Introduction

Grasslands, along with forests and shrublands, cover most of the Earth’s terrestrial surface (Buchhorn et al. 2020). Grasslands are mostly disturbance-dependent ecosystems, meaning that they often require human management, such as mowing or grazing, to maintain their diversity and productivity (Blüthgen et al. 2012; Schils et al. 2022). Thus, the changes in grasslands area extent (e.g. afforestation), as well as the changes in grassland species composition, are strongly related to human activities.

Grasslands also provide a variety of ecosystem services (Isselstein and Kayser 2014; Bengtsson et al. 2019; Zhao et al. 2020) including provisional (e.g. food supply, freshwater supply, fuel supply, regulating (e.g. climate regulation, carbon sequestration, soil erosion control, water regulation) and cultural ecosystem services (e.g. aesthetic appreciation and cultural heritage).

Historically, grassland dynamics in Europe was primarily linked to its interchange with croplands (Fuchs et al. 2015). Such interchange between both the agricultural land use systems likely resulted either from the short-term crop rotation practices, or from the long-term land use changes, such as abandonment and recultivation. In recent decades, however, urbanisation, afforestation and the abandonment of marginal areas have been the primary drivers of declining grassland extent in Europe (Huyghe et al. 2014).

The persistence and changes in the grassland cover in Europe are strongly affected by the national policies (Prishchepov et al. 2017; Commission et al. 2020) (Prishchepov et al. 2017). For instance, the widespread abandonment of the agricultural land in Eastern Europe following the collapse of socialism (Müller et al. 2009; Munteanu et al. 2014) and later countered by recultivation triggered by Common Agricultural Policies (CAP) in the EU member countries (Pazúr et al. 2020). Grassland development is also further by the changes in the livestock population and feeding habitats, the implementation of milk quotas, a shift between hay and silage making and adoption of weed control and irrigation measures (Peeters 2012).

The importance of grassland ecosystems in Europe is underlined by various initiatives and projects related to monitoring of grassland biodiversity and management (Kolecka et al. 2018; Griffiths et al. 2019). Three landcover mapping initiatives, supported mostly by the European Union’s Earth observation programme (Copernicus Programme 2021), provide multitemporal spatial information on grassland distribution. The Land Use/Cover Area Survey (LUCAS) is a point-based mapping initiative, that combines the visual interpretation of aerial images with massive in situ data collecting. The data from LUCAS are currently available for 2006, 2012 and 2018 (Eurostat 2015). The LUCAS datasets provide in situ information with high precision and reliability, including landscape photos and different environmental parameters (Weigand et al. 2020; d’ Andrimont et al. 2020). Another mapping initiative, Copernicus high-resolution layer (HRLs), provides grassland maps based on radar and optical data in 20-m and 10-m resolution for 2015 and 2018, respectively (Copernicus Land Monitoring Service 2020). Long-term changes in the European landcover, however, are only covered with the CORINE land cover dataset (Feranec et al. 2010; Bielecka and Jenerowicz 2019), which provides land cover maps for 1990, 2000, 2006, 2012 and 2018. A limitation of CORINE land cover dataset is its insensitivity to small-scale changes due to the minimum mapping units (25 ha for areal phenomena, a 100 m for linear phenomena and 5 ha for changes of land cover class) and areas of change assessed (Diaz-Pacheco and Gutiérrez 2014; Pflugmacher et al. 2019).

Besides these mapping initiatives supported by the European agencies, grassland management intensity at the EU level was mapped using MODIS NDVI time series from 2000 to 2012 (Estel et al. 2018). On a national scale, the frequency and timing of grassland mowing events were mapped in Germany with the use of Landsat and Sentinel-2 satellites (Griffiths et al. 2019). A similar evaluation was performed at the district scale in Switzerland (Kolecka et al. 2018). Using Landsat satellite imagery for mapping abandonment and recultivation trends on grassland areas was the focus of various research projects related to the agricultural land cover change in Central and Eastern Europe (Alcantara et al. 2012; Griffiths et al. 2013). However, an encompassing assessment of long-term dynamics of grasslands across the whole Europe is so far missing. Such information may bring important insights into the dynamics of the grassland ecosystems, reveal regional differences and indicate potential factors influencing the dynamics of the world’s biggest biome.

This paper aimed to map the grassland change trajectories and their spatial patterns at the regional and national scale in Europe. Starting in 1990, we covered most of the European countries and their grassland development within the periods of 1990–2000, 2000–2006, 2006–2012 and 2012–2018.

Specifically, we asked the following research questions:

  1. a)

    What were the grassland cover development trajectories in Europe since 1990?

  2. b)

    What was the spatial distribution of different trends of grassland dynamics in Europe?

