The use of radar data is a powerful tool to improve rainfall spatio-temporal estimation. Geostatistical techniques are well suited to combine both raingage and radar measurements for this purpose. The main problem of this application, particularly in the context of real time estimation, is the definition of a positive definite model of cross-correlation between radar and rainfall. We propose the direct use of the experimental surface variogram, after filtering the spectra and cross-spectra in the frequency domain to ensure positive definiteness of the model. This technique, which has been proposed in the literature, is suitable for its introduction in a real time forecasting system in which fast estimation of the rainfall spatial distribution is needed. A case study shows its application with a real data set corresponding to the Barcelona radar and its watershed pluviographs.