Speaker
Description
The PUNCH (Polarimeter to UNify the Corona and Heliosphere) mission has three heliographic Wide-Field Imagers (WFIs) and one coronagraphic Narrow-Field Imager (NFI). Soon after launch, it was discovered that NFI has a very strong and unexpected dynamic stray light component. Driven by Earth-shine that has found a route through the optical system, this stray light is by far the dominant source of measured light. Showing a strong dependence on both the land mass and cloud cover under the satellite, this stray light varies significantly from image to image. This defeats the usual methods of removing stray light (e.g., running minimum subtraction) or ignoring stray light (e.g., difference imaging). One year after launch, the PUNCH team is nearing completion of the calibration pipeline for WFI, but NFI remains largely un-calibrated and un-used due to the difficulty of addressing this stray light. Early efforts showed that principal component analysis (PCA) is able to fit and remove the dynamic stray light (as well as the F corona) remarkably well, revealing that clear coronal signals are present and recoverable. This technique was automated and deployed in the QuickPUNCH pipeline (a low-latency pipeline intended for space weather forecasting), but the quality of the subtraction fell far below that of the initial proof of concept. Recently, the PCA approach has received renewed effort, with the goal of producing a reliable method for automated pipeline deployment. While not deep learning, PCA is a type of machine learning, and machine learning has not been commonly used for this sort of fundamental calibration in a production pipeline by corona- or heliographic missions. This presentation will present the method, show the results, discuss the various complications encountered, and highlight steps we've included to ensure we do not fit and remove the coronal signal of interest.