Linear woodlots increase wild bee abundance by providing additional food sources in an agricultural landscape

Mid‐field woodlots play an important role in maintaining biodiversity in agricultural landscapes. However, it is not clear whether non‐linear or linear woodlots are most beneficial for wild bee conservation. We assessed the attractiveness of two common types of woodlots in an agricultural landscape in northern Poland (non‐linear and linear: 7 and 9 sites, respectively) in terms of wild bee abundance, species richness, and functional diversity. Linear habitats had higher abundance of wild bees. However, woodlot type did not affect wild bee species composition or functional trait composition. Species composition responded significantly to measures of syntaxonomic heterogeneity and landscape heterogeneity. Woodlot area, landscape context (isolation and landscape heterogeneity), and syntaxonomic heterogeneity explained most of the differences among habitats (non‐linear vs. linear) in wild bee abundance and species richness, regardless of the habitat type. The higher attractiveness of linear woodlots was due to increased food availability in the herbaceous layer in the spring–summer (June) and summer (July–August) periods. Linear woodlots have the potential to be used as tools for integrating agricultural production with biodiversity conservation and ecosystem services.

Wooded patches and semi-natural habitats without woody plants, such as meadows and grassy field edges in agricultural areas are often the only refuges of wildlife in such constantly disturbed environments (Forman & Baudry, 1984). Studies show that mostly linear woodlots are subject to degradation or complete destruction (Croxton et al., 2002;Dajdok & Wuczy nski, 2016;Marttila-Losure, 2013;Robinson & Sutherland, 2002). Protection and management of non-crop habitats, such as wooded networks, that is, networks composed of non-linear and linear woodlots, can be one of the management policies (Baudry et al., 2000;Merckx et al., 2012;Morandin & Kremen, 2013) included in the EU Biodiversity Strategy for 2030 (EC, 2020).
Many studies emphasize the important role of non-crop habitats in the agricultural landscape for preservation of beneficial species diversity of various groups of arthropods (Diekötter & Crist, 2013;Garibaldi et al., 2014;Hendrickx et al., 2007;Kells et al., 2001;Knapp &Řezáč, 2015). The existence of various types of mid-field woodlots has a positive influence on species richness and abundance of wild bees as well as pollination services (Alison et al., 2022;Banaszak & Szefer, 2014;Hannon & Sisk, 2009;Kremen & M'Gonigle, 2015;Morandin & Kremen, 2013). Such habitats provide wild bees with nesting sites and a continuous source of forage (Banaszak, 1983;Banaszak & Cierzniak, 2002;Croxton et al., 2002;Rivers-Moore et al., 2020). In contrast, crops generally do not enable nesting, although they are usually a rich but asynchronous and short-term source of forage Proesmans et al., 2019). However, to the best of our knowledge, small non-linear woodlots (not classified as forest) and linear woodlots (except for hedgerows which in the European agricultural landscape have a long history of being managed; and shelterbelts that are planted to protect an area, especially a farm field, from strong winds and the soil erosion) within rural landscapes have not been compared so far.
For example, Banaszak and Cierzniak (2002) focused on bees of shelterbelts, forest islands (any forest patch surrounded by fields, irrespective of its size; remnants of extensive woodlands or aggregations of trees and shrubs formed spontaneously in areas excluded from cultivation), and roadsides of agricultural landscapes.
