| | 8 | I've been frustrated over the past several weeks by relative photometry failing to determine good solutions for some areas. I've continued to look into this, and I've identified what I think are the remaining causes of problems in the current code. First, I had been failing to select all of the images appropriate to a given region if that region included the 0,360 boundary. The details are a bit obscure, but the bottom line is that in those regions, only a portion of the images were being selected. This problem had been persisting for some time. Second, I found a couple of loop index errors in the code when I threaded the photometry analysis, which was causing problems over the last 2 weeks or so. Those are now fixed (including one annoying segfault). I've finally started to get reasonable answers across the entire sky. The plots show that, when looking at only the ubercal measurements, there are very clean residuals across the sky, as Eddie & Doug showed. When non-ubercal images are included, the scatter increases for some images, but not excessively (up to 1-2%). The scatter is mostly coming from 2D structure in the effective zero point across an exposure. Since I am fitting only a single constant per exposures, the presence of clouds leaves behind a small residual pattern (often just a tilt) with a mean of zero, but non-zero edges. This is not surprising, and shows that I am getting the mean zero points right, but also shows that we can consider fitting a 2D variation of the zero points for exposures that are (a) sufficiently bad to warrant it and (b) well-enough connected to photometric (or well-calibrated) exposures to have a sensible tie down. |
