Machine learning models for ionospheric foF2 prediction: A comparative analysis

Abstract

Precise prediction of the ionospheric F2-layer critical frequency (foF2) is crucial for a range of technological applications, including high-frequency radio wave propagation, satellite communications, and global navigation satellite systems (GNSS). This study assesses the efficacy of various machine learning (ML) models, such as Support Vector Machines (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), and Linear Regression (LM), in forecasting foF2, and contrasts their performance with the traditional IRI-2020 empirical model. The analysis was conducted at five ionosonde stations representing both high and low latitudes during the solar minimum (2019) and maximum (2023). The findings indicate that the SVM and RF models surpassed the IRI-2020 model, achieving reduced error rates and enhanced explanatory power. The geographical location and volume of the training data significantly influence the model performance. This study underscores the potential of ML models, particularly SVM and RF, for more precise ionospheric parameter predictions, particularly during periods of elevated solar activity.