Many local and landscape factors can affect wild bee abundance and species richness in mid-field woodlots. Significant factors include isolation (Steffan-Dewenter & Tscharntke, 1999), habitat area (Steffan-Dewenter, 2003), food resources (Alison et al., 2022;Kratschmer et al., 2019), and landscape heterogeneity (Jauker et al., 2009;Twerd & Sobieraj-Betli nska, 2020). Generally, isolation combined with low landscape heterogeneity and small food resources can have a negative effect on the local abundance and species richness of bees. The importance of continuous provision of floral resources for bee communities during the growing season was confirmed in many studies (Oertli et al., 2005;Scheper et al., 2014;Timberlake et al., 2019). Linear woodlots are characterized by a long edge zone, where herbaceous edge communities can develop if ploughing and other agrotechnical practices are not conducted under the tree canopy (Cierzniak, 1996). In this zone, plant communities typical of ecotones are found (Matuszkiewicz, 2008), which are very valuable sources of bee forage (both pollen and nectar). Woodlot cores often contain few or no flowers after canopy closure, while the edges often have rich floral communities throughout the season. Moreover, higher attractiveness of linear woodlots in comparison to non-linear ones in agricultural landscapes may result from greater variation in the vertical structure of vegetation: the tree, shrub, and herbaceous layers. In general, wild bees are more numerous and diverse when farmlands are close to non-crop habitats, for example, hedgerows, shelterbelts, or tree lines (Banaszak & Cierzniak, 1994;Castle et al., 2019;Morandin & Kremen, 2013;Rands & Whitney, 2010). For example, Hannon and Sisk (2009) showed that hedgerows attract more diverse and abundant bee communities than woodlots (non-linear habitats) on farms or unmanaged native woodland off farms, because of poor food resources in the latter habitats. However, direct comparisons of how floral resources change throughout the growing season between nonlinear and linear woodlots have not yet been studied.
Various functional traits of bees may be associated with how species use various components of landscape (Hall et al., 2019;Luck et al., 2012). These traits include, for example, sociality, nesting location, dietary specialization, body size, phenology, and rarity of species (Hall et al., 2019;Kratschmer et al., 2019). Apart from species richness, their functional diversity is an important component of diversity. Because of this, increasing attention is paid to functional diversity indices (functional evenness, functional dispersion, Rao's quadratic entropy Q, and functional divergence), taking into account various taxa and habitats, including bees (Brito et al., 2018;Forrest et al., 2015;Hall et al., 2019;Hoehn et al., 2008;Sydenham et al., 2016). The functional diversity of bees allows more precise prediction of pollination effectiveness than species diversity alone (Fründ et al., 2013;Hoehn et al., 2008). For bees of mid-field woodlots (non-linear and linear), as far as we know, little progress has been made in evaluation of functional diversity. Only one melittological study is focused on linear networks (i.e., roadsides with trees and streams with trees) and scattered trees in rural landscapes of north-central Victoria, Australia (Hall et al., 2019).
In this study, we compare two types of mid-field woodlots, distinguished on the basis of their physical structure, that is, shape (non-linear woodlots vs. linear woodlots). We analyzed if the abundance, species richness, and functional diversity of wild bees differed between non-linear woodlots and linear ones. We hypothesized that mid-field linear woodlots are more attractive for wild bees in respect of species richness, abundance, and functional diversity than nonlinear woodlots. Considering the above, in our study, we attempted (1) to evaluate which type of woodlots is more favorable for the preservation of a high abundance and species richness of wild bees; (2) to test whether the combined effect of multiple vegetation layers can provide food sources for wild bees throughout the season; (3) to examine how local and landscape factors (habitat area, isolation, syntaxonomic heterogeneity, and landscape heterogeneity) relate to wild bee abundance, species richness, species composition, and functional trait composition; and (4) to evaluate differences in functional diversity between non-linear and linear woodlots.

Study area and study sites
Field research was conducted in the agricultural landscape of northern Poland near the city of Bydgoszcz, west of its northern districts (53 N, 17 E). The study area (about 453 ha) is located in a large farm owned by the Potulicka Foundation (CGFP), which applies the principles of precision agriculture in an area of over 5000 ha. Farmland dominates in the study area, including arable fields (85%) and meadows (only 0.6%). The major crops are: maize for grain and silage, wheat, oilseed rape, and smaller fields of lupin and pea (CGFP, 2021).
Field research was conducted in 7 non-linear woodlots and 9 linear woodlots ( Figure 1 and Figure S1; Table S1). Non-linear and linear woodlots were randomly distributed in the study area. We classified as woodlots any patches covered with trees and/or shrubs and linear structures with woody plants (Knioła, 2016) covering at least 20% of their area.
