Michael Begert

An automated procedure to detect discontinuities; performance assessment and application to a large European climate data set

Within the framework of the EU-project ENSEMBLES, an automated procedure to detect shift inhomogeneities in climatological time series was developed and used for the homogeneity assessment of a European data set including values for temperature (min, mean, max), precipitation and air pressure covering at least 45 years. The automated process combines VERAQC (Vienna Enhanced Resolution Analysis Quality Control) output with Alexanderssons Standard Normal Homogeneity Test. VERAQC is designed to find outliers, systematic errors and meteorological biases in a time series comparing the data with concurrent observations of neighbouring stations. The resulting deviations can be analysed as time series for a single station and used as an input for relative homogeneity testing. Shift inhomogeneities are detected in accordance with statistical significance. An iterative procedure ensures that multiple break points are detected. The performance of the automated homogenization method was tested comparing its findings to results of a manually homogenized Swiss data set. The latter is based on a relative homogenization procedure in combination with an in-depth analysis of station history information. It is shown that the new automated procedure is able to detect shift inhomogeneities in temperature, precipitation and air pressure series but several limitations exist. Mainly the number of false alarms as well as missed break points is comparatively high at least in the Swiss Alpine test region with its complex topography. The homogeneity assessment of the ENSEMBLES data set for the period 1960-2004 revealed that 12 % to 59 % of the series were homogeneous depending on the variable. By far the highest number of shifts was detected in air pressure series followed by temperature and precipitation.