We reviewed the literature for information on floral preferences of polylectic solitary bees with the focus on O. bicornis. Polylectic bee species collect pollen and nectar from a range of plant species for the provisioning of their brood. Although not specialised on a specific plant taxon, polylectic bees do exhibit preferences for some floral resources rather than collecting pollen and nectar indiscriminately from plants blooming in their surroundings (Bosch et al. 2001; Danforth et al. 2019). However, the preferences of polylectic solitary bee species are often not well known because they require analysis of pollen composition of brood provisions across habitats. Most published studies reporting pollen compositions of brood provisions focus on Megachilidae bees, including species of the genera Megachile and Osmia (Raw 1974; Cripps and Rust 1989; O’Neill et al. 2004; Teper 2008; O’Neill and O’Neill 2011; Gresty et al. 2018; Nagamitsu et al. 2018; Söderman et al. 2018; Bertrand et al. 2019; Killewald et al. 2019; Vaudo et al. 2020). The most data on pollen compositions were found for the red mason bee, O. bicornis L., a polylectic, solitary bee species native to Europe and an important pollinator of crops (Bosch et al. 2008). The bee is active in the spring when the females collect pollen and nectar to provision brood cells in existing above-ground cavities. Studies reporting pollen compositions of O. bicornis provisions were conducted in multiple regions in Europe, including locations in Poland, Czechia, Germany, Sweden, Switzerland and Great Britain (Raw 1974; Teper and Bilinski 2009; Jauker et al. 2012; Haider et al. 2014; Peters et al. 2016; Gresty et al. 2018; Persson et al. 2018; Ruddle et al. 2018; Söderman et al. 2018; Bertrand et al. 2019; Ryder 2019; Šlachta et al. 2020; Splitt et al. 2021; Yourstone et al. 2021; Bednarska et al. 2022; Misiewicz et al. 2023). Landscapes were mostly agricultural with several studies conducted in areas with oilseed rape fields. However, landscape compositions and the availability of various flowering crops and wild plant species varied widely between studies. Thus, the plant taxa identified from the bee-collected pollen provided a general picture of the floral preferences of O. bicornis independent of floral resources available at a single location.

We compiled the reported pollen compositions from the studies of O. bicornis and summarised the findings (Table 1). Thereby, we summarised reported plant species into higher-level taxa because pollen was identified to various taxonomic levels within and across studies. Pollen from oak (Quercus), Brassicaceae (including oilseed rape) and buttercups (Ranunculaceae) were commonly found in O. bicornis provision samples in all studies. In addition, pollen from Rosaceae (including non-cultivated as well as orchard trees and berries) and other tree species were reported across most studies. We further compared the studies reporting pollen compositions of O. bicornis provision samples quantitatively across identified plant taxa (Peters et al. 2016; Persson et al. 2018; Ruddle et al. 2018; Bertrand et al. 2019; Ryder 2019; Splitt et al. 2021; Yourstone et al. 2021; Bednarska et al. 2022; Misiewicz et al. 2023). The three most commonly observed plant taxa (Quercus, Brassicaceae and Ranunculaceae) also were the ones making up a high percentage (> 50%) of pollen in some samples. In studies that sampled brood provisions at multiple time points, pollen compositions shifted during the study period, reflecting the flowering phenology of the plant species (Persson et al. 2018; Ruddle et al. 2018; Bertrand et al. 2019; Ryder 2019; Yourstone et al. 2021; Bednarska et al. 2022). Percentages of Brassicaceae pollen (identified in the studies as Brassicaceae, Brassica or Brassica napus = oilseed rape) in O. bicornis provision samples ranged from 0% to 73% across studies providing detailed sample composition data. In four studies, flowering oilseed rape fields in the landscape were explicitly part of the study design (Peters et al. 2016; Ruddle et al. 2018; Yourstone et al. 2021; Bednarska et al. 2022). O. bicornis were released at nest boxes within or adjacent to flowering oilseed rape fields or at defined distances from the fields. In these studies, the average and maximum proportion of Brassicaceae pollen in provision samples was higher compared to the studies that did not focus on oilseed rape. For a quantitative comparison between studies, we assigned the reported pollen taxa to four categories: deciduous trees (not including Rosaceae), Brassicaceae (assumed to correspond mainly to oilseed rape, Brassica napus), Rosaceae (including, but not limited to, fruit/orchard trees and berries that may be cultivated) and non-cultivated herbaceous plants (including Ranunculaceae) (see Suppl. material 1). The proportion of pollen in provision samples originating from the four categories varied considerably and was likely strongly dependent on the study design, landscape composition and timing of the sampling. In most studies, deciduous trees (often mostly Quercus) made up the highest proportion on average (Persson et al. 2018; Ruddle et al. 2018; Bertrand et al. 2019; Splitt et al. 2021; Yourstone et al. 2021; Bednarska et al. 2022). However, Misiewicz et al. (2023) reported Brassicaceae at highest percentage across samples, Peters et al. (2016)Rosaceae pollen and Ryder (2019) non-cultivated herbaceous plants. Yourstone et al. (2021) did not identify any Rosaceae pollen in their samples and Brassicaceae pollen was absent or observed at very low proportion (< 3%) at some study sites from Ryder (2019), as well as Splitt et al. (2021). It should be noted that pollen from annual (field) crops other than Brassicaceae were uncommon in the review of O. bicornis pollen analyses. However, Ryder (2019) reported pollen from Raphanus (genus including cultivated radishes) in a subset of pollen samples, with a maximum of 70% of Raphanus pollen observed in a single sample. Bertrand et al. (2019) found pollen from the genus Vitis (including cultivated grapes) with sample proportions between 0 and 80%. Misiewicz et al. (2023) reported Poaceae pollen in some samples at proportions > 50%, but it was not resolved whether the pollen originated from cultivated cereals or other grasses. Poaceae were also reported making up 30% of pollen extracted from scopae of O. bicornis collection (museum) specimen (Haider et al. 2014). The information from the review of literature on pollen preferences of O. bicornis was used to generate the input to the SolBeePop model (see section Methods).

