Developing strategies to prevent collapse of the Amazon rainforest

By generating its own rainfall regionally and suppressing occurrence of wildfires locally, the Amazon rainforest promotes the conditions required for its own stability. Hence, removal of forest reduces the stability of the remaining forest. Studies estimate that a 20-25% deforestation combined with climate change could induce a large scale collapse of the remaining forest to tropical savanna. This would cause a loss of much of its 10% share of global biodiversity, disrupt the regional water cycle and further accelerate global climate change. The current amount of deforestation is 17%, with recent droughts seen as possible first signs of an approaching collapse.
Rapid state shifts that are disproportionate to the driving changes in conditions are commonly called tipping points. Identification and classification of tipping points is based on the analysis of governing equations. Yet, one can only obtain these equations by taking a so-called mean-field approximation, which expresses the system's dynamics in terms of the means of its macro-scale quantities (such as forest cover fraction).
This works when the number of system units (such as plant patches) is large, the environmental conditions are approximately spatially homogeneous, and random influences average out, making fluctuations small. In an ecosystem such as the Amazon rainforest these assumptions are violated: spreading processes such as plant dispersal or fire can generate large fluctuations and environmental conditions (such as rainfall patterns) are often highly heterogeneous.
This project will introduce a method that can extract tipping criteria from simulations of large systems of coupled units with random interactions, even when the model is heterogeneous or the assumptions behind mean field approximations break down (at so-called continuous phase transitions).
The method runs the simulations in a non-conventional way by introducing artificial feedback control and then extracting information about the original uncontrolled system from observations of the controlled system.
This new approach will be developed and tested on probabilistic cellular automata. We will then apply the new method to a model of the Amazon rainforest with realistic heterogeneities (including climatic gradients, soil quality and human impact). We will study its tendency to collapse under various deforestation scenarios and determine which reforestation strategies would be required for prevention of or recovery from collapse.