Investigation of trends in synoptic patterns over Europe with artificial neural networks

Abstract. The present study is a comprehensive application of a methodology developed for the classification of synoptic situations using artificial neural networks. In this respect, the 500 hPa geopotential height patterns at 12:00 UTC (Universal Time Coordinated) determined from the reanalysis data (ERA-40 dataset) of the European Centre for Medium range Weather Forecasts (ECMWF) over Europe were used. The dataset covers a period of 45 years (1957–2002) and the neural network methodology applied is the SOM architecture (Self Organizing Maps). The classification of the synoptic scale systems was conducted by considering 9, 18, 27 and 36 synoptic patterns. The statistical analysis of the frequency distribution of the classification results for the 36 clusters over the entire 44-year period revealed significant tendencies in the frequency distribution of certain clusters, thus substantiating a possible climatic change. In the following, the database was split into two periods, the "reference" period that includes the first 30 years and the "test" period comprising the remaining 14 years.