GENETIC ALGORITHMS: TOWARDS THEIR USE IN THE HOMOGENIZATION OF CLIMATOLOGICAL RECORDS
Alexander Jann · University of Zagreb University Computing Centre (SRCE) · 2006
The prerequisite of the homogenization of climatological time series is to find the points of time where the series underwent abrupt shifts in the climate variable.Since normally no information about the (potentially large) number of such shifts is available beforehand, this is a highly challenging optimization problem for the resolution of which a genetic algorithm approach was proposed.In the present paper, important steps towards its applicability to real climatological data are made.An outlier test is incorporated in the genetic algorithm framework.Another typical climatological component that can be fully integrated into the genetic algorithm scheme is the exploitation of metadata containing information about a station's history.In a conceptually similar manner, the simultaneous investigation of several, possibly interrelated, time series is made feasible.The advantages of online analysis, where the change-point detection is carried out not only for the entire n-element series but also for the partial series of its first n-1, n-2, … elements, are discussed.Such supplementary analyses can regularly be crucial when change-point signals are detected in the analyses of partial series while random noise accidentally obscures these signals in the full series.