| | 8 | I spent this week working on the 3pi reference photometry database. I have implemented DVO scripts to measure the quality of the photometric calibration by using the location of the stellar locus in 2color plots. I extracted a stellar locus from the MD fields, so this essentially measures discrepancies between the 3pi dataset and the MD fields. The stellar locus measurement does not allow us to completely constrain the system zero points : a single free parameter may remain, and the accuracy is only ~2-3% for the g-r color (which is constrained by the turn up at late spectral types). However, it does give an independent check on the photometry and allows me to catch problems with the relative photometry analysis. |
| | 9 | |
| | 10 | I quickly learned that some of the database had very poor relative photometry results: forget the zero points, the relative photometry calculation was not resulting in generally consistent values for stars in certain areas. I tracked this down to a bookkeeping problem associating the images appropriate to the analysis with the region of interest. With this fixed, I found much more sensible relative photometry results. However, the stellar locus test showed that the derived photometry was not in good agreement with the MD fields. Looking into this, I realized that the relative photometry analysis was allowing negative clouds as a valid solution. I updated the code to force images to have only positive clouds (within a small error margin of 2.5%), propagating the effect back to the images with real clouds. This significantly cleaned up the photometry across large areas. |
| | 11 | |
| | 12 | The one drawback is that the analysis does a better job of tying down the cloud levels if a particular analysis has enough images with photometric data. This drives me to use large portions of the sky, rather than looping over small area. To enable this, I have made 3 changes to relphot. First, I have allowed a user-set limit on the density of stars used for any region in the sky: this fix prevents the Galactic Plane regions from dominating the memory and forcing 100s of GB of data in active memory. Second, I have adjusted the code to use an internal structure for detections and average objects with a limited subset of the parameters stored in the database. This allows me to nearly triple the number of stars used in the analysis, while keeping the memory footprint manageable. Finally, I updated the code to allow the relative photometry analysis to be performed on multiple filters at a time (previously, only a single average filter system could be processed in one pass). When processing the data on the entire sky, the analysis spends most of the time on I/O. Since it has to read the entire database for each filter, allowing multiple filters means we can spend that time once. |
| | 13 | |
