Habitat composition near linear landscape structures across Poland: perspectives on pollinator conservation

Databases and definitions of the LLSs

To identify the surrounding landscape composition of LLSs in Poland (Fig. 1A), we used freely available databases from Geoportal (Geoportal, 2023a) and the European Space Agency (ESA; European Space Agency, 2023). Geoportal provides a database (Topographic Object Database: BDOT10k) containing the spatial location of topographic features in Poland (Figs. 2A–2C) along with descriptions of their properties (Geoportal, 2023b). ESA provides a database (WorldCover V2 2021) with global land cover products for 2021 at 10 m resolution, which were developed and validated in near-real time based on Sentinel-1 and Sentinel-2 satellite data (Zanaga et al., 2022). Land cover was independently validated with a global overall accuracy of 76.7% (Zanaga et al., 2022). The WorldCover database, which covers the whole world, enables applying the approach presented in the study to other countries and regions with a possibility to compare results. Thus, we obtained Geoportal-based datasets for railway, highway, and levee identification. We divided Poland into 16 administrative regions (voivodeships) and used ESA-based datasets for the assessment of land cover surrounding LLSs.

We used the objects of layers SKDR01 (highways) and SKDR02 (express roads; Fig. 2A) for the study based on the BDOT10K database. The railways were defined as objects of layer SKTR0, excluding railway sidings (Fig. 2B). Levees were objects of the layer BUZM02 (Fig. 2C).

LLS sampling

We used stratified random sampling to assess the variation in landscape composition of the LLSs across Poland. A total of 7,500 random sampling points were generate, stratified across the 16 strata (16 regions of Poland: voivodeships; Figs. 2D–2F). The number of points per stratum was determined by the length of an LLS in a stratum as a proportion of the total length of the LLSs for Poland. The median point number per stratum was 32 for highways (range: 11–73; total: 573; Fig. 2D), 201 for railways (125–483; 3,821; Fig. 2E), and 163.5 for levees (48–424; 3,106; Fig. 2F). The minimum distance between sampling points within each stratum was two km (distance to nearest point (median); railways: 2,873 m; levees 3,259 m; highways: 4,194 m).

For each sampling point, we extracted the coordinates (longitude and latitude), LLS class (railway, highway and levee), administrative region (16 voivodeships), and landscape characteristics.

Measuring landscape characteristics of the LLSs

We summarised the surrounding land use by creating three buffers (250 m, 500 m, 1,000 m) around the overall LLS lines of each type as well as around each sampling point (Fig. 1B) and then extracting the percentage of a land use class within the buffers using WorldCover V2 2021. Selected buffers correspond to distances covered by most pollinators (Zurbuchen et al., 2010), especially bees, which are central place foragers (Michener, 2000). Moreover, landscape characteristics affect most pollinators at referred scales of 250–1,000 m (Steffan-Dewenter et al., 2002; Saturni, Jaffé & Metzger, 2016; Lajos et al., 2021; Misiewicz, Mikołajczyk & Bednarska, 2023; Kammerer et al., 2024). Distance covered by pollinators is another important life history trait affecting pollination efficiency (Abrol, 2012). We treated the 250 m buffer as the reference scale because many small-bodied species cover only short distances, i.e., of less than 300 m (Gathmann & Tscharntke, 2002; Zurbuchen et al., 2010).

We used all 11 land use classes of WorldCover V2 2021 to give an overall percentage area within each buffer that consisted of land use categories: “bare/sparse vegetation”, “built-up”, “cropland”, “grassland”, “herbaceous wetland”, “mangroves”, “moss and lichen”, “permanent water bodies”, “shrubland”, “snow and ice”, and “tree cover” (Zanaga et al., 2022). However, at the scale of Poland, there is no “mangrove” class, and the cover for some classes is lower than 5 km² (“moss and lichens”–4.85 km², “shrubland”–0.25 km², “snow and ice”–0.06 km²). Thus, these classes were excluded from the analysis and visualisation of the data. The most predominant land use classes were tree cover (127.0 * 10³ km), crop cover (109.8 * 10³ km), grass cover (60.1 * 10³ km), and built cover (8.7 * 10³ km).

