In the field of spatial statistics, it is often desirable to generate a smooth surface for a region over which only noisy observations of the surface are available at some locations, or even across time. Kriging and kernel estimations are two of the most popular methods. However, these two methods become problematic when the domain is not regular, such as when it is rectangular or convex. Bivariate B-splines developed by mathematicians provide a useful nonparametric tool in bivariate surface modeling. They inherit several appealing properties of univariate B-splines and are applicable in various modeling problems. More importantly, bivariate B-splines have advantages over kriging and kernel estimation when dealing with complicated domains. The purpose of this dissertation is to develop a nonparametric surface fitting method by using bivariate B-splines that can handle complex spatial domains. The dissertation consists of four parts. The first part of this dissertation explains the challenges of smoothing over complicated domains and reviews existing methods. The second part introduces bivariate B-splines and explains its properties and implementation techniques. The third and fourth parts discuss application of the bivariate B-splines in two nonparametric spatial surface fitting problems. In particular, the third part develops a penalized B-splines method to reconstruct a smooth surface from noisy observations. A numerical algorithm is derived, implemented, and applied to simulated and real data. The fourth part develops a reduced rank mixed-effects model for functional principal components analysis of sparsely observed spatial data. A numerical algorithm is used to implement the method and tested on simulated and real data.

Scattered data collected at sample points may be used to determine simple functions to best fit the data. An ideal choice for these simple functions is bivariate splines. Triangulation of the sample points creates partitions over which the bivariate splines may be defined. But the optimality of the approximation is dependent on the choice of triangulation. An algorithm, referred to as an Edge Swapping Algorithm, has been developed to transform an arbitrary triangulation of the sample points into an optimal triangulation for representation of the scattered data. A Matlab package has been completed that implements this algorithm for any triangulation on a given set of sample points.

This paper is concerned with a study of approximation order and construction of locally supported elements for the space S41 (Δ) of Cl quartic pp (piecewise polynomial) functions on a triangulation Δ of a connected polygonal domain Ω in R2. It is well known that, when Δ is a three-directional mesh Δ(1), the order of approximation of S41(Δ(1)) is only 4, not 5. Though a local Clough-Tocher refinement procedure of an arbitrary triangulation A yields the optimal (fifth) order of approximation from the space S41(Δ) (see [1]), it needs more data points in addition to the vertex set of the triangulation A. In this paper, we will introduce a particular mixed three-directional mesh Δ(3)) and construct so-called mixed three-directional elements. We prove that the space S41(Δ(3)) achieves its optimal-order of approximation by constructing an interpolation scheme using mixed three-directional elements.

In the representation of scattered data by smooth pp (:= piecewise polynomial) functions, perhaps the most important problem is to find an optimal triangulation of the given sample sites (called vertices). Of course, the notion of optimality depends on the desirable properties in the approximation or modeling problems. In this paper, we are concerned with optimal approximation order with respect to the given order r of smoothness and degree k of the polynomial pieces of the smooth pp functions. We will only consider C1 pp approximation with r = 1 and k = 4. The main result in this paper is an efficient method for triangulating any finitely many arbitrarily scattered sample sites, such that these sample sites are the only vertices of the triangulation, and that for any discrete data given at these sample sites, there is a C1 piecewise quartic polynomial on this triangulation that interpolates the given data with the fifth order of approximation.