2015 年 71 巻 2 号 p. 77-89
Crop phenology models play a pivotal role in predicting yields under climate change. Cultivar-specific model parameters are essential for accurate prediction, but their estimation generally requires elaborate and laborious experiments, and such parameter sets have therefore been available only for a small number of cultivars. We propose methodology for estimating phenological parameters, combining a stochastic parameter estimation method (genetic algorithm) with the use of a database comprising 30 years of records from variety trials conducted at experimental stations across Japan. Optimal parameter sets were selected based on the results of cross-validation tests. This methodology allowed us to estimate phenological parameters objectively. We estimated phenological parameters for the 10 leading cultivars currently planted in Japan, and showed that these parameters reflect the cultivars' sensitivity to temperature and/or photoperiod. The proposed methodology can be used to provide quantitative evaluations of the environmental responses of rice cultivars, without relying on elaborate and laborious experiments, and substantially improves the efficiency of phenological trait phenotyping.