Unbiased inference of plant flowering phenology from biological recording data

Abstract

Phenology is a key indicator and mediator of the ecological impacts of climate change. However, studies monitoring the phenology of individual species are moderate in number, taxonomically and geographically restricted, and mainly focused on spring events. As such, attention is being given to nonstandard sources of phenology data, such as the dates of species' biological records. Here, we present a conceptual framework for deriving phenological metrics from biological recording data, while accounting for seasonal variation in the level of activity by recorders. We develop a new Bayesian statistical model to infer the seasonal pattern of plant 'recordability'. The modelled dates of maximum recordability are strongly indicative of the flowering peaks of 29 insect-pollinated species monitored in two botanic gardens in Great Britain. Conversely, not accounting for the seasonality in recording activity results in biased estimates of the observed flowering peaks. However, observed first and last flowering dates were less reliably explained by the model, which probably reflects greater interspecific variation in levels of recording before and after flowering. We conclude that our method provides new potential for gaining useful insights into large-scale variation in peak phenology across a much broader range of plant species than have previously been studied.