A Deep Learning Approach to Dual-Polarization Radar Rainfall Estimation

Traditionally, radar rainfall algorithms are derived through nonlinear regression of rain rates and simulated radar observables from raindrop size distribution (DSD). The performance of such empirical relations is highly dependent on the physical model of DSD and the parametric relation between the physical model and radar parameters. Such algorithms also have large uncertainties that need to be adaptively adjusted based on local DSD properties. In this research, we propose an alternative approach to dual-polarization radar rainfall estimation. In particular, a non-parametric machine learning model is designed and trained using simulated radar data based on DSD measurements in different climatological regimes. The trained model is applied to real radar measurements to produce rain rate estimates. Preliminary results show the promising performance of this novel method compared to traditional parametric rainfall relations.