Temporal and Density-dependent Variability as Source for Uncertainty in Fish Population Dynamics
Year: 2018
Type: dissertation
Abstract: Considering the uncertainties about model parameters and structure, observations, and the resulting predictions is crucial when managing natural resources. This thesis is compiled of four research articles that deal with uncertainties and change in marine fish populations. In all the articles, the Bayesian statistical modelling framework is adopted to account for various types of uncertainty. When modelling complex systems, one must recognize that all models are only approximations of reality and are based on the limited understanding about the system at that moment and on previously observed behaviour. Even if a model explains the recent behaviour and predicts how the system will respond in the near future, there is always a possibility that the system changes thus rendering the old model useless. Approaches to respond to sudden changes in the system’s dynamics are vital for successful resource management. Fish populations' renewal ability is mainly determined by their reproductive success. Thus, it is of utmost importance to be able to understand and model the reproductive dynamics of marine fish populations. In this thesis, fish reproduction is studied using models that link together the number of new individuals entering the adult population (recruits) and the size of the reproducing component of the population (spawning stock), namely, the stock-recruitment relationship. The focus in this work is on temporal and density-dependent variability in the stock-recruitment relationship. The temporal variability is studied using Bayesian change-point detection methods applied to detecting changes in the per capita reproductive output of four Atlantic cod populations by analysing recruit-per-spawner ratios, and in the parameters of the stock-recruitment relationship of four fish species in the southern Gulf of St. Lawrence. In this work, novel Bayesian methods are utilized to improve the handling of uncertainties about the parameters, the timings of the changes, and in short term predictions. This thesis presents computational methods for analysing temporal variability in linear-Gaussian problems where analytical solutions are available, and extend the methodology to analytically intractable non-linear and non-Gaussian problems by utilizing numerical approximate solutions. Traditionally, compensatory population dynamics have been assumed in marine fish populations. The validity of the assumption of compensatory stock-recruitment relationship is examined by studying the density-dependence of nine Atlantic herring populations at low population sizes using models that better capture the uncertainty related to low-abundance dynamics and allow for depensatory dynamics caused by Allee effects. The Allee effect has been largely ignored in marine fish population models. Here methodology for detecting Allee effects is developed, and it is concluded that Allee effects might be more common than previously thought. In Bayesian models, expert elicitation is used to construct prior distributions for the model parameters. When observational data is scare or not available, informative prior distributions are crucial for conducting inference. However, experts might not be calibrated and can have significant biases in their assessments about key parameters of the models. Statistical models are developed for the use of cacalibration data to infer and to correct for experts’ biases in their assessments.
Source: Työväentutkimus Vuosikirja
Author: Tommi Perälä
Language: English
Subfield: Aquatic Science
Domain: Life Sciences
SDG: Life below water
Open Access status: green
OpenAlex ID: W2794898405