For the simulations with SolBeePop, we used data from a set of six published field studies with O. bicornis (Ruddle et al. 2018) which were conducted in the context of pesticide risk assessments. Thereby, control nest boxes were installed in untreated oilseed rape fields and treatment nest boxes in fields treated with an insecticide, whereby field sizes ranged between 1.9 and 2.7 ha. For the purposes of the current study, we used only data from the untreated controls. In the studies, incubated O. bicornis cocoons were introduced at the nest boxes at the onset of oilseed rape flowering. After bee emergence, post-emergent female bees present in the nest boxes at night were monitored every three days along with the brood cells produced. At the end of oilseed rape flowering, the nest boxes were closed, preventing further nesting activity by the bees. Brood was brought into the laboratory over winter. A subset of cocoons was incubated to emergence in the following spring, with counts of female and male emergences reported.

Three studies were conducted in 2014 and the other three in 2015. Study sites in 2014 were near Kraichtal (Baden-Württemberg, Germany), Tübingen (Baden-Württemberg, Germany) and Fortschwihr (Alsace, France). In 2015, study sites were located near Niefern (Baden-Württemberg, Germany), Tübingen (Baden-Württemberg, Germany) and Celle (Niedersachsen, Germany). In 2014, 30 female and 60 male cocoons were introduced on a first introduction date and 120 female and 210 male cocoons were introduced three days later at each of the eight control nest boxes per study. In 2015, 60 female and 90 male cocoons were introduced on a single date at the eight control nest boxes per site. The percentage of females and males emerging from the introduced cocoons was recorded. In Suppl. material 2: table S1, the study locations and dates are listed along with cocoons released and their emergence rates. In 2015, semi-field trials were conducted in addition to the field trials at the study sites (Ruddle et al. 2018). Those data were used previously for calibration and validation of SolBeePop (Schmolke et al. 2023). Although the semi-field trials were not the focus of the current paper, we show the study data in the plots for reference. Number of nesting females and cumulative brood cell production from the studies are plotted in Suppl. material 2: figs S1, S2).

We used the population model for solitary bees, SolBeePop (Schmolke et al. 2023, 2024), for the simulations of the field studies with O. bicornis. SolBeePop is a trait-based model capable of simulating different representative species of solitary bees, whereby species-specific trait values correspond to model parameters. The individual-based population model captures the life cycle of solitary bees. Each individual corresponds to a bee that passes through the life cycle from egg to active (post-emergent) adult in daily time steps. Developing bees (in-nest life stages) are assumed to consume their provision entirely and do not otherwise interact with their environment. Mortality occurs in the model at time of emergence of the adult bee, summarising the probability of death of pre-emergent life stages. Post-emergent females consume pollen and nectar on a daily basis. They reach maturity after a set time-span. At this time, they experience a chance of mortality that also represents bees dispersing from the simulated population and bees that never start nesting. Mature females produce offspring, based on a species-specific offspring production rate, resources available to provision the offspring and the nesting female’s post-emergent age. Mature females have a daily chance of mortality and cannot live beyond a set maximum life span. Males are simulated by the model, but their interaction with the environment after emergence is not modelled explicitly. Post-emergent males die after a set life-span. Four daily time series (model inputs) determine the environmental conditions: weather-dependent foraging, floral resource availability from a focal crop or resource patch, resource availability from non-crop (e.g. mixed natural and semi-natural flowering plants) and the proportion of foraging on the focal crop. In combination, these time series define how much resource a female bee can collect from the landscape on a given day and, in turn, how efficiently she can provision brood cells. Accordingly, the model is not spatially explicit, but uses an aggregated landscape representation. An example input file can be found Suppl. material 2: table S9 and all input files used for the simulations in the current paper are available from Schmolke (2025). For the simulations presented in the current paper, we used the model version SolBeePop_ecotox (Schmolke et al. 2024), but we did not simulate exposures or effects of pesticides. A detailed description of the model is available from Schmolke (2025).

