A Hybrid CNN–Transformer Network to EnhanceSolar Magnetogram Resolution for Flare PredictiveAnalytics

Why It Matters

By harmonizing magnetogram resolution across decades, MagRes‑Net improves the fidelity of solar activity databases, directly supporting more accurate flare forecasting models. This advancement reduces data gaps that have limited space weather research and operational forecasting.

Key Takeaways

• Hybrid CNN‑Transformer boosts magnetogram spatial resolution
• Preserves magnetic flux and gradient consistency
• Increases active‑region coverage toward SDO/HMI levels
• Outperforms interpolation and pure CNN methods on PSNR
• Enables more reliable long‑term solar flare forecasts

Pulse Analysis

Solar physicists have long grappled with the resolution gap between historic SOHO/MDI magnetograms and modern SDO/HMI observations. The older MDI data, while spanning two solar cycles, lack the fine‑scale detail needed for precise magnetic field analysis, creating discontinuities in long‑term studies of solar activity. Bridging this disparity is essential for constructing homogeneous datasets that can feed both research and operational space‑weather models, especially as flare prediction increasingly relies on subtle magnetic signatures.

MagRes‑Net addresses the challenge with a novel hybrid design that couples a convolutional neural network for local feature extraction with a transformer‑based self‑attention module that captures global magnetic gradients. Trained on thousands of perfectly aligned MDI‑HMI pairs, the network incorporates physics‑aware losses that enforce magnetic flux conservation and gradient consistency, ensuring that upscaled images remain physically plausible. In benchmark tests, the model delivers superior PSNR, lower RMSE, higher SSIM, and reduced LPIPS compared with traditional interpolation or pure CNN approaches, while boosting active‑region coverage from 12.23 % to 13.25 %—a figure that closely approaches the 14.48 % reference from native HMI data.

The practical implications are significant. A unified, high‑resolution magnetogram archive enables more accurate statistical analyses of solar cycles, improves the training data for machine‑learning flare predictors, and supports real‑time space‑weather forecasting services used by satellite operators and power‑grid managers. As the solar community moves toward integrated, AI‑driven forecasting pipelines, tools like MagRes‑Net will become foundational, reducing reliance on ad‑hoc data stitching and fostering greater confidence in long‑term solar risk assessments.