Bayesian Methods for Great Lakes Fisheries Risk Assessment and Management
I applied Bayesian inference and techniques to analyze risk in two case studies of fisheries in the Great Lakes basin. Specifically, I developed and applied graphical Bayesian inference (GBI) to assess fishery status, its uncertainty and the risk of overexploitation (Chapter 1), and a dynamic Bayesian decision network (DBDN) to estimate the value of reducing uncertainty (Chapter 2) about population dynamics of Lake Whitefish in Lake Nipigon, Ontario. I also used a Bayesian approach to assess the risk of overexploitation due to the harvesting of Lake Erie Walleye before, versus after, the spring spawning period (Chapter 3). The state-space surplus production-based assessment of the Lake Nipigon Lake Whitefish fishery showed that there is considerable uncertainty about the carrying capacity, current biomass and risk of overexploitation. Using the most conservative and parsimonious model/data scenario, I showed that the fishable biomass in 2013 (B2013) was well above the biomass at maximum sustainable yield (Bmsy) and that the risk of overexploitation at recent harvests levels was very low. Based the GBI approach, quota adjustments lagged behind changes in biomass, potentially increasing risk of overexploitation in the early 2000s. In Chapter 2, I developed and applied a dynamic Bayesian decision network (DBDN)-based approach to estimate the value of new information (VoI) to reduced uncertainty about stock status through enhanced stock assessment and/or new research. Based on a 20 year simulation, the expected value of the cumulative total fishery operating profit, given perfect information about the population dynamics, was $3.3M, and maximum expected value (MEV) without perfect information, $2.43M, yielding an expected value of perfect information (EVPI) of $0.87M, or 36% of the MEV of the cumulative operating profit. Thus, there was considerable value in reducing uncertainty about the true state of the population dynamics of Lake Nipigon Lake Whitefish. In Chapter 3 I assessed the risk of overexploitation associated with alternative seasonal, in this case pre-spawn, harvest policies for the industrial scale freshwater commercial Walleye gill net fishery of Lake Erie. Bayesian methods were used to analyse uncertainty and estimate parameters in the stock-recruit relationship. I compared the risk associated with combinations of fishing mortality rates (F) and spring allocations as a percentage of the total allowable catch (TAC). At higher rates of constant fishing mortality (F>0.3), there was little difference in the risk of overexploitation for initial allocations ranging from 5% to 50% of the total allowable catch of the TAC. In Chapter 4, I discuss how (1) the utility of Bayesian approaches has the potential to go beyond the realm of statistics, parameter estimation and model checking; (2) how, though under-appreciated in this sense, Bayesian approaches are commensurate with the hermeneutics of the social sciences and humanities, and thus offer a pragmatic means to overcome the epistemic and cultural barriers between the natural and social sciences; and (3) how Bayesian methods can foster interdisciplinarity and a pragmatic form of fisheries risk governance better equipped to surmounting the barriers between fisheries science, policy and politics.