Monthly median foF2 modelling COST 251 area by neural networks

The use of a neural network to model the monthly median ionospheric foF2 fr equencies has been tested in order to establish a new long-term prediction procedure to support ionospheric radiowave propagation at frequencies above 2 MHz. The neural networks (NN) have been trained with the foF2 measured d ata from the European ionospheric stations in three separate cases: (i) a s ingle station model at Poitiers (46 degrees 0 N, 00 degrees 0 E) build with the classical multi-layer perceptron (MLP) with 3 inputs: hour, month and solar activity index; (ii) a modular neural network with the same inputs; ( iii) a 2D model build over Europe with additional inputs: geographical lati tude and longitude. In this last case, the problem of the spatial interpola tions between ionospheric stations is also studied. The results are compare d with those of the classical PRIME and ITU-R models. (C) 1999 Elsevier Sci ence Ltd. All rights reserved.