Performance of two stochastic approaches for forecasting water quality and streamflow data from Yesxilirmak River, Turkey

This study evaluates the forecasting performance of two modeling approaches, ARIMA and Thomas–Fiering, for selected water quality constituents and streamflow of the Yesxilirmak River at Durucasu monitoring station. For this purpose, 13-year (1984–1996) monthly time series records were used to obtain the best model of each water quality constituent and streamflow from both modeling approaches. The comparison of the mean and variance of 5-year (1997–2001) observed data vs. forecasted data from the selected best models showed that the pH model from Thomas–Fiering, and EC and Cl models from ARIMA modeling approaches should be used with caution since the forecasting values from these models does not preserve the basic statistics of observed data in terms of mean. Also the results of forecast accuracy measures including root mean square error and mean absolute error calculated for two approaches indicated that between two approaches, for Yesxilirmak River Thomas–Fiering model presents more reliable forecasting of water quality constituents and streamflow than ARIMA model.