The woodlots were not managed by systematic pruning. We could not clearly determine whether the woodlots were planted or grew spontaneously. Linear woodlots were borders between adjacent crop fields or along more or less frequently used roads. The non-linear woodlots were environmental islands surrounded by farmland. It should be noted that the linear woodlots were not typical shelterbelts or hedgerows, and the non-linear woodlots were not forests, but small patches (< 2 ha) of noncrop habitats with trees and shrubs. The nearest distances among neighbouring sites ranged from 12.30 to 1692.42 m (mean AE SE = 625.63 AE 34.57 m). The following criteria of woodlot classification based on their physical structure were adopted ( Figure S2): • non-linear woodlots-patches with trees (coverage 1%-80%) and shrubs (coverage 15%-95%), if wider than 20 m (predominantly), irrespective of the length to width ratio, and covering up to 2 ha; or if predominant woodlot width was up to 20 m and the length to width ratio was up to 5 (Knioła, 2016); • linear woodlots-narrow belts or rows of trees (coverage 15%-40%) and shrubs (coverage 10%-75%), which were predominantly up to 20 m wide and had a length to width ratio higher than 5 (Knioła, 2016).

Local and landscape factors
At each site, we determined the environmental variables at the level of habitat and landscape (predictors: C1-C9) (Table S1) , so its circum- p , and after simplification πA p , with site area in this case expressed in m 2 . The more its shape differs from a circle, the higher is the value of CBD (Kujawa, 2006).
To estimate the degree of isolation (r i ), we designated a buffer zone with a radius of 500 m around the borders of each plot. This distance is associated with predicted range of wild bee foraging, while a close location of suitable nesting sites and foraging within a distance of several dozen metres is crucial for survival of their populations (Greenleaf et al., 2007;Zurbuchen et al., 2010). Isolation of woodlots was calculated using the formula r i ¼ 1 n Â P j¼n i¼1 d ij (Forman & Godron, 1986), where: d ijdistance between the ith woodlot and a neighbouring jth woodlot; and n-number of neighbouring woodlots of the ith woodlot.
All plant species and plant communities were recorded at the study sites, including the transects analyzed. Vegetation data were collected from the same transects as the bees. Every week, all blooming species of flowering plants were recorded as potential sources of F I G U R E 1 Representative photographs of study sites: (a, b) non-linear woodlots; (c, d) linear woodlots food-nectar and pollen for wild bees (Table S2). Names of plant species and families follow Rutkowski (2008). On the basis of floristic data, we assessed the following factors for each woodlot: mean coverage of food plants in the tree layer (C4), mean coverage of food plants in the shrub layer (C5), and mean coverage of food plants in the herbaceous layer (C6).
The applied measure of heterogeneity (H) is based on the Shannon diversity index (Shannon & Weaver, 1963). To calculate it, we transformed the alphanumeric codes of cover-abundance according to the Braun-Blanquet scale into numbers: r into 1; + into 2; 1 into 3; 2 into 5; 3 into 7; 4 into 8; and 5 into 9 (van der Maarel, 1979). The heterogeneity of plant communities was calculated using the formula: of syntaxa in the woodlot; and p i -cover-abundance of ith syntaxon in relation to the sum of cover-abundance of all syntaxa in the woodlot.
The list of plant communities is presented in Table S3, where their systematics follow Raty nska et al. (2010).
Landscape structure was analyzed around each woodlot within a 500-m buffer zone from its borders, using an orthophotomap on a scale of 1:500 (source: the Main Office of Geodesy and Cartography) and on the basis of own field research. Within the buffer zone, boundaries of individual patches of habitats were drawn, and their surface area was measured using QGIS 2.14.10 Essen software (QGIS Development Team, 2020). Land use was classified into 11 classes: forest, woodlot, cropland, grassland, built-up area, orchard, wasteland, roadside, surfaced road, unsurfaced road, and water. This allowed us to determine the landscape heterogeneity in the buffer of 500 m radius (C8) ( Table S1). The In addition, the percentage of oilseed rape cover in the landscape buffer of 500 m (C9) was estimated, that is, the area of rape fields in the buffer zone divided by the total buffer zone area (Table S1).

