Ionosphere estimation and space weather prediction using machine learning

Abstract

The precise estimation of Ionospheric TEC plays a pivotal role in optimizing satellite navigation systems, communication networks, and space weather monitoring. This thesis presents a comprehensive exploration of ML-based ionospheric TEC prediction and optimal receiver placement strategies for sparse and dense receiver locations. Focusing on the precise estimation of TEC, which is crucial for optimizing satellite navigation systems, communication networks, and space weather monitoring. This work utilized ANN models to forecast missing TEC data across various latitudes and longitudes in the Indian ionospheric regions, with a specific emphasis on enhancing accuracy in low-latitude areas. Five distinct ANN models were developed, each employing different training and testing strategies. These strategies included random partitioning, regional divisions, and specific considerations near the Equatorial Ionization Anomaly (EIA). The model trained with data from both the northern and southern parts of India (Model 4) emerged as the most accurate, particularly at the EIA, demonstrating exceptional predictive accuracy. This project highlighted the effectiveness of region-specific training strategies for TEC estimation and provided valuable insights into the varying sensitivities of different models under diverse ionospheric conditions.