@misc{Le, type = {Master Thesis}, author = {Le, Nhien}, title = {MODEL SELECTION AND OPTIMIZATION FOR SWEET CHERRY BLOSSOM AND NORWAY SPRUCE BUDBURST IN THE PHENOLOGICAL GARDEN - HAUS RISWICK KLEVE}, school = {Hochschule Rhein-Waal}, abstract = {Phenological models are tools used to predict the timing of phenological events based on environmental factors such as temperature, precipitation, and daylength. This research focuses on enhancing the accuracy of phenological event prediction by applying and refining models for two distinct plant species: Sweet cherry (SC) and Norway spruce (NS Early and late provenances). The analysis employs six models included different kinds of models for each species and ultilizes weather and daylength data of Kleve spanning 2013 to 2021. The objective is to evaluate model performance in the context of Haus Riswick; optimizing their precision through the factor inclusion, parameterization or ensembling method to localize the models to Haus Riswick in Kleve. Comparative assessment reveals that forcing models based on the Growing Degree Day concept or single-phase empirical models outperform chilling-forcing models in predicting the onset of SC blossoming and NS bud bursting models. Notably, the inclusion of the daylength term significantly influences the accuracy of sweet cherry models, whereas this term is not as pronounced in Norway spruce models. After being tested, applied models were calibrated to fit the models into the phenological observed data of Haus Riswick; the calibration process yields substantial improvement in both species' projection, reducing Root Mean Square Error (RMSE) to approximately 2-3 days for sweet cherry and 3-5 days for Norway spruce. Although some uncertainties persist in the simulation of these calibrated models, this methodology contributes to rending the models for both species in Haus Riswick more congruent with local conditions, enhancing their credibility. Ensembling approach is adopted for the Norway spruce budbursting models, where six diverse models are combined to compensate individual limitations. This ensembling method, involving averaging the forecasts of individual models, yields a more refined and dependable prediction (RMSE reduced to 5 days).}, language = {en} }