Integrating complexity science and artificial intelligence: GIS, agents and reinforcement learning for modeling forest cover change

Forest cover change is a complex spatially dynamic phenomenon involving the interaction of numerous processes leading to emerging forest patterns over time. This is especially true when considering forestry operations that attempt to harvest trees for maximizing short-term profits while contending with natural disturbances, fluctuating economies, and the need to conserve long-term ecosystem functions. Conventional computer models assist harvesting activities by generating forest cover strategies that satisfy both economic and ecological objectives. However, such models ignore the dynamic forces that govern the harvesting process, and as such produce strategies that can be in direct conflict with emerging patterns. The purpose of this dissertation is to enhance existing modeling approaches by bridging complex systems theory and artificial intelligence in order to incorporate spatial and temporal complexities of forest harvesting. Specifically, this dissertation introduces a novel approach for integrating geographic information systems (GIS), agent-based modeling (ABM) and reinforcement learning (RL) for developing intelligent agents that can represent stakeholder behaviours and their influence on forest cover change. Agents embedded with RL algorithms possess learning mechanisms that allow them to gain knowledge from their experiences in a dynamic forest environment as represented by GIS digital data structures. Agents learn where and when harvesting activities should take place in the forest in order to satisfy different and at times conflicting objectives that exist at varying spatial scales. These objectives are achieved amidst fluctuating timber prices, the presence of natural disturbances, and the actions of other agents. Model results provide forest management with suitable harvesting strategies that satisfy conflicting objectives, information regarding the relationship between stakeholder interactions and emerging forest cover patterns, and the ability to evaluate the tradeoffs between different harvesting objectives. The developed approach is implemented in the context of forest management in British Columbia using datasets representing forest cover in the Chilliwack Forest District. This dissertation provides novel contributions to the fields of Geographical Information Science and Land Use/Cover Change by enhancing contemporary ABM approaches for simulating complex systems, and to the discipline of Forest Management by improving methods for understanding how to develop suitable forest cover patterns.;Keywords: Geographic information systems; complex systems theory; artificial intelligence; agent-based modeling; reinforcement learning; forest cover change.