Short- and Multi-Horizon F10.7 Forecasting Using Elastic Net Regression

The solar radio flux at 10.7 cm (F10.7) is a key proxy used in thermospheric density and atmospheric drag models and is therefore critical for low Earth orbit prediction. Forecasting F10.7 remains challenging due to its strong day-to-day persistence and the difficulty of anticipating rapid increases associated with enhanced solar activity. In this study, we formulate F10.7 forecasting as a multi-horizon regression problem, incorporating time-derivative information and solar-rotation weighting to capture short- and medium-term variability. Forecasts at 1-, 3-, and 7-day lead times are generated using a regularised multi-output Elastic Net model, chosen to stabilise multi-horizon regression in the presence of strongly correlated lagged, derivative, and rotation-weighted predictors. The model is trained on rolling windows and refitted as new observations become available. Performance is evaluated over the period 2020–2025, showing that during low solar activity conditions the model achieves RMSE values as low as ~5 SFU, increasing to approximately 10–12 SFU during higher activity periods. Across rolling test issue dates, the proposed approach consistently outperforms persistence, reducing RMSE by about 15–20%, with the largest improvements occurring at longer lead times.