Data analysis and visualisation

The inverse Simpson (D) index of landscape cover diversity was calculated for each sampling point to quantify landscape diversity. The value of this index starts with one representing a landscape containing only one land cover type. The higher the value is, the greater the diversity. The maximum value is the number of land cover types; here seven land cover types in this study.

For assessing differences in landscape covers between LLSs we estimated effect sizes using Hedge’s g or Cohen’s d formulas. The Hedge’s g formula with bias correction was used when the sample size was less than 30, i.e., for surrounding landscape of LLSs vs. Poland’s landscape; otherwise, we used Cohen’s d formula (Khan & McLean, 2024). The paired Cohen’s d was used for analysing landscape of LLSs according to three spatial scales; otherwise, we used unpaired formulas. Effect sizes were estimated with 95% confidence intervals (CI).

Differences in distributions of standardised covers between LLSs were investigated with Kolmogorov–Smirnov test. By, visualising differences between empirical distribution functions (ECDF) of LLSs landscape covers, we could identify at what cover values the differences were the greatest. The shape of distributions may be described with help of skewness, which is a measure of distribution asymmetry. A right-skewed distribution shape indicates that most of the surrounding landscape of sampling points are characterised by relatively small cover values, i.e., less than mean/median. In contrast, a left skewed distribution shape indicates that most of surrounding landscape of sampling points are characterised by cover values larger than mean/median.

The database management and map visualisation were performed using ArcGIS 10.8.2 (Environmental System Research Institute, 2020) and QGIS 3.22 (QGIS Development Team, 2022).

All data analysis and visualizations were undertaken using R ver. 4.3.0 (R Core Team, 2023) and the following packages: Durga (Khan & McLean, 2024), dyplr (Wickham et al., 2023), ggplot2 (Wickham, 2016), ggpubr (Kassambara, 2023), jpeg (Urbanek, 2022), moments (Komsta & Novomestky, 2022), readr (Wickham, Hester & Bryan, 2024), and vegan (Oksanen et al., 2022).

Surrounding landscape of LLSs vs. Poland’s landscape

Land cover within buffers of 250 m around highways, railways, and levees corresponded to the overall proportion of land cover in Poland (Fig. 3). The landscapes most different from the median land cover for Poland were tree cover within highway buffer (0.41 vs. 0.25, Poland vs. LLSs; Fig. 3A), grass cover within levee buffer (0.18 vs. 0.28; Fig. 3B), crop cover within levee buffer (0.30 vs. 0.22; Fig. 3C), built cover within highway buffer (0.03 vs. 0.13; Fig. 3D), bare cover within highway buffer (0.0009 vs. 0.0076; Fig. 3E), water cover within levee buffer (0.01 vs. 0.05; Fig. 3F), and wetland cover within levee buffer (0.002 vs. 0.014; Fig. 3G).

Surrounding landscape of LLSs according to spatial scales

The cover of the most common land types (cover of tree, grass, crop and built) in selected buffers (250 m, 500 m, and 1,000 m) around LLS sampling points changed mostly in a unidirectional manner. The land cover of trees (0.147, 0.212, and 0.267 for 250 m, 500 m, and 1,000 m buffers, respectively), and crops (0.271, 0.351, and 0.387) around highway buffers increased with the buffer size (Figs. 4A, 4C). In contrast, grass (0.224, 0.194, and 0.187) and built cover (0.135, 0.080, and 0.057) decreased (Figs. 4B, 4D). The rail buffer cover of tree (0.342, 0.331, 0.347) and grass (0.141, 0.157, 0.163) were not affected by buffer size (Figs. 4E, 4F); however, crop (0.103, 0.165, 0.212) cover increased (Fig. 4G) and built (0.041, 0.038, 0,038) cover decreased with buffer size (Fig. 4H). The levee buffer area did not affect tree (0.326, 0.313, 0.326) cover (Fig. 4I); however, the cover of crops (0.078, 0.151, 0.214) and build (0.000, 0.007, 0.015) increased (Figs. 4K, 4L) but grass (0.218, 0.212, 0.206) decreased with buffer size (Fig. 4J). Across LLS types, only crop cover uniformly changed, i.e., increased with buffer size (Figs. 4C, 4G, 4K).

