Forecasting Spread F at Jicamarca

Equatorial Spread F is a phenomenon that occurs in the F layer of the ionosphere and is characterized by plasma depletions. It is governed by highly non-linear mechanisms and has been the object of study for many years, but it has not yet been fully understood. Moreover, it is of interest to predict its occurrence because of the negative impact it has on communications and navigation systems.

In this work, we present a data-driven approach to forecast Spread F occurrence by harvesting the data that has been collected from 2000 to 2020 at Jicamarca. We processed a number of geophysical parameters and JULIA radar’s Signal-to-Noise Ratio measurements to produce a dataset which was later used to train a neural network that makes occurrence predictions on a daily basis. The architecture proposed is the result of a number of trials conducted with the aid of Optuna. Furthermore, we estimated the impact of the different geophysical parameters on the model’s predictions using SHAP values. Our model, which obtained accuracy of 81%, was compared to the Forecasting Ionospheric Real-time Scintillation Tool (FIRST).