Global and regional climate models, such as those used in IPCC assessments, are the best tools available for climate predictions. Such models typically account for large-scale land-atmosphere feedbacks. However, these models omit local vegetation-environment feedbacks that may be crucial for critical transitions in ecosystems at larger scales. In this viewpoint paper, we propose the hypothesis that, if the balance of feedbacks is positive at all scales, local vegetation-environment feedbacks may trigger a cascade of amplifying effects, propagating from local to large scale, possibly leading to critical transitions in the large-scale climate. We call for linking local ecosystem feedbacks with large-scale land-atmosphere feedbacks in global and regional climate models in order to improve climate predictions. © 2011 Elsevier B.V.