Data and methods

We utilised the CORINE land cover datasets to assess the dynamics of grasslands in Europe considering the land cover changes in 1990, 2000, 2006, 2012 and 2018. CORINE land cover contains 44 different land cover classes at the third hierarchical level of the nomenclature. The hierarchical structure of the nomenclature divides the five main land cover classes at the first hierarchical level: artificial surfaces, agricultural areas, forest and seminatural areas, wetlands and water bodies. The classification system is well established in European-based studies and may link to different classification systems, such as FAO, through its metadata (O’ Brien C.D. et al. 2021). The minimum polygon mapping area is 25 ha and the minimum land cover change mapping area is 5 ha (Heymann 1994; Bossard et al. 2000; Feranec et al. 2016).

Our analysis is based on two land cover classes related to grasslands:

  • pastures and meadows (CORINE land cover class 231) referring to grassland areas used for agricultural purposes such as mowing and livestock grazing (it includes permanent and temporal grassland and does not include the grassland of rotation system) (Bossard et al. 2000);

  • natural grasslands (CORINE land cover class 321) including natural grasslands such as alpine grasslands, arid limestone areas or steppes (Bossard et al. 2000);

Assessment of the grassland dynamics

We assessed the dynamics and the development trajectories of grasslands using 1 km × 1 km square grid spanning the entire European terrestrial area. For further analysis, we selected all grid cells containing more than 0.05% (0.5 ha) of grassland areas. The trajectory of changes in these grid cells was compared to the reference land cover dataset for 1990 dataset. Furthermore, we classified the grid cells based on their development trajectories using the following criteria:

  • increasing development trajectory (A) - areas of grasslands in 2000, 2006, 2012 and 2018 are larger than in the reference year of 1990 (see Table 1);

  • decreasing development trajectory (B) - areas of grasslands in 2000, 2006, 2012 and 2018 are smaller than in the reference year 1990

  • mixed development trajectory (C) - areas of grasslands in all years of observation have mixed extent within the squares (a combination of increasing, decreasing and equal extent as compared with the reference year of 1990).

Table 1 Illustration of the possible development trajectories of the grassland area. The rows refer to the reference period of 1990, while the columns stand for any period different from 1990 (2000, 2006, 2012, 2018). The arrow represents either the (↑) increasing, (↓) decreasing trajectory or its combination (↑↓) is the mixed trend. The (=) sign refers to the stability of the grassland area in the particular grid and period of change
Full size table

To further investigate the specific land cover changes that contributed to the observed development trajectories, we evaluated the transition matrix between different land cover classes (CORINE land cover 2nd classification level). For this purpose, we selected 12 hotspot areas of 25 km × 25 km, four for each analysed grassland development trajectory (increasing, decreasing, mixed). Hotspot areas were identified using a moving window approach quantifying the presence of the development trajectory within a 25 × 25 km2 window weighted by the proportion of grasslands within that specific window. For each hotspot area, only land cover changes that were directly related to grasslands or accounted for more than 25% of the selected square area were considered relevant to describe the trends in the selected landscape.

Results

Grassland development trajectories in Europe

In total, 31% of all 1 km × 1 km grids across Europe contained the defined minimum grassland coverage in 1990. Most of these grids were stable from 1990 to 2018 (41%), followed by decreasing (35%), increasing (20%) and mixed (3%) trajectories of grassland development (Table 2). Most of the stable grasslands were concentrated in the northern part of Europe (Fig. 1), in the countries with various proportion of grassland areas (Norway, Sweden, Belgium, Denmark) (Fig. 2). The appearance of both the increasing and decreasing development trajectory in a single country was found in Ireland, in the region of Central Europe (Austria, Germany), Eastern Europe (Romania) as well as in the Mediterranean region (Northern Macedonia, Albania) (please refer to the supplementary material (Appendix A)). Decreasing trajectories showed hotspots in Mediterranean Europe, particularly in southern Italy and western Spain. Decreasing grassland development trajectories were also observed in the Baltic countries (Lithuania, Latvia, Estonia), as well as in Slovakia, Ukraine and western England. Increasing grassland development trajectory was found on more than 35% of the analysed squares in the Czech Republic, especially in its border regions. Hotspots of grassland increase were also found in the central parts of Germany, Austria and western England.

