MLR analysis for forecasting seasonal streamflow using large scale climate drivers: a case study for NSW

High inter-annual variability of streamflow resulting from the extensive topographic variation and climatic inconsistency cause immense difficulties to the water planners and managers of New south Wales which is one of the major contributors of Australia’s agricultural production. Therefore, in this study an attempt is made to develop a skillful seasonal streamflow forecast method considering two major influential SST (Sea Surface Temperature) anomalies of north-east New South Wales; NINO3.4 and Pacific Decadal Oscillation (PDO) as predictors. Single lagged correlation analysis is performed to identify their individual interactions with spring streamflow till nine lagged months and this is exploited as the basis for developing Multiple Linear Regression (MLR) models to examine the extent of the combined impact of these two climate drivers on forecasting spring streamflow several months ahead. Several research works were carried out to forecast streamflow and rainfall for different parts of Australia using the climate indices as potential predictors but none of those apply the Multiple Regression analysis to explore the combined impact of climate indices on long lead seasonal streamflow forecast for New South Wales. Three streamflow stations (Hunter River at Singleton, Goulburn River at Coggan and Namoi River at North Cuerindi) from north-east New South Wales are selected as a case study based on their recorded data length with fewer missing values. The developed models with all the possible combinations show significantly good results in terms of Pearson correlation(r), Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Willmott index of agreement (d) where the best models with lower errors give statistically significant correlations as high as 0.71 for Hunter River station, 0.72 for Goulburn River station and 0.69 for the Namoi River Station. It is evident that every time the combined model outperforms the model considering single climate variable in terms of Pearson correlation(r) which ascertains the better predictive skills of MLR models to forecast spring streamflow several months ahead for the study region.