Performance of stochastic approaches for forecasting river water quality

This study analysed water quality data collected from the river Ganges in I ndia from 1981 to 1990 for forecasting using stochastic models. Initially t he box and whisker plots and Kendall's tau test were used to identify the t rends during the study period. For detecting the possible intervention in t he data the time series plots and cusum charts were used, The three approac hes of stochastic modelling which account for the effect of seasonality in different ways, i.e. multiplicative autoregressive integrated moving averag e (ARIMA) model, deseasonalised model and Thomas-Fiering model were used to model the observed pattern in water quality. The multiplicative ARIMA mode l having both nonseasonal and seasonal components were, in general, identif ied as appropriate models. In the deseasonaliscd modelling approach, the lo wer order ARIMA models were found appropriate for the stochastic component. The set of Thomas-Fiering models were formed for each month for all water quality parameters. These models were then used to forecast the future valu es. The error estimates of forecasts from the three approaches were compare d to identify the most suitable approach for the reliable forecast. The des easonalised modelling approach was recommended for forecasting of water qua lity parameters of a river. (C) 2001 Elsevier Science Ltd. All rights reser ved.