130 Weather Radar Information and Distributed Hydrological Modelling (Proceedings of symposium ITS03 held during 1UGG2003 at Sapporo. July 2003). IAHS Publ. no. 282, 2003. The effects of radar-derived rainfall uncertainties on forecasts from a distributed hydrological model JONATHAN J. GOURLEY & BAXTER E. VIEUX School of Civil Engineering and Environmental Science, University of Oklahoma, Norman, Oklahoma 73019, USA Jonathan,gourlev(S'nssl. noaa.gov Abstract The advent of weather radar has provided the potential to estimate rainfall accurately at high spatial and temporal resolutions. When these estimates are input to a distributed hydrological model, forecasts of streamflow may be used to anticipate, and thus mitigate, the potential hazards associated with a flash flood. In hydrological modelling, forecast uncertainty has traditionally been a function of the uncertainty in the model parameters, and in some cases the model structure. The study presented here uses the physics-based, distributed r. water.feci hydrological model with modifications to address the impact of uncertainties in the input rainfall estimates on streamflow predictions using an extension of the Generalized Likelihood Uncertainty Estimation methodology. The ensemble modelling approach allows us to evaluate the accuracy of different rainfall algorithms independently at the scale of an integrating watershed. The sUidy plan and some initial results are presented. Key words distributed hydrological model; ensemble hydrological prediction; parameter estimation; radar quantitative precipitation estimation; rainfall uncertainty INTRODUCTION It is well known that the accuracy of streamflow predictions from a hydrological model is heavily dependent on the accuracy of the rainfall inputs. Several efforts are underway (e.g. Gourley et ai, 2002) to improve quantitative precipitation estimation (QPE) by understanding the situations in which radar estimates can be erroneous and utilizing data from multiple sensors (e.g. infrared satellite, raingauges, numerical weather model output, and lightning flashes). As QPE algorithms are being formulated from emerging radar technologies such as polarization diversity, it is vital to the developers to know the error characteristics associated with the estimates. Tradit- ionally, this has been done by comparing the remotely-sensed QPEs to raingauges at collocated grid points. In addition to the measurement errors associated with raingauges, it has been noted that the sampling sizes between a typical radar pixel and a raingauge orifice differ by about eight orders of magnitude (Droegemeier et al, 2000). A methodology is proposed herein that provides the framework to evaluate potential QPE improvements at the scale of application, a watershed. In addition, the probabilistic approach enables uncertainty estimates or confidence intervals to be assigned to the predicted hydrological variables.