MSc Seminar: Devin Rose

Title: Describing an Environment-Agent Based Model for Improving Computational Efficiency Using a Case Study Investigating Anthropogenic Effects on a Local Fishery

Abstract:

Computer simulations reproducing observed behaviour of real-world systems have experienced a lot of interest as data processing techniques and computer processing speeds have improved over the years. One approach to modelling capable of producing accurate results is using an agent-based model (ABM). ABMs accurately simulate the decisions of any finite number of individual autonomous agents using an appropriate set of parameters, but with all the attention to detail, the simulation of many agents is very computationally expensive. The growth in complexity (increasing runtime) of ABMs effect the development and use of the model during three stages, finding the appropriate set of parameters accurately modelling the behaviour from available data, testing individual components of the model during development, and performing a risk assessment with statistical significance. Thus, the Environment-ABM (Enviro-ABM) model paradigm is proposed to address the problems with the computational expense associated with environmental ABMs. Using a case study of an ABM investigating the effects of anthropogenic activity on lake whitefish (Coregonus clupeaformis) in the main basin of Lake Huron, Canada, the enviro-ABM paradigm is explained. Studying the effects of anthropogenic activities through a population risk assessment can quantify the impact on lake whitefish and is vital for understanding how commercial fisheries can return to a sustainable harvest. With the requirement of isolating variables between simulations, the model was developed with a set of parameters affecting the anthropogenic mortality, spawning, harvest total, seasonal harvest zones (commercial and recreational), and weekly agent movement decisions affected by the time of year. With the high level of biological detail incorporated into the model for hundreds of thousands of agents, the enviro-ABM was developed to reduce runtimes. The enviro-ABM splits the environment into spatially indexed cells allowing agents to move freely throughout adjacent cells in the environment while agents remain sorted by age and location through the runtime. On top of efficient access to agents, the calculations for each agent in a single cell can be calculated avoiding redundant computation in the same cells. The third benefit of using an enviro-ABM is more predictable computational requirements because the runtime is more dependent on environment size rather than population total. Although the model is applied specifically to the lake whitefish case study, it was built with the intent of application to other environment models so the enviro-ABM can be easily generalized. Applying the enviro-ABM to the case study successfully reduced average simulation runtimes from 18-24 hours to 3-5 hours. The reduced runtimes can drastically speed up the development process and assist environmental management decision making in a timely fashion.

Advisor: Dr. Daniel Gillis
Advisory Committee Member: Dr. Gary Grewal