Generalized additive mixed models for disentangling long‐term trends,local anomalies,and seasonality in fruit tree phenology

Quantifying temporal patterns of ephemeral plant structures such as leaves, flowers, and fruits gives insight into both plant and animal ecology. Different scales of temporal changes in fruits, for example within‐ versus across‐year variability, are driven by different processes, but are not always easy to disentangle. We apply generalized additive mixed models (GAMMs) to study a long‐term fruit presence–absence data set of individual trees collected from a high‐altitude Afromontane tropical rain forest site within Bwindi Impenetrable National Park (BINP), Uganda. Our primary aim was to highlight and evaluate GAMM methodology, and quantify both intra‐ and interannual changes in fruit production. First, we conduct several simulation experiments to study the practical utility of model selection and smooth term estimation relevant for disentangling intra‐ and interannual variability. These simulations indicate that estimation of nonlinearity and seasonality is generally accurately identified using asymptotic theory. Applied to the empirical data set, we found that the forest‐level fruiting variability arises from both regular seasonality and significant interannual variability, with the years 2009–2010 in particular showing a significant increase in the presence of fruits‐driven by increased productivity of most species, and a regular annual peak associated occurring at the end of one of the two dry seasons. Our analyses illustrate a statistical framework for disentangling short‐term increases/decreases in fruiting effort while pinpointing specific times in which fruiting is atypical, providing a first step for assessing the impacts of regular and irregular (e.g., climate change) abiotic covariates on fruiting phenology. Some consequences of the rich diversity of fruiting patterns observed here for the population biology of frugivores in BINP are also discussed.