Surrounding landscape of the LLSs according to cover distribution shapes

The distributions of tree, grass, crop, and built cover were highly right-skewed for all LLSs 250 m buffers around the sampling points (Fig. 5). This distribution shape type indicates that most of the surrounding landscape of sampling points have relatively small cover values, i.e., less than mean. The tree cover distribution shape was similar for rail (skewness = 0.483) and levee (0.572) buffers; however, for highways (0.971), the share of tree cover decreases more steeply (Fig. 5A). The distribution shapes of grass cover within 250 m buffers of highway (0.900), rail (1.333), and levee (0.976) sampling points were similar to each other (Fig. 5B). Considering the distribution of crop cover around the 250 m sampling point buffers, the distributions of rails (0.955) and levees (1.326) are similar, but highways (0.356) were characterised by a larger share of 0.25–0.75 crop cover (Fig. 5C). Although the distribution shape of built cover within 250 m buffers rapidly decreased for levees (4.166), the declines for highways (1.578) and railways (1.839) were lower (Fig. 5D). Differences between ECDF of LLS landscape covers were the greatest at the low cover values (below 50%; Supplement S1), except tree cover for rails and levees where the ECDF differences were the greatest at the high tree cover (above 50%; Supplement S1C).

Surrounding landscape diversity of LLSs

The inverse Simpson index revealed that the landscape diversity of seven land covers (Fig. 3) within a 250 m buffer around sampling points was the highest for highways (D = 2.44; Fig. 6), low for levees (D = 2.23; Fig. 6), and the lowest for railways (D = 2.03; Fig. 6).

Surrounding landscape of LLSs vs. Poland’s landscape

Landscape composition surrounding LLSs well mirrored land use types at the national scale (Fig. 3). Most pollinator hotspots in temperate climates are linked with open habitats, i.e., grass cover. Grasslands are under severe pressure, as traditional land use is increasingly replaced by either highly intensive agricultural practices or abandonment (Noordijk et al., 2009). Thus, mostly grassy levee and highway verges can potentially support diverse communities of pollinators as a supplementary habitat due to the larger proportion of grass cover in the surroundings compared to Poland’s cover (Fig. 3B). However, woodlands can also act as a habitat for pollinator communities (Alison et al., 2022); therefore, mostly grassy rail verges can be considered as a supplementary habitat likely boosting pollinator diversity due to the larger proportion of wood cover in the surroundings compared to Poland’s cover (Fig. 3A).

Wetlands are threatened habitats worldwide due to changes in land use, especially in the EU and North America (Davidson, 2014). The highest proportion of the habitat consisted of buffers surrounding levees (Fig. 3G). Thus, actions considering wetland conservation and restoration should include dryer and unflooded levee verges as a supplementary habitat. Interestingly, although most bee species are ground nesters, wetland areas may be rich in many bee species (Moroń et al., 2008).

From food safety and economy perspectives, pollinator services are mostly linked to commercial crops that are dependent on pollination (Potts et al., 2010; Abrol, 2012). Among the three studied LLSs, highway verges may provide pollination services to the largest proportion of crop land cover (Fig. 3C). Thus, landscape managers should identify the potential of highway verges as a supplementary habitat to increase pollinator services provided to nearby crops. Additionally, highways are surrounded by landscapes with a higher proportion of built cover than the other LLSs. In light of climate change and predicted food security concerns (Van der Sluijs & Vaage, 2018), growing food in and around cities could also be a partial solution to sustainably increasing food production in an urbanised world (Nicholls et al., 2023). Thus, the potential of highway verges should be considered as a potential supplementary habitat in human-dominated landscapes.

