Shortcomings of classical phenological forcing models and a way to overcome them

A theoretical study proves that the common Spring-Warming model, which is widely used in phenological studies and frequently described in the literature, has systematic defects that do not allow a reliable projection of phenological stages for the future (e.g., up to 2100). When calculating spring phenological phases (e.g., beginning of blossom or leaf unfolding, etc.), defects occur because either the advance in blossom is included implicitly in the model and cannot be calibrated sufficiently to observations, or the model parameters attain unphysiological values or lie in a range so that a prognosis for the far future cannot be accomplished. Therefore, the introduction of a daylength term is suggested, which improves the Spring-Warming model and eliminates almost all of the discussed shortcomings. The performance of this improved model is demonstrated by calculating the beginning of apple blossom in Germany. For this purpose, we compared the improved model (M1) with three different versions of the original Spring-Warming model (M2–M4). The models were calibrated (optimized) using observed blossoming and temperature data (1962–2009), which have been regionalized on a 0.2° grid. The optimization was done for a representative grid point. The performance of the various model versions in predicting the beginning of apple blossom was compared with observations from independent years, which were not used in the optimization. Also, the beginning of blossom and its possible future changes were calculated with these models, using temperatures from the Regional Climate Model REMO-UBA with GHG emission scenario A1B (2001–2100). The new daylength term improved the performance of model M1 remarkably, and the model calibration automatically led to model parameters with meaningful values. These results, which were confirmed by other fruit tree species and locations, provided strong evidence that the conventional Spring-Warming models in phenology must be extended by photoperiodic sensitivity, at least for species which are photosensitive.

► We showed that the common growing degree day models have systematic defects.
► They do not allow a reliable projection of phenological stages for the future.
► These models can only be calibrated if unphysiological parameter values are used.
► For improvement, an additional day length term is introduced.
► GDD models are improved remarkably by implementing photoperiodic sensitivity.