A Data-Driven Assessment of the Sensitivity of Global Ecosystems to Climate Anomalies

Vegetation is a central player in the climate system, constraining atmospheric conditions through a series of feedbacks. This fundamental role highlights the importance of understanding regional drivers of ecological sensitivity and the response of vegetation to climatic changes. While nutrient availability and short-term disturbances can be crucial for vegetation at various spatiotemporal scales, natural vegetation dynamics are overall driven by climate. At monthly scales, the interactions between vegetation and climate become complex: some vegetation types react preferentially to specific climatic changes, with different levels of intensity, resilience and lagged response. For our current Earth System Models (ESMs) being able to capture this complexity is crucial but extremely challenging. This adds uncertainty to our projections of future climate and the fate of global ecosystems. Here, following a Granger causality framework based on a non-linear random forest predictive model, we exploit the current wealth of satellite data records to uncover the main climatic drivers of monthly vegetation variability globally. Results based on three decades of satellite data indicate that water availability is the most dominant factor driving vegetation in over 60% of the vegetated land. This overall dependency of ecosystems on water availability is larger than previously reported, partly owed to the ability of our machine-learning framework to disentangle the co-linearites between climatic drivers, and to quantify non-linear impacts of climate on vegetation. Our observation-based results are then used to benchmark ESMs on their representation of vegetation sensitivity to climate and climatic extremes. Our findings indicate that the sensitivity of vegetation to climatic anomalies is ill-reproduced by some widely-used ESMs.