Springtime phenological events are often used as biological indicators of climate change and can be useful for predicting future community responses to global change. Assessment of phenological responsiveness is often accomplished using linear regression, with temperature or temperature proxies in the preceding months used to explain observed events. While linear regression approaches remain useful, they require somewhat arbitrary delineations of antecedent weather windows used to model observed events. Our objective is to explain observed differences in flowering time over the last 30 years among a group of 16 closely related cherry taxa (Cerasus spp.) at the Tama Forest Science Garden (Hachioji, Japan) using a survival analysis approach to better understand flowering cues and how these may vary among taxa. We utilize a survival regression approach that models phenological events over the observation intervals with both time-varying and time-invariant environmental covariates. By using this approach, we are able to quantify differences among species through time in their probability of flowering and incorporate temperature-derived variables (e.g., chill and heat units) on a fine time scale. By gaining a more detailed understanding of the timing of flowering, we hope to provide better forecasts of future flowering events and ultimately how species will respond to climate change.
Results/Conclusions
The cumulative probability curves of first flowering of Tama Forest cherries demonstrate clear taxon-level differences in shape and rate of these probability functions. Some species, such as Cerasus speciosa, show an extended early period of low flowering probability followed by a later, steeper cumulative probability curve. Others, such as C. jamasakura, show much later and compressed cumulative probability curves. Cox proportional hazards models indicate that taxa differ in their response to temperature cues and that the physical environment, such as slope and aspect, also influence flowering times. Variation in cumulative probability estimates is not consistent among species, indicating that some taxa may respond more coherently to temperature cues than others. These results allow the direct use of temperature-related covariates at the same temporal scale as observations, thereby allowing the data to dictate phenology-temperature relationships.