The study was carried out in the Trentino province, an area in Northern Italy covering about 6,214 km². The area is generally mountainous, but it is characterized by a considerable landscape heterogeneity, with about 80% semi-natural areas, 15% agricultural areas, and 5% urban areas. As a result, the climate is highly variable. The mean annual temperature is about 9°C, and the mean annual precipitation at 200 m a.s.l. is 1,200 mm.

We selected 13 sampling locations, which were characterized by great variability in landscape composition (Supplementary Table 1; Figure 1). Three sites were close to apple orchards, three were close to vineyards, three were in an urban setting, and three were far from agricultural areas. The average elevation of the sampling locations was 533 m a.s.l. (min = 93 m a.s.l., max = 1,481 m a.s.l.). The mean air temperature at the sampling sites during the sampling periods, i.e., from April to September of 2019 and 2020, was 17°C.

In 2019 and 2020, we placed one small apiary consisting of two honey bee colonies at each sampling location. All colonies were originated from the livestock managed by Edmund Mach Foundation and had sister queens of A. m. carnica x A. m. ligustica. The colonies were managed directly by Edmund Mach Foundation personnel according to the local beekeeping practice. From April to September, we carried out pollen samplings every month, for a total of six pollen samples collected per colony per year. At some locations, however, the number of pollen samples was lower due to adverse climatic conditions. In particular, in 2019, at one location only three samples could be collected, and only five at other two. In 2020, only five samples were collected at three locations. Pollen samples were collected activating pollen traps at the hive entrance for 48 h, Pollen samples were then stored at −20°C.

For each sampling location, we extracted the cover of the main habitat types using the CORINE Land Cover (CLC) database (© European Union, Copernicus Land Monitoring Service 2018, European Environment Agency) at three spatial scales, i.e., the local foraging scale of honey bees (1 km radius buffer around the sampling locations) and two landscape foraging scales of honeybees (3 and 5 km radius buffers around the sampling locations). Following the CLC classification, we considered a total of 24 land-use classes (Supplementary Table 2). Landscape composition was heterogeneous across sites. In 3 km radius buffers around the sampling locations, an average of 33% of the land was covered by agricultural areas (min = 7%, max = 52%), 51% by semi-natural areas (min = 26%, max = 92%), 13% by urban areas and other artificial surfaces (min = 0, max = 30%), and 2% by other areas (min = 0, max = 13%) (Supplementary Table 1).

As most of these classes were highly correlated, we performed a Principal Component Analysis (PCA) to extract the landscape composition at each of the three spatial scales. We extracted the first two eigenvalues, PC1 and PC2, which explained 38, 48, and 45% of the total landscape variability at 1, 3, and 5 km radius buffers around the sampling locations (Supplementary Table 3). PC1 was positively related to semi-natural areas, in particular coniferous forests, natural grasslands, and areas with sparse vegetation, and negatively related to intensive areas, in particular urban areas and vineyards (Supplementary Table 3). Therefore, high values of PC1 can be interpreted as a high proportion of semi-natural areas within the landscapes.

Moreover, we calculated the Shannon diversity index for landscape composition using the 24 land-use classes at each of the three spatial scales. Shannon diversity index quantifies the heterogeneity of landscapes, taking into account both richness and evenness of land-use classes, with low values of the index indicating a low landscape heterogeneity. Shannon diversity index was calculated using the R package vegan (Oksanen et al., 2020). All statistical analyses were performed using the R software version 3.6.1 (R Core Team, 2019).

From each pollen sample, we extracted two grams of pollen pellets, which were dissolved in distilled water and mixed using an advanced vortex mixer (VELP Scientifica, ZX3). We took 20 μl of the obtained suspension and placed it on a microscopic slide. Once the suspension of water and pollen was dry, we placed a drop of glycerin jelly on top of the sediment, and we covered it with a slide. Pollen was then observed under the optical microscope (Optika, B500PPH). For each sample, we counted about 500 pollen grains applying the “transect” method (Tamic et al., 2011). Pollen grains were identified at the lowest possible taxonomic level according to available literature (Ricciardelli d'Albore, 1998; Bucher, 2004; El-Labban, 2020) and palynological databases (PalDat, 2000; PollenAtlas). The identified pollens were classified following the “pollen types” nomenclature proposed by Persano Oddo and Ricciardelli d'Albore (1989).

For each pollen sample, we calculated the Shannon diversity index. As for landscape composition, the Shannon index for pollen reflects both richness and evenness of pollen samples, with lower values indicating a lower diversity in pollen sample composition. Moreover, to understand how pollen composition changed throughout the flowering season, we calculated the mean beta richness and replacement at each location over the six sampling months, based on presence/absence data. All pollen indices were calculated using the R package vegan (Oksanen et al., 2020).

First, to determine the effect of landscape composition and seasonality on pollen diversity, we built two linear mixed-effect models for each spatial scale using the R package nlme (Pinheiro et al., 2019). In all models, the response variable was pollen Shannon index. Selected explanatory variables were collection month, year, landscape Shannon index, and the interaction between month and landscape Shannon index for the first model, and collection month, year, PC1, PC2, and the interactions between month and PC1 and between month and PC2 for the second model. We also included the sampling location as random factor in all models. Starting from each full model, we used a backward deletion procedure, removing one-by-one the interactions with p-value > 0.05, and re-ran the model to correctly interpret model main effects.

