Graduation Date
Spring 2026
Document Type
Thesis
Program
Master of Science degree with a major in Natural Resources: option Environmental Science and Management
Committee Chair Name
James Graham
Committee Chair Affiliation
Cal Poly Humboldt Faculty or Staff
Second Committee Member Name
Tawanda Gara
Second Committee Member Affiliation
Cal Poly Humboldt Faculty or Staff
Third Committee Member Name
Bill Trush
Third Committee Member Affiliation
Community Member or Outside Professional
Keywords
Resolution, Classification and regression tree, Random forest, Support vector machine, Land cover classification, Remote sensing
Subject Categories
Natural Resources
Abstract
Ecosystem services and functions are prone to water resource exploitation resulting in cascading effects that include decrease of biodiversity and loss of riparian vegetation, a keystone habitat in desert riparian ecosystems. Mono Lake is a prime example of overexploitation of water resources leading to legal action and eventually a legally mandated long-term monitoring and restoration project. Monitoring and restoration projects can benefit from techniques such as remote sensing and machine learning algorithms to generate accurate land cover classification maps for calculating land cover change over time. However, the spatial resolution of remote sensing imagery and the machine learning algorithms chosen can affect classification accuracy. In this research, we compiled and digitized historical land use/land cover maps along Lower Rush Creek in the Mono Basin to determine how the land use/land cover has changed over the last hundred years. Aerial and ground surveys were conducted to collect necessary data for building a current land use/land cover model. Data was resampled to 50-centimeter, 75-centimeter, 1-meter, 1.125-meter, 1.25-meter, 1.375-meter, 1.5-meter, 2-meter, and 5-meter resolutions. A basic stacked ensemble model was created using a minimum distance classifier and used as a covariate within classification and regression tree, random forest, and support vector machine (SVM) models. The 1.25-meter support SVM model performed the best, suggesting that very high-resolution imagery is not necessary within this study area. The 1.25-meter SVM model was then used to calculate percent composition of each class in comparison with the historical data. Future research could focus on streamlined flight plans, tuning of model hyperparameters, and inclusion of additional model covariates to further increase model accuracy. Overall, these findings can be used as protocol for land use/land cover annual surveying and mapping along Lower Rush Creek as required in the long-term restoration and monitoring project that is being conducted by Los Angeles Department of Water and Power. By comparing the composition of each class for all historical maps and the best SVM map, we determined there have been increases in water over time. In addition, woody riparian vegetation and herbaceous riparian vegetation have increased since 2004, but further increases are still needed to return to 1929 conditions. Changes in the methodology of reporting desert vegetation and bare ground led to drastically different values in 2025 compared to earlier years.
Citation Style
APA
Recommended Citation
Suoja, Jessica R., "Land cover classification using optimized imagery resolution and machine learning algorithms for a long-term monitoring and restoration project" (2026). Cal Poly Humboldt theses and projects. 2588.
https://digitalcommons.humboldt.edu/etd/2588