Evaluation of downscaled POAMA M24 for monthly and 3-monthly streamflow forecasts

Abstract

This paper evaluates skill of ensemble forecasts of monthly and three-monthly streamflows for three catchments from different hydrological regions, using a conceptual hydrological model - the GR4J. The latest available POAMA-M24 rainfall predictions for the period of 1980-2008 are downscaled and used as forcing inputs of the model to produce streamflow forecasts. In dealing with model uncertainty, 200 parameter sets derived through BATEA are used for each downscaled rainfall forcing. The results show that skill scores are both catchment and season dependent. In the Biggara catchment (SMD region), Jan., Apr., and Nov. are the months with the highest skill scores; for the Picnic Crossing catchment (QLD region), the best forecasts are for June and July, while for the Tinderry catchment (SEC region), the best forecast months are Sept. to Nov. and Feb. as well. The skill scores of monthly steramflow forecasts are higher than that of three-monthly forecasts except for reliability. Slight difference between M24-E33 and M24-E99 is found when additional forcings with lead times of 1 and 2 months are used in the ensemble forecasting.