Second, to explore the effect of landscape composition on temporal beta diversity of pollen, we built four linear models for each spatial scale. We selected beta richness and replacement of pollen as response variables, PC1 and PC2 as explanatory variables for the first model, and landscape Shannon index as explanatory variable for the second model.

Honey bees, despite being extraordinarily polylectic, usually select a limited number of flowering plant species to forage on (Lau et al., 2019). Here, we observed a strong temporal turnover in the composition of pollen collected by honey bees, which partly reflects honey bee foraging preferences, and partly reflects plant phenology and pollen availability in the study area, at least in the early and late flowering season. In particular, trees were revealed to be a key resource for honey bees in spring. Salix spp. in April, Malus/Pyrus spp. in May, and Castanea sativa in June were the main pollen taxa collected by honey bees. The importance of trees for honey bees is well-known (Donkersley, 2019), as they are often among the early-flowering species. We showed that Compositae such as Taraxacum spp., Helianthus spp., and Senecio spp., and the legume Trifolium repens strongly supported honey bees throughout spring and summer. In August, Artemisia spp. pollen was highly represented, as this taxon is common in fallows and urban areas. In September, almost all pollen was collected on Hedera spp., which was the most abundant flowering plant species in late summer in the study area (Prosser et al., 2019).

Pollen diversity was also shaped by seasonality. We observed that the start and end of the flowering season, i.e., April and September, were characterized by a dearth of floral resources, while we observed a peak of pollen diversity in May. Interestingly, despite the high percentage of agricultural areas in certain landscapes, we did not highlight a strong effect of mass flowering crops such as apple, which flowers in April and May, which can potentially reduce the diversity of pollen collected by honey bees, as bees tend to focus on these resources. The diversity of pollen collected in August was surprisingly high, given that resources are usually relatively scarce at the end of summer (Garbuzov et al., 2015; Requier et al., 2015; Danner et al., 2017; Sponsler et al., 2020). The high diversity found in August could be explained by the mid-elevation of sampling sites, which causes a shift in pollen decline from August to September.

Since many of the pollen taxa that we collected were grouped at the family level, and almost all the remaining ones were identified at the genus level, pollen diversity could have been even higher in some seasons, if we had been able to achieve species-level identifications of pollen.

Our results highlighted that diversity of pollen collected by honey bees was influenced by the interaction between collection month and landscape composition, i.e., the proportion of semi-natural areas.

Pollen diversity was independent of the proportion of semi-natural areas from April to August. Several studies found that pollen composition was not affected by landscape composition (Danner et al., 2017; Guzman et al., 2019; Simanonok et al., 2020; Jones et al., 2021). This can be explained by both the structure of selected landscapes and the foraging behavior of the honey bee. Even if some landscapes were strongly modified by anthropogenic activities, they always included a certain proportion of semi-natural habitats, ranging from 25 to 83%, which comprised both open habitats and forests. Moreover, many agricultural areas, which ranged from 7 to 52%, were intermixed with semi-natural habitats. In all landscapes, honey bees were therefore able to collect pollen in areas that offered a high amount of resources, at least until mid-summer. Moreover, honey bees can travel more than 10 km from their hive, although they usually forage <1 km away from the hive (Von Frisch, 1967; Visscher and Seeley, 1982; Tautz, 2008; Seeley, 2019). Several studies highlight that landscape composition affects the distance to which honey bees forage, and in particular that their foraging distance increases in simplified landscapes (Steffan-Dewenter and Kuhn, 2003; Abou-Shaara, 2014; Danner et al., 2017). A study from the UK reports that in landscapes dominated by the common heather (Calluna vulgare), the average distance of foraging honey bees strongly changed during the season: while in May it was about 1 km, in August, during the flowering period of the common heather, the average foraging distance increased up to 5.5 km (Beekman and Ratnieks, 2000). In our study, however, we did not collect data on how far honey bees traveled to collect pollen. In some areas, honey bees may need to travel much further to obtain food resources, with potentially negative consequences for colony fitness.

On the other hand, in September, pollen diversity increased with increasing proportion of semi-natural habitats at 3 and 5 km radius buffers. While honey bees are able to collect heterogeneous pollen independently of landscape composition from spring to mid-summer, the scarcity of floral resources in late summer may turn semi-natural areas into key habitats. In this part of the season, when nectar sources are decreasing (Tew et al., 2022), honey bees search for the most diverse pollen sources in order to breed winter individuals that must develop adequate fat bodies (Frias et al., 2016). Semi-natural areas in the landscape can promote late-season pollen protein and winter survival of honey bee colonies (Kuchling et al., 2018; Simanonok et al., 2020; Rutschmann et al., 2022). Therefore, the higher number of resources offered by these habitats can be crucial for honey bees at such a critical stage of the colony cycle.