Bağımsız Bileşenler Analizi ile Çok Değişkenli Jeoistatistiksel Kestirim

Abstract

This work is prepared with the support of TÜBİTAK Project, named ‘’Çokdeğişkenli Jeoistatistiksel Kestirimde Dikleştirilmiş Bileşenli Yeni Yöntemlerin Geliştirilmesi’’ with the code number 111M218. Geostatistical estimation of multvariate data is much more difficult than univariate estimations. There are two main reasons for this difficulty: Modeling spatial cross-correlations between variables and using model parameters in geostatistical estimations. If there is no spatial cross-correlation between variables, multivariate estimation is reduced to simple univariate estimation. The main purpose of this thesis is to develope methods that reduce multivariate problems to univariate ones. One approach is to transform spatially cross-correlated variables to orthogonal factors which do not show spatial cross-correlations with each other. In this study, two new methods were developed to generate orthogonal factors: (1) Independent component analysis and (2) minimum spatial cross-correlation method. Components derived from each method are estimated at unknown locations and estimated values are back-transformed into original space. Perfrmance of vi traditional cokriging estimation method is compared to Indedepedent Component Kriging (ICK) and Minimum Spatial Cross-correlation Kriging (MSCK). ICK and MSCK methods were also used for determination of exploitable blocks of an andesit quarry. The study results show that the ICK and the MSCK methods are good alternatives to traditional cokriging estimation method.