A neural network approach to empirically model the lower ionosphere

The ionised part of the mesosphere is also known as the ionospheric D- and E-regions. These parts of the ionosphere crucially depend on the composition of the neutral atmosphere. Hence a realistic climatological model of the lower ionosphere is not only important for radio wave propagation, but today increasingly constitutes one of the references with which the validity of atmospheric models can be assessed. The empirical data required for such modelling efforts are very limited and most of these data were compiled by staff of the Technical University Graz. Hitherto empirical models of electron density, effective recombination rate or negatve ions were successfully established in Graz using conventional fitting methods. The shortcoming of this approach is that one needs to make a-priori assumptions concerning the anticipated variations. In contrast, neural networks do not require assumptions to be made pertaining to the dependencies of the parameters which can be described by a conceivably large number of available input parameters. The aim is to develop new empirical ionospheric models for the lower ionosphere using neural networks. A thorough comparison of the results of the two approaches will be made and an investigation into the residual dependencies hidden in the data will be undergone. The International Reference Ionosphere (IRI) model is the most widely used global ionospheric model. The outcome from this research project will contribute significantly towards improving the current IRI model`s lower ionosphere predictions.