We generated three scenarios (model inputs) of resource availability in the landscape during the field phases of the six field studies, based on different assumptions and data (Table 2). From the field studies, detailed data were available that could provide information for the resource availability in the landscape (Ruddle et al. 2018), including quantitative data on pollen composition of O. bicornis provisions from samples taken at 3–4 time points during the studies. Flowering growth stages (BBCH) of the focal oilseed rape field were reported (see Suppl. material 2: tables S2–S7). The landscape composition around the study locations was derived from satellite data. We generated SolBeePop input files, reflecting scenarios of floral resource availability in the landscape using increasingly detailed data. Table 2 provides an overview of assumptions and data used to generate the time series in the model input scenarios. For all scenarios, daily weather data were used, combined with the same assumptions about weather conditions amenable to foraging by O. bicornis. Weather-dependent foraging is addressed in more detail in section Weather-dependent foraging.

The low detail scenario was generated using only information related to the focal oilseed rape field. The field trials were conducted during flowering of the focal oilseed rape fields, corresponding to growth stages of the oilseed rape plants between BBCH 60 and 69. The flowering BBCH stages indicate the percentage of flowers open whereby peak flowering (BBCH 65) is reached when 50% of the flowers are open (Meier 2018; d’Andrimont et al. 2020). For the purpose of the floral resource assumptions in the SolBeePop input file, we assume that full flowering corresponds to optimal resource availability (Quality_crop = 1 in the model input) from the focal field. The floral resource availability at other growth stages was scaled to the percentage of flowers open of each BBCH stage. If a range of BBCH stages were reported in the Ruddle et al. (2018) study, the average between the highest and lowest of the range was used. For instance, BBCH 62 corresponds to 20% of flowers open in oilseed rape (d’Andrimont et al. 2020). If the plants in the field were reported to be in BBCH stages 62–65, it was assumed that, on average, 35% of flowers were open. Scaled to full flowering (BBCH 65 with 50% flowers open), this corresponds to the assumption that the proportional resource availability from the field corresponds to Quality_crop = 0.7. As BBCH stages were not reported for all days of the field study phase by Ruddle et al. (2018), the days without reported BBCH were interpolated linearly between the previous and following BBCH stage. Oilseed rape BBCH stages and the derived proportional resource availability from the crop are listed for each field study in Suppl. material 2: table S2–S7.

In the low detail scenario, detailed data related to the landscape and the pollen composition of bee provisions were not used and the simplifying assumption was applied that no resources were available other than the focal oilseed rape field. Accordingly, the resources from non-crop were set to 0 in the input file (Quality_nat = 0) corresponding to no foraging on resources other than crop (Prop_foraging_crop = 1).

For the generation of the medium detail scenario, floral resources were considered that were available from the landscape in addition to the focal oilseed rape field. Landscape composition data were used that were specific to the study sites. Landscape composition data are available for many regions from public data bases although with varying resolution, detail in land-cover types and years for which the data can be retrieved. We used land-cover data from the European Union’s Copernicus Land Monitoring Service Information, ‘Corine Land Cover + Backbone’ (CLC+Backbone) from 2018, with a 10 m resolution (European Environment Agency 2023). Data from this public inventory were available for all six study sites with the high spatial resolution. The dataset from 2018 was available at the time of our simulation study. Although the year does not correspond to the years when the field studies were conducted (2014 and 2015), the land covers were assumed to be constant over multiple years. In the Corine Land Cover data, raster cells with 10 m edge length are assigned to one of 11 classes (Table 3; see also Suppl. material 2: figs S3–S8 for the land-cover maps of the six sites).

