Phenomenological pattern recognition in the dynamical structures of tidal sediments from the German Wadden Sea

A self-organizing map (SOM) is trained by a data set on the sediments of the back barrier tidal flats of Spiekeroog Island (southern North Sea). The data set comprises 170 samples from two seasonal situations (May/July 1992) and from different locations on a 250 m-grid at three depths. Additional samples were taken on a 4 cm-grid at three locations. The microbiological and geochemical conditions of the sediments were determined by the variables total bacterial numbers, viable counts of aerobe-heterotrophic bacteria, nitrate-reducing-bacteria, sulphate-reducing-bacteria, proteins, extracellular polymeric substances, total and dissolved carbohydrates, lipids, and water content. SOMs are capable of displaying both the typical features of an input data set and the neighbourhood relationships between the structural units. In this study it is demonstrated that the 11-dimensional data manifold representing the different sediments can be mapped onto a 2-dimensional subspace preserving all relevant features of the data set. The interpretation of the SOM leads to cross-sectional phenomena of tidal sediments. In a modelling approach the vulnerability of certain sediments in consequence of simulated external distortions was estimated. The analysis of SOM responses revealed attainable domains of coherent nodes indicating that the sediments can only alter in traced out directions. The SOM application to tidal sediments demonstrates the efficiency of the SOM technique for the analysis of multivariate data sets of complex natural systems.