Abstract

This dissertation presents an investigation into the changes of evergreen forest cover within the Great Salt Lake Basin and its implications for regional water resources. Utilizing a Random Forest Classifier (RFC) model trained on Landsat imagery, this research developed a methodology to map and track changes in evergreen forest coverage over the period from 1984 to 2019. The initial phase of this study focused on the development of model trained within the Bear Lake sub-basin, demonstrating the feasibility of the method for analysis of coniferous coverage. The model incorporated summer and winter Normalized Difference Vegetation Index (NDVI) to enhance the distinction between evergreen and deciduous vegetation. Expanding the scope, the dissertation applied this trained model to all 21 sub-basins of the Great Salt Lake Basin, revealing a general trend of increasing evergreen forest cover through time. This increase was hypothesized to potentially impact regional water supply due to the higher sublimation rates associated with evergreen forests compared to the subcanopy. Preliminary estimations of sublimation losses were conducted for the Wasatch Range basins, where the model demonstrated higher reliability. A critical component of this research involved a thorough evaluation of the National Land Cover Database (NLCD), which was used as the primary source of labeled data for model training. Comparisons with concurrent high-resolution NAIP imagery in the Rush Tooele Valley sub-basin revealed that NLCD tends to liberally classify sparser vegetation types, such as pinyon-juniper woodlands, as evergreen forest. This discrepancy led to an underestimation of evergreen cover by the developed model in certain arid regions when compared to NLCD. However, visual inspection suggested that the model provided a more accurate representation of densely packed evergreen forests, particularly in mountainous regions, which were the primary focus for assessing impacts on water resources. The findings of this dissertation highlight the importance of understanding the specific classification criteria and potential limitations of existing land cover datasets like NLCD, especially when used for training remote sensing models for specific vegetation covers. The research demonstrates that cultivating training data based on specific features of interest can lead to more accurate and targeted land cover classification, ultimately improving the reliability of analyses concerning long-term environmental change and resource management.

Degree

PhD

College and Department

Ira A. Fulton College of Engineering; Civil and Environmental Engineering

Rights

https://lib.byu.edu/about/copyright/

Date Submitted

2025-04-21

Document Type

Dissertation

Handle

http://hdl.lib.byu.edu/1877/etd13589

Keywords

remote sensing, Random Forest classifier, Landsat, forest cover, land cover classification, National Land Cover Database (NLCD), Google Earth Engine (GEE), Great Salt Lake Basin, Water Resources

Language

english

Download

Included in

Engineering Commons

Share

COinS