Skip to main content

Data from: Climatic conditions and landscape diversity predict plant-bee interactions and pollen deposition in bee-pollinated plants.

Cite this dataset

Sydenham, Markus Arne Kjær et al. (2024). Data from: Climatic conditions and landscape diversity predict plant-bee interactions and pollen deposition in bee-pollinated plants. [Dataset]. Dryad.


Climate change, landscape homogenization and the decline of beneficial insects threaten pollination services to wild plants and crops. Understanding how pollination potential (i.e. the capacity of ecosystems to support pollination of plants) is affected by climate change and landscape homogenization is fundamental for our ability to predict how such anthropogenic stressors affect plant biodiversity. Models of pollinator potential are improved when based on pairwise plant-pollinator interactions and pollinator´s plant preferences. However, whether the sum of predicted pairwise interactions with a plant within a habitat (a proxy for pollination potential) relates to pollen deposition on flowering plants has not yet been investigated. We sampled plant-bee interactions in 68 Scandinavian plant communities in landscapes of varying land-cover heterogeneity along a latitudinal temperature gradient of 4–8 C°, and estimated pollen deposition as the number of pollen grains on flowers of the bee-pollinated plants Lotus corniculatus, and Vicia cracca. We show that plant-bee interactions, and the pollination potential for these bee-pollinated plants increase with landscape diversity, annual mean temperature, plant abundance, and decrease with distances to sand-dominated soils. Furthermore, the pollen deposition in flowers increased with the predicted pollination potential, which was driven by landscape diversity and plant abundance. Our study illustrates that the pollination potential, and thus pollen deposition, for wild plants can be mapped based on spatial models of plant-bee interactions that incorporate pollinator-specific plant preferences. Maps of pollination potential can be used to guide conservation and restoration planning.

README: Climatic conditions and landscape diversity predict plant-bee interactions and pollen deposition in bee-pollinated plants.

The ModelMetaComNetDF.csv file was used to model plant-bee interactions using a generalized mixed effects model using the formula (R syntax):

glmer(Interaction occurrence ~ Bee Social status × Annual mean temperature + Bee Social status × Landscape Shannon diversity + Plant abundance + BeeDCA1 × PlantDCA1 + BeeDCA2 × PlantDCA2 + BeeDCA3 × PlantDCA3 + BeeDCA4 × PlantDCA4 + Bee Phenology × Plant Phenology + square root (Distance to Sandy soils) + Regional Commonness + (1|Site identity), family = binomial)

To predict onto sites we used a leave-one-out cross validation where for each of the 68 sites, data from the site was removed from the data while fitting the model which was then used to predict the likelihood of occurrence of all bee-plant interactions in the site.

the MetaComNetPollen.csv contains information on the number of pollen grains deposited on flowers within sites and was used to test if the sum of predicted occurrences across all bee species was positively related to the number of pollen grains on flowers. Tests were run using the formula (R syntax):

glmmTMB(Pollen grains on stigma ~ Predicted pollination potential × Plant species + (1| Site identity), ziformula=~1, family=nbinom1)

and to test if the spatial predictors of plant-bee interactions predicted pollen deposition a second model was run using the formula (R syntax):

glmmTMB(Pollen grains on stigma ~ Plant abundance + Landscape Shannon diversity + Annual mean temperature + Plant species + square root(Distance to Sandy soils) +(1|Site identity),  ziformula=~1, family=nbinom1)

The R code for running the analyses is stored in the 'R script for running analyses and statistical tests.R' file.

Description of the data and file structure

The dataset contains the files: "ModelMetaComNetDF.csv" used for modelling and predicting plant-bee interactions, "MetaComNetPollen.csv" used for testing if predicted occurrence of pollinators on plants relates to pollen deposition, "Spatial Predictors for map.tif" used for making spatial predictions of pollen deposition, "region sat image.tif" used to for producing maps with locations of study sites, "Denmark sat image.tif" used to for producing maps with locations of study sites, and "Satelite image Norway.tif" used to for producing maps with locations of study sites and pollen deposition prediction maps. Satellite imagery from Map data ©2023 Google via QGIS 2023.

Columns in ModelMetaComNetDF.csv

Site_id: identifier for the 68 study sites

Pollinator_species: identifier for each bee species

Plant_species: identifier for each plant species

SitePlant: combination of site and plant identifiers

PlantFreq: number of 1m vegetation plots a plant was recorded in at a site

Occurrence: occurrence (1) or absence (0) of interaction between plant and bee within the site 

SitePlantPollinator: combination of site, plant, and bee species identifiers

PlantDCA1-4: plant species score along DCA axis 1-4

meanPlantPhenolgyDay: mean julian day for interactions recorded for the plant species

BeePhenology: activity period for bee species

Eusocial: TRUE for bee species within the Bombus genus, FALSE for non-Bombus species.

