Brightness evolution of LEO Starlink mega-constellation satellites from 2021 to 2023: a multiyear ground-based photometric study

We report a multi-epoch V-band campaign (2021–2023) with uniform processing and a common (Formula presented) normalization across Starlink generations. Median magnitudes (68 per cent confidence intervals; N in brackets) are: v1.0: 5.365, [5.084, 5.553] (34), Gen-2: 6.012, [5.801, 6.173] (6), v1.5: 6.106, [6.065, 6.154] (164), VisorSat: 6.618, [6.403, 6.804] (54), DarkSat: 8.431, [5.916, 10.947] (2). This yields the ordering (Formula presented), i.e. mitigation-era designs are typically fainter than the original v1.0. The Gen-2 – v1.5 difference is small (0.094 mag) and Gen-2 has limited coverage ((Formula presented) ), so the trend is not strictly monotonic. We use an open, python-based processing pipeline built on astropy for standard image calibration (bias, dark and flat-field correction), astrometric and photometric calibration against Gaia DR3, and derivation of the viewing geometry (solar phase angle, range, elongation, and airmass). Satellite tracks are identified from TLE-based ephemerides, matched to the detections, and then used to measure a brightness value for each track in a reproducible way. At a common height, the medians are 5.37 (v1.0), 6.01 (Gen-2), 6.11 (v1.5), 6.62 (VisorSat), and 8.43 (DarkSat; (Formula presented) ), which are (Formula presented) 0.4–1.6 mag brighter (numerically smaller) than the target of (Formula presented). Thus, mitigation shows clear progress but does not yet meet the IAU CPS/SATCON goal of (Formula presented) at (Formula presented). These benchmarks can guide future satellite designs, survey planning and avoidance strategies, with the main uncertainties arising from the very small samples for DarkSat and Gen-2.