Table 2 Summary of trajectories of grassland change in the period 1990–2018
Full size table
Fig. 1
figure 1

Grassland development trajectories in Europe in the period 1990-2018. (increasing ↑; decreasing ↓; mixed ↑↓; == stable). Rectangles illustrate the allocation of the selected representative areas of: increasing (A), decreasing (B), mixed (C) grassland development trajectories. The two-letter country codes are based on the ISO 3166 standard for country identification

Full size image
Fig. 2
figure 2

Grassland development trajectories in the different European countries covered in this study. The two-letter country codes are based on the ISO 3166 standard for country identification

Full size image

Representative areas of different trajectories of grassland change

The landscapes of selected 25 km2 areas were mostly composed of multiple patches of agricultural areas and forests (Fig. 3). The dynamics between the land cover classes (ranging between 8 and 49 % of the total area) were mostly linked to the interchange between arable land, grasslands and heterogenous agricultural areas. The increase in grassland extent was observed mostly as a transition from arable land and heterogenous agricultural areas (Fig. 4). The decrease of grassland was related to the opposite transition to arable land and heterogenous agricultural areas but also to the transformation to urban areas. Mixed development trajectories within their selected hotspots were related to the interchange between cropland, grassland, heterogeneous agricultural areas, shrubs and open spaces with no vegetation.

Fig. 3
figure 3

Land cover in selected areas representing A increasing, B decreasing, C mixed grassland development trajectories. Grassland classes are highlighted in green colour, the grey shades stand for non-grassland land cover classes. The land cover classes are (11) urban fabric, (12) industrial, commercial and transport units, (13) mine, dump and construction sites, (14) artificial, non-agricultural vegetated areas, (21) arable land, (22) permanent crops, (23) pastures, (24) heterogeneous agricultural areas, (31) forests, (32) scrub and/or herbaceous vegetation associations, (33) open spaces with little or no vegetation, (41) inland wetlands and (51) inland waters

Full size image
Fig. 4
figure 4

Flowcharts of land cover change trajectories on selected areas representing A increasing, B decreasing, C mixed of grassland development trajectories. Grassland classes are highlighted in green colour, the grey shades stand for non-grassland land cover classes. The codes represent the land cover as follows: (11) urban fabric, (21) arable land, (22) permanent crops, (23) pastures, (24) heterogeneous agricultural areas, (31) forests, (32) scrub and/or herbaceous vegetation associations, (33) open spaces with little or no vegetation

Full size image

Discussion

Identified trends in grassland change across Europe

Using the land cover datasets, we evaluated different grassland development trajectories across Europe during the period of 1990–2018. We summarised the occurrences and changes in grassland areas within 1 km2 grids, which enabled us to identify hotspots of grassland development trajectories across Europe. Although numerous studies have focused on the European-wide evaluation of land cover change hotspots (Kuemmerle et al. 2016), persistence (Lieskovský and Bürgi 2018), land change trajectories (Stürck et al. 2018), archetypal patterns and trajectories (Levers et al. 2018), our research adopts a unique approach, focused on the trajectories of specific land cover class.

We focused on grasslands, being one of Europe’s most widespread land cover classes and are largely influenced by the dynamics of nature and human decisions. Prior research has explored the extent of grassland at the European scale in the context of the intensification and abandonment of agricultural production (Feranec et al. 2010; Alcantara et al. 2013; Estel et al. 2015). Both processes are important indicators of grassland change as grassland biodiversity faces a strong decline due to the intense change of agricultural production (Kuhn et al. 2021). In this paper, we used a different approach by focusing on the long-term development trajectories in gains and losses of grassland areas across Europe.

We found that grassland persistence was the most prevailing across Europe, with hotspots in intensively used lowland areas, such as the agricultural land in Belgium and the Netherlands. These grasslands are notable for high demand on multiple ecosystem services such as provisioning of fodder, grazing areas, recreational use, ecological corridors or cultural values (Gulickx et al. 2013; Estel et al. 2018). The observed spatial patterns are in line with the spatial patterns of persistence of human appropriation of net primary production derived based on CORINE land cover datasets (Plutzar et al. 2016) but showed different patterns from long-term European land cover persistence with the hotspots in the Alps, north and south Italy, north Ireland, Sweden, south Finland, Romania and the UK (Lieskovský and Bürgi 2018).

We also found that from 1990 to 2018, Europe has continuously lost a significant part of its grassland areas. The hotspots of declining trajectories were found in the former Eastern Bloc countries (such as the Baltic countries, Slovakia and Romania), where agricultural production collapsed after the breakdown of socialism and partly recovered only after decades (Griffiths et al. 2013; Pazúr et al. 2014). The recovery of agricultural production largely varied between countries and was largely related to the support gained from different subsidy schemes, such as the Common Agricultural Policies. The hotspots of declining trajectories were found in western England, northern Germany and the border regions of Austria. The trend of grassland decrease in western England was confirmed by the more detailed national-wide dataset and has been attributed to the expansion of the woodland and settlement structures (UKCEH 2020). In the case of Germany, similar areas of the northern part of the country have experienced in recent years an increase in land use intensities (mowing events) (Schwieder et al. 2022). Interestingly, an increase of the land use intensities within the areas of decreasing grassland extents was also found on selected study areas where grasslands have been transformed to permanent crops (e.g. fruit trees), arable land or were lost due to the expansion of settlement areas. Considering the natural values and ecosystem services the grassland provides, such losses negatively affect the species richness (Hilpold et al. 2018) and release the carbon stored in the grassland soils (Reinermann et al. 2020). Similarly, grassland transition towards other land cover classes often comes at cost of losses of the cultural heritage or cultural landscape (Plieninger et al. 2015).

