Forecasting Continuum Intensity for Solar Active Region Emergence Prediction Using Transformers

Early and accurate prediction of solar active region (AR) emergence is crucial for space weather forecasting. Building on established Long Short-Term Memory (LSTM) based approaches for forecasting the continuum intensity decrease associated with AR emergence, this work expands the modeling with new architectures and targets. We investigate a sliding-window Transformer architecture to forecast continuum intensity evolution up to 12 hr ahead using data from 46 ARs observed by Solar Dynamics Observatory/Helioseismic and Magnetic Imager. We conduct a systematic ablation study to evaluate two key components: (a) the inclusion of a Conv1D front-end and (b) a novel Early Detection architecture featuring attention biases and a timing-aware loss function. EarlyDetect, our best-performing model, uses the Early Detection architecture without the Conv1D front-end and achieves an RMSE of 0.1189 (10.6% lower than the LSTM baseline) and, on average, captures emergence onset 9.24 hr in advance, even under a stricter emergence criterion than previous studies. While EarlyDetect demonstrates superior aggregate timing and accuracy, this comes at the cost of higher variability in lead-time estimates than the LSTM baseline. Our results demonstrate that Transformer architectures modified with early detection biases, when used without temporal smoothing layers, provide a high-sensitivity alternative for forecasting AR emergence that prioritizes earlier precursor timing over statistical smoothness.