Medium-term storage volume prediction for optimum reservoir management: A hybrid data-driven approach

Abstract

A hybrid regressive and probabilistic model was developed that is able to forecast, six weeks ahead, the storage volume of Little Nerang dam. This is a small elevated Australian drinking water reservoir, gravity-fed to a nearby water treatment plant while a lower second main water supply source (Hinze dam) requires considerable pumping. The model applies a Monte Carlo approach combined with nonlinear threshold autoregressive models using the seasonal streamflow forecasts from the Bureau of Meteorology as input and it was validated over different historical conditions. Treatment operators can use the model for quantifying depletion rates and spill likelihood for the forthcoming six weeks, based on the seasonal climatic conditions and different intake scenarios. Greater utilization of the Little Nerang reservoir source means a reduced supply requirement from the Hinze dam source that needs considerable energy costs for pumping, leading to a lower cost water supply solution for the region.