), which is in line with the studies of Best (2006) and Caporali et al. (2003). As shown by the study of Bichler and Häring (2003), land with high shares of organic agriculture tends to show higher shares of grassland and lower shares of maize (Fig. 7), especially in Brandenburg where there is a low livestock density. Edge density tends to be higher in areas with a high share of organic farming and a low share of maize. In contrast to other studies, we could not find significantly higher soil qualities in areas under organic production; however, agglomeration effects of organic farming have been noted (Schmidtner et al. 2012). One reason for this finding could be that organic agriculture is possible even in close proximity to nature preserves, which cover larger coherent areas in Brandenburg (Venghaus and Acosta 2018). The low fragmentation (N = 34) and low intensity (N = 43) clusters did not show a high degree of spatial autocorrelation and were distributed across the state. The peri-urban (N = 54) and high fragmentation (N = 71) clusters showed medium spatial autocorrelation and seemed randomly spatially distributed over the state, whereby the peri-urban cells were concentrated around Berlin.

The increase of maize cultivation in Brandenburg in recent years has led to areas with larger plot sizes, hence the lower edge densities and potentially intensive management (represented by clusters 3 and 4, N = 808 which accounted for 30% of all hexagons). Lüker-Jans et al. (2016) emphasise that ‘intensive land use is connected to landscapes with rather favourable site conditions for arable cultivation such as relatively flat and fertile land’, which corresponds with our findings, particularly for cluster 3 low fragmentation and cluster 4 high intensity. Consistent with Lüker-Jans et al. (2016), using k-means clustering, we identified similar agricultural types based on cropland share, with maize as a focal crop. However, in contrast to our hexagons, which provided a smooth, homogeneous surface that enabled unambiguous identification of neighbourhoods for the study area, Lüker-Jans et al. (2016) analysed metrics on a municipal level, which resulted in a higher variance in shape and size than grid-based analysis.

General Discussions

Other studies which identify landscape types consider similar indicators such as plot size, share of cropland, built-up or linear landscape elements (Levers et al. 2018; Tieskens et al. 2017; van der Zanden et al. 2016). In relation to the European archetypes by Levers et al. (2018), our identified types are in line with the intensity classifications for Brandenburg (i.e. large share of cropland or as landscape mosaic). However, our clusters considered additional, region-specific plot-based information with a focus on agricultural landscape structure, diversity and management characteristics. Similar to van der Zanden et al. (2016), Tieskens et al. (2017), who also take into account landscape composition, structure and management indicators but on a European level, our types range from small to large farming scale as indicated by median plot sizes, and can be further differentiated by edge density (potential linear elements) and land management intensity (approximated by the share of cropland and maize). Several studies showed that Brandenburg is characterised by agriculture under medium to large scale arable land (Andersen et al. 2013; Levers et al. 2018; van der Zanden et al. 2016). In addition to the identification landscape types, our study analysed the spatial patterns of characteristics and types showing that particularly high intensity and organic farming tend to be spatially concentrated.

Our results complement and broaden information about agricultural landscapes, such as the agro-ecological zones of Brandenburg (Landbaugebiete), that have been given a suitability rating for crop production potentials (Ackerzahl; Landesamt für Ländliche Entwicklung, Landwirtschaft und Flurneuordnung 2016) and the maps available in the Thünen Atlas, including the distribution of crop types or grassland on a municipal scale (Thünen Institut, 2014). Using a plot-based analysis, we add new information to the existing data regarding composition, configuration, diversity, and management, including intensity, which may become relevant also from a perspective of resilience and climate smartness of landscapes. Earlier typologies of Brandenburg’s agriculture have been based mainly on farmers’ decisions and referenced renewable energy production (Venghaus and Acosta 2018). These authors considered farmers as decision makers as the designers of agricultural landscapes; however, the explicit landscape composition scale as we do here was not considered explicitly. In this study, we used different metrics as inputs for typologising agriculture on a landscape level and thus complement the results of Uthes et al. (2020) and Venghaus and Acosta (2018), who focused their analysis on the farm level.

Our approach innovatively integrates different metrics into a new land-use typology, which we consider an improvement over the use of single indicators (e.g. soil quality). The newly provided information improves the understanding of the agricultural landscape structure in Brandenburg and helps identifying regions specified support measures may be required. Additionally, the region-specific types can be used for monitoring changes over time or assessing changes after e.g. policy measures have been implemented from an outcome-based perspective. Resulting changes may for instance include the frequency and spatial distribution of identified types, but may also generate new types that are not included in the six used in this study.

