2009/07/24 - Michael Wood-Vasey

Here's a summary of my looking over the past couple of weeks at the PS1 MD fields from the July data release of June (and some previous) images. I've been looking from the perspective of detecting transient and variable objects. I've also been looking at astrometry in MD fields with Dave Monet and will provide an update on that later.

Overall notes:

• The subtractions are only fair and need to be better. The most common issue is an effective PSF/astrometric mismatch leading to dipoles.
• The Transient Classification Server is going to need more information about candidates than currently available. (see below)
• The stacks appear over-smoothed. This leads to positive and negative excursions covering many pixels that masquerade as PSFs.
• Sky subtraction over-subtracts near large extended galaxies such as NGC 4258 (MD06 skycell.073).
• Masking of center of galaxies seems excessive. Galaxies that should not be near the saturation limit still have their cores masked. Need to collect more specific examples for analysis.
• Using the bad flags (as specified in stripcmf2good.py) apparently helpful but not sufficient. The lack of known "good" difference objects makes it hard to calculate the false negative rate from any set of cuts. A clear next step should be the use of fake sources to check this.

TCS More Information Requests

• Aperture and PSF flux at locations of diff.cmf objects in template and input images.
  • Verify that flux is from positive excursion from sky in input and not from negative excursion from sky in template
  • I don't want to cut out pixels that are negative outliers in the template (as proposed for 3Pi analysis), but I do want to know if the region in the template was a negative outlier.
• Aperture mag curve of growth for candidates. This should help with dipoles (which roughly conserve flux).
• Addition of calibration zeropoints to CMF files: populate CAL_PSF_MAG and store overall corresponding zeropoint in header. (Is zeropoint for given PSF or "infinite" aperture 1 count/sec?).
• Mis-subtracted lobes from bright stars/artifacts tend to produce clusters of detections. Should compute a quantity that gives the number of
• Distance between difference image detection and nearest SNR>5 detection in template.