3D radar data downscaling

High-resolution precipitation fields are needed for many applications. The traditional ways for obtaining them are using pure stochastic rainfall models, or using real data measured by an instrument as a base of a downscaling process. Synthesizing rainfall fields with higher resolution than observed, reproducing the variability of the rain at all scales, is a challenge due to the complexities of the rain. Here, a 3D downscaling technique for radar data, based on modeling the precipitation fields using multifractals, is proposed. The technique is based on three independent steps: the first one consists on downscaling the first PPI observed by the radar applying two techniques (one based on wavelet analysis and the other based on the “String of Beads” model). The second step consists on downscaling the rest of PPIs through a homotopy of the observed vertical profiles of reflectivity [VPR], and the last one is the transformation of the polar values to the requested Cartesian grid. In the wavelet technique, radar data is decomposed using the Haar wavelet base. Then the fluctuation wavelet components normalized by the scaling component are considered to be Gaussian distributed. It is observed that a scaling law can be adjusted to the standard deviation of these distributions. Therefore, the variability of the high-resolution scales can be obtained through extrapolation. Finally, applying inverse wavelet transform, this variability is introduced when generating the new scales. On the other hand, the String of Beads model assumes the radar rainfall to be lognormal distributed. The two parameters of this distribution and the slope of the Fourier power spectrum are extrapolated to the new scales. Comparisons of those two models will be shown. The proposed VPR-homotopy process to downscale the upper elevations conserves the original observed profiles, and allows to create complete artificial elevations form the original data (for example between the real elevations, to increase the vertical density), as well as obtaining CAPPIs. The Introduction of a random component in the homotopy process is under investigation. The final interpolation from polar coordinates to Cartesian coordinates is done applying closest neighbor interpolation algorithm to preserve the extreme values generated in the downscaling process. Other interpolation techniques will be investigated in future. The full downscaling process including some examples will be presented.