An integrated SOM-based multivariate approach for spatio-temporal patterns identification and source apportionment of pollution in complex river network.

In this study, three classification techniques (self-organizing maps, hierarchical cluster analysis and discriminant analysis) were applied to identify spatial water pollution levels, temporal water quality response delay phenomena (WQRDP), source pollution types (point, urban non-point, or agricultural non-point). Two models (principal components analysis (PCA), and positive matrix factorization (PMF)) were used to do the further quantitative source apportionment studying. The 27 inflow rivers in spatial were divided into three pollution levels (A, high; B, medium; C, low). The primary pollution pattern in spatial Clusters A, B, and C were point, urban non-point and agricultural non-point separately, in consideration of simultaneous land use types. Source apportionment results identified five typical factors in spatial Cluster A and six typical factors in spatial Cluster B and C as responsible for the data structure, explaining 80%-90% of the total variance of the dataset. Crown