Forecasting Geomagnetic Disturbances in Solar Cycle 25 for Space Weather Prediction

Accurate forecasting of geomagnetic disturbances is essential for mitigating space weather effects on critical infrastructure. We present an optimized Long Short-term Memory neural network (LSTM+), refined through a reforecast procedure and hyperparameter tuning using data from Solar Cycle 24, achieving high predictive skill. For Solar Cycle 24, the model demonstrates a Nash–Sutcliffe efficiency of 0.95 for Ap and 0.78 for Dst, with Pearson correlation coefficients of 0.983 and 0.901, respectively. Furthermore, the error rate for predicting the peak Ap and trough Dst values was low, at 4.16% and 7.77%, respectively. This LSTM+ model is used to predict the Ap and Dst indices for Solar Cycle 25. The model forecasts a peak in Ap around 2026 January and a trough in Dst around 2026 February, consistent with the observed lag of up to 3 yr between geomagnetic indices and the sunspot number maximum in prior cycles. Validation against historical data confirms the model’s reliability in capturing these temporal offsets. These projections offer a predictive framework for improving space weather preparedness and safeguarding technological systems.