Abstract
Reliable phenological observations are needed to quantify the impact of climate change on tree phenology. Ground observations remain a prime source of phenological data, but their accuracy and precision have not been systematically quantified. The high subjectivity of ground phenological observations affects their accuracy, and the high within-population variability of tree phenology affects their precision. The magnitude of those effects is unknown to date. We first explored the inter-observer variability in the timing of bud development and leaf senescence in trees using a unique dataset of seven observer inter-calibration sessions. Then, using tree phenological data collected in three European forests (n = 2,346 observations for budburst, n = 539 for leaf senescence), we quantified how the 'observer uncertainty' (accuracy of the observations) and the 'population sampling uncertainty' (precision of the observations) combine to affect the estimates of the budburst and the leaf senescence dates. The median observer uncertainty was 8 days for budburst (BBCH = 7) and 15 days for leaf senescence (BBCH = 95). As expected, the population sampling uncertainty decreased with increasing sample size, and was about 6 days for budburst and 10 days for leaf senescence for a sample of 10 individuals monitored per population (corresponding to the median sample size in the phenological literature). As a whole, the overall uncertainty of phenological observations could reach up to 2 weeks for budburst and 1 month for leaf senescence.Synthesis. This paper quantifies for the first time the accuracy and precision of ground phenological observations in forest trees and as such offers tables to estimate the uncertainty of phenological data. We show that reliable estimates of budburst and leaf senescence require three times (n = 30) to two times (n = 20) larger sample sizes as compared to sample sizes usually considered in phenological studies. We further call for an increased effort of observer inter-calibration, required to increase the accuracy of phenological observations. These recommendations reduce the uncertainty of phenological data, thereby improving the estimation of phenological trends over time, the response of phenology to temperature or the inference of phenological model parameters.