Modeling of dissolved oxygen concentration of the Tigris river with artificial neural networks

Studies on water pollution, which is one of the important problems in an increasingly globalized world, are increasingly continuing. Contribution of water quality studies is important in controlling pollution. Analysis studies are carried out on water quality due to the requirement that the waters comply with the standards in terms of quality. Addressing dissolved oxygen, which is an important parameter of water quality, which can directly affect the balance of the aquatic environment, is very important in the interpretation of water quality. In this study, the dissolved oxygen parameter of the Tigris sub-basin was modeled with the Artificial Neural Network method, using the data obtained from the 10th Regional Directorate of State Hydraulic Works (DSI). The data of the Dicle dam and Ongözlü bridge station points were taken as stations, processed with the help of Matlab program as input, and ANN models with different structures and combinations were created for dissolved oxygen. In order to get maximum efficiency from the data, the normalization technique was applied to the raw data. Prediction models were created by processing the normalized data in artificial neural Networks. Forecast models are twelve models in total for each station, and values such as R2 (coefficient of determination), MAE (mean absolute error) and MSE (mean square error) were calculated and interpreted for the performance analysis among them. The models that give the most favorable output are Feed Forward Backpropagation Neural Network (FFBPNN) at both observation points and Levenberg – Marquardt backpropagation(trainlm) as an algorithm. In the modeling of river water dissolved oxygen, it has been seen that artificial neural networks create successful and efficient outputs in terms of the number of data used, time and cost. On the other hand, analysis studies of water bodies can easily determine the pollutant parameters by making use of neural networks independently of water quantity. By determining the essential pollution sources in its region of each study, the analysis of the parameter can be carried out efficiently through neural networks.