Shifting phenology, the timing of life history events, is a primary response of organisms to changes in their environment, particularly related to climate [1, 2]. Phenological patterns vary across space and time, often in ways that are predictable based on gradients in temperature and precipitation [3, 4] or static cues like photoperiod [5]. Yet not all species respond to environmental shifts in the same way [2] and so when multiple species interact, differential shifts in phenology may cause mismatches in seasonal timing [6]. These mismatches may have demographic or even ecosystem consequences and phenological mismatches are currently a major focus of ecological and evolutionary research (e.g., [4, 7, 8, 9]).
Insect phenology has been shown to be particularly sensitive in terms of how organisms shift their timing to adjust to changing environments [3, 10]. This is because insects are ectothermic and their developmental rate is thus driven largely by ambient temperature [10]. Because of their importance for agricultural systems, models that accumulate degrees within certain ranges (called growing degree day, or GDD, models) have been developed to predict local phenology of insects and have proven to be highly effective for both pest and non-pest species [11]. Butterflies are an excellent group for the study of insect phenology; their biology is well-known and they are primary consumers, the trophic level that has been found to be the most sensitive in terms of animal phenology [4].
Studies of butterfly phenology range from detailed, mechanistic single-species studies (e.g., [12,13,14,15,16 ]) to broader examinations of whole communities that assess consistency of responses across scales while also capturing species-specific variability in sensitivity (e.g., [17–25] ). Consistently and not surprisingly, these studies have found that many, but not all, butterfly species fly earlier in warmer years. Traditional timing of emergence (e.g., spring, summer, or fall flyers) has been found to be an important factor in phenological shifts [17–20, 23], with earlier flyers showing more sensitive shifts forward and later flyers sometimes shifting later. Overwinter stage was also shown to be important [17, 19, 20, 23] with the later developmental stages when overwintering (adult vs. pupa vs. larvae vs. egg) being associated with earlier and more sensitive emergence timing. Other traits that have been studied, such as hostplant breadth, mobility, and voltinism have had more mixed results.
Phenological shifts are difficult to estimate because their detection is strongly influenced by the timing and structure of monitoring events. Thus, a critical component for all phenology studies is choosing an appropriate metric (or “yardstick”) and data set to detect change at different points along seasonal time-courses [6, 26]. One metric of particular interest is onset, the first emergence of adults each season. Yet onset is particularly difficult to estimate because it occurs, by definition, when population levels are at their smallest [26, 27]. This challenge is compounded by monitoring data emerging from a variety of different observation protocols, each with particular biases that may obscure this hard-to-detect event. One solution is to focus on an arbitrary threshold when a certain portion of earlier-emerging individuals have been recorded (e.g., 10% or 25%). Alternatively, mid-season metrics, (e.g. mean or median period) are often more robust to variation in data type and density [26], but may be less meaningful in terms of ecological dynamics, such as mismatch.
Here, we focus on the Northeastern US butterfly community, comparing early (10%) adult emergence and mid-season timing (50%) for data generated from two types of community (“citizen”) science inventories. Our goal is to determine how these two popular and growing monitoring resources inform adult phenology at broad, regional scales as a means to understand their shared utility and potential for future, analytical integration. The first type of data emerges from networks of volunteers who carry out repeated surveys on established transects using academic-like protocols that were designed specifically to track broad patterns in butterfly abundance and timing [28]. These programs typically provide high quality data, including all observed target species, abundances, and metrics of effort; yet, such surveys are generally limited geographically because of the effort to initiate them, recruit volunteers, and retain them [29]. Of particular relevance to phenological studies, the timing of survey initiation each year will influence the ability to capture early-season dynamics [30].
A second class of community science resource is incidental observations of butterflies posted to online platforms such as iNaturalist or eButterfly [31, 32]. These platforms have few restrictions for inclusion and growth in participation has been phenomenal, leading to the highest spatial density of records compared to other monitoring programs, although their recent initiation means that the temporal scope of data is currently limited [33]. iNaturalist.org, for example, has nearly doubled the number of records collected every year since its inception in 2008. By 2014, participation had been slowly growing and, by that year, 1,738 community scientists added 19,598 butterfly observations globally. However, by 2020, 121,470 community scientists across the globe reported 838,080 butterfly observations, a greater than 40-fold increase in observations in just 6 years [34]. Almost all of these reports include a digital photograph voucher, and a sizable proportion have at least two agreed-upon identifications by other members of the iNaturalist community. iNaturalist considers these records as “research-grade” [35].
This recent, explosive growth of incidental data provides significant potential for use in phenological analysis. However, accounting for variable effort across time and space for these resources is a substantial challenge [36–38]. Data without repeated site visits and where no information on effort or reports of absences (“presence-only” data) must account for recording bias, often by aggregating records at coarse grains (i.e. 10 km or higher) to achieve sufficient data density, but this also obscures local phenological variability [39]. Despite the challenges, with sufficient data density, deriving insights about phenology from presence-only data holds promise [26, 40]. For example, Karlsson [17] obtained high density of presence-only data from Sweden’s popular community science web portal (https://www.artportalen.se/), and found results consistent with other analyses of European butterflies (e.g., [19, 23, 25]).
Given potential biases in both data collection methods (structured surveys with inconsistent start dates or incidental reports that are presence-only), it is not clear which data set should be considered the “right” one (i.e., the standard to compare other data sets), and so consistency of results between data sets is one benchmark that can be used to compare the validity of findings of multiple sources of data. Another benchmark to consider is whether findings conform to patterns reported elsewhere; results that diverge substantially from typical observations should be carefully rechecked for unaccounted data biases and the selection of appropriate modeling frameworks [38]. We present findings on phenological sensitivity from two sources of butterfly monitoring data. Our specific a priori hypotheses are that flight phenology will advance where and when temperatures are warmer. Additionally, we predict that timing will be earlier and more sensitive for summer species that overwinter at more advanced stages (adult, pupae, larvae, then egg). We also explore the relationship between shifting phenology and other traits that have been found or suggested to potentially be important, including mobility, habitat association, hostplant breadth, and voltinism.