ML - FArSide Trained Active Region Recognition (FASTARR) DataSet

These datasets are used for training the FArSide Trained Active Region Recognition (FASTARR) ML-model, which aims to improve active region (AR) identification on the Sun's far hemisphere. It comprises of far-side helioseismic phase-shift maps and corresponding AR masks. 1. Data Sources 1.1. Helioseismic Phase-Shift Maps: - Generated by the National Solar Observatory’s (NSO) Global Oscillation Network Group (GONG). - Represent 24-hour averaged observations at a 6-hour cadence. - Provided in longitude and sin(latitude) coordinates with a spatial resolution of 0.72°/pixel (longitude) and 0.01/pixel (sin(latitude)). 1.2. Far-Side EUV/304 Å Observations: - Obtained from the Solar TErrestrial RElations Observatory/Extreme UltraViolet Imager (STEREO/EUVI). - Used as a ground truth reference for far-side AR detection. 1.3. AR Masks: - GONG-derived AR masks: Created using helioseismic phase-shift measurements. - EUV/AR masks: Generated from STEREO/EUVI observations. - Both mask types are validated and cross-referenced for spatial and temporal consistency. 2. Dataset Pre-processing & Filtering - The dataset spans from May 2010 to May 2016, corresponding to the far-side observational limits of STEREO satellites. - Raw Data Filtering: - Excluded datasets with partial far-side coverage or data artifacts. - Reduced the dataset to 2,381 high-quality pairs (from an initial 3,057 pairs). - Image Processing: - All maps are centered on the far-side central meridian. - Cropped to exclude off-limb regions. - Resized to 256 × 256 pixels, with interpolation of missing polar data. - Ensured consistency in format for machine learning training.