Modeling approaches to predict conditions of water quality using physical, chemical, and hydrological data focused on biological contaminations

The quality of water in coastal areas is a vital component to human activities as well as the natural ecosystem. Beaches in California have been suffered with their pollutions and the state has dedicated substantial resources to ensure beach water quality through monitoring and maintenance programs. Indicator bacteria have been adopted as the index of water quality for recreational water and the State of California chose three indicator bacteria, Total Coliform (TC), Fecal Coliform (FC), Enterococci (ENT). The state developed the monitoring system to maintain the water quality based on the level of these indicator bacteria and beach posting/closures also depends on this monitoring system. The timing gap, however, between testing water samples for indicator bacteria and natural cleaning system of bacterial contaminations still exist. In addition to this timing gap, the current monitoring system is labor and cost intensive. To solve these issues, appropriate methods, which could provide information for bacterial concentrations in a timely manner with cost and labor efficient procedures, are needed to develop for a suitable management of the water quality.; The dissertation, therefore, focused on finding such methods, modeling approaches, and consisted of following steps; (1) Relationships between indicator bacteria and several physical and chemical parameters, which could serve as surrogates for predicting concentrations of indicator bacteria, were investigated. Principal Component Analysis was also adopted to find relationships between indicator bacterial concentrations and all physical and chemical parameters together. (Chapter 2) (2) Decision Tree Analysis was tested as an extension to the research of previous chapter, Chapter 2, to investigate whether or not the approach could be used as a tool for prompt predictions of bacterial concentrations with selected parameters. Parameters, which could be detected easily and promptly, were used for the approach. The approach focused on dry seasons because physical and chemical parameters were available only for dry seasons (Chapter 3) (3) An Artificial Neural Network (ANN) was adopted as a lumped modeling approach focused on rainfall events since rainfall events might be the major factor of increasing bacterial concentrations during the wet seasons. However, the approach would focus on predicting bacterial concentrations for yearly periods while decision tree approach focused on only dry seasons. (Chapter 4) (4) Distributed Modeling Approach was investigated focused on land use type of the study area to find effects of rainfall distribution over each land use because each land use would respond to rainfall events in its own way and these differences would bring dissimilar affects on bacterial concentrations. The distributed modeling approach would also be compared with lumped approach (Chapter 5); The results of dissertation may yield several advantages. First of all, the study makes it possible to predict bacterial concentrations in a timely manner, so the gap between testing time for indicator bacterial concentrations and natural cleaning of a system could make up. Second, the study would help to cover long term observations, daily or weekly sampling data, throughout predicting finer time scales of bacterial concentrations, for example 6hrs bacterial concentrations. With advantages of this study, new monitoring system, combining modeling approaches and current in situ sampling monitoring systems, would be proposed, so problems of current monitoring system could be resolved. Modeling approach would provide timely mannered information for bacterial concentrations and also cover no sampling periods, between observations and results from current monitoring system could be used for adjusting model parameters.