Towards the development of a new global foF2 empirical model using neural networks

In this work, the use of neural networks (NN) for predicting F2-layer critical frequencies at any location and time has been tested in order to develop a global foF2 empirical model. Over the years a large number of global, regional and station-specific models have been developed to predict the value of F2-layer critical frequency with a set of special variables limited to geographic coordinates, universal time (UT), sunspot number (R12), and a modified-dip latitude (Modip). Models that use geographic coordinates as their basis have the problem that ionospheric data is not available for the vast areas occupied by the oceans. This paper investigates a model based on parameters other than geographic coordinates, on which the ionosphere is known to depend, and which are more evenly spread over the available data gridpoints. These parameters are: solar zenith angle (x), magnetic activity, angle of meridian relative to subsolar point (Local Time), magnetic dip angle (I) and angle of declination (D). Data from 31 stations across the globe based on availability are used to train a neural network. The predicted values of foF2 from this network are compared with observed and IRI model values for five selected stations that were not part of the stations used for training. The root mean square (RMS) error difference between the two models and measured station data were calculated. The NN model compares favourably with the IRI model, based on RMS error.