We use a multifractal characterization of two heavy rainfall events in the London area to quantify the uncertainty associated to the rainfall variability at scales smaller than the usual C-band radar resolution (1 km2 x 5 min.) and how its transfer to sewer discharge forecasts. The radar data are downscaled to a higher resolution with the help of a multifractal cascade whose exponent values correspond to the estimates obtained from the radar data. A hundred of downscaled realizations are thus obtained and input into a semi-distributed urban hydrological model. Both probability distributions of the extremes are shown to follow a power-law, which corresponds to a rather high dispersion of the results, therefore to a large uncertainty. We also discuss the relationship between the respective exponents. In conclusion, we emphasize the corresponding gain obtained by higher resolution radar data.