Early warning signals of simulated Amazon rainforest dieback

We test proposed generic tipping point early warning signals in a complex climate model (HadCM3) which simulates future dieback of the Amazon rainforest. The equation governing tree cover in the model suggests that zero and non-zero stable states of tree cover co-exist, and a transcritical bifurcation is approached as productivity declines. Forest dieback is a non-linear change in the non-zero tree cover state, as productivity declines, which should exhibit critical slowing down. We use an ensemble of versions of HadCM3 to test for the corresponding early warning signals. However, on approaching simulated Amazon dieback, expected early warning signals of critical slowing down are not seen in tree cover, vegetation carbon or net primary productivity. The lack of a convincing trend in autocorrelation appears to be a result of the system being forced rapidly and non-linearly. There is a robust rise in variance with time, but this can be explained by increases in inter-annual temperature and precipitation variability that force the forest. This failure of generic early warning indicators led us to seek more system-specific, observable indicators of changing forest stability in the model. The sensitivity of net ecosystem productivity to temperature anomalies (a negative correlation) generally increases as dieback approaches, which is attributable to a non-linear sensitivity of ecosystem respiration to temperature. As a result, the sensitivity of atmospheric CO2 anomalies to temperature anomalies (a positive correlation) increases as dieback approaches. This stability indicator has the benefit of being readily observable in the real world.