Advanced Doppler Polarimetric Weather Radar Simulation and Processing

Doppler-polarimetric weather radars represent the state of the art in the measurement and forecasting of atmospheric phenomena. Polarimetric features have been largely used to infer the bulk hydrometeor types and bulk amounts. However, the absence of an exact ground-truth for the validation of the related algorithms is a strong limitation. Simulators provide a controlled environment in which develop and assess the performance of novel processing tools. For this reason, several approaches for the simulation of weather radar-like signals have been developed for the last forty years. Several European projects foresee the use of X-band Doppler-polarimetric avionic weather radars for the optimization of the flight trajectories, increasing environmental performances and safety of civilian aircraft. Unfortunately, this type of sensors is not available on the market yet thus, the simulation of such systems is an indispensable approach. The novel POWER simulator, which is based on the jointly use of the Weather Research and Forecast weather prediction model together with the T-Matrix electromagnetic code, is presented in this thesis. Models for raindrops, hailstones, graupel, snow aggregates and mixtures of them were developed at S, C and X band and are here shown. The POWER simulator follows a statistical approach for generating I&Q time series as correlated Gaussian random processes according to the propagation-modified polarimetric covariance matrix. Doppler effects are modelled with Gaussian shaped spectra. In addition, realistic land and sea clutter radar returns are generated. The POWER simulator can reproduce ground-based or avionic weather radars, for which all the radar subsystems were modelled and are arbitrary configurable by the user. A complete signal processing chain was developed and performance assessment is shown. Incoherent polarimetric decomposition was used to develop a novel feature set which outperformed the traditionally used features in terms of class separability power. A large database generated by the POWER simulator was used to preliminary validate the simulator and results show a good agreement with previous studies. Future developments will address the refinement of all the proposed models, together with the validation of the POWER simulator and of the signal processor on real data.