Skip to Main Content
 

Global Search Box

 
 
 
 

Files

ETD Abstract Container

Abstract Header

Multi-Resolution Statistical Modeling in Space and Time With Application to Remote Sensing of the Environment

Johannesson, Gardar

Abstract Details

2003, Doctor of Philosophy, Ohio State University, Statistics.
Analyzing massive spatial and space-time environmental datasets can be demanding. A central example used in this dissertation is the analysis of Total Column Ozone (TCO) data remotely sensed from a satellite. There are a number of issues that need to be resolved. These include computational issues, the challenge of modeling and predicting nonstationary spatial processes, and developing realistic temporal dynamics for a space-time processes. In this dissertation, we look at the problem of (1) representing and fitting a large-scale spatial trend surface to massive, global datasets; (2) variance-covariance modeling and estimation for multi-resolution spatial models (MRSMs); and (3) developing a dynamic MRSM with special emphasis on the development of the temporal dynamics. One can argue that the large-scale spatial features of massive, fine-resolution spatial data can be obtained from the coarser-resolution aspects of the data. Consequently, we propose a sequential-aggregation procedure that yields more manageable data at coarser resolutions and use these for spatial trend surface fitting. Assuming that the trend surface belongs to the class of linear combinations of smooth basis functions, we investigate a new trend-surface-fitting approach based on penalized weighted-least-squares regression, where the penalty term is data-adaptive. Extensive comparisons are made to standard fitting procedures based on a day's worth of TCO data. Multi-resolution spatial models (MRSMs) have been shown to be successful at modeling massive spatial datasets. The MRSM models the spatial dependence indirectly through a coarse-to-fine-resolution process model, where it is necessary to specify the variance parameters. We propose a spatially smooth model for the variance parameters, outline computationally fast, resolution-specific-likelihood-based methods for parameter estimation, and apply the statistical methodology to a day's worth of TCO data. The MRSM is a spatial-only model. An extension of the MRSM is given that incorporates temporal dynamics at the coarsest spatial resolution of interest, yielding a dynamic (space-time) MRSM. A physics-based flow model is proposed for the coarse-resolution dynamics, which maintains the computational advantage of the MRSM. An application to month's worth of TCO data is given.
Noel Cressie (Advisor)
238 p.

Recommended Citations

Citations

  • Johannesson, G. (2003). Multi-Resolution Statistical Modeling in Space and Time With Application to Remote Sensing of the Environment [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1051282014

    APA Style (7th edition)

  • Johannesson, Gardar. Multi-Resolution Statistical Modeling in Space and Time With Application to Remote Sensing of the Environment. 2003. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1051282014.

    MLA Style (8th edition)

  • Johannesson, Gardar. "Multi-Resolution Statistical Modeling in Space and Time With Application to Remote Sensing of the Environment." Doctoral dissertation, Ohio State University, 2003. http://rave.ohiolink.edu/etdc/view?acc_num=osu1051282014

    Chicago Manual of Style (17th edition)