From habitat to management: a simulation framework for improving statistical methods in fisheries science

Abstract

Monte Carlo simulation consists in computer experiments that involve creating data by pseudo-random sampling and has shown to be a powerful tool for studying the performance of statistical methods. In this thesis Monte Carlo simulation was used to improve statistical methodology related to three different fields of fisheries science: 1) Species distribution models (SDM) field, where focusing on regression-based models, we proposed using shape-constrained generalised additive models (SC-GAMs) to build SDMs in agreement with the ecological niche theory imposing concavity constraints in the linear predictor scale and testing their performance trough Monte Carlos simulation, 2) stock assessment models field, where uncertainty estimation methods for statistical catch-at-age models with non-parametric effects on fishing mortality were compared through simulation in addition to the comparison of two available stock assessment models to an ad-hoc Bayesian approach, and 3) management advice field, where a full-feedback management strategy evaluation (MSE) is developed for the sardine in the Bay of Biscay, incorporating the official Stoch Synthesis assessment model within the Monte Carlo simulation, and introducing gradually different sources of uncertainty such as process, parameter and observation error in order to study their effect in management advice. Monte Carlo simulation was an adequate tool to accomplish the objectives of this thesis that definitely could not have been achieved using only available real data or analytical solutions.