Surrounding landscape of LLSs according to spatial scales

Bees, the most important pollinators in the temperate zone (Abrol, 2012), as central placed foragers, use only part of the available landscape because they cover limited distance (Zurbuchen et al., 2010). Distances realised by foraging bees are a function of the species body size, which translates into longer foraging routes of large-bodied bees (Greenleaf et al., 2007). Typically, most bee species cover up to 500 m from the nest; however, some social species, i.e., Apis spp. or Bombus spp., are able to cover more than one km (Steffan-Dewenter et al., 2002).

Flies and butterflies are not central place foragers and are potentially able to cover longer distances to search for a suitable patch. However, these pollinators avoid crossing a patch boundary (Moroń, Skórka & Lenda, 2019) and leave a patch only once the cost of continued foraging outweighs the amount of energy gained (Tillman & Adelman, 2023). This finding translates to the high range of distances covered by different butterfly species (Skórka et al., 2013b; Nowicki et al., 2014; Stefanescu et al., 2016); for flies, no such systematic studies are known.

Many studies indicate that environmental factors affecting pollinator populations operate locally, i.e., floral diversity (Moroń et al., 2014), or concern the closest landscape, i.e., distance to the nearest semi-natural habitat (Skórka et al., 2013a), rather than at a broader landscape scale. Our study shows that the land cover surrounding LLSs is relatively uniform at the 250 m, 500 m, and 1,000 m scales (Fig. 4). This means that conservation programs at LLSs can target pollinators with different life history traits connected to habitat preferences, (e.g., nesting in wood) and mobility (e.g., body size). Additionally, geographic information system (GIS)-based plans that consider the suitability of LLSs for pollinators as supplementary habitats may be limited to one spatial scale, e.g., 250 m, saving time and reducing the analysis cost (Rands & Whitney, 2011).

Surrounding landscape of LLSs according to cover distribution shapes

The distribution of land cover is highly right-skewed (Fig. 5), and there are many sampling points whose surrounding landscape is characterised by relatively small cover values, i.e., less than mean. Moreover, the shapes of the distributions are more similar between railways and levees than for highways (Fig. 5). Taking into consideration the shape of the land cover distributions, railways and levees seem to be equivalent, whereas highways have a more distinct distribution of land cover proportions. Thus, plans targeting pollinator diversity and their ecosystem services in LSSs as supplementary habitats should consider the potential differences in landscape cover distribution patterns and the feasibility of using different types of LLSs as alternatives, given that the three studied LLSs share a structure. The relatively larger proportion of LLSs with smaller values of the covers (i.e., right-skewed distributions) may imply that most LLS management schemes may affect several land cover types.

Surrounding landscape diversity of LLSs

An important characteristic of landscapes that support biodiversity is habitat complexity (Bottero et al., 2023). If a landscape is more diverse, more potential food resources and nesting sites are available for pollinators (Gathmann & Tscharntke, 2002). Moreover, among pollinator species, many have narrower food or nesting preferences; thus, available habitats are even more needed for specialised species (Hadley & Betts, 2012). Here, we show that the probability of containing more land use types in the near surroundings is higher for highways than for other studied LLSs (Fig. 6).

Unfortunately, predicting the effects of additional habitat types in the landscape on pollinator richness and abundance is not trivial, mostly because ecological processes seem to be idiosyncratic, i.e., include nonlinear relationships (Moroń et al., 2019). However, knowledge about the more diverse land cover of the surrounding LLSs can be incorporated into ecological management programs. On the basis of our findings, we emphasise that highways should be prioritised as a supplementary habitat to sustain pollinator populations and their ecosystem services, considering the landscape diversity.