We assigned each land-cover type to plant taxa reported to be collected by O. bicornis (see Section Floral preferences of O. bicornis; Table 1). For the current study, we assumed that oilseed rape from the focal fields were the only non-permanent crop providing floral resources for O. bicornis. Different flowering times occur across the range of plant species potentially providing floral resources per land-cover type. In particular, many tree species flower early in the spring for a limited time period. Ranunculaceae and other herbaceous plants tend to start flowering later in the spring. Thus, we assumed different levels of floral resource availability from the non-crop resource-providing land-cover types. In addition, we assumed that the resource availability was dependent on time of year. In Table 3, we list the categorisation of resource availabilities applied per land-cover type and date range. The categorical assignments were based on reported flowering times of the plant taxa in Germany and the occurrence of their pollen in O. bicornis provisions. Details of the categorical assignments are provided in Suppl. material 2.

The land-cover composition was analysed within the presumed foraging radius of O. bicornis around the nest locations. Maximum foraging distances across multiple Osmia species were reported to range between 400 m and 1200 m (Greenleaf et al. 2007; Zurbuchen et al. 2010). Hofmann et al. (2020) reported average and maximum foraging distances in O. bicornis of 100 m and 250 m, respectively. Bednarska et al. (2022) found high percentages of oak (Quercus) pollen in sampled O. bicornis provisions and identified the nearest oak tree from each study location. The furthest distance between O. bicornis nest locations and an oak tree was 810 m, suggesting that the bees’ maximum foraging ranges extend at least to that distance. As the reported foraging distances of O. bicornis (and Osmia in general) varied considerably between publications, we included three radii around the Ruddle et al. (2018) study locations in the analysis of the land cover: 400 m, 800 m and 1200 m.

We translated the land-covers within each foraging radius to model input time series using two approaches: based on distance to the nearest land-cover patch with resources (“dist”) or based on the proportion of area with resource-providing land-covers (“prop”). For “dist”, the minimum distance, d, (within the applicable foraging radius, r) was determined for each of the land-cover types providing non-crop floral resources, woody broadleaved deciduous trees and permanent herbaceous. As a generic proxy for the resource quality dependent on distance, we calculated q = 1 – d/r for each land-cover type. In case the land-cover type did not occur within the foraging radius, q was set to 0. We then multiplied q with the factor from Table 3 corresponding to the date. The maximum of the two resulting values was used in the input time series Quality_nat. Thus, “dist” is a simplification of the landscape that only considers the land-cover that provides bee resources closest to the nest location. It corresponds to the simplifying assumption that the bees forage in one resource patch on a given day and that the resource availability is not limited by patch size. For the input files generated using “prop”, the area of each of the two relevant land-covers within the foraging radius was divided by the total area. The proportion was then multiplied by the factor listed in Table 3 for each day. The sum of the two resulting numbers was used in the input time series Quality_nat. Accordingly, “prop” reflects the simplifying assumption that bees use multiple resource patches within their foraging radius on any given day and that resource availability is linked directly to patch size. Note that, for all study sites, at least one of the land-covers occurred within 400 m radius, i.e. Quality_nat > 0 for all study-specific inputs during the field phases for both “dist” and “prop”. See also Suppl. material 2: tables S13, S14 for the summary of the land-cover distances and proportions per site.

For the generation of input files for the high detail scenario, study-specific data on pollen composition of brood provisions were used whereby the provisions were sampled 3–4 times during the field phase of the studies (Ruddle et al. 2018). In the three studies conducted in 2014, samples were taken from unfinished brood cells on three dates during the field phase. Pollen was identified as Brassica napus, Quercus or other and the percentage in the samples of each were reported. In 2015, two sites were sampled four times (S15-01802, Niefern and S15-01803, Tübingen) and one was sampled three times (S15-01803, Celle). Pollen was identified as Brassica, Quercus, Acer, Ranunculus, Juglans, Aesculus, Rosaceae, Ligustrum or other. Detailed pollen composition data from Ruddle et al. (2018) can be found in Suppl. material 1.

The percentage of Brassica pollen reported in the provisions was assumed to directly reflect the proportion of foraging on crop by the bees. The days between pollen samples were interpolated linearly to generate the daily values of ‘Prop_foraging_crop’ time series in the input file. As pollen provisions were not sampled on the first and last day of the field study phase, we assumed that, on the first day of the field phase, half of the proportion of Brassica pollen reported on the earliest sampling date applies to the proportion of foraging on crop. For the last day, we assumed that half of the proportion of Brassica pollen reported on the latest sampling date applied to the proportion of foraging on crop. The generic assumptions of half the proportion of Brassica pollen of the subsequent or previous sampling, respectively, reflects that oilseed rape was flowering at the start of the field phase in the field studies and the field phase was ended when oilseed rape flowering was coming to an end. These assumptions were necessary to allow interpolation of proportion of foraging on crop for all days during the study field phases.