Agricultural production in particular creates pressure on the environment whereby appropriate farming systems help preserve landscapes and habitats (Lütz and Felici 2009). From the European policy perspective, the supply of ecosystem services and biodiversity conservation within farmland is fostered by the EU Biodiversity Strategy to 2020 and within the CAP’s greening measures (Weissteiner et al. 2016, p. 317). Typologising the agricultural landscape in Brandenburg allows for the comprehensive assessment of the potential prioritization of areas for the supply of environmental measures, such as the implementation of green infrastructures to support landscape diversity and connectivity supported by multifunctional agriculture (Oberlack et al. 2019). Through typologising agricultural landscapes, along with estimates of the provision of ESS by type offers a step forward towards landscape-type functioning assessment. Exemplarily, Type 3 (low fragmentation) and Type 4 (high intensity) could indicate low habitat diversity due to large median plot sizes, low edge density and a high share of cropland and maize. Contrary, Type 2 (high fragmentation) and Type 5 (low intensity) indicate a potentially higher habitat diversity through a comparably low share of cropland and agriculture in general as well as through high edge densities. At the same time, in agricultural landscapes with a high share of organic agriculture (particularly Type 6) the farming systems are potentially offering enhanced ESS, e.g. biodiversity, soil quality or pollination services, and at the same time show low environmental impacts of agricultural production (Bavec and Bavec 2015). Based on this approach, likewise, typologising could be combined with climate smartness assessments, and help policy makers in defining and evaluating respective agricultural landscape feature goals.

Although our results are shown for an exemplary case study of Brandenburg, Germany, the methodology can be applied to other regions where sufficient data is available such as other regions in Germany and the EU. The integration of metrics via cluster analysis may result in different (number of) typologies in other areas, which is, however, one advantage of utilizing small-scale region-specific data rather than generalized types. The landscape focus enhances a more integrated assessment of agricultural landscapes than the focus on pure farm size and farm-based characteristic. Furthermore, a typology based on landscape structure, diversity and management is independent of the area of application and can thus be ubiquitarily applied as a general framework for the characterization of agricultural land.

Limitations and Further Research

Similar to Lomba et al. (2017), Uthes et al. (2020) and Lüker-Jans et al. (2016), we were able to show the potential of IACS data for analysing agricultural land use. Future backing through remote sensing data, such as crop type mapping (Griffiths et al. 2018), crop yield mapping (Lobell et al. 2015) or landscape pattern analysis (Weissteiner et al. 2016), would increase the potential for this approach to be applied to areas in which frequent land-use monitoring is not available. In this study, we did not consider the temporal dimension of land use, e.g. crop rotation and crop diversity over time. Applying our proposed method to different time slices would make it possible to address changes in the set of indicators and the resulting clustering. This would reveal processes that occur in the agricultural landscape and could help identify how changes in boundary conditions would impact the composition of a landscape. In Brandenburg, two prominent examples of such processes are the increase of maize in the crop portfolio and the construction of biogas plants in direct response to implementation of the Act on Granting Priority to Renewable Energy Sources in 2000 (Federal Environmental Ministry 2000). Such developments would be revealed by an analysis with multiple time periods of IACS data. Furthermore, our results do not represent a full set of potential agricultural landscape types, e.g. across the whole EU under different landscape structure, diversity and farmland management characteristics. We argue, however, that the methodological approach is highly transferable to other regions in the EU, where IACS data are available; missing soil quality assessments could for instance be replaced by increasingly available remote sensing data.

A common problem in ecological analyses of spatial indicators is the spatial scale and the unit of analysis known as the Modifiable Area Unit Problem (MAUP) (Wu 2004) which is not quantitatively analysed within this study. Scale dependence (of metrics and number of clusters) could be addressed by performing a sensitivity analysis of e.g. changing grid cell size in future studies. In earlier studies, however, landscape metrics have proven to be a suitable tool for landscape analysis, even though there are limitations when it comes to up- and down-scaling of the generated results (Schlesinger and Drescher 2018).

Conclusions

This paper focuses on the methodological suitability of standard and landscape metrics as an input for cluster analysis within a hexagonal grid. One of the advantages of using IACS data is the potential to transfer the approach to other study regions in which similar monitoring data are available.

Our findings reveal six different types of agricultural landscapes and their respective spatial patterns. We conclude that Brandenburg is characterised by highly fragmented agriculture and a high degree of spatial clustering of intensive agriculture and organic production. The chosen landscape metrics derived from IACS data have proven to be adequate for improving the understanding of agricultural landscapes. The approach could potentially be applied for measuring agricultural landscape structure and diversity in terms of plot composition and configuration at the EU level since IACS data are available across the EU. Our paper proposes an approach at the landscape level, which, according to Thomson et al. (2019), provides a fundamental connection between the diverse array of relevant socio-economic and biophysical conditions and processes and can inform particularly regional decision-making.

In addition to performing spatio-temporal analysis, future work should address the relations among different types of agricultural landscapes and land price development, ownership patterns and trade-offs, for example between food and energy production, particularly on different units of analysis in regard to decision-making units.

Availability of data and material (data transparency)

Not applicable.

Code availability (software application or custom code)

Not applicable.