Using Self-Organising Maps to investigate long-term changes in deep Adriatic water patterns

The paper attempts to document long-term changes in deep Adriatic water patterns by applying the Self-Organising Maps (SOM) method to temperature, salinity, dissolved-oxygen content, orthophosphate and total inorganic nitrogen profiles sampled at a single deep station in the South Adriatic Pit (SAP) over a half century (1957–2009). Seasonality observed in upper layers has been removed by the least-squares fitting of the annual and semi-annual sinusoidal functions. The sensitivity of the SOM to various parameter combinations reveals the importance of temperature, salinity and dissolved oxygen for mapping different water patterns, while nutrients have less influence on quality and applicability of SOM solutions to the extraction of characteristic SAP water profiles. The quality of fit obtained for different combination of the measured parameters introduced to a SOM suggests that the incomplete combinations of input parameters increase an imperfection in the applicability of SOMs to the dataset. Two modes of long-term changes in the SAP obtained by the SOM analyses are discussed with respect to the processes that drive the variability in the area, e.g., the Adriatic-Ionian Bimodal Oscillation: where the first mode is characterised by rapid changes in the transition of SAP water masses, observed before 1980s (less adoptable by the SOMs), and the second mode is characterised by steady transitions (better adoptable by the SOMs), observed in the 1990s and the 2000s. The SOM method is found to have certain advantages when compared to other methods that have previously been used to distinguish the Adriatic water masses, as it does not depend on predefinition of water mass sources and allows for gaps in series.

Research highlights
► Novel mathematical method (Self-Organising Maps) has been used to map the characteristic patterns of various parameters in the deep waters of the south Adriatic.
► The patterns are not sensitive to nutrients but are determined mostly by temperature, salinity and DO content.
► Two modes of the Adriatic-Ionian BiOS are detected within the SOM solutions.
► The SOM method exhibits a number of advantages compared to density- or buoyancy-based definitions of water masses.