Classification of fish stock-recruitment relationships in different environmental regimes by fuzzy logic with bootstrap re-sampling approach

The analysis of stock-recruitment (SR) relationships is a basic step in developing and evaluating fishery policies, such as establishing optimal escapement goals for salmon or optimal size of spawning stocks at maximum sustainable yield (MSY). Traditional SR analyses assume that there is a functional relationship between the size of the stock spawning biomass and the biomass of fish that recruit in the future. Numerous models have been discussed for this functional relationship. A comprehensive summary can be found from Ricker (1975), Hilborn and Walters (1992) and Quinn and Deriso (1999). In the search for better tools for fish stock assessment, there has recently been a growing interest in the use of machine learning models (such as neural network models, fuzzy logic models and genetic algorithms) for research and management of natural resources (Lek et al. 1995; Mackinson et al. 1999 and Tang et al. 2000). It has been demonstrated that these models offer substantial advantages over traditional SR methods in model fit and forecast (Saila 1996; Chen and Ware 1999 and Chen et. al. 2000). In this paper, the utility of fuzzy logic model with a hybrid global learning algorithm is explored to classify the SR relationships under different regimes for environmental and fishery management interventions. A bootstrap re-sampling scheme is also proposed to address the lack of uncertainty estimation in the machine-learning methods. The scheme produces a sampling probability distribution for the SR parameters related to fishery management policies so that the associated uncertainty measures (such as, variance, standard error, or confidence interval) can be obtained. Two SR applications: 1) southeast Alaska (SEAK), USA, pink salmon, and 2) west coast Vancouver Island (WCVI), BC, Canada, herring, are examined to demonstrate the advantages of this new model to the traditional approaches. In both examples, the annual mean sea-surface temperature (SST) is incorporated as an environmental intervention.