There is a clear and pressing need for improvements in our ability to generate operational flash flood forecasts in the semiarid southwestern United States and other dry regions around the world. This study investigated the predictive uncertainty of the physically based, distributed semiarid rainfall-runoff model KINEROS2 driven by high-resolution radar rainfall input for this purpose. Uncertainty sources considered were rainfall estimates, model parameters, and initial moisture conditions. The variance-based Sobol' global sensitivity analysis method was used to investigate dominant sources of uncertainty. The flash flood forecasting system was implemented using the GLUE methodology to facilitate operational assimilation of incoming event information. The approach is applicable to any model amenable to a Monte Carlo framework, and can be implemented operationally for a fixed number of factor sets without requiring resampling or optimization. The Monte Carlo framework utilized factors applied to the rainfall, initial soil moisture and model parameters.