Quantitative forecasting of near-term solar activity and upper atmospheric density

Autoregressive algorithms are developed to forecast solar activity on timescales of 1 to 10 days and utilized to forecast upper atmospheric densities using the NRLMSIS density specification model. Quantitative assessment of solar activity observations and forecasts made over 27 years (from 1980 to 2006) indicates that the chromospheric Mg index is superior to the coronal F10.7 radio flux, both as a proxy for the day‐to‐day EUV irradiance variations that drive density changes and as an input to empirical models for density forecasts. For 1‐ to 10‐day Mg forecasts, the average of the root‐mean‐square error (evaluated in 81‐day windows and averaged over the 27 years) increases from 3% to 13%; for F10.7, the corresponding forecast uncertainty increases from 5% to 20%. We demonstrate how the use of Mg instead of F10.7 reduces the errors in forecasting upper atmospheric density changes by comparing orbit‐derived and forecast density changes along the tracks of two low earth‐orbiting objects. For the Yohkoh orbit, the 3‐day Mg and F10.7 forecasts have average root‐mean‐square errors of 29% and 30%, respectively. For the 10‐day solar activity forecasts, the errors are 37% and 41%. Although the improvement using Mg is evident, density uncertainties arising from errors in the solar activity forecasts are on average three to five times smaller than the uncertainties in the combined NRLMSIS density specification model and density data. To improve density forecasts, better characterization of the upper atmospheric response to solar and other drivers is needed.