Bivariate Gaussian random fields : models, simulation, and inference

Moreva, Olga

Vorschau

PDF
DissertationOlgaMoreva.pdf - Veröffentlichte Version
Download (5MB)

URL:	https://ub-madoc.bib.uni-mannheim.de/45380
URN:	urn:nbn:de:bsz:180-madoc-453800
Dokumenttyp:	Dissertation
Erscheinungsjahr:	2018
Ort der Veröffentlichung:	Mannheim
Hochschule:	Universität Mannheim
Gutachter:	Schlather, Martin
Datum der mündl. Prüfung:	22 Juni 2018
Sprache der Veröffentlichung:	Englisch
Einrichtung:	Fakultät für Wirtschaftsinformatik und Wirtschaftsmathematik > Applied Stochastics (Schlather 2012-)
Fachgebiet:	510 Mathematik
Normierte Schlagwörter (SWD):	Geostatistik , Kovarianzfunktion , Positiv-definite Funktion
Freie Schlagwörter (Englisch):	circulant embedding , cokriging , compactly supported covariance function , cut-off embedding , multivariate covariance function , multivariate Gaussian random field , multivariate geostatistics
Abstract:	Spatial data with several components, such as observations of air temperature and pressure in a certain geographical region or the content of two metals in a geological deposit, require models which can capture the spatial dependence structure of individual components and the relationship between them. In a wealth of applications, multivariate Gaussian random fields are sensible models for multivariate spatial data and their second order structure specifies the marginal correlations and the cross-correlations between the components. In this thesis we focus on covariance models and simulation techniques for bivariate fields. In Chapter 2 we summarize some definitions and facts from univariate and multivariate Geostatistics which are essential for the subsequent chapters. Chapter 3 introduces two novel bivariate parametric covariance models, the powered exponential (or stable) covariance model and the generalized Cauchy covariance model. Both models allow for flexible smoothness, variance, scale, and cross-correlation parameters. The smoothness parameter is in (0, 1]. Additionally, the bivariate generalized Cauchy model allows for distinct long range parameters. The results are based on general sufficient conditions for the positive definiteness of 2×2-matrix valued functions. These conditions are easy to check, since they require only computing the derivatives of a bivariate covariance function and calculating an infimum of a function of one variable. We also show that the univariate spherical model can be generalized to the bivariate case with spherical marginal and cross-covariance functions only in a trivial way. Circulant embedding is a powerful algorithm for fast simulation of stationary Gaussian random fields on a rectangular grid in R^n , which works perfectly for compactly supported covariance functions. Cut-off circulant embedding techniques have been developed for univariate random fields for dimensions up to R^3 and rely on the modification of a covariance function outside the simulation window, such that the modified covariance function is compactly supported. In Chapter 4 we propose extensions of the cut-off approach for bivariate Gaussian random fields. In particular, we provide a method for simulating bivariate fields with a bivariate powered exponential covariance model and the full bivariate Matérn covariance model for certain sets of parameters. On the way we extend the cut-off circulant embedding method even for univariate models. In Chapter 5 we illustrate the use of the bivariate powered exponential model for a data example.