mini-uber-calibration

One of the achievements of the edgy (tm) and underground (tm) data-management operation of which I am a part is the uber-calibration of the SDSS data. In this project (led by Padmanabhan), we (in what follows when I say we I mean Padmanabhan) found all overlapping parts of the SDSS data (at field edges and where the survey-strategy great-circle stripes converge) and cross-identified detected sources. We then simultaneously fit for all of the calibration parameters for all fields, and all of the magnitudes of all of the stars in all of the overlaps (plus some clever smoothness priors on the atmospheric terms suggested by Roweis). This fit—which involves optimization over more than 10⁸ parameters) did a nice job of improving calibration and made possible the high precision of this paper among others.

Today, Moustakas and I realized that we could make a mini version of this uber-calibration for any multi-image dataset like the one he has been reducing this week. It may in fact be the right thing to do.