| | 3 | Based on Nigel's email this morning (details at http://ps1sc.ifa.hawaii.edu/PS1wiki/index.php/DRAVG_telecon#Correlated_bias_structure_between_cells), I spent part of the afternoon looking at this structure again, starting from Nigel's point that adjacent cells seemed to have the same structure. |
| | 4 | |
| | 5 | The OTA chosen by Nigel is OTA57, which has PATTERN.ROW applied to columns 0,1,2,6,7, leaving 3,4,5 with no PATTERN.ROW correction. The oversubtraction PATTERN.ROW leaves is clearly visible in the first image, in the cell one up and right of the bottom left corner. As columns 345 had lower row-to-row offsets, these cells were left unmodified. It's also clear from this image (and other investigations) that the cells we do not apply PATTERN.ROW to still have some level of row-to-row variations. |
| | 6 | |
| | 7 | My study used bias images from OTA67, as I had them on disk locally (an MHPCC power outage prevented grabbing new data). For this study, I constructed row models by subtracting the median value for each cell from that cell, and then calculating the median "stack" across a row of cells. This results in 8 cell sized models appropriate for removing the bias variations from a row of cells. Subtracting this row model from each cell should remove the row-to-row variations. The following image shows these models for one bias frame, showing the variations between different cell-rows for one image. |
| | 8 | |
| | 17 | This method seems to do a decent job of minimizing row-to-row variations, without the oversubtraction around bright stars that PATTERN.ROW introduces. However, there are still some issues that need to be considered before this can be implemented. |
| | 18 | |
| | 19 | 1. The first science image shows a ghosting due to bright pixel columns that cause an oversubtraction in the background level in vertical bands. This can probably be mitigated by including the static mask data when making the median stack. |
| | 20 | 2. This structure appears to change between each bias frame that I examined, suggesting that this is not something that can be precalculated. This variation makes sense, as any static feature would be included in the dark model. |
| | 21 | 3. These tests were done on raw images, without any dark correction. Some thought needs to be done to determine where in the detrending process this kind of correction should be applied. |
| | 22 | 4. Bright extended objects will need to be checked closely to ensure that this does not damage those objects. |
