The Bureau of Meteorology is soon to release a new ensemble streamflow forecasting service, jointly developed with CSIRO. The service relies on calibrated Numerical Weather Prediction (NWP) precipitation forecasts to force an initialised hydrological model and an error model. The calibration of NWP forecasts downscales NWP forecasts to the spatial scale required by the hydrological model, reduces NWP forecast errors and biases, and quantifies uncertainty in rainfall predictions in the form of an ensemble. The service has largely been tested using the Bureau's ACCESS-G NWP as forcing. ACCESS-G has the benefit of long lead times (240 h), but trades this off with relatively coarse spatial resolution (~40 km). In this study, we assess the benefits of using the more finely resolved ACCESS-R (~12 km) and ACCESS-C (~1.5 km) NWP models.

Despite the increase in resolution, we find that all NWP models produce similarly accurate streamflow forecasts in all catchments. Increased NWP resolution reduces biases in NWP predictions, particularly in mountainous regions. However, statistical calibration is also extremely effective at removing biases, obviating this advantage in high resolution NWP models.  Further, the rivers used in this study attenuate rainstorms, as catchment water stores (soils, etc.) fill and slowly discharge water. Forecasts from the ACCESS-R and ACCESS-C models are only available to lead times of 72 h and 36 h, respectively. This means that improvements in the temporal resolution of rainfall forecasts do not necessarily improve the simulation of streamflow events, even for large floods, because the hydrological response to rainstorms can be >36 h (and in some cases, >72 h). We note, however, that in catchments where streamflow is more responsive to rainfall (e.g., urban areas), the benefits of high-resolution NWP models for streamflow forecasting are likely to be more evident.