Introduction

Tropical forests hold close to 250 Pg of carbon, with Latin America contributing half of this (Saatchi et al. 2011). Although the rates of deforestation appear to have decreased over the past decade, tropical deforestation still represents the bulk of the c. 1.1 PgC yr−1 of C emissions due to land-use change (Friedlingstein et al. 2010). The direct impact of deforestation and degradation has the potential to be mitigated through a performance-based mechanism such as REDD (Reducing Emissions from Deforestation and Forest Degradation), by monetising carbon held in both managed and unmanaged forests (Agrawal, Nepstad & Chhatre 2011). Additionally, tropical forests contribute to a terrestrial carbon sink (Le Quéré et al. 2009), offsetting fossil carbon emissions into the atmosphere through a physiological response of the vegetation (Lewis et al. 2009; Lloyd & Farquhar 2008). Thus tropical forests offer critically important ecosystem services by reducing the short-term effect of anthropogenic carbon emissions into the atmosphere.

However, in the face of global climate trends, the resilience of tropical forests has been called into question (Cox et al. 2000). South America is sensitive to a number of large-scale climatic anomalies, including the El Niño Southern Oscillation, the Pacific Decadal Oscillation and the North Atlantic Oscillation. All of these contribute to displacing the yearly course of the Inter-Tropical Convergence Zone (ITCZ) and increase the strength of the dry season in some regions (Garreaud et al. 2009; Marengo 2004). The 2005 and the 2010 climatic events over Amazonia have exemplified these atmospheric regime shifts, and these may occur more frequently during the twenty-first century. As a result of the increased likelihood of severe droughts, models suggest that Amazonian forests may shift by 2100 to a different biome type akin to a woodland savanna or dry forest (Cox et al. 2000, 2004; Huntingford et al. 2008; Malhi et al. 2009; Poulter et al. 2010). Aside from the radical implications for human and wildlife populations inhabiting Amazonia, this new biome type would have far less potential to hold carbon, and such a shift would have important consequences for global life support services. Some of the predictions of this ‘Amazon dieback’ scenario have been empirically tested (Phillips et al. 2009). Theoretical work has also attempted to understand whether a shift showing alternative stable states is a likely outcome.