Featuring 10 phenological estimators using simulated data

Abstract

Summary1. Studies reporting phenological changes in response to climate change are numerous and concern all groups of living organisms. Phenological changes could cause mismatching in food chains, thus inducing important perturbations in ecosystem functioning. Nevertheless, the relevancy of the conclusions drawn from phenological studies strongly depends on the estimation accuracy of such phenological changes. Many different estimators exist and some have already raised major criticism, although they continue to be used. Therefore, there is a crucial need for an extensive study documenting the behaviour of phenological estimators. 2. Here, we compare the estimation efficiency of 10 phenological estimators: different first appearance dates, mean dates, different percentile dates and a smoothing method based on spline functions using simulated phenological data. Root mean-squared errors and bias of the phenological estimations are calculated in relation to different parameters of the simulated phenological data. 3. Results show that first appearance dates behave as a very inaccurate and biased estimator regarding any phenological data set. Mean dates and estimates calculated using the smoothing method provided in general the most accurate estimates of phenological shifts. They were also the most robust to variation in sample sizes and to imperfect detectability. 4. Our results allow us to warn against the use of first appearance dates in future phenological studies and to recommend using mean dates or smoothing techniques to estimate phenological change of entire distributions. We also provide advice concerning phenological monitoring effort. These recommendations should most importantly apply to studies aiming at comparing phenological variation among sites or among species.

Species' traits predict phenological responses to climate change in butterflies

“…Although some have criticized the use of the absolute date of first appearance (which can be biased as a result of systematic changes in sampling effort and population abundance over time; van Strien et al 2008), we used the mean date of first appearance averaged across all study transects for a given species to mitigate potential bias. A common alternative metric, peak date of appearance, is generally less sensitive to sampling effort and population trends (Moussus et al 2010), but is difficult to interpret when comparisons are being drawn across taxa (e.g., butterfly species) that differ in their number of annual generations. We emphasize that the main goal of our analysis is to examine relative differences in the degree of phenological change with respect to species' traits, rather than to obtain unbiased estimates of the magnitude of phenological change.…”

Climate Change and Flowering Phenology in Worcester County, Massachusetts

“…As noted in several recent articles, however, first flowering date is just one aspect of flowering phenology (Sparks et al 2005;CaraDonna et al 2014). Other metrics that have been used in studies of distributions of phenological records include mean, median, peak, last date, range in dates, and quantiles of the distribution (Sparks et al 2005;van Strien et al 2008;Moussus et al 2010;CaraDonna et al 2014). The use of mean, median, or peak flowering dates has been encouraged as a means of overcoming potential difficulties in interpreting first flowering dates related to different population sizes or sampling frequencies CaraDonna et al 2014).…”

Estimating the onset of spring from a complex phenology database: trade-offs across geographic scales

“…For example, there can be considerable variation in the frequency of records collected by observers at different sites. Further, in the cases where observers are encouraged to record both the presence and absence of phenophases, as in Bstatus monitoring^ (Denny et al 2014), trends may be skewed towards later onset dates when there is a greater timespan between the last observation of absence and the first observation of presence (Miller-Rushing et al 2008;Keatley and Hudson 2010;Morellato et al 2010;Moussus et al 2010). In addition, observers may only record presence of a phenophase and thus fail to capture observations of absence, reducing our ability to quantify precision or uncertainty of phenological metrics such as onset date or duration of phenophase activity (Cole et al 2012;Ferreira et al 2014).…”