BeeDCA1-4: Bee species score along DCA axis 1-4

RegionalCommonness: Number of 10km grid cells bee species have been recorded in

coords.x2: latitudinal coordinates for study site (EPSG:3035)

coords.x1: longitudinal coordinates for study site (EPSG:3035)

ShannonH: Landscape diversity within 1km buffer around site

bio1: Annual mean temperature at site

DistSnd: Distance in meters to soil deposits classified as sand dominated

Columns in MetaComNetPollen.csv

siteplant: combination of site and plant identifiers

MCNsiteID: identifier for the study site where a flower was collected for pollen grain analyses

plant: identifier for the plant species a flower was collected (TT = Lotus corniculatus, FV = Vicia cracca)

plantSpecimen: plant specimen the flower was collected from

TubesPicture: number of pollen tubes in style counted from pictures of the flower (if NA no picture was taken)

PollenPicture: number of pollen grains on stigma counted from pictures of the flower (if NA no picture was taken)

TubesMic: number of pollen tubes in style counted through the microscope

PollenMic: number of pollen grains on stigma counted through the microscope


All analyses were run using R (R Core Team 2022)  with the following R packages: terra (Hijmans 2023), effects (Fox and Weisberg 2019), DHARMa (Hartig 2022), MASS (Venables and Ripley 2002), glmmTMB (Brooks et al. 2017), MuMIn (Bartoń 2023), ncf (Bjornstad 2022), ape (Paradis and Schliep 2019), car (Fox and Weisberg 2019), and boot (Canty and Ripley 2019, Davison and Hinkley 1997).


  • Bartoń K (2023). MuMIn: Multi-Model Inference. R package version 1.47.5,
  • Bates, D. et al. (2015). Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software, 67(1), 1-48. doi:10.18637/jss.v067.i01
  • Bjornstad ON (2022). ncf: Spatial Covariance Functions. R package version 1.3-2,
  • Canty, A. and Ripley, B. (2022). boot: Bootstrap R (S-Plus) Functions. R package version 1.3-28.1.
  • Davison, A. C. & Hinkley, D. V. (1997) Bootstrap Methods and Their Applications. Cambridge University Press, Cambridge. ISBN 0-521-57391-2
  • Hartig, F. (2022) DHARMa: Residual Diagnostics for Hierarchical (Multi-Level / Mixed) Regression Models. R package version 0.4.6
  • Hijmans R (2023). terra: Spatial Data Analysis. R package version 1.7-46,
  • Fox, J. and Weisberg, S. (2019). An R Companion to Applied Regression, 3rd Edition. Thousand Oaks, CA
  • Brooks et al. (2017). glmmTMB Balances Speed and Flexibility Among Packages for Zero-inflated Generalized Linear Mixed Modeling. The R Journal, 9(2), 378-400. doi: 10.32614/RJ-2017-066.
  • Paradis E, Schliep K (2019). “ape 5.0: an environment for modern phylogenetics and evolutionary analyses in R.” Bioinformatics,

35, 526-528. doi:10.1093/bioinformatics/bty633
* Venables, W. N. and Ripley, B. D. (2002) Modern Applied Statistics with S. Fourth Edition. Springer, New York. ISBN 0-387-95457-0
* R Core Team (2022). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL


We sampled plant-bee interactions in 68 semi-natural, forb-dominated, plant communities. We used linear open landscape features such as roadsides as a model system and established one 50×2m transect for our surveys in each site. To cover the main flowering period, we sampled plant-bee interactions once in May, June and July at each site. To standardize sampling times across sites and countries, timing of the first sampling was determined by the peak flowering of dandelions (Taraxacum officinale). All flower-visiting bees were collected from flowers and stored in 96% EtOH prior to identification. Specimens within the B. sensu stricto subgenus were treated as one morpho-species. Each transect observation lasted 30 minutes, adding 30 seconds per collected specimen, to account for handling time. Sampling only took place on days with temperatures > 15°C, local wind speed < 5 m/s, with little to no cloud cover and no rain to standardize sampling conditions between networks.

In late June to early July 2021, we sampled the plant communities by placing ten 1m2 vegetation plots regularly along each transect. In each 1m2 vegetation plot, we recorded the occurrence of bee-pollinated plant species, regardless of life stage, within four 0.25m2 subplots. Conducting vegetation surveys separately from plant-pollinator interaction surveys was necessary for us to be able to sample plant-bee interactions at all 68 sites within the same time periods. Despite being long after the flowering period of the earliest flowering plants, such as Tussilago farfara, these plants were recorded during our survey. Our approach allowed us to obtain estimates of plant abundances (i.e. number of plots, or subplots occupied by a plant), from just one survey per site, regardless of plant phenologies.