Increasing development trajectories of the grassland areas have been found in central France, southern Germany and Ireland. These hotspots coincide with high mowing frequencies, as identified by the MODIS satellite (Estel et al. 2018), which may indicate the grasslands expansion in areas suitable for agricultural use. Such expansion may also be driven by the subsidy schemes where grassland management turned out to be the most supported agricultural land use strategy. For example, along the borders of the Czech Republic were croplands transformed into grassland mostly due to the subsidy schemes and ownership change (Kupková et al. 2021).

Observed grassland development trajectories demonstrate that national-wide policies likely formed the heterogeneity of grassland development across Europe. While the resulting maps highlighted spatial gradients in the grassland change across landscapes in Europe, we also identified the heterogeneity of the changes in the trends of grassland cover between the national borders. Examples may be found in the border region of the Netherlands and Germany, where the state border divided the observed trend between stable grasslands and grassland expansion, respectively. Similarly, between the state borders of Austria and Switzerland, we found that grassland simultaneously increased in the flat areas likely suitable for agriculture and has been overgrown by woody vegetation on some of the alpine meadows (Buchgraber et al. 2011), while alpine grasslands with similar bioclimatic conditions in eastern Switzerland appeared to be more stable.

Limitation of our study

Several factors may influence the accuracy of our maps and the interpretation of the outcomes. Firstly, the resulting maps are influenced by the mapping methodology of CORINE land cover datasets. The CORINE land cover dataset has a minimum mapping unit of 25 ha and a minimum identified change area size of 5 ha, which means that it is insensitive to small-scale changes (Diaz-Pacheco and Gutiérrez 2014; Pazúr et al. 2014; Cegielska et al. 2018). Additionally, the CORINE land cover mapping methodology has evolved over time, which can lead to inconsistencies in comparing the land cover patches across different periods (García-Álvarez and Camacho Olmedo 2017). Despite these limitations, the CORINE land cover has been proven to be a suitable land cover data source for land cover change mapping at the European level (Feranec et al. 2010; Hatna and Bakker 2011; Anderson and Mammides 2020). Secondly, our maps were produced on the aggregated data from their native 100-m pixel resolution to a 1 km2 grid. This means we did not evaluate the inner dynamics of grassland change within each square grid. Consequently, some areas that were indicated as stable may exhibit dynamics within those 1km square grids. However, we assume that such patterns do not substantially influence the spatial variation of the mapped trajectories across Europe. Thirdly, the CORINE land cover datasets do not provide a continuous mapping but only cover specific points. This means that the mapping of the trends is based on five different time frames. Although available satellite-based data nowadays allows for continuous mappings of the land surface (Woodcock et al. 2020) based on the solid training data (EUROSTAT 2017) and robust classification models (Belgiu and Drăguţ 2016), we believe that for the time frame of 1990–2018, the CORINE land cover brings better accuracies on the European scale. This is because the CORINE land cover datasets result from a rigorous process of mapping and state-wide validation, which is managed by the European Union’s Earth observation programme.

Considering the scale of this study, we certainly missed out on particular hotspots and important local specificities of grassland change. To encourage the readers to investigate the different trends in the CORINE land cover dataset, we have provided a link to a code in the supplementary material (Appendix B) that may be used to visualise the different CORINE land cover datasets and grassland datasets and download the European-wide maps of development trajectories of grassland cover. The code and related links are executable in the Google Earth Engine platform (Gorelick et al. 2017).

Conclusion

The proposed study highlights the different developmental trajectories of the grassland cover across Europe from 1990 to 2018. We used a specific methodology that summarised the trajectories across spatial and temporal scales. We found that, besides the persistence of grassland, the decrease of grassland areas was the most prevailing development trajectory, followed by the increased development trajectory of grassland areas. Using the Earth Observation data revealed that the patterns of the grassland development trajectories have certain spatial gradients across Europe. These gradients are often influenced by the national policies. Such findings indicate a need to define an EU-wide strategy better to target the preservation of grassland ecosystems in Europe. The Earth observation data can provide the necessary spatial information to better target policies on different hotspots of landscape change.