TRACKING WETLAND DYNAMICS: REMOTE SENSING AND DEEP LEARNING FOR COASTAL WETLAND CLASSIFICATION IN THE HAWAIIAN ISLANDS
Date
2024
Authors
Contributor
Advisor
Department
Instructor
Depositor
Speaker
Researcher
Consultant
Interviewer
Narrator
Transcriber
Annotator
Journal Title
Journal ISSN
Volume Title
Publisher
Volume
Number/Issue
Starting Page
Ending Page
Alternative Title
Abstract
Wetlands are critical ecosystems in our environment. They cover only 6% of the Earth’s surface but about 40% of all plants and animal species live in wetlands for some portion of their life (United Nations 2024). The objective of this research study was to explore the relationship between machine learning complexity and the duration of data variations on the accuracy of wetland identification for six wetland classes in the Hawaiian Islands. The research aimed to compare the models’ classification maps across three distinct durations of imagery data. This exploration used remote sensing imagery and machine learning to classify diverse types of wetlands across three regions of the island of O’ahu. A U-Net convolution neural network was used as the architecture for the model to conduct three experiments: one year of data, two years of data, and three years of data. The only variable to change in the model is the input imagery data. Each experiment used Python programming language.This study advanced the understanding of how temporal changes affect machine learning models used in environmental studies. By determining the optimal duration of imagery that enhances model accuracy, this study contributes to effective monitoring of coastal wetlands in Hawai’i. Enhancing conservation strategies would allow for more targeted, timely, and focused management of wetland ecosystems. Furthermore, the findings could provide insights into the dynamic changes occurring within wetland ecosystems over time. This research fills a gap in previous studies which have not systematically compared how dataset duration influences machine learning models’ ability to accurately delineate coastal wetlands in Hawai’i.
The results of the study showed that the overall accuracy improved as more data was added. Experiment one had an overall accuracy of 56.57%. The overall accuracy increased to 60.46% in experiment two. Experiment three had the highest overall accuracy of 61.27%. These results indicate that adding more years of imagery provides more context into the deep learning modeling to learn more details about the characteristics of wetlands and their change over time. Using the F1 score to measure each class finds pond, forested wetland, and “other” returned the highest improvement from one year of imagery to three years of imagery. The insights gained from this study may provide a foundation for further research to focus on refining the model to better distinguish difficult classes. Future research should focus on improving the overall accuracy of the wetland classes for the Hawaiian Islands.
Description
Keywords
Geography, Remote sensing, Artificial intelligence, Artificial Intelligence, Deep Learning, Image Classification, Machine Learning, Remote Sensing, Wetlands
Citation
Extent
113 pages
Format
Geographic Location
Time Period
Related To
Related To (URI)
Table of Contents
Rights
All UHM dissertations and theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission from the copyright owner.
Rights Holder
Local Contexts
Collections
Email libraryada-l@lists.hawaii.edu if you need this content in ADA-compliant format.