This Is AuburnElectronic Theses and Dissertations

Modeling Complex Climate Change Effects on Fluctuating Populations of Fish Communities in the Northern Pacific Ocean

Date

2019-07-23

Author

Correia, Hannah Elizabeth

Type of Degree

PhD Dissertation

Department

Biological Sciences

Abstract

The main purpose of this study was to bring new mathematical and statistical methods to the ecological community and highlight novel application of these methods to practical scientific questions. Deepwater marine systems provide a challenge to understanding population dynamics, and commercial fisheries are particularly interested in understanding how valuable fish populations change with variable climate conditions. Several interdecadal climate modes and their effects on marine systems have been studied, and many of those climate modes are found to contribute to variability in the northern Pacific Ocean system. Studies on how climate and marine environment affect commercially important fish populations in this region often lean too heavily on the potential effect of climate modes. The importance of climate variability induced by unknown sources in marine systems must also be considered when understanding population changes in marine vertebrates. This study aimed to incorporate a variety of marine and atmospheric variables to model population-level changes in seven groundfish species. The response of these commercially valuable fishes to changes in climate and their marine environment are not well understood, as they live in deep waters ($>300$ m) making experimental studies on adults difficult. Such models involved considering population responses over space and time for multiple variables, increasing model complexity beyond the capacity of basic statistical methods. Several useful methods in statistics and mathematics allow for modeling of high-dimensional data without assumptions on population distributions. While previous fisheries research relied heavily on time series analysis, the past decade has seen a move to generalized additive models (GAMs) as a nonlinear method of modeling fish populations using smooth coefficient functions. The method provides the flexibility of fitting high-dimensional functions to allow incorporation of space and time in a non-additive way. The single-index model combines the interpretability of generalized linear models (GLMs) with GAMs, encasing a GLM in an unspecified link function. An extension of this is the varying-coefficient single-index model, which models several common covariates within a single-index model and uses varying smooth coefficients of the single-index model to quantify the relationship between other additive covariates and a response. Quantifying the relationship of individual species' responses to changes in climate using these methods provides a starting framework to consider how these groundfish species interact over time and space. A new mathematical tool called convergent cross mapping (CCM) can factor in multiple variables and map dynamic causal relationships extracted from time series data. So far the method has been applied to a two-species sardine-anchovy system off the coast of California, but there is the potential for this method to be expanded to measuring spatiotemporal effects involving more than two species. The goal of this study was to (1) quantify groundfish responses to climate and ecosystem fluctuations using multiple indicators of ocean variability through improved statistical methods described herein, and (2) detect causal factors and complex interactions involved in changes in populations numbers found in Alaskan groundfish species by applying CCM to this complex groundfish ecosystem. A combination of multiple sources of ecological, environmental, climatological, and geographic data were used to investigate potential causal factors and attempt to explain visible changes in groundfish populations. It is hypothesized that changes in fish populations will be best explained by multiple interacting variables that, when modeled correctly, will provide a more accurate system for monitoring and predicting the health of fish communities. Not only does this understanding have the potential to contribute to the development of more informed management practices for wild fish populations, it also provides greater insight for modeling changing ecosystems over time and creating more accurate prediction models describing complex systems. Over twenty years of fishery longline surveys collected by the NOAA were merged with corresponding climate data from ICOADS, COPEPOD, WODA13, and WOD13. Multivariate analysis methods were used to discover potentially significant clustering within the data, examine relationships and dependence among variables, and uncover potential correlations for further investigation. Robust and efficient nonparametric statistical procedures provided inference for small samples. Single-index models (SIMs) including a host of environmental variables were used to determine the most important environmental effects on groundfish catch rates. Single-index varying coefficient models (SIVCMs) provided framework for including latitude and longitude to environmental variables that vary spatially, allowing for estimation and prediction of groundfish catch over space and time without the `curse of dimensionality'. SIMs and SIVCMs also permitted exploration of the effects of trophic and habitat interactions by co-occurring groundfish on the species of interest. CCM was applied to the ecological system to search for signals that indicate potential causal effects of common environmental forces on the fish populations. All statistical analyses were performed using the free statistical computing software environment \textbf{\textsf{R}}. Results from the analyses and modeling of interactions were evaluated with the following main questions: (1) Do certain sources of variation influence fish population dynamics more heavily? (2) Which model(s) tested herein most accurately predict future ecological fluctuations? (3) Can CCM be applied to different complex systems efficiently and precisely? Application of causal analyses such as CCM to marine ecosystem data has the potential to provide explanations for changes in catch and improve prediction models with applications to broader ecological modeling that can inform wildlife policies and fisheries management. Considering that the first application of the CCM analytical protocol was on a complex sardine-anchovy system over multi-year scales \citep{Sugihara12}, the longline surveys on groundfish populations in the northern Pacific Ocean and related environmental data from sources such as ICOADS are ideal candidates for CCM and comparisons of successful method deployment. Based on the results of CCM and statistical analyses, I will create models to explain the dynamic changes in the fish populations and test model prediction capabilities and limitations utilizing cross-validation as well as data from future longline surveys. The models will include three main factors theorized to affect population dynamics: time, climate, and species interactions. CCM techniques will be refined to improve the detection accuracy of causal relationships. Further research into the connections behind the complex ecological interactions of the system may be explored using graph theoretic techniques. Future plans for modifications to the CCM technique will increase accuracy of quantifying causes and creating models to predict future population responses to potential ecosystem variations.