Document Type

Dissertation

Degree

Doctor of Philosophy

Major

Physics

Date of Defense

9-2-2021

Graduate Advisor

Sonya Bahar

Co-Advisor

Alexey Yamilov

Committee

Ricardo Flores

Wendy Olivas

Julia Medvedeva

Abstract

This study is an exploration of phase transition behavior in evolutionary and population dynamics, and techniques for predicting population changes, across the disciplines of physics, biology, and computer science. Under the looming threat of climate change, it is imperative to understand the dynamics of populations under environmental stress and to identify early warning signals of population decline. These issues are explored here in (1) a computational model of evolutionary dynamics, (2) an experimental system of decaying populations under environmental stress, and (3) a machine learning approach to predict population changes based on environmental factors. Through the lens of critical phase transition behavior, the non-equilibrium continuous transition of a neutral agent-based model is shown to exhibit power-law-like behavior for two control parameters in the critical regime. The model does not fall within the directed percolation universality class, despite exhibiting some characteristics of directed percolation. The results also compare a system exhibiting quenched randomness with one that does not. Experimentally, the impact of two stressors, temperature and NaCl stress, are examined in S. cerevisiae. Increased levels of NaCl in growth media result in a smooth transition from a survivable to an uninhabitable environment, whereas increased temperature stress results in a transition with signs of critical behavior. Lastly, population data from the Living Planet Index and weather data from NOAA are used to predict population changes based on weather attributes using classification and regression machine learning models. Results indicate that a machine learning approach is viable, but additional data and environmental factors are needed to improve the predictive models.

Share

COinS