Determination of seasonal changes in Filyos Stream water quality by artificial neural network

Water is in an endless cycle, which is source of life for human beings. In this cycle, human being can use water again and again. During this cycle, substances that are contaminated in water cause physical, chemical or biological alteration of the water's natural features, that leads to water pollution and therefore causes the environmental balance to deteriorate over time. The water quality in resources can be changed by either domestic or industrial wastes being discharged without any treatment and the pesticides or fertilizers used in agriculture being contaminated with natural water due to rain and similar reasons. This quality changes cause deteriorations in ecosystem. For this reason, it is important to investigate the water quality in rivers and water reservoirs which are close to settlement areas. Observations and measurements on a river, give the necessary information about how to benefit from the river. Obtaining the necessary information depends on long-term data collection. In this study, surface water quality measurements were carried out at five observation stations along the main line of the Filyos stream, which forms the largest sub-basin in the western Karadeniz Basin, at intervals of thirty days in one year period. In the scope of the study, measurements in the field survey (temperature, pH, dissolved oxygen and electric conductivity) and in the laboratory (suspended solids, chemical oxygen demand, turbidity, total organic carbon, ammonium, calcium, magnesium, hardness, phosphate, nitrite, nitrate, aluminium, manganese, iron, chromium, lead and zinc ) analyses were carried out. According to the water pollution control regulation, water classification was also performed within the scope of the study. Finally, estimation of the turbidity parameter based on parameters of chromium, chemical oxygen demand, iron, aluminium, suspended solids, manganese, zinc, lead and calcium was performed by artificial neural networks. Estimation is designed in the form of two scenarios. In the first scenario at the analysis, the parameters of chromium, chemical oxygen demand, iron, aluminium, suspended solids, manganese, zinc, lead and calcium were used as input in the models to determined that best predicts turbidity parameter. Then the models that are created by adding the best other parameters to this parameter each time and the most effective model was determined on basin basis. In the second scenario at the analysis, selected parameters were used as input for each station in the models to determined that best predicts turbidity parameter. Then the models that are created by adding the best other parameters to this parameter each time and the most effective model was determined. For both scenarios the model results are shown in tables for each station separately and the topologies with the best performance is charted.