Material and Methods
We conducted this study in a subalpine meadow at the University of
Colorado’s Mountain Research Station (40°01’48”N 105°32’26”W), located
at 2900 m of elevation 22 km west of Boulder, CO, USA. The meadow faces
east and is surrounded by aspen and spruce-fir forest. We collected
interaction data weekly during the entire flowering period from 2015 to
2019. The flowering season at the study site typically starts after
snowmelt in late May to early June and extends to late September.
Interactions were recorded on 16 – 18 weeks per year. Sampling was done
on average 6.95 days apart (SD = 1.17 days).
On each sampling date, we sampled plant–pollinator interactions in six
20-30 m × 2 m plots (five 30 m × 2 m and one 20 m × 2 m) by observing
flowers (plant-centred sampling). Sampling was conducted in fair weather
during mornings between 8:00 am and 12:00 pm, a time range when
pollinator activity is high and before the onset of thunderstorms that
often occur at mid-day during the summer in the Rocky Mountains. We
sampled plant–pollinator interactions within each plot (in random
order) by doing 15-min surveys in which we carefully observed all
flowers for visitors while walking the periphery of plots to minimize
trampling. When we observed an interaction, defined as a pollinator
contacting the reproductive structures of a flower, we recorded the
identity of the plant and pollinator species. Insect pollinators were
collected with aspirators or aerial nets for later identification in the
laboratory. Expert entomologists (see Acknowledgements) assisted with
insects that are difficult to identify. To assess the thoroughness of
the sampling effort, we compared observed richness of plants,
pollinators, and interactions to values from three commonly used
richness estimators: Chao2, first-order jackknife (Jack1), and
bootstrapped values (Chacoff et al. 2011; Gotelli & Colwell
2011).
We recorded temporal and spatial variation in abiotic conditions.
Temperature is fundamental in constraining pollinator foraging (Corbetet al. 1993; Willmer & Stone 2004) and plant flowering (Schemskeet al. 1978). To measure temporal variation in temperature at the
site, we compiled air temperature data from every morning (8 am–12 pm)
during which sampling was conducted from a weather station located
~1 km away at a similar elevation, 3020 m (SNOTEL Site
Niwot, 663). Soil moisture affects plant growth and reproduction (Fayet al. 2000). To quantify aspects of spatial variation in abiotic
conditions among plots we measured ground temperature and soil moisture
every 2 meters along sampling plots on one occasion at 7:55–8:38 am on
6 July 2018. To measure ground temperature, we used a handheld infrared
thermometer. To measure soil moisture, we used a time-domain
reflectometry (TDR) moisture sensor.
To test the relationship between the network position of interactions
and their temporal or spatial persistence we used Spearman’s rank-order
correlation tests between the proximity to the core of the nested
network and each variable of temporal or spatial persistence: the number
of years, the span of days (phenophase), and the number of plots in
which interactions were recorded. The position of each interaction was
calculated using a maximally nested matrix. To this end, we first
compiled all the observations from the study into a plant–pollinator
interaction matrix, sorting rows and columns to maximize matrix binary
nestedness. This sorting organizes plant species (in rows) and
pollinator species (in columns) from most specialist to most generalist
such that generalist species are packed into the top left corner of the
matrix. We then used this matrix organization to assess the relationship
between proximity to the core of the nested network and persistence
values for each temporal and spatial variable. The proximity of
interactions to the core of the nested network was calculated as one
minus the standardized Euclidean distance of each interaction to the
upper-left cell in the nested matrix with the distance between each
adjacent cell equal to one (as in Chacoff et al. 2018). We tested
for the number of modules present to determine if the network could have
multiple cores.
To assess the ecological significance of our results, we compared the
observed Spearman’s correlation coefficients for the relationships
between proximity to the core of the nested network and temporal or
spatial persistence values and compared these coefficients to those from
null models. Null models generated 1000 randomized matrices by shuffling
persistence values within matrices while fixing marginal totals and
connectance, which we judged to represent a conservative null model. To
determine how temporal and spatial variables are related to one another,
e.g., whether interactions that have longer phenophases tend to be more
persistent across years, we tested for correlations between each
combination of temporal or spatial persistence variables with Spearman’s
rank correlation tests.
We used linear regressions with a 2nd degree
polynomial to assess how plant and pollinator species’ proximity
to the network core (row or column number in the nested network divided
by the total number of rows or columns) relate to inter-annual
occurrence), intra-annual occurrence, and inter-plot occurrence. Species
proximity to the nested core was correlated with degree, the number of
interacting partner species, a common metric of species specialization
(for plants: Pearson’s r = 0.87; for pollinators: Pearson’s r = 0.70).
Finally, to link species’ phenophases to their environmental tolerances,
we correlated the range of days in which pollinator or plant species
were recorded as interacting with the range of temperature recorded in
the mornings during those ranges of dates.
All analyses were performed in R version 4.0.2 (R Core Team 2020). We
used the vegan package version 2.5-6 for calculating richness estimates
(Oksanen et al. 2010), the bipartite package version 2.15 for
network analyses, visualization, and null models (using the swap.web
function) (Dormann et al. 2008), and the reshape2 package version
for data formatting (Wickham 2007).