In order to identify the most relevant environmental parameters that regulate flowering time of bulbous perennials, first flowering dates of 329 taxa over 33 yr are correlated with monthly and daily mean values of 16 environmental parameters (such as insolation, precipitation, temperature, soil water content, etc.) spanning at least 1 yr back from flowering. A machine learning algorithm is deployed to identify the best explanatory parameters because the problem is strongly prone to overfitting for traditional methods: if the number of parameters is the same or greater than the number of observations, then a linear model can perfectly fit the dependent variable (observations). Surprisingly, the best proxy of flowering date fluctuations is the daily snow depth anomaly, which cannot be a signal itself, however it should be related to some integrated temperature signal. Moreover, daily snow depth anomaly as proxy performs much better than mean soil temperature preceding the flowering, the best monthly explanatory parameter. Our findings support the existence of complicated temperature sensing mechanisms operating on different timescales, which is a prerequisite to precisely observe the length and severity of the winter season and translate for example, 'lack of snow' information to meaningful internal signals related to phenophases.
Keywords: climate change; daily snow depth anomaly; environmental explanatory variables; flowering time fluctuations; machine learning fit; phenology of bulbous taxa; soil temperature.
©2020 The Authors. New Phytologist ©2020 New Phytologist Foundation.