Wild bee community sampling and characteristics
Research was conducted after obtaining oral consent from managers of the land and in accordance with applicable law. Insects were caught in Moericke traps (Moericke, 1951) and additionally, we used the transect method (Banaszak, 1980). This was motivated by the low effectiveness of Moericke traps for larger bee species, especially of the genus Bombus (Geslin et al., 2016). Using both methods, we collected insects in weekly intervals from early April till late September in 2016, along a transect. The transect was located in a representative part of each study site, that is, it was positioned in the south-facing edge of the given woodlot (determined after exploration and visual evaluation). In the case of non-linear woodlots, the transect was located at the woodlot edge, but at the middle of a long edge. We did not catch honeybees Apis mellifera Linnaeus, 1758 because the abundance of this managed species is likely to be related to beekeeping (Rivers-Moore et al., 2020). Finally, all bees sampled by the transect method were added to the data obtained from Moericke traps, as these methods can be seen as complementary (Lerman & Milam, 2016;Normandin et al., 2017;Prendergast et al., 2022).

Bowl trapping
Moericke traps were yellow plastic bowls (20 cm in diameter and 9 cm deep) filled to 3 /4 with a mixture of water (94.2%), ethyl glycol (

Aerial netting
In each study site, the wild bees were caught along a transect, which was 200 m long and 1 m wide (Banaszak, 1980). During the sampling process, each transect was walked slowly and wild bees were collected with an aerial insect net. In order to maintain uniform collector bias throughout the study, all transect walks were done by one surveyor . Sample collection lasted about 30 min per transect. The clock was not stopped when insects were caught. The bees were collected when the weather was favorable for Apiformes activity, that is, with no or little wind (< 3 on the Beaufort scale) (Krauss et al., 2009) and there was at least 70% sunshine (cloudless sky). During field research, the temperature was always above 16 C, and never exceeded 25 C.

Identification and community description
The collected specimens were pinned and identified to the species level by A. S.-B, based on our bee reference collection and taxonomic literature. Species names followed Kuhlmann et al. (2021) (Table S4).

Data analysis
All analyses were performed using pooled data obtained by both sampling techniques (i.e., pan traps and aerial netting). Statistical analyses were made using R version 4.1.1 (R Development Core Team, 2021).
Individual-based bee species accumulation curves were generated to examine sampling effort for non-linear and linear woodlots. We aggregated wild bee data within each type of woodlots and calculated accumulation curves with the package iNEXT (Hsieh et al., 2016). Following MacGregor-Fors and Payton (2013), we used 84% confidence intervals to determine statistical significance of differences between accumulation curves with an error rate of 0.05. To estimate true species diversity in two types of woodlots, we used the Chao1 estimator (Chao, 1984); 1000 randomizations were employed.
Given the relatively short distances between adjacent sites, community samples are likely subject to spatial autocorrelation. We used Moran tests for abundance and species richness values normalized by a log transformation to examine this issue. We found that Moran's test was not significant for abundance (p = 0.131) nor species richness (p = 0.071).
We performed two sets of analyses using generalized linear mixedeffect models (GLMMs). The month of collection and site identity were taken as random factors. This allowed us to evaluate differences in wild bee abundance and species richness between the two types of habitats that are independent of phenological and spatial effects. In the first analysis, we evaluated differences in abundance and species richness of wild bees between non-linear and linear woodlots (i.e., only woodlot type used as a fixed predictor). In the second set of analyses, we tested how much of the differences in bee communities between habitat types can be ascribed to local and landscape factors (habitat area, isolation, syntaxonomic heterogeneity, and landscape heterogeneity). We repeated the first analysis, controlling for major determinants of wild bee abundance and species richness: area (C1), isolation (C3), syntaxonomic heterogeneity (C7), and landscape heterogeneity (C8) of the woodlots. These co-variates were selected a-priori for their well-known effects on wild bee abundance and species richness. Moreover, they had to be not directly related to the process of shape-based habitat type selection, not correlated strongly with each other, and their Variance Inflation Factor (VIF) values had to be lower than 5 (low to moderately correlated) in a model including habitat type ( Figure S3). Further, we performed four analyses where we removed individual co-variates (e.g., only C1), while keeping other co-variates in the model, to evaluate the effects of the removed co-variate on the significance of the habitat type variable in the model. We used glmm.nb function from the package MASS (Venables & Ripley, 2002) with the negative binomial distribution for abundance, and glmer function with the Poisson distribution from the lme4 library (Bates et al., 2015) for species number.
