Sequential Parametrization of Phenology Model for Predicting Cherry Blossom.

Cherry blossom festivals are often organized on peak bloom of cherry trees (Prunus x yedoensis ‘Yoshino’) to celebrate the onset of spring. Therefore, it is critical to predict peak bloom date in regards with local weather forecast. A two-step model was proposed to estimate flowering time after dormancy release, which were describe by consecutive steps of chill and heat unit accumulations. While this approach demonstrated reasonable accuracy for cherry trees in the Tidal Basin, Washington D.C., exhaustive grid search for model parameterization had several shortcomings. Its inherent computational cost often enforced a coarse search grid with less accuracy. Multimodal parameter space also led to many ambiguous solutions which required another inspection task. In this presentation, a novel sequential method is proposed to remedy these issues. By exploiting previously unused records on intermediate development stages, we construct an additional constraint on the amount of heat accumulation between two particular stages. Then each parameter can be solved in a series by a regular optimization method within a fraction of time. The result was almost identical to the previously known solution and its calibration error was even smaller.