Understanding dynamic of biogeochemical properties in the northern Adriatic Sea by using self-organizing maps and k-means clustering

[1] The dynamic of biogeochemical properties in a coastal area of the northern Adriatic Sea (Gulf of Trieste) is analyzed through (1) identification of a small number of water typology classes and classification of samples, obtained by means of a novel multivariate classification procedure based on a combination of Artificial Neural Networks (ANN) and "traditional'' clusterization algorithms, (2) interpretation of each class based on biogeochemical properties and ecological phenomena likely to occur in the water body, and (3) discussion of time evolution and spatial distribution of water classes which summarized and provided indications on the system's space and time evolution. Basing itself on a multivariate comparison, the Self-Organizing Map (SOM) grouped 1292 samples collected in a 3-year-long monitoring program in 187 sets and identified a representative synthetic sample for each group. These groups were further classified in seven clusters, which identified the water typology. The complexity of the space and time coevolution of 12 variables was so reduced to variation of one categorical variable. Results included an objectively derived typology of water masses and their typical temporal succession, a spatial dividing based on biogeochemical processes, a conceptual scheme of biogeochemistry in the Gulf. Results clearly indicated the importance of river input in triggering plankton blooms and pointed out that trophodynamics followed current paradigms of marine ecosystem functioning, with shifts from conditions dominated by classical food chain to situations in which most of the energy flowed through the autotrophic and heterotrophic parts of the microbial food web.