We sampled wilted flowers from two legumes (Vicia cracca, Lotus corniculatus) along the transects in Norway in July 2021, and counted the number of pollen grains deposited on stigmas as a measure of total pollen deposition. Flower sampling was conducted independently from surveys of plant-pollinator interactions. These legumes were selected because they are mainly bee pollinated and had been found in at least 10 sites during the vegetation surveys. Vicia cracca, and Lotus corniculatus occurred at 26, and 25 sites, respectively. We collected one flower from 10-15 individuals per species per site. Flowers were fixated with 4% formaldehyde alcohol acetic acid and stored at 4°C until further processing. Entire flowers were softened in 5M NaOH for 24 hours. After softening, the NaOH was removed by pipetting and gently washed out with ddH2O five times. Then, several drops of 0.001g/ml Aniline Blue Fluorochrome (ABF) in 0.1M K2HPO4 pH 10 buffer were dropped on microscope slides. Gynoecia were carefully removed from flowers with tweezers, immersed in ABF on the microscope slides, covered with aluminum foil, and left for one hour. Following staining, squashing, and mounting, we counted the number of pollen grains using a Leica DM2500 fluorescence microscope.

We used the annual mean temperature at sampling locations, obtained from the WorldClim database (Fick & Hijmans 2017). As a proxy for landscape homogenization (or its inverse, landscape diversity), we calculated the Shannon landscape diversity within circular buffers with radii of 1 km surrounding each site from a European 10 m resolution land cover map (Venter & Sydenham 2021). As a proxy for availability of high-quality nesting substrates for ground nesting bees, which account for the majority of bee species in our region, we included the geographic distance to sand dominated geological deposits, estimated as the distance to the nearest spatial polygon classified as having a high or moderately high infiltration capacity in Norway (Geological Survey of Norway 2011) or explicitly classified as being sand dominated in Denmark (Landbrugsstyrelsen 2019). We did not include specific predictors for cavity nesting bees, because we expected the availability of nesting substrates for this group to correspond with landscape diversity.

We used the number of 1 m2 vegetation plots in which a species occurred. Plants not recorded during the vegetation survey but on which bees had been observed were assigned a plant abundance value of 1, corresponding to the lowest value recorded for plants during the plant surveys. To account for forbidden links due to phenological mismatches, we included the mean Julian day on which bees were observed on the plant across all sites as a proxy for plant peak flowering time. As a proxy for flowering time duration, we included the standard deviation of Julian days on which interactions between the plant and bees had been observed. We used pre-existing data on plant and bee associations (Rasmussen et al. 2021) to derive plant-bee association scores for both plants and bees. We assembled a matrix with information on the number of plant genera for each of 62 plant families that 281 bee species from Denmark and Norway are known to visit (Rasmussen et al. 2021, Wood et al. 2021). We used a detrended correspondence analysis (DCA) in the ‘vegan’ package in R (Oksanen et al. 2022) to ordinate the plant family-bee species matrix and used the plant family and bee species DCA-scores along the four axes returned by the decorana() function (DCAs 1-4), resulting in four scores for each plant species from our surveys (Plant DCA1-4). The DCA scores separated plant species according to differences in which bee species they are known to interact with.

We included the number of 10 km grids within a spatial extent (Longitude = [7.97, 13.84], Latitude = [54.37, 61.81]) slightly exceeding that of our study region (Longitude = [9.47, 12.34], Latitude = [55.87, 60.32]) where bee species sampled during our study had been previously recorded (GBIF, 2024) as a proxy for regional commonness. We assigned each bee species a categorical trait with two levels to distinguish between eusocial bumblebees and solitary or facultatively social bees. We used data from the European bee fauna (Scheuchl & Willner 2016) to assign each bee species a phenological trait which was a categorical variable with four levels: ‘Spring-mid-summer’ indicating activity from April to July; ‘Early-late summer’ indicating activity from May to August; ‘Mid-late summer’ indicating activity from June to August; Entire summer’ indicating activity from April to August. Although the phenology of species variesy across climatic gradients the relative phenological difference among species is likely to be fairly constant. As for plants, each bee species was assigned a floral preference score (Bee DCA1-4), extracted from the DCA analysis.


Fick, S.E. and Hijmans, R.J. 2017. WorldClim 2: new 1-km spatial resolution climate surfaces for global land areas. International Journal of Climatology 37: 4302–4315.

GBIF 2024. Derived dataset (4 June 2024) Filtered export of GBIF occurrence data

Geological Survey of Norway (2011). Løsmasser WMS. Retrieved from

Landbrugsstyrelsen. (2019). Jordbundskort 2019 Landbrugsgeodata.;jsessionid=E97078277812AF9C3C5A6FD7BA8516E8#/metadata/da22d09d-eec3-44f6-a470-c551e03d512b

Oksanen J et al. 2022. vegan: Community Ecology Package. R package version 2.6-4, <>.

Rasmussen, C. et al. 2021. Evaluating competition for forage plants between honey bees and wild bees in Denmark. PLoS One 16: e0250056.

Scheuchl, E., and Willner, W. 2016. Taschenlexikon der Wildbienen Mitteleuropas: Alle Arten im Porträt. Quelle & Meyer Verlag.

Venter, Z.S. and Sydenham, M.A.K. 2021. Continental-scale land cover mapping at 10 m resolution over Europe (ELC10). Remote Sensing 13, 2301.

Wood, T.J. et al. 2021. Global patterns in bumble bee pollen collection show phylogenetic conservation of diet. Journal of Animal Ecology 90: 2421–2430.


The Research Council of Norway, Award: 302692