Razvoj i primena hemometrijskih metoda za klasifikaciju i procenu kvaliteta vode / The development and application of chemometric methods for the classification and assessment of water quality

Modern society is characterized by intensive industrialization and urbanization, leding to the depletion of natural resources and increasing threat for the environment. In terms of global development, concern about water is the same as a matter of survival of civilization. Therefore, water management and water quality control are essential social needs. Teams of experts from different scientific fields are searching for the right patterns and models for modeling of water quality parameters and prediction of variables responsible for water quality in order to discover the key variables that lead to groups of similar locations and objects, temporal and spatial variations and to identify the sources of pollution and to assess the distribution of pollution. The aim of this study was to optimize the system for monitoring and protection of water resources. Laboratory and on-line analysis performed daily provide a huge number of data on the quality of water which should be evaluated, processed and reduced in order to obtain information that would be sufficient to indicate the water quality and that would serve as the basis for efficient modeling and management of water resources. The goal of this investigation was to apply multivariate statistical techniques and choose the most applicable one for this case. The statistical evaluation and multivariate analysis are very important in all fields, ranging from the economics, marketing, psychology, social studies, to medicine, pharmacy, and especially chemistry and environmental protection. Regular monitoring of parameters of the process water, with the possible reduction of the number of on-line parameters that should be monitored and their frequency of measurement is important for the maintanance services in the industry. From key variables which are measured on-line, and by which indirectly the information on other parameters are recieved, based on their correlations, simplification of the monitoring system is enabled with less consuming activities and the same level of preventing accidents. Within this Ph.D. thesis the concentration of ionic species and the cruical physicochemical parameters in the process water in the water-steam systems are monitored. A complete evaluation of the measurement was performed. The ion chromatography (IC) method for the determination of ions at trace level (ppb-ppt) was tested on real samples, and the method was optimized and validated. Different chemometric methods such as principal components analysis (PCA), factor analysis (FA), cluster analysis (CA) and discriminant analysis (DA) were applied, for the ultrapure water. The optimal combination of multivariate techniques and the selection of one or more techniques that can be applied for the reduction of data, as well as the choice of key variables and the prediction of the parameters which are the most responsible for variations in water quality and the design of future monitoring system was investigated. The locations with the highest pollution, in respect to the analysed elements, as well as their possible origin and source was defined and investigated. The relationships between the elements and mechanisms by which these impurities get into the water-steam cycle were discovered based on the correlation coefficients. The monitoring system with a reduced number of parameters and reduced frequency of measurements is proposed as a result of this research. The second object of the analysis within this study was the raw water (surface and groundwater) which the Belgrade Waterworks system used for procesing into the drinking water. In order to define the parameters that are responsible for the change of water quality and to obtain current information on the qualitative and quantitative composition of the water, the chemometrics methods can be applied to assess the quality of surface water (Sava river ) and groundwater. Based on the measured water quality parameters of the Sava river, the database (data set) that was subjected to multivariate analysis was created. The chemometric methods which give the best interpretation of the spatial/temporal variations and the most accurate prediction and modelling parameters responsible for the variations in water quality were adopted. The causes of variation in all aspects: location, seasonal and temporal were investigated. The parameters that led to all forms of variability in the water quality were revealed. The modelling based on the DA method and artificial neural network (ANN) was performed.