Answers are grounded in this paper and cite their sources.
Lightning radiates electromagnetic energy over a very wide frequency range, and the low-frequency component travels to global distances inside the Earth–ionosphere waveguide. How far such a signal reaches, and the highest frequency a given path will carry, are set by the state of the ionosphere at the time of the discharge, which varies by the hour. We forecast that propagation environment over lightning-active regions and convert the forecast into the quantities a radio link actually needs: the F2-layer critical frequency (foF2), the maximum usable frequency (MUF), and the one-hop ground range. A central methodological point is addressed explicitly. Deriving foF2 and MUF from Total Electron Content (TEC) and then training a model to predict them from TEC is circular because the label is a closed-form function of an input feature; any model then recovers R² ≈ 1.0 while learning nothing. We therefore pose the problem as a forecast of TEC at t + ∆t, which is genuinely unobservable at t, and apply the physics to the forecast output rather than to the label. Using 381,858 ISS-LIS flashes, 17,665 fifteen-minute global ionosphere maps, and 8,760 hourly solar and geomagnetic records spanning March–August 2018, five models are compared against a persistence benchmark at 1-, 2-, and 6-hour lead times on a strictly chronological split. All five beat persistence at every lead time. Histogram gradient boosting is the best-performing model, reaching RMSE 3.15 TECU, R² = 0.68, and skill = +0.68 at six hours, where persistence attains R² = −2.11. Skill increases with lead time because persistence decays across the diurnal cycle while solar geometry does not. Feature importance shows the model relying on the current ionospheric state at short lead times and switching to solar zenith angle at the target epoch at long lead times, recovering the dominant physical driver without supervision. A case-by-case analysis qualifies the headline figure: at one-hour lead, the model wins only 51.9% of individual cases and acts mainly as a variance-reducer. Lightning activity was found not to predict TEC, and we report that negative result rather than omit it.
No takes yet. Share an insight, caveat, or question.
Alam et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: