Introduction
Climate change has caused widespread shifts in the reproductive periods of populations across species, which may result in profound consequences across levels of ecological organization. To date, the majority of phenological studies has focused on magnitudes of phenological responses in mean flowering time to climate conditions. However, many of the ecological effects of phenological changes are caused by changes in the duration of a plant species’ synchrony with pests or pollinators, or the duration over which a species is exposed to adverse conditions during vulnerable phenophases such as flowering or fruit production (Inouye 2008, Park et al. 2020). Mean flowering dates are not necessarily useful in evaluating these processes, as changes in climate affect not only the mean timing of flowering, but also flowering duration (CaraDonna et al. 2014). In such cases, phenological shifts in flowering duration may alter the synchrony between interacting taxa, affecting plant-pollinator interactions (Bodley et al. 2016), interspecific competition for pollinators (Harris 1977, Waser 1978, Anderson and Schelfhout 1980, Rathcke 1988, Forrest et al. 2010), and susceptibility to herbivory (Asch and Visser 2007, Singer and Parmesan 2010) in ways that are not apparent when considering only shifts in their mean timing. Therefore, more fully determining the ecological consequences of phenological shifts attributable to climate change requires that we develop the ability to forecast changes in the duration of each phenophase (e.g., flowering) within local populations by modeling changes in the dates of population-level onsets and terminations for that phenophase. However, field-based phenological records documenting the onset and termination of phenophases across multiple species are limited in geographic or taxonomic scope (Sherry et al. 2011, Bock et al. 2014), and frequently focus on repeated observation of specific individuals rather than local populations. To date, this has limited our ability to assess climate-driven shifts in phenological synchrony across regions and taxa, highlighting the need for taxonomically and spatially extensive data sources that offer the capacity to estimate the duration of targeted phenophases.
Herbarium records and other specimen-based data represent the most taxonomically, geographically, and temporally extensive source of phenological information for wild and naturalized species (Davis et al. 2015, Willis et al. 2017). Moreover, herbarium specimens have been widely used to estimate phenological responses to climate in temperate regions (Davis et al. 2015, Rawal et al. 2015, Jones and Daehler 2018, Park and Mazer 2018, Park et al. 2019, Taylor 2019, Ramirez-Parada et al. 2022) and have captured patterns of phenological variation that are similar to those observed in the field (Miller-Rushing et al. 2006, Ramirez-Parada et al. 2022).
Despite their growing use, the utility of herbarium specimens for estimating phenological onset and termination dates is unclear due to several limitations. Crucially, herbarium records are sampled opportunistically, providing single snapshots of the phenological status (e.g., flowering) of an individual at a given place and time. As such, it is rarely possible to discern whether a specimen was collected immediately after the onset of a given phenophase, at its peak, or shortly before its termination. Similarly, it is typically not possible to determine whether the sampled individual represents a collection from an early- or late-flowering individual within its local population. Due to these sources of uncertainty, the phenological dates of individual specimens may not reflect the date on which any specific individual- or population-level phenological event occurred. This limitation has the potential to restrict their utility for measuring the precise timing of a given phenophase at the individual level or for estimating the responses to climate of the extremes of a population’s temporal phenological distribution (i.e., the onset and termination of a phenophase at the population level).
Additionally, while validation studies have shown that herbarium-based estimates of the temperature sensitivity of mean flowering dates typically match those derived from field observations (Robbirt et al. 2011, Ramirez-Parada et al. 2022), estimates of the first (and last) occurrence of a phenophase are more subject to the effects of outliers and to variation in sampling intensity, population size, and other confounding effects than estimates of mean flowering (Tryjanowski et al. 2005, Miller-Rushing and Primack 2008). These qualitative limitations of specimen data may intrinsically limit the accuracy with which population-level flowering onset and termination can be predicted even when plant phenology responds strongly to well documented aspects of climate. Similarly, it is possible that the number of specimens required to overcome these limitations and produce accurate phenological estimates from these data are prohibitive.
Finally, relative to living plants that are intensively monitored for their phenological transitions, herbarium records may be subject to several forms of bias when used to estimate the timing of phenological events. First, some species may be preferentially collected during the early or late portion of their local population-level flowering displays. This is most likely among the earliest and latest-flowering species, which flower partially outside of the typical growing season. If specimen collection efforts in general are highest when most species are in flower at a given location, then collection effort is likely to be relatively high in the later portions of early-flowering species’ flowering period (i.e., when most other species are flowering) and inearly portions of late-flowering species’ flowering period. Similarly, species are likely to be less frequently collected during portions of their flowering period that frequently overlap with inclement weather or storm events, as poor weather is associated with reduced collector activity (Daru et al. 2017).
Alternatively, collectors may preferentially collect specimens from individuals within certain portions of their individual flowering period. Evaluations of collections across multiple species have found that collectors often preferentially collect specimens from individuals that are close to their peak flowering date, when the largest numbers of flowers are present (Primack et al. 2004, Davis et al. 2015, Panchen et al. 2019). Conversely, species with fragile flowers or that are subjected to high rates of herbivory may be preferentially collected shortly after the onset of flowering, when petals and other delicate structures are most likely to be intact. Additionally, collectors who prefer specimens bearing both flowers and fruits may collect specimens shortly before flowering termination, when both structures are likely to be present.
Despite such biases (Daru et al. 2017, Panchen et al. 2019), recent work by Ramirez-Parada et al. (Ramirez-Parada et al. 2022) found that herbarium- and field-based estimates of flowering sensitivity to temperature closely agreed in magnitude and direction despite substantial differences in the timing, location, and associated climate conditions captured by both types of data. However, as this work examined mean flowering time, the implications of these forms of bias for predictions of the timing and duration of the local flowering period for each species remain unknown.
Nevertheless, forecasting changes to the entire distribution of phenological events within a population—rather than simple changes in mean timing—is essential to understanding the effects of climate change on seasonal floral resource availability as well as on a host of ecological processes from pollinator activity to floral vulnerability to frost damage. Determining whether predictions of population-level flowering onset and termination are less accurate than predictions of median flowering, or require larger sample sizes is therefore necessary to confidently leverage the unparalleled taxonomic and spatiotemporal scope of natural-history collections. Despite this, no studies to date have sought to validate herbarium-based estimates of phenological onsets and terminations, likely because such assessments require a suitable reference dataset of population-level phenological timings against which the accuracy of phenological predictions derived from specimen data can be tested. Unfortunately, extensive field datasets of population-level phenological events across several locations throughout the range of a species are exceedingly rare, limiting our ability to empirically validate such herbarium-based estimates.
In this study, we used simulated phenological data to robustly and systematically assess the accuracy of models of population-level flowering onset and termination derived from opportunistically sampled data (henceforth, “phenoclimate” models). These data incorporated uncertainty or bias in the timing of specimen collection relative to the start and end of the flowering period of the sampled individual, and in the relative timing of flowering of the individual relative to its source population. By assessing the accuracy of estimated population-level flowering onsets and terminations of simulated plant taxa exhibiting different flowering characteristics and sampling biases, we then assessed the effects individual plant flowering duration, intrapopulation variation in flowering time among individual plants. We then determined the relationship between data availability and model performance, from which we inferred the number of specimens required to produce reliable phenoclimate models of population-level flowering onset and termination. Finally, we evaluated the effects of A) biases towards collection of early or late individuals within local populations and B) biases towards collection of individuals proximate to their flowering onset or termination dates.