Time Series Forecasting of Sefidrood River Water Quality Using Linear Stochastic Models

In the management of river basins, prediction of river water quality is essential to maintain water quality within standard limits. This study performed a time-series analysis of the prediction of chlorine concentration and electrical conductivity time series data for the period of 1991-2005 from Sefidrood River in northern Iran. The seasonal prediction of chlorine and electrical conductivity time series data was done using the linear stochastic model known as the multiplicative seasonal autoregressive integrated moving average (SARIMA). Initially, Mann-Kendall and Box-Pierce tests were used to identify the trend and stationary state of the time series, respectively. The results showed that there was no significant trend in these time series, but that 12-month seasonal patterns were observed. As a result, seasonal patterns were removed from both time series data using first-order differencing. SARIMA modeling was performed in three steps: model identification, parameter estimation and diagnostic checking. Different models of SARIMA were identified according to the ACF and PACF time series results and the model with the minimum AIC criterion was selected. For parameter estimation, model parameters were estimated using a least squares optimization algorithm that minimized the residual sum of squares. The results of diagnostic checking then indicated that the residuals were independent, normally distributed and homoscadastic. The selected SARIMA model was then used to predict chlorine concentration and electrical conductivity time series data for 2003-2005. There was a good unanimity between the predicted and observed data. For model verification, the mean and variance of the predicted and observed data were compared. The RMSE for Cl and EC were 2.2 and 278.9, respectively. The results showed that there was no significant difference between the predicted and observed time series data. This study showed that the SARIMA model can be used reliably to predict chlorine concentration and electrical conductivity time series data in Sefidrood River.