To compare temporal patterns in wild bee abundance, species richness, and cover of food plants at three layers of vegetation (tree, shrub, and herbaceous) between non-linear and linear woodlots, we used generalized additive mixed models (GAMMs) with site identity as a random factor. Abundance was log-transformed prior to the analysis. To evaluate variation in temporal change in abundance and species richness of wild bees, we simulated 99 values of log-transformed abundance and species richness from the fitted GAMMs for each time point and each woodlot type, and calculated difference between predicted values from the two habitat types. We used the gratia R library (Simpson, 2021) for data generation from the GAMMs.
To determine the influence of the distinguished local and landscape factors on wild bee occurrence, we used redundancy analysis  To explore relationships between bee functional traits and local and landscape variables (C1, C3, C7, and C8), we performed a fourth-corner analysis (Legendre et al., 1997) using the 'traitglm' function from the R mvabund library (Wang et al., 2021). The data were combined at the site level; no monthly division was included. Multivariate generalized linear fourth-corner models were fitted with a negative binomial distribution and a least absolute shrinkage and selection operator's (LASSO) penalty (i.e., method = glm1path). LASSO is an approach which penalizes coefficients that do not reduce Bayesian Information Criteria to zero. Model deviance was estimated using a Monte-Carlo resampling procedure (1000 resamples) to evaluate the global significance of traitenvironment relationships.
Using GLMMs, we compared individual indices between the two woodlot types (non-linear vs. linear). The data were combined at the site level; no monthly division was included. Because values of the above indices are bounded between zero and one, we used the beta distribution with the logit link function for the errors as implemented in the glmmTMB library (Brooks et al., 2017).

Characteristics of the wild bee fauna in mid-field woodlots
We recorded 7320 individual wild bees belonging to six families (Colletidae, Andrenidae, Halictidae, Melittidae, Megachilidae, and Apidae) and 134 species (see Table S4 for a complete list of species).
They account for about 28% of all bee species reported from Poland.  (Table S4).
The majority of individuals in mid-field woodlots were solitary, ground-nesting, polylectic, medium-sized, common bees emerging in spring (  (Table S4).
The species accumulation curves comparing the two types of woodlots reflected a good sampling effort, but the curves still did not reach saturation ( Figure S4). The expected bee species richness (esti-

Attractiveness of non-linear and linear woodlots for wild bees
Abundance and species richness of wild bees (non-linear vs. linear woodlots) Non-linear woodlots had higher mean coverage of food plants in the shrub layer (C5), lower mean coverage of food plants in the herbaceous layer (C6), syntaxonomic heterogeneity (C7), and coefficient of border development (C2) ( Figure S5 and Table S7). There were strong positive correlations between C6 and C7 and between C2 and C6, C7, and C8 (landscape heterogeneity) ( Figure S3). GLMMs controlling only for the phenology (sampling month) and site identity (site) showed significantly higher abundance (Z = 2.26, p = 0.024) but not species richness of wild bees (Z = 1.87, p = 0.062) in linear woodlots ( Figure 2 and Table 2) compared to non-linear ones. When area (C1), isolation F I G U R E 2 Comparison of observed values (gray points) of (a) abundance and (b) species richness of wild bees between non-linear and linear woodlots. A significant difference is indicated with an asterisk (*p < 0.05); black points indicate the means; whiskers indicate 95% bootstrapped CI's (C3), syntaxonomic heterogeneity (C7), and landscape heterogeneity (C8) were included in the model, the difference in bee abundance between habitats was no longer significant (Table S8). Habitat type was significantly, positively correlated with the wild bee abundance in the model with co-variates only if syntaxonomic heterogeneity was removed (Table 2).
Abundance and species richness of wild bees (nonlinear vs. linear woodlots) by sampling period Comparisons of temporal trends between two types of woodlots by using GAMM (site taken as a random factor) showed that during the spring-summer period (June) and partly during summer (July), wild bees  Table 3).
Differences between the predicted values of the fitted models indicated significantly higher wild bee abundance and species richness in linear woodlots in June and July, followed by a slower decrease in wild bee abundance later in the growing season (August and September) (non-linear smooth terms: F = 198.30, p < 0.001, and F = 221.80, p < 0.001, respectively; Figure 3c-d, Table S9).
