A Self-Organizing Maps Multivariate Spatio-temporal Approach for the Classification of Atmospheric Conditions

This work demonstrates the potential of Self-Organizing Maps (SOM) as a multivariate clustering approach of spatio-temporal datasets in atmospheric physics. A comprehensive framework is proposed and the method is applied and assessed for its performance in the field of synoptic climatology within a specific region at southeastern Mediterranean. The results indicate that the SOM can be a powerful tool for the identification and classification of atmospheric conditions, allowing an analytical description of the principal atmospheric states. The coupling of sea level pressure (SLP) and 500hPa geopotential (F500) in a synoptic-scale domain with the wind, the specific humidity and the air and dew point temperature in the chosen mesoscale subdomain, allows the SOM algorithm to define the relevant atmospheric circulation patterns. The corresponding patterns are well documented and the method accounts for their seasonality. Furthermore, in the resulting two-dimensional lattice the similar patterns are mapped closer to each other, compared to more dissimilar ones.