APPLICATION OF SELF-ORGANISING MAP (SOM) ARTIFICIAL NEURAL NETWORKS FOR THE CHEMICAL ASSESSMENT OF SEDIMENT QUALITY
Sustainable process-product development & green chemistry
Environmental Engineering & Management (T1-3P)
Keywords: sediments, sustainable management, chemical assessment, Self-Organising Map (SOM), Artificial Neural Network (ANN)
Sediments comprise an essential component to take into account when analysing the environmental status of marine ecosystems [1]. The physico-chemical characterisation of estuarine sediments is a basic initial step in the design and implementation of environmental management actions in coastal zones. However, due to the fact that many variables have influence on the sediment quality [2, 3], the use of tools that allow the analysis and interpretation of these variables becomes necessary in a global characterisation approach; then, this will be also useful in a decision-making framework for determining the application of the best engineering processes that lead to the sustainable management of sediments.
The Kohonen Self-Organizing Map (SOM) is one of the most well-known artificial neural networks (ANNs) with unsupervised training algorithms. The aim of this work is to apply the SOM for the integrated assessment of physico-chemically characterised sediments from different zones and with different levels of contamination, as well as to compare the results of the SOM with those of traditional multivariate statistical techniques (hierarchical cluster (HC) and principal component analysis (PCA)). The SOM Toolbox 2.0 for Matlab® was used to create and visualise the SOMs, and all statistical analyses were performed using SPSS 14.0 for Windows.
The data used in this study included 40 samples of superficial sediments from 3 estuarine zones of Cantabria (Northern Spain): Santander, Suances and Santoña, exposed to different levels of anthropogenic pressure due to urban, industrial and port activities. A set of 12 physico-chemical variables were studied in each sample: % of silt+clay, Total Organic Carbon (TOC), 8 metals (As, Cr, Cu, Fe, Mn, Ni, Pb and Zn) and 20 polycyclic aromatic hydrocarbons (PAHs), grouped in sum of 7 low molecular weight PAHs, (S7LPAHs) and in sum of 13 high molecular weight PAHs (S13HPAHs).
A 42-unit SOM (7x6) allowed a proper classification of the 40 sediment samples into 4 groups according to their different level of contamination. The HC analysis, which was carried out as a comparison, identified 3 clusters; although two of the groups were very similar to those found in the SOM, HC grouped in a third cluster both contaminated and clean samples that the SOM was able to separate appropriately in two groups. As well as providing a better classification of the sediment samples than HC, it should be emphasized that SOM allows an easier visualisation of multidimensional data; the very useful graphic representations that SOM offers help in the task of interpreting the data involved in the sediment quality assessment and management.
The SOM was also applied to look for associations of variables analysed in the sediments. In the 14-unit map (7x2) that was obtained, 3 groups of variables could be distinguished: 1) the heavy metals Fe, Mn, Cr, Ni and Cu; 2) Pb, Zn, As and S13HPAHs; 3) the physical variables (silt+clay and TOC) and S7LPAHs. The application of HC leaded to a dendrogram where these same 3 clusters were identified. However, the results obtained using PCA were slightly different and perhaps showed the interrelations between the variables more clearly; specifically, PCA discovered a component that included Pb, Zn, As and S7LPAHs, but the three metals with positive coefficients and S7LPAHs with a negative one.
Acknowledgements
This research project was supported by the financial help of the Spanish Ministry of Education and Science Project CTM 2005-07282-C03-03. M. Álvarez-Guerra was funded by the Spanish Ministry of Education and Science by means of an F.P.U. fellowship.
References
[1] Commission of the European Communities, Proposal for a Marine Strategy Directive, COM(2005) 505 final, Brussels, 24.10.2005.
[2] O. Rodríguez-Obeso, M. Álvarez-Guerra, A. Andrés, T.A. DelValls, I. Riba, M.L. Martín-Díaz, J.R. Viguri, TrAC-Trends Anal. Chem. (in press).
[3] J.R. Viguri, M.J. Irabien, I. Yusta, J. Soto, J. Gómez, P. Rodriguez, M. Martinez-Madrid, J.A. Irabien, A. Coz, Environ. Int. (in press).
Presented Monday 17, 13:30 to 15:00, in session Environmental Engineering & Management (T1-3P).
