Species respond idiosyncratically to environmental variation, which may generate phenological mismatches. We assess the consequences of such mismatches for solitary bees. During nine years, we studied flowering phenology and nesting phenology, and demography of five wood-nesting solitary bee species representing a broad gradient of specialization/generalization in the use of floral resources. We found that the reproductive performance and population growth rate of bees tended to be lower with increasing nesting–flowering mismatches, except for the most generalized bee species. Our findings help elucidate the role of phenological mismatches for the demography of wild pollinators, which perform key ecosystem functions and provide important services for humanity. Furthermore, if climate change increases phenological mismatches in this system, we expect negative consequences of climate change for specialist bees.

Methods

Study area and sites

The study was conducted in ttwo 2-ha sites located at ca. 1250 m above sea level in Villavicencio Nature Reserve and its vicinity (32° 32' S, 68° 57' W), Las Heras Department, Mendoza Province, Argentina. Each site consisted of a 100 m × 200 m plot. Because data come from several projects running consecutively at the same sites, there are minor differences in the sampling methods used in different years.

Flower abundance

Our estimate of flowering phenology of the plant species present in our study sites is based on weekly estimates of the floral density in fixed plots and transects, conducted during the flowering season of the majority of plant species (September–December). During the flowering season, floral density was measured weekly at fixed quadrats/transects: forty regularly spaced 2 m × 2 m quadrats in 2006, five fixed 50 m × 2 m transect belts in 2007, and four 8 m × 20 m plots plus two 2 m × 50 m transects in the remaining years (2008–2014). Flower density was estimated multiplying the mean number of flowers per individual by the total number of flowering individuals in the transect or plot when individuals could be distinguished (shrubs and some herbs); we estimated the number of flowers per individual in at least ten individuals of the site. When it was not possible to identify flowering individuals (some herbaceous species), all flowers in a plot or transect were counted. We included in the study all flowering plant species that were assumed to be animal pollinated (excluding only grass species).

Bee nesting, bee reproductive success, and nest pollen identification

We used wooden trap nests to study bee nesting phenology and reproductive success. Each trap nest consisted of two wood pieces bound together with paper tape with a longitudinal hole drilled in between them. We used this design so that we could open the nests regularly to examine their contents without damaging the brood cells. We used trap-nests of 5 mm in internal diameter and 14 cm in tunnel length, 8 mm diameter and 14 cm length, and of 11 mm diameter and 28 cm length; the latter diameter was not used during the first study year. These trap nests were placed in the field in bundles with eight trap-nests of each diameter (24 total) at multiple sampling points in each study site. Trap bundles were attached to shrubs in the first two study years, and to metal poles in the remaining years. In each site, we placed trap nests at 30 points separated at least by 20 m from each other (2006 and 2007) or at twelve paired points separated 100 m from each other (2008–2014). Although the study design varied slightly between the first two and the remaining seven study years, the number of trap nests of each size per site was similar among years, and so this variation is unlikely to affect our estimate of the number of brood cell per nest, the main reproductive variable used in our study.

Bees included in this study belonged to four genera and five species: the carder bee Anthidium vigintipunctatum; the petal-cutting bees Megachile leucographa and M. nigella (identified as M. ctenophora in previous publications in the same study area); the wood-borer bee Trichothurgus laticeps; and the carpenter bee Xylocopa atamisquensis. Other bee species also nested in our trap-nests, but their abundance was too low to allow conducting the study at the broad temporal and spatial scales reported here.

Trap nests were checked weekly during the bee breeding season. Once the construction of a nest had ended, the occupied traps were removed from the trap bundles and replaced by empty traps. During the first three study years, we took to the lab all traps with nests for rearing and identification purposes, while during the remaining years we took to the lab only traps for which identification was not possible in the field; other nests were removed from the bundles but left in the field, a few meters away from the original nesting site. For the nests taken to the laboratory, traps were opened to record the number of cells; whenever the nest had more than one brood cell, during the first three study years one cell was extracted for pollen identification, and the remaining cells were kept until adult emergence. Taxonomic identification of pollen was conducted by light microscopy, in comparison with a reference collection. The latter included all flowering plant species recorded in our study plots. For taxonomic identification of pollen grains we used the natural method, comparing with a reference collection that included all flowering plant species occurring in our sites. We kept some emerged adults of each nest for identification and returned the remaining individuals to their site of origin in the field. After the first three study years (i.e., from 2009 onwards), whenever possible, we identified the bee species in the field without taking the nests to the laboratory. The number of brood cells and their taxonomic identity were recorded in all nests. 

