Ecosystem physio-phenology revealed using circular statistics
Abstract. Quantifying responses of vegetation phenology to climate variability is a key prerequisite to predict shifts in how ecosystem dynamics due to climate change. So far, many studies have focused on responses of classical phenological events (e.g. budburst or flowering) to climatic variability for individual species. Comparatively little is known on physio-phenological events such as the timing of the maximum gross primary production (DOYGPPmax). However, understanding this type of physio-phenological phenomena is an essential element in predicting the response of the terrestrial carbon cycle to climate variability. In this study, we aim to understand how DOYGPPmax depends on climate drivers across 52 eddy-covariance (EC) sites in the FLUXNET network for different regions of the world. Most phenological studies rely on linear methods that cannot be generalized across both hemispheres and therefore do not allow for deriving general rules that can be applied for future predictions. Here we explore a new class of circular-linear (here called circular) regression approach that may show a path ahead. Circular regression allows relating circular variables (in our case phenological events) to linear predictor variables (e.g. climate conditions). As a proof of concept, we compare the performance of linear and circular regression to recover original coefficients of a predefined circular model on artificial and EC data. We then quantify the sensitivity of DOYGPPmax to air temperature, short-wave incoming radiation, precipitation and vapor pressure deficit using circular regressions. Finally, we evaluate the predictive power of the regression models for different vegetation types. Our results show that the DOYGPPmax of each FLUXNET site has a unique signature of climatic sensitivities. Overall radiation and temperature are the most relevant controlling factors of DOYGPPmax across sites. The circular approach gives us new insights at the site level. In a Mediterranean shrub-land, for instance, we find that the two growing seasons are controlled by different climatic factors. Although the sensitivity of the DOYGPPmax to the climate drivers is very site specific, it is possible to extrapolate the circular regression model across vegetation types. From a methodological point of view, our results reveal that circular regression is a robust alternative to conventional phenological analytic frameworks. In particular global analyses can benefit, where phase shifts play a role or double peaked growing seasons may occur.