Detection of rain/no rain condition on ground from radar data using a Kohonen neural network

A Kohonen self-organizing feature mapping (SOFM) neural network based scheme for rain/no rain classification on the ground using radar data is described. Vertical reflectivity profiles of radar observations are used as input variables to the rain/no-rain classifier. “Winner take all” unsupervised learning algorithm is used by the Kohonen neural network during the training process. Ground raingage measurements corresponding to the input data are used to label the class of each neuron in the output layer of the network as rain or no-rain by a voting scheme. There will be only one winning neuron when a vertical reflectivity profile of radar observation is applied to the classifier. If the winner is a rain neuron, the corresponding ground condition is classified as raining, otherwise, as no-rain. This rain/no-rain classifier has been applied to classify the radar data collected by the Melbourne NEXRAD system, Florida, and raingage measurements from the TRMM raingage networks located at the same area were used to validate the classification results. Experiment results are presented in this paper