CHARACTERIZING PATTERNS AND IDENTIFYING PREDICTION MODELS TO SIMULATE SOIL MOISTURE

Simple item page
dc.contributor.advisorTsang, Yinphan
dc.contributor.authorHeu, Cherryle Noelani
dc.contributor.departmentNatural Resources and Environmental Management
dc.date.accessioned2025-02-20T22:36:49Z
dc.date.available2025-02-20T22:36:49Z
dc.date.issued2024
dc.description.degreeM.S.
dc.identifier.urihttps://hdl.handle.net/10125/110209
dc.subjectNatural resource management
dc.subjectMachine Learning
dc.subjectSoil Moisture
dc.titleCHARACTERIZING PATTERNS AND IDENTIFYING PREDICTION MODELS TO SIMULATE SOIL MOISTURE
dc.typeThesis
dcterms.abstractSoil moisture is known to be a critical component in our Earth’s system and serves a pivotal role in evaluating water balance. Its dynamic behavior is essential for understanding hydrological processes and water budgets. Despite its significance, there is still much to learn about the key drivers of soil moisture dynamics and the variables that influence these patterns. This study analyzed hourly soil moisture data from 16 sites in Canterbury, New Zealand, to identify wetting and drying events over a period of three to five years. To characterize the dynamic behavior of soil moisture during each event, the study proposed five distinct sets of soil moisture signatures. For each signature, statistical analyses such as Principal Component Analysis (PCA) and Pearson Correlation were conducted to find what antecedent or climate variables could statistically explain a signature’s value or variability. After extensive examination, no significant drivers were found. Additionally, this study used a machine learning technique, Long Short-term Memory (LSTM) to generate 12-hour to 72-hour soil moisture predictions for depths 10, 40, and 80cm. When comparing its performance to that of Autoregressive Integrated Moving Average (ARIMA), ARIMA was found to perform better overall, with the advantages of easy hyperparameter tuning. However, the LSTM models demonstrated a unique ability to predict soil moisture at sites not included in the model’s training dataset and were most successful in predicting soil moisture when the model was trained from multiple sites’ datasets. Both LSTM and ARIMA faced limitations, particularly a tendency for predictions to lag behind observed values. The simplicity and adaptability of ARIMA make it well-suited for site-specific applications, while LSTM's strong performance with cumulative data indicates its potential for broader applications, such as regions with limited historical soil moisture data. Further refinement and development of LSTM models could enhance their applicability and accuracy, providing greater potential for more robust soil moisture predictions.
dcterms.extent76 pages
dcterms.languageen
dcterms.publisherUniversity of Hawai'i at Manoa
dcterms.rightsAll 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.
dcterms.typeText
local.identifier.alturihttp://dissertations.umi.com/hawii:12388

Files

Original bundle

Now showing 1 - 1 of 1
Thumbnail Image
Name:
Heu_hawii_0085O_12388.pdf
Size:
1.93 MB
Format:
Adobe Portable Document Format
Download

Collections

M.S. - Natural Resources and Environmental Management