BT2PB: High-cadence polarized brightness of the solar corona from LASCO C2 total brightness observations using deep learning

Abstract BT2PB is a convolutional neural network (CNN) that increases the effective cadence of calibrated polarized brightness (pB) observations of the Large Angle and Spectrometric Coronagraph (LASCO) C2 instrument aboard the Solar and Heliospheric Observatory (SOHO) from one or a few per day to the total brightness (B t) cadence of tens of images per day, allowing new science and operations dependent on high time-resolution calibrated data. LASCO C2 has been observing the solar corona almost continuously for over three decades. The pB observations offer a measurement of the polarized K-corona with drastically reduced signal from the F-corona and stray light components that dominate the B t images, thus are more useful for quantitative coronal science. In this work we aim to estimate pB directly from B t images using a convolutional neural network (CNN) trained on thousands of pB and B t observation pairs made over the past 25 years. The CNN operates on brightness profiles extracted at multiple heliocentric heights between 2.3 and 6.0 R , learning to predict log 10 (pB/B t) as a function of position angle. Observing geometry metadata including the spacecraft roll angle, spacecraft heliocentric distance, and heliographic latitude are provided as auxiliary inputs, allowing the network to account for time-varying instrumental and orbital effects. When evaluated on held-out test observations, the model achieves a mean absolute relative error of approximately 8-12% on the predicted pB across heliocentric heights from 2.3 to 6.0 R , with accuracy varying by a few percent as a function of solar cycle. We estimate the error arising solely from model prediction as ≈4-8%. The approach enables high time-resolution studies of dynamic events including coronal mass ejections (CME), the potential for more reliable mass estimates of dynamic events, and improved time resolution for coronal density diagnostics including tomography. Furthermore, SOLA: ms.tex; 12 October 2026; 17:22; p. 1 Morgan & Gandhi the quality of these results allow highly efficient observation strategies for current or future coronagraphs with the capability for measuring B t and pB.