Theses and Dissertations
Date of Award
7-1-2026
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Mathematics
First Advisor
Hansapani Rodrigo
Second Advisor
Rafael M. Almeida
Third Advisor
Nathan Oliveira Barros
Abstract
Human-made aquatic ecosystems such as reservoirs and excavated ponds used for hydropower and aquaculture are significant and growing sources of greenhouse gas (GHG) emissions, including CO2, CH4, and N2O. These emissions are temporally variable, and field measurements are costly and time-consuming. The data for these systems is sparse and is also a main challenge for traditional statistical models, whose performance is highly sensitive to data quantity and quality. This research proposes a generative artificial intelligent (AI) framework based on conditional variational autoencoders (CVAEs) for predicting GHG fluxes under sparse data conditions. The framework is supported by formal theoretical error bounds that provide interpretable guarantees on reconstruction uncertainty under sparse sampling. In addition, we leverage globally available geospatial and landscape-level high dimensional data from the Hydro ATLAS database to develop random forest (RF) models for predicting CH4 and CO2 fluxes from lakes and reservoirs across twelve spatial basin resolutions. This work establishes a pathway for applying generative modeling to scientific domains with limited data, which could be useful for environmental monitoring, climate modeling, and sustainable aquaculture management.
Recommended Citation
Mejia Garcia, M. (2026). Machine Learning Approaches to Model Greenhouse Gas Emissions and Other Effects of Human-Made Aquatic Ecosystems [Doctoral dissertation, The University of Texas Rio Grande Valley]. ScholarWorks @ UTRGV. https://scholarworks.utrgv.edu/etd/2008

Comments
Copyright 2026 Maria Camila Mejia Garcia. All Rights Reserved. https://proquest.com/docview/3409016490