Skip to main content
Dryad

MetaComNet: A random forest-based framework for making spatial prediction of plant-pollinator interactions

Cite this dataset

Sydenham, Markus Arne Kjær et al. (2021). MetaComNet: A random forest-based framework for making spatial prediction of plant-pollinator interactions [Dataset]. Dryad. https://doi.org/10.5061/dryad.n02v6wwzn

Abstract

1. Predicting plant-pollinator interaction networks over space and time will improve our understanding of how environmental change is likely to impact the functioning of ecosystems. Here we propose a framework for producing spatially explicit predictions of the occurrence and number of pairwise plant-pollinator interactions and of the species richness, diversity, and abundance of pollinators visiting flowers. We call the framework ‘MetaComNet’ because it aims to link metacommunity dynamics to the assembly of ecological networks.

2. To illustrate the MetaComNet functionality, we used a dataset on bee-flower networks sampled at 16 sites in southeast Norway along with random forest models to predict bee-flower interactions. We included variables associated with climatic conditions (elevation) and habitat availability within a 250m radius of each site. Regional commonness, site-specific distance to conspecifics, social guild, and floral preference were included as bee traits. Each plant species was assigned a score reflecting its site-specific abundance, and four scores reflecting the bee species that the plant family is known to attract. We used leave-one-out cross-validations to assess the models’ ability to predict pairwise plant-bee interactions across the landscape.

3. The relationship between observed occurrence or absence of interactions and the predicted probability of interactions was nearly proportional (GLMlogistic regression slope = 1.09), matching the data well (AUC = 0.88), and explained 30% of the variation. Predicted probability of interactions was also correlated with the number of observed pairwise interactions (r = 0.32). The sum of predicted probabilities of bee-flower interactions were positively correlated with observed species richness (r = 0.50), diversity (r = 0.48), and abundance (r = 0.42) of wild bees interacting with plant species within sites.

4. Our findings show that the MetaComNet framework can be a useful approach for making spatially explicit predictions and mapping plant-pollinator interactions. Such predictions have the potential to identify areas where the pollination potential for wild plants is particularly high, and where conservation action should be directed to preserve this ecosystem function.

Methods

We sampled bee-flower networks along 16 roadsides (sites) in Southeast Norway in 2017. Eight of the study sites were located on sandy sediments and the remaining eight were located on clay dominated sediments (Skoog 2018). At each study site, flower-visiting bees were collected during one hour by two observers along a 50m transect, once during early to mid-July and once during early to mid-August in 2017. A total of 910 interactions between wild bees (n = 45 species) and plants (n = 44 species) were observed. We only included non-parasitic bees (n = 39) in our analyses because parasitic bees do not visit plants for nectar and pollen. We estimated the site’s abundance of flowering plant species by recording the number of occurrences (presence/absence) of each plant species in a grid of four 25x25 cm squares within six 1 m2 quadrats placed in a 2 by 30 m grid along the roadside, totaling ninety-six squares per site.

Plant and bee DCA scores were extracted from a detrended correspondence analysis on a binary matrix of bee species and plant family interactions. 

#    The csv file: "Sydenham et al MetaComNet data frame.csv"
##        Contains the model data frame for running the analyses presented in the paper. The data frame consists of the following columns.
###            "Site" = Identity of study site
###            "SiteBee" = Identity of study site and bee species
###            "SitePlant" = Identity of study site and plant species 
###            "SitePlantBee" = Identity of study site and plant species and bee species
###            "Number" = Number of recorded interactions between bee species and plant species at the study site
###            "Occurrence" = Occurrence or absence of recorded interactions between bee species and plant species at the study site
###            "DCA1" = Plant species DCA score on axis 1
###            "DCA2" = Plant species DCA score on axis 2
###            "DCA3" = Plant species DCA score on axis 3
###            "DCA4" = Plant species DCA score on axis 4
###            "BeeDCA1" = Bee species DCA score on axis 1
###            "BeeDCA2" = Bee species DCA score on axis 1
###            "BeeDCA3" = Bee species DCA score on axis 1
###            "BeeDCA4" = Bee species DCA score on axis 1
###            "Solitary"= Bee species is solitary or social (Bombus)
###            "PlantFreq" = Frequency of plant species in site (number of 0.25cm subplots in which the species was recorded)
###            "MASL" = Elevation of study site
###            "LnscpH" = Landscape shannon diversity within a 250m radius around the site
###            "LndscpGR" = Proportion of grassland within a 250m radius around the site 
###            "DistSand" = Distance to sandy soil deposits (deposits with a high infiltration capacity)
###            "NearestOcc" = km to the nearest known occurrence (from GBIF) of the bee species
###            "RegionalCommonness" = number of 10km grid cells within the bee species occurs within the region

#     The shapefile: "Site locations.shp"
##        Contains a spatial points dataframe with points for each study site

#     The csv file: "GBIF wild bee occurrence records.csv"
##        Contains occurrence records for wild bee species within the wider study region downloaded from GBIF using the gbif() function in R package dismo. Records were first downloaded on 16.07.2021, and then downloaded again on the 17.10.2021 to obtain all metadata. These data are used for producing the prediction maps in figure 4, and were used for estimating regional commonness and distance to nearest occurrences in the metacomnet model data frame.


#    The raster file: "StackedPredsForMAP.tif" 
##        contains a rasterstack for producing the prediction maps in figure 4. Note that the raster was aggregated from 10m to 100m resolution when making prediction maps. 
###        "MASL" = Elevation
###        "LnscpH" = Landscape shannon diversity within 250m of pixels
###        "LndscpGR" = Landscape shannon diversity within 250m of pixels
###        "DistSand" = Distance to sandy soil deposits
###        "CellID" = Identity of raster cell (first cell in top left corner)

#### MASL was based on a digital elevation model with a 50m resolution (Norwegian mapping authority, 2016)

### LnscpH and LndscpGR were calculated from Venter & Sydenham (2021)

### DistSand was obtained from the infiltration capacity from geological survey maps (Geological Survey of Norway, 2011) 

#    The raster file: "MosaicBlock_UTM32N_T32VPM.tif"
##        Contains a satellite image (rgb) for producing figure 4 in the manuscript
###             Copernicus Sentinel-2 data (2019)/processed by the Norwegian Mapping Authority, downloaded from www.geonorge.no

Funding

The Research Council of Norway, Award: 302692