Estimating transition dates from status-based phenology observations: a test of methods

The scale of phenology research has expanded due to the digitization of herbarium specimens and volunteer based contributions. These data are status-based, representing the presence or absence of a specific phenophase. Modelling the progress of plant dormancy to growth and reproduction and back to dormancy requires estimating the transition dates from these status-based observations. There are several methods available for this ranging from statistical moments using the Julian day to newly introduced methods using concepts from other fields. Comparing the proficiency of different estimators is difficult since true transition dates are rarely known. Here I use a recently released dataset of flowering observations where, due to high sampling frequency and unique physiology, the transition dates of onset, peak, and end of flowering are known. I used a Monte Carlo analysis to test estimators across two scales using a range of sample sizes and proportion of flowering presence observations. I tested estimates of onset, peak, and end of flowing at the population level, and estimates of onset and end for individual plants. Overall a recently introduced method using a Weibull curve performed the best for population level onset and end estimates, but other estimators may be more appropriate when there is a large amount of absence observations relative to presence observations. For individual estimates a method using the midway point between the first flower presence and most prior flower absence, within 7 days, is the best option as long as the restriction does not limit the final sample size. Otherwise the Weibull method is adequate for individual estimates as well.