Although the pollen composition data of the brood provisions provided study-specific data, it cannot be derived from the data to what extent each floral species was available in the landscape, its distance from the nest or the bees’ effort involved in extracting the resource from the flowers. As this information was lacking, we used the same information and assumptions as in the medium detail scenario about the resource availability, based on proportions of relevant land-cover types within the foraging radius (“prop”) or smallest distance from the nest (“dist”), respectively (see also Table 2).

Weather is an important factor in the reproductive output of solitary bees: if the weather prevents nesting females from foraging, they cannot provision offspring. Weather conditions allowing foraging vary between species (Stubbs et al. 1994; Vicens and Bosch 2000; Drummond 2016). In the SolBeePop model, we assume that the life span of bees is independent of weather and, thus, lost time for brood cell production due to inclement weather cannot be compensated for by individual simulated bees. For O. bicornis, published data on weather-dependent foraging are limited. In a study with O. bicornis, no foraging activity was reported at temperatures below 10 °C, during rain and strong wind (Franke et al. 2021). The study was not focused on weather-dependent foraging and no further details were addressed. In a study focused on blueberry pollination, no bee activity from any species was observed on days with precipitation of 25.4 mm or more (Drummond 2016; Drummond et al. 2017). As default, we assumed that days with maximum measured air temperatures < 10 °C did not allow foraging by O. bicornis, as well as days with precipitation ≥ 25.4 mm (Prop_foraging_day = 0). No other weather conditions were considered and all days with maximum temperatures above the temperature threshold and below the precipitation threshold were assumed to allow foraging (Prop_foraging_day = 1). Given that these thresholds were based on unspecific data, we tested an alternative set of plausible thresholds for weather-dependent foraging. Seidelmann et al. (2010) used weather-dependent foraging thresholds for an analysis of O. bicornis study data. Although the authors did not specify the sources for their assumptions, we used their assumptions as a plausible alternative scenario of weather-dependent foraging in the species. The assumed thresholds by Seidelmann et al. (2010) addressed air temperature (18 °C) and relative air humidity (60%). We applied 18 °C as minimum temperature threshold for foraging whereby we compared it to the maximum daily air temperature. Similarly, as a condition for foraging, the minimum daily relative humidity had to be 60% or less. As no threshold for precipitation was defined by Seidelmann et al. (2010), we used the threshold of 25.4 mm for this scenario as well.

Separate input files were generated for each scenario and study, defining the five daily time series (Table 2). Thereby, six input files per study were generated to represent the medium and high detail scenarios because two versions of representing the landcovers (“dist” and “prop”) and three alternative foraging radii (400 m, 800 m and 1200 m) were applied as alternatives. For the exploration of weather-dependent foraging, we conducted additional simulations using one study (Tübingen 2015), combined with the low detail scenario and the high detail scenario “dist” with 1200 m foraging radius. The input files for these simulations were generated using different assumptions about weather-related foraging in O. bicornis. Model parameter settings corresponded to trait values for O. bicornis derived from literature and addressed in Schmolke et al. (2023). For a subset of parameters, average values were used that reflect the calibration of SolBeePop to semi-field study data with O. bicornis. The semi-field studies were conducted in 2015 at the same study locations as the 2015 field studies used in the current paper (Ruddle et al. 2018). The calibration to the semi-field studies was reported in Schmolke et al. (2023) and can also be found in Schmolke (2025). Study-specific dates were used in the simulations of each study, including the introduction of the O. bicornis cocoons at the nest boxes and the end of the field phase (when nest boxes were closed and further brood cell production was prevented). The parameters and settings used for the simulation and the input files are listed in Suppl. material 2: tables S11, S12, respectively. Each combination of study and scenario was simulated using 50 repetitions, i.e. simulations that differed only in the random number seeds and covering the stochasticity of the model.

For each simulation set (of 50 repetitions), we calculated the average, minimum and maximum outputs for number of post-emergent females and cumulative brood cells produced. For the comparison of the simulations to the study data, we calculated normalised root mean square errors (NRMSE). We used the NRMSE to identify the scenarios that corresponded best to the cumulative brood cell production observed in the field studies. For the generation of model input files, model output analysis and plotting, we used R (R Core Team 2022). Scripts and inputs and are provided in Schmolke (2025).