Classification and quantification of hydrometeors based on polarimetric radar measurements: development of fuzzy logic and neuro-fuzzy systems and in-situ verification

Abstract

One of the main applications of polarimetric radar is to retrieve the characteristic information about the hydrometeors in the radar resolution volume. The information extraction from polarimetric radar data includes identification of hydrometeor type, quantitative estimation of precipitating hydrometeors, and forecast of hydrometeor evolution. Some Artificial Intelligent (AI) methods, such as fuzzy logic and Neural Network techniques, are proposed in this research to address the issues of hydrometeor classification and quantification. Fuzzy logic and Neuro-Fuzzy systems for the classification of hydrometeor type based on polarimetric radar measurements is developed. The hydrometeor classification system is implemented by using fuzzy logic and neural network, where the fuzzy logic is used to infer hydrometeor type, and the neural network learning algorithm is used for automatic adjustment of the parameters of the fuzzy sets in the fuzzy logic system. Five radar measurements, namely, horizontal reflectivity (ZH), differential reflectivity (ZDR), differential propagation phase shift (KDP), correlation coefficient (PHV(0)), and linear depolarization ration (LDR), and corresponding altitude have been used as input variables to the hydrometeor classifier. The output is one of the many possible hydrometeor types, namely 1) drizzle, 2) rain, 3) dry and low density snow, 4) dry and high density crystals, 5) wet and melting snow, 6) dry graupel, 7) wet graupel, 8) small hail, 9) large hail, and 10) mixture of rain and hail. The Neuro-Fuzzy classifier is more advantageous than a simple Neural Network of a fuzzy logic classifier because it is transparent rather than a "black box" (unlike a neural network), and can learn the parameters of the system from the past data (unlike a fuzzy logic system). The neuro-Fuzzy hydrometeor classifier has been applied to several case studies and the results are compared against in-situ observations. An adaptive neural network scheme for quantitative precipitation estimation is developed in this research. The neural network is a non-parameteric method for representing the relationship between radar measurements and rainfall rate. The relationship is derived directly from a data set consisting of radar measurements and raingage measurements. The effectiveness of rainfall estimation by using a neural network can be influenced by many factors, such as the representativeness and sufficiency of the training data set, the generalization capability of the network to new data, season change or location change, etc. To achieve the best performance, the neural network may have to be refined to accommodate changes such as change of seasons. In this study, a novel scheme of adaptively updating the structure and parameters of the neural network for rainfall estimation is presented. This adaptive neural network scheme enables the network to implement the non-stationary relationship between radar measurements and precipitation estimation with change of season and other environment conditions, and also can incorporate new information, without re-training the complete network from the beginning. This precipitation estimation scheme is a good compromise to the dilemma of accuracy and generalization. Data collected by a WSR-88D radar and network of raingages were used to evaluate the performance of the adaptive neural network for rainfall estimation. It was shown that the adaptive neural network can reach the same estimation accuracy compared with the neural network which is trained with all the available date, but the implementation of the network is much faster, more efficient and convenient for real time rainfall estimation to be used with WSR-88D. Another important issue for the application of radar rainfall algorithm addressed in the dissertation is the detection of rain/no-rain condition on the ground. A Radial basis Function (RBF) neural network based scheme for the rain/no-rain determination of the ground using radar data is described in this research. Vertical reflectivity profiles of radar observations are used as input variables to the rain/no-rain determination. Radar data and ground raingage measurements are used to train the neural network. This rain/no-rain classifier is evaluated using the radar data collected by the WSR-88D radar over central Florida for two different years. Results indicate that rain/no-rain condition on the ground can be inferred from the procedure developed in this paper fairly accurately. It is shown that by using rain/no-rain classification scheme (prior to radar rainfall estimation) the accuracy of rainfall accumulation estimates can be improved greatly.