Predicting return stroke peak currents from high-speed lightning videos

We present a linear regression analysis using total-frame-luminosity derived features from high-speed video recordings to predict return-stroke peak current in negative cloud-to-ground lightning. High-speed video sequences (15,000 fps) of 265 first-stroke events were processed to obtain summed-frame Digital Numbers (DNs; luminosity). For each sample, each frame was background-subtracted to isolate lightning emission and scaled by the square of the camera–stroke distance to compensate for geometric intensity fall-off. Corresponding peak-current values, obtained from the South African Lightning Detection Network, were linear regressed against the derived luminosity features. The resulting model explained 48% of the variance ( R 2 ) in peak current with a mean absolute error of 6.11 kA (MARE ≈ 32.7%) between fitted and observed values. Forward stepwise regression identified the return-stroke luminosity decay rate as the dominant predictor, with stepped-leader duration providing the largest incremental gain in explained variance when added to the model. Return-stroke decay magnitude and decay rate each independently explain more than one-third of the variance and are strongly correlated with each other, indicating that peak current is coherently encoded in multiple decay-phase luminosity metrics arising from shared post-attachment channel thermodynamics. These findings demonstrate that total-frame-luminosity derived features provide a quantitative optical basis for estimating return-stroke strength.