In general, the coverage of food plants in the tree, shrub, and herbaceous layers in both types of woodlots was positively correlated with abundance and species richness of wild bees. The correlations were the strongest for tree and shrub cover and often weak or lacking for the herbaceous layer coverage ( Figure S6). However, these two habitat types differed in their coverage only in the herbaceous layer, which was higher in the linear woodlots throughout the year ( Figure   S7 and Table S10).
Factors affecting species composition and functional trait composition of wild bees in mid-field woodlots At the site level (all bee individuals collected throughout the year at each site), wild bee community composition did not differ between the two habitat types (RDA with habitat as predictor: F = 0.97, p = 0.468). The model including variables C1 (area), C3 (isolation), C7 (syntaxonomic heterogeneity), and C8 (landscape heterogeneity) was also not significant overall (F = 1.14, p = 0.217), although landscape heterogeneity was significant ( Fourth-corner analysis did not reveal associations between the abundance of different bee functional traits, local and landscape variables (multivariate test, fourth corner, p = 0.182; Figure S8).
T A B L E 2 Results of generalized linear mixed-effect models comparing wild bee abundance and richness in non-linear and linear mid-field woodlots Note: Two types of models were studied: (1) using habitat type as the only explanatory variable; (2) using selected variables as co-variables. In both models site and month of collection were treated as random factors. Variables statistically significant at p < 0.05 are bolded.
Functional diversity of wild bees (non-linear vs. linear woodlots) No differences were detected in the four computed indices of wild bee functional diversity at the community level ( Figure S9 and Table S11).

DISCUSSION
The results of the present study show that linear woodlots compared to non-linear woodlots in the agricultural landscape were characterized by significantly higher abundance but not species richness of the Apiformes. However, when controlling the phenology (random factor: month), site identity, site area, isolation, syntaxonomic heterogeneity, and landscape heterogeneity, no significant differences in wild bee abundance and species richness were observed. For wild bee abundance, woodlot type was significant in the model with co-variates only if syntaxonomic heterogeneity was removed. Considering that abundance is crucial for performing ecosystem services and is the best measure of species condition (Naeem & Wright, 2003;Woodcock et al., 2019), it can be supposed that linear mid-field woodlots create more favorable conditions for the Apiformes.
Because of the seasonal variation in bee fauna (Williams et al., 2001), it is important to investigate the use of various habitats during the whole growing season, rather than only in total, at the bee community level (Dar et al., 2017;Hannon & Sisk, 2009). A direct comparison of phenology in the two habitat types showed that linear woodlots were characterized by significantly higher values of abundance and species richness of bees in the spring-summer period F I G U R E 3 Log-transformed values of (a) abundance and (b) species richness of wild bees during the sampling period (spring: April-May; spring-summer: June; summer: July-September) in non-linear and linear woodlots; smoothers show differences between two habitats; both are statistically significantly different (p < 0.05); shading represents 95% confidence interval's. Differences in values of (c) abundance and (d) species richness of wild bees between non-linear and linear woodlots predicted from generalized additive mixed models with habitat type as a discriminating factor; site as a random factor and with a smoother fitted along time; values above the black line (zero difference) indicate a higher predicted value of a given characteristic (abundance or species richness) in linear woodlots; grey points indicate predicted differences (June) and partly in summer (July). This was most probably associated with the higher coverage of herbaceous food plants at the edges of linear woodlots. Besides, in linear habitats the spring peak of bees was higher (in early April), and after a significantly higher abundance and species richness of bees in June and July, bee abundance declined less quickly at the end of the growing season, that is, in August and September. The high bee abundance in early April was associated with massive food flights and courtships flights of mostly solitary bees, for example, A. minutula and A. nigroaenea in both types of woodlots, and A. haemorrhoa in non-linear ones. In linear woodlots, the peak of bee abundance in June and July was mostly determined by L. pauxillum and the second generation of A. minutula.
On a local scale, bee occurrence is significantly affected by habitat quality, characterized primarily by coverage of food plants.
