Last modified 10 years ago Last modified on 12/22/2009 04:01:41 PM

The bulk of DC3b input data will be generated by the LSST Image Simulator. Each image will be constructed from detailed models of the astronomical sky, of the atmosphere, and of the LSST telescope and camera system. The full set of images is about 3% of the volume of image data expected for LSST DR1. Within that volume, there is wide latitude in trading off number of fields (sky coverage) versus number of epochs and filters observed for each field. Driven by the need to realistically test deep detection and measurement, and photometric and astrometric calibration, the preliminary plan is to cover this pattern on the sky:

  • An equatorial stripe approximately 1 CCD in width
  • A stripe running from RA=0, DEC=0 over the south pole to RA=12h, DEC=0, approximately 1 CCD in width
  • A region of full LSST fields, approximately 15 degrees on a side, centered on the equatorial stripe at RA=0

The observing pattern and filter selection is driven by the output of output_opsim3_61 between MJD=49353 and 49718.

This results in the following statistics:

  • 290 fields, with an average of 46 ccd's per field (some of them are narrow strips, only one ccd wide)
  • 30144 field exposures, including all 6 filters
  • 1.4e6 ccd-exposures
  • 47 TB of raw image data (16 bit integer)

Very roughly, this leads to a catalog with about 1.5e8 objects, and about 100 measurements per object. FaintSources for all measurements will be saved. 1/3 of the measurements will be saved as Sources.

A preliminary list of fields for simulation, taken from output_opsim3_61, is given here.

As one indicator of the processing time that will be required, if we were to use 1/2 of Abe - 600 8 core nodes - and processed images through IPSD at DC3a rates, IPSD processing alone would take approximately 1.4e6 ccd-exp *16 amp/ccd *5 min/(600 nodes*8 core/node * 60 min/hr) = 389 hours, or just over two weeks of solid running. Obviously, the data release production will take far more time than IPSD, so we are likely talking about several months to run the entire data set. Since I doubt we can realistically have 1/2 of Abe for this length of time, we have to define some intermediate performance tests (PT1 - PT3), driven by the data required to test astrometry and photometry:

  • Astrometry
    • PT1: Set of overlapping fields near SCP over entire year; tests basic approach. does not test seasonal gaps in data, or widely varying airmass
    • PT2: Add sets of overlapping fields near celestial equator; tests seasonal gaps in data, airmass variation
  • Photometry
    • PT1: same as astrometry
    • PT2: same as astrom
    • PT3: add set of linked fields over full sky for full-up test of global self-calibration (as given at the top of this page)

As a suggestion for the number of fields and exposures for the intermediate stages:

  • PT1: 10 full-FOV overlappin SCP fields, all 6 filters (approximately 12% of the PT3 fields)
  • PT2: 3 groups of 10 full-FOV fields each, spaced at 8hr intervals along the celestial equator, all 6 filters (approximately 35% of the PT3 fields)