| | 7 | |
| | 8 | I spent the week working on stack photometry issues. I have generated a set of MD04 stacks from raw data using my current development branch with 8 exposures per filter. I am trying to run psphotStack on the resulting stacks, but have been getting side-tracked by issues in the stacks and the in the photometry analysis of stack images. (The full psphotStack program is heavy enough to run that I would like to address issues in the single-stack psphot analysis before braving the full psphotStack analysis.) |
| | 9 | |
| | 10 | I found that the PSF model for the stacks is not very well represented by the standard single-image PSF (PS1_V1, with a power-law slope of r^3.33^). For the 8-way stacks, at least, much better fit can be had with the 'QGAUSS' model, with a power-law slope of 2.25. For the generic stack, it may be necessary to use the 'RGAUSS' model, in which the model slope is a free parameter. Tests showed that this model worked nearly as well for the 8-way stacks as QGAUSS, but it should have the flexibility to handle the much deeper stacks as well. The impact of the model is the reduce the amplitude of the aperture-psf correction, and improve the scatter somewhat. |
| | 11 | |
| | 12 | I also explored the question of weighted vs unweighted fits. As suspected, the unweighted fits do not show the substantial trend of the aperture-psf correction with magnitude (bias in the psf fits). We will use unweighted fits for the full-scale stack photometry analysis as well. |
| | 13 | |
| | 14 | I identified the source of the error that is caused by having CELL.SATURATION set incorrectly. This value is used to mark objects as saturated if their fitted peak is higher than the saturation level. This information is then used in the aperture-correction analysis to exclude saturated stars. For deep stacks, this results in nearly all stars being excluded, and only a small number of poorly measured stars were actually used. The errors from those poorly measured stars were carried into the 2D aperture correction, with the result that all PSF magnitudes were then tainted. Fixing the value of CELL.SATURATION help, but this investigation also showed that the aperture correction analysis needed to be made more robust against catastrophic failures. |
| | 15 | |
| | 16 | I addressed some cases in ppSub where the Kron analysis of the stamps failed to converge: the measurement is used to determine the stamp sizes, but if the image has insufficient clean stars, it will not converge. This situation is now detected as a bad-data case. |
| | 17 | |
| | 18 | I also ran a number of tests to see if we can improve the rejection of bright objects in the background model. I added code to psphot to mask pixels significantly above the background model, and to re-measure the model. Even pushing the rejection to 2 sigma above background, I did not find a significant improvement in the background model: very bright stars and galaxies continue to bias the background model. I believe this is because nearly all of the pixels in the corresponding supercell have contributions from these objects near the sky level. The bias is not due to the extreme outliers, but rather to the plentiful insignificant outliers. My only possible suggestion to reduce this rejection would be to use larger superpixels (say 512 instead of 128). |
