Monthly and seasonal streamflow forecasts using rainfall-runoff modeling and POAMA predictions

Improved streamflow forecasts a month or season ahead are essential for water resource management and planning. This paper explored the skills of forecasts for monthly and three-monthly total streamflows with a dynamic approach using a conceptual rainfall-runoff model SIMHYD for 31 catchments located in east Australia. For all the catchments, the SIMHYD was calibrated in a moving mode, i.e. using all the data prior to the forecast year. Retrospective forecasts of streamflow totals were generated from 1981 onwards using the calibrated SIMHYD model together with three types of forcings: The observed daily rainfall - this option uses observed (real) rainfall data, the prediction skill of the model reflects the performance of the model in the verification period with real forcings, i.e., the top limit of the skills of the model-based forecasting. The daily rainfall data from all the years prior to the prediction year. For example, if there were 50 years of data available before the prediction year, the model was run 50 times, each with the daily rainfall data from each of the previous 50 year. An ensemble of 50 forecasts of daily streamflow was generated. The donwnscaled rainfall predictions from the Predictive Ocean Atmosphere Model for Australia (POAMA). An ensemble of 11 daily rainfall series for the prediction period from 1985 to 2006 were generated through downscaling POAMA forecasts (10 ensemble forecasts with the ensemble mean) using an analogue approach. The results show that the SIMHYD model was able to capture the rainfall-runoff relationships in majority of months/seasons of studied catchments, once it was properly calibrated. However, the model performance varied in different months/seasons of the year and across catchments. It was relatively poorer in drier period of the year, i.e, winter-spring time in the northern catchments and summer-autumn time in the southern catchments. The dynamic forecasting approach based on conceptual rainfall-runoff modelling provides a potential way to improve streamflow forecasting at monthly and seasonal lead time in east Australia. Using POAMA forecasts as forcing for the rainfall-runoff model improved the forecasting skills as compared to using forcing sampled from history for monthly streamflow forecasts, but not for three-monthly forecasts. Across all the study catchments, use of POAMA ensemble as forcing led to an increase in total number of months with NSE>0.0 by 14% (164 to 187 months with NSE>0.0), but a decrease by 9% (112 to 102 months with NSE>0.0) for three-monthly forecasts. Possible improvements in forecasting skills through further bias-correction approaches are also discussed