One‐ to 10‐Day Forecasts of Ionospheric Total Electron Content Using a Statistical Model

A new, rapid, nonassimilative technique is demonstrated for forecasting the ionosphere's vertical total electron content (TEC) on time scales longer than 1 day. The approach uses a statistical model constructed by regressing solar extreme ultraviolet irradiance and seasonal, diurnal, and geomagnetic predictors at multiple lags against the 2‐hourly International Global Navigation Satellite Systems Service observations, with a formulation that accounts for solar modulation of the seasonal oscillations and solar and seasonal modulation of the diurnal oscillations. Solar irradiance inputs to the statistical model are forecast at successive 2‐hr intervals from 1999 to 2015 using an autoregressive model of irradiance variations during the prior 100 days. As the forecast time increment increases from 1 to 10 days, the average over the globe of the mean absolute error of TEC observations and the forecasts increases from 2.5 to 3.2 TECU (total electron content unit, 1 TECU = 10 16 el/m 2 ); the root‐mean‐square error increases from 3.7 to 4.8 TECU. Averaged over the equatorial ionization anomaly region (30°S–30°N) the mean absolute error of the forecasts increases from 3.2 to 4.3 TECU and the root‐mean‐square error increases from 4.6 to 6.4 TECU. The skill of the TEC forecasts at time increments of 3, 5, and 8 days ahead exceeds persistence by 9%, 13%, and 15% and climatology by 9%, 12%, and 10%, respectively. Forecast skill is higher in April than in July. Long‐range, multiyear forecasts from 2018 to 2030 are demonstrated based on current expectations that solar activity in cycle 25 will be comparable to that in cycle 24.