Estimation of bee generalization

Bee generalization was defined as Shannon’s diversity index of the pollen found in all nests of each species in the three years in which pollen studies were conducted (2006–2008). To calculate Shannon’s diversity index we started from the number of pollen grains of each plant species in each nest. To this end, pollen from each extracted brood cell was suspended in an ethanol-water solution, from which three samples were taken for pollen quantification under a microscope. Fifteen microscope fields were counted for each sample, and 45 fields total for each brood cell. With these data we calculated the proportion of pollen grains of each plant species in each nest, which we used as a standardized measure of plant abundance per nest, so as to remove the effect of the total number of pollen grains counted per nest. We then summed the standardized abundance of each plant species across all nests as a measure of the abundance of that plant species in the population of a bee species. This is the abundance we used to calculate Shannon’s diversity index.

Estimation of phenologies and nesting–flowering phenological mismatch

We used the weekly data on bee nest and flower abundance for the pooled data for the two study sites. At the level of individual nests, we estimated phenology based on the end date of nest construction, which was the most accurate date we could record in our weekly visits to the sites for three of our study species (A. vigintipunctatum, the two Megachile species), as females usually build the nests in less than a week. T. laticeps appears to take slightly longer than a week, as we sometimes found nests under construction, while X. atamisquensis takes substantially longer (several weeks), but we decided to use also the end date of nest construction for these species to make data more comparable across species. At the population level, we estimated phenology using the abundance-weighted arithmetic mean date (WMD) of nesting or flowering, respectively, calculated with function weighted.mean of the stats package of R statistical software [34]. WMD is considered one of the best phenological estimators and has been widely used in previous studies. Dates were represented as the number of days from July 1 each year, and the weighted average of flowering date was calculated by multiplying each date by the relative abundance of flowers in that date. For example, if flower abundance was 10, 60, and 30 flowers on days 60, 67, and 74 from July 1, the WMD of flowering is calculated as 0.1 × 60 + 0.6 × 67 + 0.3 × 74 = 68.4.

We used the above phenological estimates to estimate nesting–flowering phenological mismatches. Phenological mismatch was calculated as the absolute difference between the end date of the nest construction and the flowering WMD of the main floral resource species (those representing 85% of the pollen found in nests of of a particular bee species). For the population-level analysis, we calculated the number of days between the nesting WMD and the flowering WMD to estimate the nesting–flowering phenological mismatch for each bee species and its main floral resource species. Because we needed to record at least one nest to estimate the nesting–flowering phenological mismatch for a given year, we excluded from the analyses years in which no nests of a particular species were recorded. Under the above definition of phenological mismatches it is reasonable to expect mismatches both for specialist and generalist bee species, as in both cases the nesting and flowering dates are likely to shift among years.

Estimation of bee reproductive success

Reproductive success for individual nests was calculated as the number of brood cells per nest, whereas at the population level, reproductive success was calculated as the average number of brood cells per nest; number of brood cells per nest is commonly used as a measure of reproductive success of female solitary bees. One species, Trichothurgus laticeps, lacks organized brood cells, as females lay bare eggs amidst a pollen mass. In turn, in some nests of Xylocopa atamisquensis adults emerged in the field before we could count the number of brood cells, and although some nests bore clear marks of the former brood cell divisions, in other nests the marks were less clear. Thus, for these species we used the length of the trap cavity occupied by pollen to estimate of the number of brood cells in nests where a direct count was not possible. To this end, we first used a subset of nests for which we could count the number of pupal cocoons (T. laticeps) or brood cells (X. atamisquensis) to fit a linear regression model between number of brood cells or pupal cocoons and length of the cavity occupied by pollen (T. laticeps) or formerly occupied by brood cells (X. atamisquensis), and then used this relationship to estimate the number of brood cells in all nests from the length of the cavity occupied by pollen (see Supplementary Methods for further information on the calculations).