Terrestrial ecosystems around the world have been losing resilience, i.e. their ability to recover after disturbances. Climatic and direct anthropogenic pressures interact with internal ecosystem dynamics to drive these resilience losses, but the specific processes, global patterns and locally most relevant aspects are not well understood. In this project, we computed different indicators of resilience loss (so-called Early Warning Signals, EWS) based on remotely sensed time series of vegetation indicators in terrestrial ecosystems around the globe. In the next step, we are now trying to understand the role different variables play to drive these resilience losses. To do so, we have collected a comprehensive global dataset of multiple potential drivers and will use this to train a range of machine learning model to predict the occurrence of the EWS. We combine the models with various techniques for model interpretability and transparency to tease out the role of different climatic and anthropogenic drivers. Ultimately, this analysis will allow us to identify globally different sensitivities of terrestrial ecosystems to drivers of change, locally most relevant drivers of resilience losses, and potential thresholds and non-linearities in ecosystems' responses to different biophysical drivers.