L* neural networks from different magnetic field models and their applicability

Abstract

The third adiabatic invariant L* plays an important role in modeling and understanding the radiation belt dynamics. The typical way to numerically calculate the L* value follows the method described by Roederer (1970), which is just a line integration method that is computationally slow and expensive. This work describes the application of an artificial neural network technique to a series of magnetospheric field models for calculating L* values in microseconds instead of seconds without losing significant accuracy, thereby delivering to the radiation belt community various L* neural networks. These neural networks will enable comprehensive solar‐cycle long studies of radiation belt processes and can also help the development of operational radiation belt models because of the speed in calculating L*. The main focus of this work is to test the applicability of each L* neural network, an aspect not addressed in the previous studies, under different interplanetary and magnetospheric conditions. Specifically, we describe the conditions when the neural network is providing a good approximation to the full numerical calculation of L* and when the traditional but more time‐consuming method should be used. These L* neural networks are available for download at http://lanlstar.net.