The R dataset consists of the bipartite plant-pollinator interaction matrices (bipartite.networks) for 18 subseas (3 subseasons per 6 years), 18 subseasonal dependence matrices (directed.networks), and the plant species (plant.id) and pollinator species (pollinator.id) identification codes. The code used for the analyses (miele-JAE.R) is also provided.

Study site and data collection

Data come from a previously published study describing a plant-pollinator network from pollinator visits to flowers in a dryland ecosystem (Chacoff et al., 2018, Ecology 99: 21-28). Data collection was conducted in sites lying at the lowlands of Villavicencio Nature Reserve, Mendoza, Argentina.

Data were collected weekly during three months during the flowering season (Austral spring and early summer, September-December) between 2006 and 2011. Pollinator visits to flowers were recorded in two 1-ha sites separated by ca. 5 km, with two additional 1-ha sites sampled in only in 2006. Data from these sites were combined into a single network to improve sampling completeness of species and interactions occurring in the region. Pollinator visits to flowers were recorded between 7:00 and 14:00 in 5-min observation periods, a representative portion of the daily activity period of pollinators in our study sites. The data include 59 plant species, 196 flower visiting insect species, and 28015 interaction events (flower visits) involving 1050 different pairs of interacting species. Plant abundance was estimated based on the density of flowers of each plant species, as flowers are the relevant plant structure for this interaction type. Flower abundance was estimated during the flowering season of all study years using fixed quadradts/transects. Several rare plant species were absent from our fixed quadrats and transects but present elsewhere in our study site; for those species we assigned an abundance of one flower, the minimum we could have detected with our sampling method.

We aggregated the data by pooling the number of visits of any pollinator to any plant into three subseasons of approximately equal length throughout the flowering season of each year (before November 1st, after November 30th and in between). Such level of aggregation allowed us to consider seasonal dynamics at a temporal grain that was not too fine nor too coarse to allow a reasonable seasonal representation of network structure. For any subseason, we built a plant-pollinator dependence network D, a directed weighted network representing the relative dependences among plant and pollinator species.