Improvement in short-term streamflow forecasting using an integrated modelling framework

Abstract

Short-term streamflow forecasting is a crucial element in any regulated river system operation. The multiple benefits from an improved streamflow forecasting capability include: (i) an enhanced ability to predict the volumes and timing of flood events, (ii) improved water use efficiency through better anticipation of river inflows (particularly associated with inflows from unregulated tributaries), (iii) a concomitant reduction in operational losses due to over releases from water storages, and (iv) fewer shortfalls in supplying water orders. Over the past few decades, many numerical streamflow prediction techniques using observed time series (TS) have been developed and widely used in water resources planning and management. Recent advances in quantitative rainfall forecasting by numerical weather prediction (NWP) models have made it possible to produce improved streamflow forecasts using continuous rainfall-runoff (RR) models. In the absence of a suitable integrated system of NWP, RR and river system models, river operators in Australia mostly use spreadsheet-based tools to forecast streamflow using gauged records. The eWater Cooperative Research Centre of Australia has recently developed a new generation software package called Source Integrated Modelling System (Source IMS), which allows a seamless integration of continuous RR and river system models for operational and planning purposes. A study was undertaken using Source IMS for a comparative evaluation of streamflow forecasting methods on a regulated section of the Murray River, a major stream in the Murray-Darling Basin, Australia. The methods include three TS based linear techniques and, RR models of two selected unregulated sub-basins that drain into the river reach. The results were compared using three statistical indicators with the actual forecasts made by the Murray river operators and the observed data. The results show that while streamflow forecasts by the river operators were reasonably accurate up to day 3 and traditional TS based approaches were reasonably accurate up to 2 days, well calibrated RR models can provide better forecasts for longer periods when using high quality quantitative precipitation forecasts. The river operators tended to underestimate large magnitude flows. This paper presents the outcomes from the case study. The uncertainty associated with forecasts using different datasets and methods and, potential operational benefits of the short- term streamflow forecasts are also discussed.