| | 7 | |
| | 8 | We have been focusing this week on three main issues of concern by the KPAT/DRAVG : photometry in stack, astrometry issues, and crosstalk. Paul has made good progress at understanding the behavior of the photometric quality of the stacks, and at this point, I believe that he and Nigel agree that the pixel values are OK: they have both shown that aperture photometry comparisons between warps and stacks have scatter which is small (where not photon limited) and linear. For stacks with good IQ input images, Paul showed that the psf photometry is substantially more stable than those with varying input IQ, though there is still more scatter in the psf photometry than expected from signal-to-noise trends. This last point highlights that the scatter is probably being driven by the mismatch between PSF and model coupled to the correlated errors. |
| | 9 | |
| | 10 | Chris has finished the crosstalk masking, and it is incorporated in the standard analysis. |
| | 11 | |
| | 12 | I have been working on the astrometry issues, both large-scale and small-scale failures. for the large-scale failures, I have made several fixes that substantially reduce catastrophic failures and give more robust solutions to avoid the small-scale errors (see below). The code for testing and retrying or rejecting badly fit chips based on the rest of the chips is working, but is not currently being used for GPC1: currently, the guess is coming from a linear model of the focal plane, which is not really accurate enough for this rejection. I need to generate a full-scale model, but I have not fully tested that the full non-linear model behaves correctly in all cases. I am going to put that on a lower priority until we have more statistics on the current catastrophic failure rates. One remaining fix that is needed in this regard is to ensure that warp is not including chips which have failed: there have been reports of known bad astrometry solutions being used in the warping analysis. |
| | 13 | |
| | 14 | There were a number of poor astrometry results showing up in difference images pairs for run 3 data, especially in the low Galactic latitude 3pi fields. After working on the iterative clipping process to be robust against outliers in these cases, I finally realized that these problems were in fact caused by a bug in PSPHOT -- these images where processed before a number of changes over the summer to the psphot source size analysis, and many stars in these images were identified as cosmic rays. The code then masked the cores of these stars, and re-fitted sources in the wings of the stars. As a result, there were large numbers of consistently offset sources at 2.5 pixels from the centers of real reference stars in one of the 4 cardinal X,Y directions. running psastro on a newer psphot analysis of the same images gave much better results. |
| | 15 | |
| | 16 | We also had a hardware incident over the weekend: the database machine ippdb02 crashed complaining about a hardware error. The machine did not reboot on power-cycle, and we were concerned that the failure may have been serious. Cindy & Bill were able to get the machine back up after re-setting the BIOS values, and it seems the database is intact. However, this gave us a good opportunity to set up an alternate Nebulous installation on ippdb00, something which has not been done for our production machines since Josh left. |
| | 17 | |
| | 18 | * psastro fixes: |
| | 19 | * improve the test for the allowed order for a given number of stars, and make the requirements tighter. |
| | 20 | * iteratively remove clumps of objects in the raw and ref tables used for the astrometric match and analysis. This avoids catastrophic failures due to nebulosity and galaxies. |
| | 21 | * add a feature to require a single (closest) reference match in the fitting after the first N clipping iterations. This prevents solutions driven by outliers in crowded fields. |
| | 22 | * some minor tuning of the clipping radii for GPC1 |
