Can a daily rainfall-runoff model be successfully calibrated to monthly streamflow data?

Abstract

Conceptual hydrological models that predict streamflow at daily time steps are widely used in water forecasting, water resources planning and operations. Typically, these models are calibrated using daily observed streamflow data. However, in several important practical applications, including the prediction of inflows into large dams, only monthly streamflow estimates are often available for model calibration. Development of robust approaches for calibrating daily rainfall-runoff models to monthly streamflow data is hence of major practical interest. This study assess the calibration of a daily hydrological model (GR4J) to monthly streamflow data and compares the resulting performance to the performance attained after calibration to daily streamflow data. Multiple performance metrics are used: fit of the daily and monthly flow duration curve, daily and monthly pattern metrics, and longterm bias. The analysis is carried for 508 Australian catchments and two evaluation periods. It is found that monthly calibration performs similar or better than daily calibration in a majority of sites and periods in terms of bias and fit of the daily flow duration curve. However, performance of monthly calibration is worse than daily calibration for daily pattern metrics such as Nash-Sutcliffe efficiency in a majority of sites and periods. This performance loss can be reduced significantly by using regionalised values for the flowtiming parameter of GR4J. Similar results are obtained for other pattern metrics. Overall, our findings suggest that monthly calibration of rainfall-runoff models to dailyrainfall/ monthly-streamflow is a viable alternative to daily calibration when no daily streamflow data are available.