Reducing Ground-based Astrometric Errors with Gaia and Gaussian Processes

Stochastic field distortions caused by atmospheric turbulence are a fundamental limitation to the astrometric accuracy of ground-based imaging. This distortion is measurable at locations stars with accurate positions provided Gaia DR2 catalog; we develop use Gaussian process regression (GPR) interpolate arbitrary in each exposure. We introduce an extension standard GPR techniques that exploits knowledge 2D curl-free. Applied several hundred 90 s exposures from Dark Energy Survey as test bed, find correction reduces variance turbulent ?12× , on average, better performance denser regions catalog. The rms per-coordinate riz bands typically ?7 mas before any and ?2 after application model. corrections validated observation their reduces, 10 5 rms, residuals orbit fit riz-band observations over yr r = 18.5 trans-Neptunian object Eris. also propose method, not yet implemented, for simultaneously estimating fields 5D stellar solutions stack overlapping exposures, which should yield further reductions future deep surveys.