Theses and Dissertations

ORCID

https://orcid.org/0009-0002-3581-3944

Advisor

Miranda, Leandro E.

Committee Member

Dunn, Corey G.

Committee Member

Boyles, Ryan P.

Committee Member

Hunt, Kevin M.

Committee Member

Evans, Kristine O.

Date of Degree

5-16-2025

Original embargo terms

Immediate Worldwide Access

Document Type

Dissertation - Open Access

Major

Forest Resources (Wildlife, Fisheries, and Aquaculture)

Degree Name

Doctor of Philosophy (Ph.D.)

College

College of Forest Resources

Department

College of Forest Resources

Abstract

Anthropogenic climate change has been associated with reservoir aging and reduced ecosystem function. Communities throughout the world depend on reservoirs for services such as hydropower, drinking water, flood mitigation, and recreational opportunities. Though reservoirs were typically constructed to meet human needs, they are now also acknowledged as supporting biodiverse aquatic communities. Reservoir managers are expected to maintain services and biodiversity demanded by diverse users despite high uncertainty surrounding how climate will influence reservoirs through time. Additionally, data describing reservoir function at broad spatial extents are often either unavailable, incompatible between agencies, or prohibitively resource-intensive to obtain. Expert opinion surveys are one alternative to in situ data, leveraging the experience and knowledge of those working on reservoir systems to identify broad patterns of impairment. I related information on reservoir impairment from expert opinion surveys covering 1090 reservoirs distributed across the conterminous United States with climate data to inform decision makers on how climate affects reservoir function. To integrate a complex data structure, I constructed a series of models using logistic regression, random forest, support vector machine, and multilayer perceptron neural networks. Models were compared by predictive and discriminatory ability to identify which could most effectively classify reservoir impairment. Support vector machine emerged as the best performing model by accuracy, Cohen’s Kappa, and area under the receiver operating characteristic curve. This model was used to project impairment scores for four time intervals and three future climate scenarios. Twelve spatially distinct clusters of reservoirs were identified by similar climate conditions, and future impairment was projected for the conterminous United States, each climatic cluster, and individual reservoirs. These results are compiled in a web application developed in R-Shiny to inform decision makers about future climate conditions relevant to their needs. My study demonstrates the information value of expert opinion surveys in contexts where empirical data are unavailable or decision makers are obliged to make a decision before such data can be obtained. I also demonstrate the potential of machine learning techniques on diverse data structures, identify metrics for evaluating and comparing models, and provide general recommendations for applying such models in ecological research.

Sponsorship (Optional)

Funding was provided by the U.S. Fish and Wildlife Service through the Reservoir Fish Habitat Partnership and the Multistate Conservation Grant Program. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Download

Share

COinS