Cascade prediction of winter wheat developmental stages considering snow cover period in Hokkaido, Japan

We propose a model to predict winter wheat developmental stages in Hokkaido, a snowy region in Japan. The model uses regression equations incorporating day of the year (DOY) data of multiple developmental stages preceding the prediction target stage and air temperature. Using DOY of snow melting dates, the timing of snow disappearance is considered explicitly. We predicted dates of panicle formation, heading, flowering, and maturity of winter wheat ‘Kitahonami’. We also applied the conventional developmental rate (DVR) model, which considers the vernalization requirement of winter wheat, and compared the obtained prediction accuracy against that of our proposed model. For the DVR model, the root mean square errors (RMSEs) were, respectively, 3.6, 1.9, 2.5, and 3.1 days for panicle formation, heading, flowering, and maturity. By contrast, for the proposed model, they were 2.8, 1.9, 1.9, and 2.6 days, respectively. Mean absolute errors (MAEs) for the proposed model were also smaller than those for the DVR model. The proposed model shows improved prediction accuracy, indicating that the use of DOYs of multiple stages and air temperature is beneficial for predicting various developmental stages.