Hedgerows are important refuges of bee food plants in rural landscapes, as they constitute a so-called stable 'food tape' for bees from early spring till late summer (Garratt et al., 2017;Hannon & Sisk, 2009;Morandin & Kremen, 2013). In this study, the coverage by bee forage plants in the tree, shrub, and herbaceous layers in both habitat types had a positive influence on the abundance and species richness of wild bees. However, it must be noted that correlations with bee abundance and species richness were the strongest for tree and shrub cover and often weak or lacking for the herbaceous layer coverage.
However, with this kind of general correlation, we do not expect to see a strong correlation, because we correlate general bee abundance and species richness with bee food plants in individual vegetation layers, which dominate at different times of the year. Nevertheless, the observed correlations suggest a strong influence of early flowering T A B L E 3 Results of generalized additive mixed models for wild bee abundance and species richness along time; with site taken as a random factor trees and shrubs and a supplementary role of the spring-summer and summer periods. In linear woodlots, the major bee food plants included both pollen-and nectar-producing species of trees and shrubs, for example, Prunus insititia L., Salix cinerea L., and Tilia cordata Mill., as well as herbaceous species, for example, Capsella bursa-pastoris (L.) Med., Lamium purpureum L., Arctium lappa L., and Ballota nigra L. The major bee forage plants in non-linear woodlots were represented by trees and shrubs, for example, Prunus avium L., P. insititia, Prunus spinosa L., S. cinerea, but also Prunus padus L., Pyrus pyraster Burgsd., Pyrus communis em. Gaertner, Acer platanoides L., and a tall herb Alliaria petiolata (Bieb.) Cav. et Grande, providing bees with nectar and pollen (Saunders, 2018). Wild bees need suitable floral resources (Scheper et al., 2015), which supply them with both nectar, as a source of energy, and pollen, as a source of protein for reproduc- Overall, the type of woodlots did not affect species composition or functional trait composition of wild bees. Woodlot community structure changed only at the sample level, when controlling for month and site identity. The lack of general differences in bee structure probably resulted from the immediate neighborhood of the woodlots. As in our study, bee community composition in hedgerows and woodlots on farms located in a mosaic of small-scale agriculture and natural vegetation in south-eastern Arizona did not differ (Hannon & Sisk, 2009). European research did not show differences in bee community structure between shelterbelts and forest patches in farmlands (Banaszak & Szefer, 2014). The structure of bee communities of mid-field woodlots in our study was significantly affected by landscape heterogeneity and syntaxonomic heterogeneity. The most heterogeneous agricultural landscapes, with large numbers of both wooded and treeless habitats are characterized by more diverse bee species (Jauker et al., 2009;Tscharntke et al., 2005). Also in our study, the higher heterogeneity of landscape around woodlots favored the occurrence of most of bee species at the level of communities. The significant impact of syntaxonomic heterogeneity on bee occurrence is probably due to the fact that plant communities varying in structure, for example, in respect of life-forms, density or height, shape a broad spectrum of microhabitats (Cierzniak, 2003). This may increase the chances that within the distance of their flights, the Apiformes will have access to all the environmental attributes necessary for their survival, that is, optimal foraging sites and nesting sites. Provision of such resources is a common objective of many activities aimed at improvement of habitats for pollinators in agroecosystems (Carvell et al., 2006;Pywell et al., 2005). For instance, the presence of grassy com-  (Banaszak & Romasenko, 2001), but in the studied woodlots their abundance was very low. A mosaic of various plant communities can also create favorable conditions for courtship flights of bees.
Apparently, courtship flights of some bee species take place in very specific places, often around tree crowns. For example, male Andrena flavipes and A. nigroaenea court around single trees and shrubs or their groups (Cierzniak, 2003). A lack of clear differences in community structure and diversity can also result from the fact that bee communities in agricultural landscapes are well mixed locally in respect of species, which is also linked with a lack of differences in functional traits.
