MSc Defence by Weiyuan Chen
Title: Machine learning approach to identifying tipping of ecosystems via self-organization and pattern formation.
Abstract:
Desertification is a threat to millions of people that live in drylands. There is a need for deeper understanding of the catastrophic effects climate have on ecosystems. Recognizing critical transitions or tipping points, contributes to this understanding. It has been suggested that spatial self-organization in ecosystems, such as the formation of regular vegetation patterns can act as a early-warning signal. However, recent findings indicate that spatial self-organization may enable ecosystems to evade tipping points and thus serve as a sign of resilience. This study uses machine learning to help bridge the gap between theory and observations. A convolutional neural network trained on satellite images showing spatially self-organized vegetation patterns was used to identify and mine similar patterns, highlighting locations where regular vegetation patterns occur. The locations were then plotted and analyzed with a bifurcation-type diagram. The results show that patterns exist in the same precipitation range with no clear boundary. Only stripes and labyrinth can be separated.
Supervisors: Johannes Lohmann and Peter Ditlevsen
Censor: Jens Olaf Pepke Pedersen