Multivariate statistical technique for spatial variation in river water quality

This paper proposes an integrated approach for the spatial variation in water quality based on the multivariate statistical analysis, i.e., the cluster analysis (CA) and the discriminant analysis (DA), aiming at identifying the spatial similarity and differences between sampling sites and optimizing the monitoring network. The main procedures of this approach include: (1) checking the normality of all parameter's distribution with the kurtosis and skewness tests, then log-transforming the original data of all parameters, (2) grouping the sampling sites based on CA which was performed on the standardized log-transformed data, and (3) recognizing the discriminant parameters based on DA which can account for most of the expected spatial variation in water quality. The proposed approach is applied to deep bay water control zone in northern Hong Kong, and the results demonstrate that: (1) the distribution of original data is improved after log-transformation, and all parameters are close to the normal distribution; (2) the sampling sites are classified into 3 clusters at (D_link/D_max) × 100 < 25, i.e, the low, moderate, and highly polluted. Moreover, the later two clusters belong to the nutrient and heavy metal pollutions, thus the domestic wastewater, livestock pollution, industrial pollution and surface runoff should be controlled; (3) in backward stepwise DA on the original data, only seven discriminant parameters (pH, NH₃-N, NO₃-N, F.coil, Fe, Ni and Zn) in spatial variation are identified and the correct assignations are 90.65% for three cluster sites; and (4) based on spatial variations in water quality, it is possible to optimize the monitoring strategy in future with only one of 3 sampling sites and seven discriminant parameters, which can decrease the number of sampling stations and corresponding costs.