Rapid Identification and Spectral Moment Estimation of Non-Gaussian Weather Radar Signal

Doppler weather radar observations of strong convective weather phenomena, e.g., tornadoes and supercells, showed that their Doppler spectra could deviate from the Gaussian shape. Classical method for identification of non-Gaussian spectral signal (NGSS) is based on spectral width and high-order spectral moments (HOSM). The DFT processing is needed to obtain the power spectrum density to calculate HOSMs, increasing the computation complexity. Besides, power, mean velocity, and spectral width are often obtained using autocorrelation method, i.e., pulse-pair-processing (PPP) method which is generally considered to be the most efficient estimator, where a Gaussian spectrum is assumed. Thus, a bias in the mean velocity and spectral width will occur if the Doppler spectrum deviates from the Gaussian shape. In this article, a generalized PPP (GPPP) method, which calculates HOSMs using autocorrelation function directly, is proposed. It outputs two additional identification parameter products, including skewness and kurtosis, which can directly identify NGSS and contribute to select the appropriate spectral moment estimation algorithm for Gaussian-or-not spectral signal. Compared with the classical method, the NGSS identification based on the GPPP method has a lower computation complexity since the DFT processing is avoided and has a better performance especially for low signal-to-noise ratio conditions. Besides, the clustering algorithms including expectation maximization and K-means algorithms are investigated for the adaptive spectral moment estimation of NGSS for the first time. The numerical simulation experiments and verifications based on real weather radar data of an actual supercell are implemented to confirm the feasibility and superiority of proposed method.