Homogeneity analysis of climatological time series: experiments and problems

Experiments with different tests for homogeneity are performed. Sensitivity of the tests to different modifications of the time series is examined. Experiments were made with artificially generated normally distributed random series, on which different modifications were imposed: trends (constant and changing), simple periodicity (sine wave), jump, and various combinations of jump plus another change. Experiments with the combined models aimed to examine whether the jump can be detected in the presence of other modifications (with a presumption that the most probable form of inhomogeneity is an abrupt change). To obtain reliable results, each experiment was repeated with 100 different random series, and the results for individual series were averaged. The experiments show that graphical tests can give an indication about the form of the change. Serious problems are the cases when two or more modifications interfere in such a way that the tests fail to indicate the change. An example with the annual precipitation series of Sofia (Bulgaria) is presented.