ASSESSMENT OF STOCHASTIC MODELS FOR PREDICTION OF RIVER WATER QUALITY IN THE BUYUK MENDERES CATCHMENT, TURKEY

This study analyzed quality parameters, including water temperature, pH, electrical conductivity, nitrate, boron and dissolved oxygen, in water samples collected from Buyuk Menderes river in Turkey from 1996 to 2004, for forecasting, using stochastic models. At first, the Box-Whisker plots and Kendall's tau tests were used to detect the trends during the study period. Marginal increasing and decreasing trends were observed for certain periods at some locations. Three approaches of stochastic modeling, deseasonalized model, multiplicative autoregressive integrated moving average model and periodic autoregressive model, were used to model the observed pattern in water quality parameters. The predicted water quality data from the stochastic modeling approaches were compared with the observed data. Considering pH, electrical conductivity, nitrate and water temperature, the accuracy measures, F-cal and Z(cal) tests indicated that the predicting results of the deseasonalized model were slightly better than the results of multiplicative autoregressive integrated moving average model and periodic autoregressive model for these parameters. Considering boron and dissolved oxygen parameters, the multiplicative autoregressive integrated moving average modeling approach produced more reliable predictions than the deseasonalized and periodic autoregressive modeling approaches.