Matzneller Philipp, Blümel Klaus, Chmielewski Frank-M
Agricultural Climatology, Faculty of Agriculture and Horticulture, Humboldt-University of Berlin, Albrecht-Thaer-Weg 5, 14195, Berlin, Germany,
Int J Biometeorol. 2014 Jul;58(5):703-15. doi: 10.1007/s00484-013-0651-1. Epub 2013 Mar 2.
Seven different model approaches to calculate the onset of sour cherry blossom for the main growing regions in Rhineland-Palatinate (Germany) were compared. Three of the approaches were pure forcing models (M1, M2, M2DL) and the remaining four models were combined sequential chilling-forcing (CF) models. Model M1 was the commonly used growing degree day (GDD) model in which the starting date of temperature accumulation (t1), the base temperature (TBF) and the forcing requirement F* were optimized on the basis of observed data. Because of a relatively late optimal starting date (t1=1 March), the model can be applied only to calculate the onset of cherry blossom for present climate conditions. In order to develop forcing models that could possibly be used to estimate possible shifts in the timing of cherry blossom due to climate change, the starting date t 1 of the models was intentionally set to 1 January (M2, M2DL). Unfortunately, model M2 failed in both the optimization and validation period. The introduction of a daylength term (DL) in model M2DL improved model performance. In order to project possible shifts in the timing of plant phenological events, combined CF-models are preferred over pure GDD-models. For this reason four CF-models were developed with (M3DL, M4DL) and without (M3, M4) consideration of daylength in the GDD-approach. The chilling requirement was calculated using chilling hours (M3, M3DL) and chill portions (M4, M4DL). Both models without daylength estimated implausible model parameters and failed model validation. However, models M3DL and M4DL showed meaningful model parameter estimations and the error between modelled and observed data was markedly reduced. Moreover, the models optimized and validated (internal validation) for one sour cherry growing region in Germany, were applied successfully to calculate the beginning of the blossom period in other regions in Europe and even at one station in North America (external validation).
对七种不同的模型方法进行了比较，这些方法用于计算德国莱茵兰-普法尔茨州主要种植区酸樱桃开花的起始时间。其中三种方法是纯强迫模型（M1、M2、M2DL），其余四种模型是组合的顺序低温-强迫（CF）模型。模型M1是常用的生长度日（GDD）模型，其中温度积累的起始日期（t1）、基础温度（TBF）和强迫需求F*是根据观测数据进行优化的。由于最优起始日期相对较晚（t1 = 3月1日），该模型仅适用于计算当前气候条件下的樱花开放起始时间。为了开发可能用于估计气候变化导致的樱花开放时间可能变化的强迫模型，模型的起始日期t1被有意设置为1月1日（M2、M2DL）。不幸的是，模型M2在优化和验证阶段均失败。在模型M2DL中引入日长项（DL）提高了模型性能。为了预测植物物候事件时间的可能变化，组合CF模型比纯GDD模型更受青睐。因此，开发了四种CF模型，其中两种（M3DL、M4DL）在GDD方法中考虑了日长，另外两种（M3、M4）未考虑日长。低温需求是使用低温小时数（M3、M3DL）和低温部分（M4、M4DL）来计算的。两种未考虑日长的模型估计出的模型参数不合理，且模型验证失败。然而，模型M3DL和M4DL显示出有意义的模型参数估计，并且模拟数据与观测数据之间的误差显著降低。此外，针对德国一个酸樱桃种植区进行优化和验证（内部验证）的模型，成功应用于计算欧洲其他地区甚至北美一个站点的开花期开始时间（外部验证）。
