Využití neuronových sítí v meteorologické predikci

Abstract

Thesis deals with nowcasting of precipitation from radar images using artificial neural networks. It presents method for obtaining data from radar images and also means for verification the accuracy of predicted precipitation. Tests are performed on neural networks trained by three approaches: backpropagation, backpropagation through time for recurrent networks and simulated annealing. The aim is to examine learning methods and determine their properties for further research, finding the most suitable parameter settings for learning and also to verify data obtaining method on actual radar images from CZRAD provided by Czech Hydrometeorological Institute.