Classification characteristics of multivariate analyses for groundwater chemistry in the nitrate contaminated area

Monday, 14 December 2015
Poster Hall (Moscone South)
Kei Nakagawa and Hiroki Amano, Nagasaki University, Graduate School of Fisheries and Environmental Sciences, Nagasaki, Japan
Groundwater nitrate pollution in agricultural field is a common problem in many parts of the world. To design effective countermeasures for the pollution, understanding of contaminant transport and source identification will be in need. Classification of groundwater chemistry is useful tool to compare between water from different sources. In this study, results of 4 multivariate analyses for groundwater chemistry were compared.

In the multivariate analyses, 277 sampling points data of major ion concentrations (Cl-, NO3-, SO42-, HCO3-, Na+, K+, Mg2+, Ca2+) were used (2011-2013). 4 multivariate analysis methods were as follows; 1) HCA (Hierarchical Cluster Analysis); Based on 8 major ion concentrations (8 dimensional 277 vectors), HCA was performed. 2) HCA with PCA (Principal Component Analysis); PCA was performed with major ion concentrations. Based on 2 principal components scores obtained from PCA (2 dimensional 277 vectors), HCA was performed. 3) HCA with SOM (Self Organized Map); SOM was performed with major ion concentrations. Obtained 84 reference vectors used for HCA (8 dimensional 84 vectors). 4) HCA, SOM with PCA; PCA was performed with major ion concentrations. Based on 2 principal component scores obtained from PCA, SOM was performed. Finally, HCA was performed with 80 reference vectors (2 dimensional 80 vectors). The number of clusters were fixed to 5, which was determined based on DBI (Davies-Bouldin Index) at the method 3).

Broadly characteristics of each cluster is same among all multivariate analysis methods. According to this characteristics, 2 of 5 clusters show nitrate pollution, which are exceeding Japanese drinking water standards (10 mg/L). This 2 clusters can be distinguished by level of nitrate concentration. Other 3 clusters can also distinguished by level of major ion concentrations. Each cluster may be related to land use, because spatial distribution of sampling points shown by clusters congregate in specific locations. In the methods 1) and 3), higher nitrate concentration points were not classified into polluted clusters. In this regards, methods 2) and 4) seem better to represent classification of the field water chemistry. Difference between methods 2) and 4) is the clusters of 2 sampling points. Both of clusters are belong to nitrate pollution cluster.