Integrating information and ecological practice: Two applications

Thumbnail Image
Date
2022-05
Authors
McCombs, Audrey Lamson
Major Professor
Advisor
Dixon, Philip
Caragea, Petrutza
Kadelka, Claus
Danielson, Brent
Rogers, Haldre
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Altmetrics
Authors
Research Projects
Organizational Units
Journal Issue
Is Version Of
Versions
Series
Department
Statistics
Abstract
This dissertation combines two separate topics that both rest on a conceptual foundation involving the integration of information (sensu Claude Shannon) and ecological theory and practice. The first part develops an integrated population model combining two data sets measuring abundance of Artemesia tridentata in the Yellowstone Northern Range. Three essential features of the study system are captured in the model: 1) repeated measures of abundance from multiple sample plots per frame, 2) a non-linear spatial trend, and 3) by-data-set differences in spatial extent, number of observations, and the amount of information contained in a single observation. We describe the model in detail and report results related to the estimation of species abundance for A. tridentata and the precision of those estimates. We show that the model successfully integrates the two data sets and estimates parameters for Beta distributions associated with each sampling frame. We assess model fit through posterior predictive checks including posterior p-values, posterior prediction intervals, and two methods to assess the fit of cover class data, one of which is novel for this study. We also investigate the statistical properties (bias and coverage) of the abundance estimates using a simulation study. We compare results of the model that integrates the two data sets with results when the data sets are modeled separately, with respect to the precision of the abundance estimates and prediction intervals. We find that model fit is generally good, although observation-level predictions of cover class are only weakly positively correlated with observed values, while frame-level measures of similarity are overall high. Finally, the model identifies, but is unable to resolve, two discrepancies between the two data sets that should be verified in the field and/or through a separate study designed to quantify sources of observation error. Because both data sources are widely used as bases for scientific research and management decisions, resolving the discrepancies between the two data sets should be prioritized. The second part of this work concerns parametric models for ecological networks. We begin by observing that the current non-parametric approach to inference on ecological networks is driven by a perceived lack of parametric models that represent random variability in the absence of ecological mechanisms. Simulating from non-parametric null models is problematic, however, because we are never sure that the null model captures the range of patterns specified by the null hypothesis, nor are we sure that the associated randomization algorithm adequately approximates the variation predicted by the null. We describe four parametric models that can be used to test scientific hypotheses about ecological networks in a statistically rigorous way because they quantify the uncertainty in the parameter estimates. We summarize the theoretical foundations of each model, discuss options for software implementation, and review the few studies in the ecological literature that make use of these models. We end with a discussion of the ecological mechanisms that generate network data and their relation to the study of ecology as a whole.
Comments
Description
Keywords
Citation
Source
Subject Categories
Copyright