SHAPE: Solar Heliospherical Analysis and Prediction of Eruptions A machine learning approach to solar flare forecasting

Abstract

Solar flares are the most energetic impulsive events in the solar system, and reliable operational forecasting of the rarest and most destructive M- and X-class events remains an open problem. We present Solar Heliospherical Analysis and Prediction of Eruptions (SHAPE), a two-phase deep-learning pipeline that bypasses sunspot proxies in favour of direct soft X-ray observations from the GOES X-ray Sensor (XRS). Phase 1 forecasts monthly flare counts, aggregated to annual totals, with a hybrid model: a cycle-shape curve anchored to published solar-cycle minima fixes the timing and mean profile, and a compact gated-recurrent-unit network with Monte Carlo dropout models the residual variability. Trained on the science-quality NOAA GOES Flare Report for 1995--2020 and validated on 2021--2025, it achieves a mean absolute error of 204 flares yr$^{-1}$, although it underestimates the 2024 solar maximum. Phase 2 forecasts the next day's strongest flare class (C, M or X) with a shared-trunk, dual-head LSTM operating on a 12-day rolling window of five intensity features, with an optional blend against an active-region physics score and a decision threshold tuned to favour recall of M-or-above (M+) events. On a 100-day chronological backtest, the classifier attains a true skill statistic of 0.351, a Heidke skill score of 0.335 for M+ events, a Brier skill score of 0.163 relative to climatology, and an expected calibration error of 0.023. A day-to-day persistence baseline is, however, competitive on both skill scores, showing that much short-term predictability in flare class is autocorrelation rather than learned structure. SHAPE shows that competitive flare forecasting is achievable with compact architectures that retain in minutes on standard hardware, without sunspot proxy inputs.