Early warning signals of simulated Amazon dieback

<p>Dieback of the Amazon rainforest has been considered a potential tipping point in the Earth system due to the belief that there is more than one stable attractor in its dynamics and for future projections within global climate models (GCMs), in some cases a huge amount of forest is lost abruptly. The rainforest is a huge carbon sink, playing a critical role in the global carbon cycle and so if dieback is going to happen over a short period of time, it is important to have some early warning that this is going to occur.</p> <p>In our paper recently published in Theoretical Ecology (http://link.springer.com/article/10.1007/s12080-013-0191-7), we test generic early warning signals (rising AR(1) and variance and changes in skewness over time) on output from an 57-member ensemble of HadCM3 where each member differs due to perturbing poorly constrained parameters. We find that the early warning we receive prior to dieback is poor apart from a rising variance, which we attribute to increasing variance observed in temperature and precipitation which drive the forest. Our lack of observing a robust signal is due to the non-linear forcing of the forest as well as the increasing variance found in the driving time series.</p> <p>The failure of the generic early warning signals led us to seek more process-based indicators of approaching dieback. The first of these looks at the change in sensitivity of NEP anomalies to temperature anomalies over time, which generally increases across the ensemble and can be attributed to a non-linear sensitivity of ecosystem respiration to temperature. Because of this, the sensitivity of CO2 anomalies (which are strongly negatively correlated with NEP) to temperature anomalies also increases over time. This measure could also be observed in the real world.