Global biosphere–climate interaction: a causal appraisal of observations and models over multiple temporal scales

Abstract

Abstract. Improving the skill of Earth system models (ESMs) in representing climate–vegetation interactions is crucial to enhance our predictions of future climate and ecosystem functioning. Therefore, ESMs need to correctly simulate the impact of climate on vegetation, but likewise feedbacks of vegetation on climate must be adequately represented. However, model predictions at large spatial scales remain subjected to large uncertainties, mostly due to the lack of observational patterns to benchmark them. Here, the bidirectional nature of climate–vegetation interactions is explored across multiple temporal scales by adopting a spectral Granger causality framework that allows identification of potentially co-dependent variables. Results based on global and multi-decadal records of remotely sensed leaf area index (LAI) and observed atmospheric data show that the climate control on vegetation variability increases with longer temporal scales, being higher at inter-annual than multi-month scales. Globally, precipitation is the most dominant driver of vegetation at monthly scales, particularly in (semi-)arid regions. The seasonal LAI variability in energy-driven latitudes is mainly controlled by radiation, while air temperature controls vegetation growth and decay in high northern latitudes at inter-annual scales. These observational results are used as a benchmark to evaluate four ESM simulations from the Coupled Model Intercomparison Project Phase\xa05 (CMIP5). Findings indicate a tendency of ESMs to over-represent the climate control on LAI dynamics and a particular overestimation of the dominance of precipitation in arid and semi-arid regions at inter-annual scales. Analogously, CMIP5 models overestimate the control of air temperature on seasonal vegetation variability, especially in forested regions. Overall, climate impacts on LAI are found to be stronger than the feedbacks of LAI on climate in both observations and models; in other words, local climate variability leaves a larger imprint on temporal LAI dynamics than vice versa. Note however that while vegetation reacts directly to its local climate conditions, the spatially collocated character of the analysis does not allow for the identification of remote feedbacks, which might result in an underestimation of the biophysical effects of vegetation on climate. Nonetheless, the widespread effect of LAI variability on radiation, as observed over the northern latitudes due to albedo changes, is overestimated by the CMIP5 models. Overall, our experiments emphasise the potential of benchmarking the representation of particular interactions in online ESMs using causal statistics in combination with observational data, as opposed to the more conventional evaluation of the magnitude and dynamics of individual variables.