INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 20: 1743–1760 (2000) REGIONAL VARIABILITY OF SEASONAL PRECIPITATION OVER TURKEY MIKDAT KADIOG  LU* Department of Meteorology, Istanbul Technical Uniersity, Maslak, 80626 Istanbul, Turkey Receied 23 February 2000 Reised 30 May 2000 Accepted 6 June 2000 ABSTRACT Principal components (PCs) of the parametric spatial Pearson and non-parametric Spearman rank correlation matrices are used to depict the seasonal precipitation characteristics or features over Turkey. Eighty-five irregularly scattered meteorology stations over Turkey were chosen to study the monthly totals of precipitation observed from 1931 to 1990 inclusive. There are similarities among the rotated and unrotated first four components of the seasonal precipitation distributions. Meteorological processes, such as cyclonic storms during autumn and winter, are represented by the first PC over the whole study region. The influences of continentality towards the inland in Anatolia and the effect of mountains in the east of Turkey are shown by the second PC. The third PC explains the marine influence of the surrounding seas on the precipitation. Finally, it is shown that in this study, a non-parametric spatial correlation matrix produces quite similar fields with those of a parametric spatial correlation matrix. Copyright © 2000 Royal Meteorological Society. KEY WORDS: Turkey; principal component analysis; Pearson and Spearman correlations; precipitation; climatology 1. INTRODUCTION Eigenvector analysis techniques, such as empirical orthogonal functions (EOFs), principal component analysis (PCA) and common factor analysis (CFA) have been applied in meteorology/climatology domains with increasing frequency to delineate patterns of temperature, pressure, drought, etc. (Gilman, 1957; Kutzbach, 1967, 1970). Principal components (PCs) are linear combinations of original variables, such that each ordered function explains the maximum variance of the remaining variance. In meteorol- ogy and climatology the variables can be grid points or station data fields of one or many weather elements. The many procedural options that exist in eigenvector analysis may produce widely differing results; examples include the selection of the dispersion matrices, such as correlation, covariance and cross-product coefficients used to relate data in these analysis (Richman, 1986; White et al., 1991). In many of these studies, the Pearson product-moment correlation coefficient was selected as the type of dispersion matrix. In order to obtain a classification, all station variability is placed on an equal basis through the correlation function so that PCs would dissect the domain rather than the magnitudes of the dispersion matrix. Pearson’s correlation is, however, quite sensitive to non-normality (Kowalski, 1972; White et al., 1991). By using the Pearson correlation matrix they, therefore, assumed that the data used is normally distributed. Because of this, in some of the studies, each individual station’s precipitation distribution was examined by quantile – quantile (Q-Q) plots. These plots are, in fact, widely used in statistics after Wilk and Gnanadesikan (1968) as a simple and effective tool for visualizing transforms (Cleveland, 1993). * Correspondence to: Department of Meteorology, Istanbul Technical University, Maslak, 80626 Istanbul, Turkey; e-mail: kadioglu@itu.edu.tr Copyright © 2000 Royal Meteorological Society