Summary:

Alexandersson test is a statistical tool for detection of a step change in mean value in normally distributed time series that identifies the change point. The test may be applied to log-normally distributed hydrological series where normality is achieved by logarithmisation. The method becomes to a tool for testing if there appeared a step change that caused the inhomogeneity of the series. In the article the efficiency in detection of the step change in mean value in log-normally distributed time series was investigated after transformation to normally distributed series and application of the Alexandersson test. As the power of the test is a measure of efficiency, the estimates of the power were calculated. The artificial time series simulated real annual log-normally distributed discharges with an abrupt shifts in the mean values. The series differed in lengths: 20, 60 and 100, heights of the jump: from 1% to 25% and coefficients of variation: from 0,1 to 1. If the length of the series is at least 60 then for the series with very low variability the power turned out to be very large for the change at least equal to 5%. For the series with medium variability the large or medium power is guaranteed if the length of the series is at least 60 and the change is over 10%. If the series is featured by a high variability, the fairy large power is observed only for series with 100 elements and for changes equal to at least 25%. For very large variability even high jumps are detected with difficulty. The changes lower than 5% are usually not identified for series of moderate variability. The test was applied to real mean and maximum annual discharges in some rivers from USA and Poland. It indicated an abrupt change in the maximum annual flows on the Nida River.