Selection and grouping of indicators of water quality using Multivariate Statistics

Abstract

Multivariate statistics techniques (Principal Component Analysis and Cluster Analysis) were employed to select the most important parameters that explain water quality variability at a rural watershed in the state of Espírito Santo (Brazil). In addition to group the waters studied for the similarity of features selected to verify the effect of type of soil cover (agriculture, livestock, forest and urban), water resource (surface and underground) and sampling period (rainy and dry seasons). Nineteen physico-chemical parameters of water quality were analyzed: pH, electrical conductivity, total solids, total dissolved solids, total suspended solids, turbidity, biochemical oxygen demand (BOD), ammoniacal nitrogen, nitrate, nitrite, total phosphorous, Ca, Mg, Fe, Na, K, Zn, Cu and total coliform. Application of Principal Component Analysis reduced the 19 parameters to three components that explained 87.53% of the total variance of data set. Water quality parameters that best explained variability of data were: electrical conductivity, total solids, total dissolved solids, turbidity, BOD, nitrate, Ca, Mg, and Na. Application of Cluster Analysis showed four different groups of water quality that differed in concentration of physicochemical characteristics and the type of water resource study, since the collection periods and the type of soil cover did not influence the segregation of groups formed.