To evaluate the uncertainty of a real-time river discharge prediction with a distributed rainfall-runoff model, an error structure of real-time rainfall prediction by a translation model is modeled as a spatial random field and predicted rainfall fields are simulated according to the characteristics of the prediction error structure. Characteristics of error structures of real-time predicted rainfall by a translation model are statistically analyzed, and a relative prediction error field is found to be modeled using a lognormal spatial random field. Then a method to generate rainfall fields with the uncertainty having lognormal spatial random field characteristics is realized with a factorization technique of covariance matrix decomposition into its square root matrix approximately by using the Chebyshev polynomials. The generated rainfall fields will be used to evaluate the uncertainty of real-time river discharge predictions with a distributed rainfall-runoff model on a Monte Carlo simulation framework.