An extension for the model IMAZ for large absorption.

Area of Science:

• Space Physics
• Atmospheric Science
• Computational Geophysics

Background:

• The empirical Ionospheric Model of the Auroral Zone (IMAZ) relies on extensive electron density data from EISCAT radar and sounding rockets.
• Neural networks (NNs) are effective for IMAZ predictions within the data's input space but yield unrealistic results for extrapolation.

Purpose of the Study:

• To develop a method for realistic extrapolation of IMAZ predictions beyond the original data's input space.
• To address the limitations of neural network models in predicting ionospheric conditions not well-represented in training data.

Main Methods:

• Utilized over 100,000 electron density profiles from the European Incoherent Scatter Radar (EISCAT).
• Integrated data from approximately 100 sounding rocket measurements.
• Applied a neural network (NN) for modeling and developed an analytical procedure for extrapolation.

Main Results:

• The neural network model provides accurate predictions within the training data range.
• The new analytical procedure allows for reasonable extrapolation of IMAZ predictions to conditions outside the input space.
• Demonstrated improved realism in predictions for under-represented ionospheric conditions.

Conclusions:

• The analytical extrapolation method enhances the reliability and applicability of the IMAZ model.
• This approach overcomes limitations of purely data-driven neural network models for ionospheric research.
• Enables more robust modeling of auroral zone ionospheric variability.