We assumed that the analysis of functional diversity of bees would provide additional information on their structure in the two types of mid-field habitats. No significant differences in indices of functional diversity of bees in the studied habitats were found at the community level. These results can be compared with the findings of Hall et al. (2019), who reported that scattered farmland trees, wooded roadsides, and wooded streams in rural landscapes in southern Australia did not differ in bee functional divergence (RaoQ) and functional richness (FRic). Our results indicate that non-linear and linear woodlots are functionally similar with respect to bees. The lack of functional differences is because the microhabitats are not different, but their proportion is just slightly different, with more edge habitat in the linear woodlots. Moreover, the lack of these differences could also result from the local scale of the study, that is, at the level of one agricultural landscape. Depending on the studied species and applied methods, bees cover distances of 25-1400 m (Gathmann & Tscharntke, 2002;Geib et al., 2015;Greenleaf et al., 2007;Steffan-Dewenter & Kuhn, 2003;Zurbuchen et al., 2010), but for most of solitary bees, the distance is about 250 m (Greenleaf et al., 2007), while for bumblebees, 25-110 m (Geib et al., 2015). The shortest distances between the study sites explored by us varied from 12.3 to 1692.4 m (on average 626 m), that is, were within the flight range of some bee species, which could enable them to move freely between the study sites in the search for food (Hannon & Sisk, 2009). Moreover, solitary bees often use several habitats to find the necessary resources, moving between their nesting sites and foraging habitats, and searching for patchy, ephemeral floral resources (Hannon & Sisk, 2009;Westrich, 1996). The ability to discover functional differences between species increases with increasing number of functional traits taken into account in functional diversity calculations (Cadotte et al., 2011;Gutiérrez-Chac on et al., 2020). Considering that as many as six functional groups of bees were taken into account in this study, the possibility of classifying a species as redundant functionally was low.
Moreover, Petchey and Gaston (2006) reported that there is no 'correct' number of traits; in contrast, all the traits important from the functional point of view should be taken into account in calculations, so we did this. Besides, research on functional diversity on the basis of functional groups has some limitations, also in our study, as it assumes functional equivalence within groups and greater interspecific variation than intraspecific variation (Goldstein, 2018).
Linear woodlots in agricultural landscapes can increase the connectivity between other isolated pollinator populations in non-linear woodlots so well-connected networks of linear and non-linear woodlots are important to increase pollinator movement. The role of woodlot type in the wild bee community dynamics is unclear. Therefore, the outcome of our evaluation of linear and non-linear woodlots for wild bee communities could be different if the proportions of these habitats in the agricultural landscape were modified. For example, in the case of a much higher or much lower proportion of linear versus non-linear woodlots, significant differences in species richness and/or functional diversity of bees could be observed.

CONCLUSIONS AND IMPLICATIONS
Pollination is a key issue in agriculture. At the time when many pollinating insects populations are in decline (Ollerton, 2017;Zattara & Aizen, 2021), both non-linear and linear woodlots should not be regarded as limitations to farming practices but as assets, which must be taken into account in general management for more environmentfriendly agriculture (Rivers-Moore et al., 2020). The European 2030 Biodiversity Strategy (EC, 2020) postulates an increasing contribution of agriculture to biodiversity, functioning of ecosystems, and provision of ecosystem services in the European Union. Considering this, to maintain and increase biodiversity on a local scale in agricultural landscapes, it is necessary to protect the existing woodlots and create new ones. Recommendations for designing and establishing woodlots to improve agricultural habitats often focus on birds and mammals (Blakesley & Buckley, 2010;Rosenberg et al., 2003), but marginalize or omit wild bees. The results of this study show that linear woodlots increase the abundance of wild bees by providing additional forage sources for bee in a rural landscape. This is due to differences in the temporal dynamics of plant communities during the growing season, possibly as a result of better developed plant communities along the edges of these habitats. Protection of existent woodlots and establishment of new linear ones (e.g., planting of trees in areas where agriculture is unprofitable or along roads, irrigation ditches, water bodies or watercourses) in intensively farmed landscapes seem to be the least problematic from the point of view of property owners, as they do not cover any arable land or only its small proportion. Such activities can be a 'low cost-high benefit solution' (Dainese et al., 2017).
Linear woodlots are also easy to establish, offering an interesting potential as a tool of integration of agricultural production with biodiversity conservation and ecosystem services (Baudry et al., 2000;Merckx et al., 2012;Morandin & Kremen, 2013). The financial inputs into maintenance and management of non-crop habitats in agricultural landscapes is a compromise between profits from those components of landscape and a loss of productive areas (Morandin et al., 2014).