Where Does Ionospheric Predictability Reside? Univariate, Multivariate, and Cross-Station Forecasting of Ionosonde Time Series

Understanding the predictability of ionospheric variability is important for both scientific interpretation and operational forecasting. In this work, we investigate short-term forecasting of Digisonde-derived ionospheric parameters, including foF2, hmF2, and h′F2, using open-source Python implementations of both statistical and machine-learning time-series models. We compare univariate forecasts of each parameter against multivariate forecasts that jointly model their coupled evolution, with the goal of quantifying differences in information content among observables. We also examine how forecast skill depends on station location and evaluate the extent to which time series from one site improve predictions at other sites, particularly along regional longitudinal chains. Finally, we assess the impact of adding exogenous drivers, including solar radiation proxies, solar wind parameters, and geomagnetic indices. This work aims to identify where predictive information is found and how it changes across geophysical conditions by using forecasting skill as a way to study ionospheric coupling and regional coherence.