A Precipitation Classification System Using Vertical Doppler Radar Based on Neural Networks

Abstract

An accurate estimation of precipitation intensity can only be achieved if the particle type of precipitation is identified. In this study, we describe the construction of a system based on neural networks that classifies precipitation particles by type. The system uses Doppler spectra obtained by a vertical pointing Doppler radar as input data and the type of precipitated particle, based on ground observations, as a teacher signal. The reliability of the classification system was evaluated by the system's F-measure after a period time. We show that the F-measure of the classification system can be 0.915, and we confirm that the system classifies precipitation particle types satisfactorily until 0.9–49.1 min after estimation.