| | 8 | I spent the week working on some code issues in the relative astrometry and photometry analysis to be ready to create the full-sky reference database. The work needed involved better handling of the per-measurement errors in the image fits and adding fields to track the systematic errors for each image. I also improved the visualization tools to be sure the code is working correctly. I ran some modest-scale tests (600 square degrees) of relastro and am getting acceptable residuals, with the bulk of images in the range 20-30mas. To be clear, this number is the scatter of all stars on a given image with S/N > 100; I do not attempt to back out the purely systematic term from the Poisson noise, so the Poisson contributions may be elevating this value a bit. A more accurate number will come from the analysis of the per-object residuals as a function of S/N, which I have not examined. Even so, this is not quite as good as we probably can do. I have not yet tried to chase down the worst outlier images or worry too much about the outlier rejection. I have the impression from residual plots that there are still systematic structures visible in the residuals as a function of position, so some improvement can probably be made. The fits may also be somewhat driven by the fact that each 2MASS measurement contributes to the fit, even though they are not statistically independent (and have worse systematics than PS1). Even though improvements can be made, I am going to use the code as is to generate the full-sky reference database to be used by the Grand Reprocessing. |
