Detector Linearity
I've regenerated all the linearity data, refit it all, and tested some correction code that seems to repair some of the issues. This reprocessing was needed to fix two errors I made the first time:
- The original data had been dark corrected, confusing the reason why things weren't linear. This analysis now uses the raw images.
- I mislabeled rows and columns because I read my own code wrong. All row/column data in this analysis is labeled xXXX or yYYY to denote the single x or y value that was held constant while the median was taken over that row or column.
As before, the deviation is measured by creating a model for the expected flux for the region under consideration as a function of exposure time.
model_flux = b_fit + m_fit * exptime
The model is calibrated by fitting the observed data between taken with exposure times between 12 and 50 seconds. The correction code samples this model once for each exposure times input, and calculates the ratio of the model to the observed fluxes. The set of (observed flux,ratio) samples from the reference data (chosen as the 2010-07-02 "linearity" data) are then linearly interpolated based on the observed flux values in the test data.
John pointed out that the edges that seem to have significant issues do not correlate with what I observed in January (EdgeRolloff). I do not currently know if this is a labeling issue or due to some camera change.
Cell Level
x001
More images
I've placed the full set of data/correction images for the edges here:
































