The Alexandersson test for homogeneity of temperature data: a case study for the Northern Great Plains

Abstract

There are many research problems facing climatologists today, such as the study of global climate change. In order to obtain the most definitive results, the time series being analyzed must be homogeneous. A homogeneous time series is one in which trends and variations are caused by effects of weather and climate. The concept of homogeneity has been addressed by some researchers, but only as a mention that the results may contain some biases due to the inhomogeneities in the data being studied. In addition, researchers that have addressed this homogeneity problem have only tested "average" time series such as the means and the annuals. This thesis utilizes a homogeneity test developed by Alexandersson (1986) and applies it to mean monthly maximum and minimum temperature data from 22 stations in the Northern Great Plains. The Alexandersson test locates where a break or discontinuity occurs and whether or not it is significant. A theoretical and numerous practical aspects are examined. The theoretical aspect studied is the question of whether or not a temperature series with one significant break has a constant variance throughout its entire record. An F-test is used to compare the variances and determine if there is a significant difference between them. The results show that some series do not have a constant variance throughout and this difference may effect the standardization procedure used in the Alexandersson test. The practical aspects studied include data type and temporal comparisons. The results of these analyses show that a break point found in a station's means or annual is not always the same break point found in that station's maximums or minimums and each individual months respectively. Finally, an adjustment procedure is discussed and applied to selected series, and it is shown that the annual adjustment is not the same as the adjustment for the individual months. In addition, the monthly adjustments show a seasonal pattern and when summed, equal the annual adjustment.