Problems in the mechanistic spring phenology models

Observational studies have reported phenological changes such as earlier springs and extended growing season length in terrestrial vegetation that may have tremendous ecological and biological consequences. Temperature is the main driver of spring phenology in most vegetation, but empirical and mechanistic models differ in precisely how temperature cues dormancy break. Recent studies identified deficiencies in some empirical approaches for measuring temperature sensitivity and advocated for mechanistic models as a potential remedy. But, current mechanistic models have considerable structural and functional diversity, and no consensus has emerged across a robust literature. This uncertainty makes it difficult to anticipate how plant phenology will respond to future climate warming. In this study, we use simulations and a variety of phenological and temperature datasets to fit and test a suite of the most commonly used mechanistic spring phenology models. All mechanistic models had the same performance, regardless of how they quantified the effect of thermal forcing. In fact, none outperformed a simple mean temperature approach. More importantly, we found instability in parameter estimates that undermines our ability to test hypotheses and predict future response. Very minor changes in training data can produce dramatic changes in estimated parameters. Simulations show that statistical optimization is incapable of simultaneously estimating the true parameters, even when spring phenology is simulated by the same model. This indicates that the problem is fundamental to the structure of the models, regardless of whether they correctly encode the true phenological process. We caution that using mechanistic models is not a panacea. In fact, their limitations may be more severe because they are harder to identify. We suggest that all future studies be cautious of fitting these mechanistic models, and take care to rigorously estimate parameter uncertainty when drawing inference from them.