Reduced Order Emulator of the physics-based for Ionospheric Irregularity Forecasting

First-principles physics-based models of the ionosphere normally provide the highest fidelity representations and are critical to achieving high confidence forecasts. These models require large computational resources that render them ineffective for real time prediction. Training a machine learned (ML) emulator would sacrifice some accuracy but increase the computational speed drastically. Here we propose to develop ML algorithms to emulate the performance of non-linear physics-based models that calculate the onset of instability and subsequent evolution of small-scale irregularities associated with radio frequency scintillations. The inputs for the physics-based models, Cornell's 3D Model and Boston College's Physics Based MODel (PBMOD), are generated by Whole Atmosphere Model - Ionosphere Plasmasphere Electrodynamics (WAM-IPE). The inputs as well as the two models outputs are used to train a model using machine learning emulator algorithms. The input and output space are extremely large and require some feature and output engineering to reduce the size of the model. This presentation describes determination of the inputs and outputs and the machine learning methods to be used in training the model. The work directly supports The SPAce Weather Research and Technology Applications (SPARTA) SWx-C, a multi-institutional, NASA-sponsored Center established to develop advanced capabilities to forecast scintillation supporting meaningful operational products responsive to user needs.