Stud. Geophys. Geod., 56 (2012), 977-992, DOI: 10.1007/s11200-011-9024-6 977 © 2012 Inst. Geophys. AS CR, Prague Multivariate statistical analysis of deformation tensors: independent vs. correlated tensor observations KHOSRO MOGHTASED-AZAR 1 , ERIK W. GRAFAREND 2 , FAROKH TAVAKOLI 3 AND HAMID R. NANKALI 3 1 Department of Surveying Engineering, Faculty of Civil Engineering, University of Tabriz, 29 Bahman Boulevard, P.O. Box 51666-16471, Tabriz, Iran (moghtased@tabrizu.ac.ir) 2 Institute of Geodesy, Geschwister-Scholl-Str. 24D, P.O. Box D-70174, Stuttgart, Germany (grafarend@gis.uni-stuttgart.de) 3 National Cartographic Centre (NCC) of Iran, Meraj Ave., Azadi Sq., P.O. Box 13185 1684, Tehran, Iran (farokh.tavakoli@ncc.org.ir, nankali@ncc.org.ir) Received: March 28, 2011; Revised: September 16, 2011; Accepted: October 14, 2011 ABSTRACT In the presence of errors in measuring a random displacement field (under the normal distribution assumption of displacement field),stochastic behaviors of principal components of deformation tensors (strain tensor and tensor of change of curvature (TCC)), based on the intrinsic assumption of geometrical modeling of surface deformation analysis, are discussed. We divided the contents into two parts: In the first, we considered independent random vectors of repeated tensor measurements. In the second step, we considered correlations among repeated measurements. Then, covariance components between tensor elements by Helmert estimator, based on prior information of variance components, are estimated. As a case study, both assumptions are applied to the estimation of principal components of deformation rate tensor observations in Zagros region (Western Iran). The results of numerical analysis showed that greatest shortening is accommodated in oblique orientation (NS) with respect to the Main Recent Fault (MRF), northwest part of North Zagros, Central Iran block and MRF, respectively. Most of the extensions occurred in the east part of the belt. The pattern of eigenspace components of TCC shows highest positive values across the NW region, nearly in orthogonal direction to the MRF and Main Zagros Fault (MZF). The pattern has insignificant values in the Central Zagros. It takes the significant negative values across the SW part, especially along the SPF and Persian Gulf shore. The effect of non- independent observations on the estimation of eigenspace components of deformation tensors (strain tensor and TCC) shows that the estimation of covariance components has influence on the confidence intervals of eigenspace components, especially in seismically active regions of the belt (along the Persian Gulf shore, NW of the belt and region between the Central Iran block and MRF). The results demonstrate the importance of considering the correlation structure among the observations on statistical behavior of principal components of deformation tensors in seismically active regions. Keywords: random tensor, principal components, (co)variance component estimation