Artificial Neural Network–based prediction of D-region electron density variations during solar flares using ground-based VLF observations over low-latitude Indian region

Abstract

The ionospheric D-region (60–90 km) critically controls very low frequency (VLF) radio wave propagation and responds rapidly to solar flare radiation. Accurate prediction of D-region electron density is therefore essential for reliable space weather nowcasting. We develop an artificial neural network (ANN) framework to predict flare-induced electron density enhancements using ground-based VLF observations from Dehradun and Indore, India (2020–2025). A total of 244 solar flares (157 C-class, 87 M-class) were analyzed. Seven physically motivated inputs: time of day, day of year, VLF amplitude perturbation, reflection height (H′), sharpness factor (β), F10.7 flux, and GOES X-ray flux were used to train a feed-forward backpropagation network. The model achieved excellent skill for C-class flares (R = 0.98

MSE = 8.735 × 10−5

RMSE = 0.0295

and MAE = 0.0182) and moderate performance for M-class flares (R = 0.74

MSE = 4.06 × 10−1

RMSE = 0.63

MAE = 0.47), reflecting increased ionospheric variability under stronger solar forcing. These results demonstrate the robustness of ANN-based approaches for real-time D-region electron density estimation and operational space weather forecasting. © The Author(s), under exclusive licence to Springer Nature B.V. 2026.