+The main difference from the DP1 astrometric calibration is an additional step that models the effect of atmospheric turbulence on the astrometry. The per-visit part of the astrometric model already described captures large-scale variation between visits, which includes some atmospheric effects. However, higher order oscillations are visible in the astrometric residuals, and these can be attributed to atmospheric turbulence (\citep{DMTN-324}). As described in \citet{2021AJ....162..106F} and \citet{2021A&A...650A..81L}, the wave-like astrometric effects caused by turbulence can be modeled by a Gaussian Process. In DP2, we do this by taking the astrometric residuals remaining after the \texttt{gbdes} fit, grouping them by visit, then modeling them with the Gaussian Process package \texttt{treegp}. We then use this Gaussian Processes model to predict the effects of turbulence on a grid. Finally, this grid is turned into a two-dimensional spline that can use the AST component \texttt{SplineMap} so that it can be concatenated with the \texttt{gbdes} model. By including this model for the higher-order atmospheric turbulence, we achieve a significant improvement in the astrometric calculation, as demonstrated in Section~\ref{sec:performance}.
0 commit comments