Show simple item record

dc.contributor.advisorHering, Amanda S.
dc.contributor.authorKazor, Karen E.
dc.date.accessioned2016-08-30T15:13:56Z
dc.date.accessioned2022-02-03T12:58:12Z
dc.date.available2017-08-29T04:18:44Z
dc.date.available2022-02-03T12:58:12Z
dc.date.issued2016
dc.identifierT 8113
dc.identifier.urihttps://hdl.handle.net/11124/170406
dc.descriptionIncludes bibliographical references.
dc.description2016 Summer.
dc.description.abstractIn this work, we evaluate statistical clustering methods with the goal of gaining insight into complex environmental processes. To identify predominant wind patterns, i.e. wind regimes, we assess various model-based clustering methods when applied to wind data simulated from a realistic two-regime model. A Gaussian mixture model (GMM) and two forms of Markov-Switching models are applied to wind data in the Pacific Northwest and are demonstrated to identify different features in the data. We then propose a short-term wind speed forecasting model that leverages the GMM regimes to better incorporate off-site information into wind speed forecasts. When compared to state-of-the-art reference models, the proposed model is demonstrated to significantly improve forecast accuracy. To improve the monitoring of a wastewater treatment process, we evaluate methods that distinguish between observations generated under normal and abnormal conditions. To capture salient relationships among 28 nonlinear process variables, monitoring methods based on principal component analysis (PCA), kernel PCA (KPCA), and locally linear embedding (LLE) are compared. Extensions to these methods that account for autocorrelation and nonstationarity in process data are evaluated along with a nonparametric thresholding approach for identifying faults. When applied to data collected from a decentralized wastewater treatment system, adaptive-dynamic versions of PCA, KPCA, and LLE all flag a strong system fault, and the use of nonparametric thresholds considerably reduces the number of false alarms. To gain insight into groundwater behavior, we propose a method for grouping observations from a spatial random process based on similarities in the relationship of response and predictor variables. The proposed Markov random field finite mixture of regressions (MRF-FMR) model extends finite mixture of regressions (FMR) models to the spatial domain. We propose the MRF-FMRlasso algorithm for fitting an MRF-FMR model and provide a method for simulating from this model. MRF-FMRlasso is evaluated in comparison to alternate methods for fitting FMR models, and we find that it accurately selects and estimates the coefficients within each component regression while also capturing the spatial structure among component assignments. When applied to groundwater data from the Missouri river basin, MRF-FMRlasso identifies physically interpretable behaviors that correspond to the geography of the region.
dc.format.mediumborn digital
dc.format.mediumdoctoral dissertations
dc.languageEnglish
dc.language.isoeng
dc.publisherColorado School of Mines. Arthur Lakes Library
dc.relation.ispartof2016 - Mines Theses & Dissertations
dc.rightsCopyright of the original work is retained by the author.
dc.subjectmultivariate data
dc.subjectspatial processes
dc.subjectclustering methods
dc.subjecttime series
dc.subjectprocess monitoring
dc.titleIdentifying clusters in multivariate temporal and spatial data with application to environmental processes
dc.typeText
dc.contributor.committeememberNavidi, William Cyrus
dc.contributor.committeememberTenorio, Luis
dc.contributor.committeememberCath, Tzahi Y.
dcterms.embargo.terms2017-08-29
dcterms.embargo.expires2017-08-29
thesis.degree.nameDoctor of Philosophy (Ph.D.)
thesis.degree.levelDoctoral
thesis.degree.disciplineApplied Mathematics and Statistics
thesis.degree.grantorColorado School of Mines
dc.rights.accessEmbargo Expires: 08/29/2017


Files in this item

Thumbnail
Name:
Kazor_mines_0052E_11087.pdf
Size:
21.82Mb
Format:
PDF

This item appears in the following Collection(s)

Show simple item record