Self-Organizing Maps method in recent Adriatic Sea environmental studies: applications and perspectives

Herein we present three recent oceanographic studies performed in the Adriatic Sea (the northernmost arm of the Mediterranean Sea), where Self-Organizing Maps (SOM) method, an unsupervised neural network method capable of recognizing patterns in various types of datasets, was applied to environmental data. The first study applied the SOM method to a long (50 years) series of thermohaline, dissolved oxygen and nutrient data measured over a deep (1200 m) Southern Adriatic Pit, in order to extract characteristic deep water mass patterns and their temporal variability. Low-dimensional SOM solutions revealed that the patterns were not sensitive to nutrients but were determined mostly by temperature, salinity and DO content; therefore, the water masses in the region can be traced by using no nutrient data. The second study encompassed the classification of surface current patterns measured by HF radars over the northernmost part of the Adriatic, by applying the SOM method to the HF radar data and operational mesoscale meteorological model surface wind fields. The major output from this study was a high correlation found between characteristic ocean current distribution patterns with and without wind data introduced to the SOM, implying the dominant wind driven dynamics over a local scale. That nominates the SOM method as a basis for generating very fast real-time forecast models over limited domains, based on the existing atmospheric forecasts and basin-oriented ocean experiments. The last study classified the sea ambient noise distributions in a habitat area of bottlenose dolphin, connecting it to the man-made noise generated by different types of vessels. Altogether, the usefulness of the SOM method has been recognized in different aspects of basin-scale ocean environmental studies, and may be a useful tool in future investigations of understanding of the multi-disciplinary dynamics over a basin, including the creation of operational environmental forecasting systems.