Mean annual runoff proved to be significantly different for deficit and non-deficit years (t-test, at the 5% significance level). We continued to investigate whether the flow variation patterns differ between these years. Grouping the weights of the AFs for different years as obtained by EOF decomposition can be used for discriminating the flow variation patterns in the observation series (Krasovskaia et al. 1999, 2003). Table 2 presents the values of the relative explained variance after a truncation level of approximately 90% and the number of AFs necessary to reach this level. It can be observed that for all three series the first AF represents 60–70% of the variation in monthly runoff. In general, the first AF reflects the average monthly variation pattern, while the second and higher AFs might indicate shifts from this pattern (Krasovskaia et al. 1999).