The application of spectral analysis and artificial intelligence methods to weather radar

Power spectral density (PSD) of reflectivity and polarimetric variables have the potential to provide the linkage between the dynamics and the microphysical properties of scatterers within the radar resolution volume. The artificial intelligence (AI) methods such as fuzzy logic and neural network have been widely used in weather radar. The main goal of this dissertation is to exploit spectral analysis and AI methods to the two specific areas of tornado detection and the retrieval of microphysical properties of rain-hail mixture.A novel approach of using both fuzzy logic and neural network, termed neuro-fuzzy tornado detection algorithm (NFTDA), is developed to integrate tornado's shear, spectral and polarimetric signatures for both regular resolution and high resolution with the goal of enhanced and robust detection. The spectral signatures are characterized by spectrum width and three additional parameters derived from the analysis of bispectrum, statistics, and Eigen-ratio. The statistical analysis from numerical simulation and real data has shown that NFTDA provides improved detection compared to the conventional shear-based detection algorithm in terms of probability of detection (POD), false alarm rate (FAR), and detection range. For the retrieval problem, a model of Doppler and polarimetric spectra is first developed for the presence of both raindrops and melting hail. The melting ratio is introduced the first time in the retrieval using weather radar. A genetic algorithm (GA) is introduced to solve the optimization of fitting the observed Doppler and polarimetric spectra to the model. Consequently, the drop size distribution (DSD) of both rain and hail, the melting ratio, the radial component of ambient wind and spectrum broadening can be retrieved. The retrieval algorithm is demonstrated and tested using numerical simulations.