A novel forecasting of continuous monthly streamflow based on lagged climate indices: uses of autoregressive integrated moving average and artificial neural network techniques

ABSTRACT This study used autoregressive integrated moving average (ARIMA) and artificial neural network (ANN) models to forecast long-term streamflow of Victoria's six stations (Acheron, Goulburn, Rubicon, Ovens, King, and Buffalo). Lagged monthly combined climate indices were used as predictors. ARIMA models validated statistical significance with p-values, while ANN models, using the Levenberg algorithm, showed better generalization. Model accuracy was assessed using Pearson correlation (R), RMSE, and MAE. The results indicated that the ENSO–IOD ANN models outperformed the ARIMA models based on Pearson correlation (R), RMSE, and MAE metrics. Among them, the Acheron ENSO–IOD ANN model demonstrated the best performance, achieving a Pearson R of 0.90, an RMSE of 0.04, and an MAE of 2.50, making it well-suited for forecasting streamflow 6 months in advance.