SV Calibration Plan

This page is a summary of activities for photometric calibration (and sky/fringe subtraction) during SV. It combines ideas from Jim Annis's Photometricity versus Calibration page as well as Douglas Tucker's docdb-6449 Calibration Commissioning plan, plus input from Gary & Paul.

Overview of SV Approach

The underlying premise is that DES images will be processed in 4 steps:
  1. Overscan / bias subtraction and trim (we will assume this is tested in Commissioning and not worry further)
  2. Division by a star flat that normalizes each pixel to the flux (more exactly, the surface brightness) of a star of nominal spectrum focused onto this pixel.
  3. In z and Y bands: subtraction of one or more scaled fringing templates, for fringing that has too much small small-scale structure to be removed in step 4.
  4. High-pass filtering to remove additive sky fluctuations.
The star flat will be produced in a two-step process:
  1. A flat produced by broadband diffuse-light illumination (probably dome flat, potentially DECal or twilight) is used to produce the flat structure on small scales.
  2. Large-scale corrections to the diffuse flat are derived by forcing internal agreement between magnitudes of objects observed on multiple dithered exposures. A functional form for the large-scale corrections is chosen, with its parameters optimized by software. The magnitudes are produced by images that have been dome-flattened; the functional form prescribes a multiplicative correction to turn the dome flat into the star flat.

See the Data Processing Overview wiki page for more details.

Setting up this process and validating that it works to produce accurate relative magnitudes across the array (in not-too-cloudy conditions) is what Jim called "photometricity" and what I've called "Relative photometric calibration." Next there is a step of assigning magnitudes on an absolute scale, which Jim called "Calibration" and which I'll call "Absolute photometric calibration" testing. For SV, we are requiring that we show that the absolute calibration of DECam have zeropoints and color terms that are consistent with the expected qualities of the hardware and atmosphere; that the variation in zeropoints and color terms is within the range expected from atmospheric variations, i.e. the camera response is not fluctuating and we understand what is needed for zeropoint determination each night; and that the observations that will be needed for absolute calibration are feasible (i.e. BD+17).

Relative Calibration Steps

  1. Scattered light: for SV, we will assume that pinhole tests, etc., that are meant to track down sources of unfocused light reaching the focal plane are done as part of DECam Commissioning. We will just assume that we have some differences between diffuse-light flats and our desired star flat that we must determine, so no SV work here.
  2. Diffuse-light flats: First SV job is to check that dome flats are providing the information that we want, namely small-scale response variations. Cal-R4 is a quantitative test:
    a. We divide dome flat by twilight flat and ask whether the quotient is constant after application of chosen high-pass filter.
    b. We take DECal flats in a band at red and blue sides of each filter. Test that ratio of these to dome (or to each other) satisfies same criterion. _[Note that DECal slit width can be broadened to make 10-nm-wide illumination to speed up the count rate if necessary to make this test feasible without the hours' worth of dark dome that normal DECal sweeps are expected to take.]
    c. Repeat this test on ratios of dome flats taken early (Commissioning) and later (through SV).
    d. If there is observed color dependence of response at small scale, we will need to re-evaluate the flattening strategy. It's unlikely that it will be something we can fix. Time dependence is an issue that should be traced back to the hardware.
    e. Open Issue: Does DESDM throw away all information from relative level of flats in different amplifiers? We probably want to keep this, unless we find CCD gains are varying at a level that requires continuous re-derivation of relative responses of different CCDs.
  3. Construct/test fringe frames: Ratio median night-sky flats to the dome flats and see whether the sky-subtraction high-pass filter leaves behind visible fringing. Expect this to be true in z and Y bands. Then need to compare fringe frames taken at different times to see if they are simple rescalings of each other, i.e. is one fringe template sufficient or are multiple needed? Cal-R5 gives a standard, and Cal-G6 a goal for fringing removal.
  4. Construct Star Flat: Next step is correction of the dome flat to stellar response flat, by taking a series of dithered exposures of a single field, dividing each by the dome flat, then fitting parameters of a large-scale response model to force internal agreement in magnitudes of all stars in the field. Repeat in each filter. Notes on this:
    a. We will not specify a grid, instead it is an action item to figure out a good dither pattern to constrain the kinds of response patterns we expect to have. How many exposures needed? Clearly the smallest dither must be smaller than the "sharpest feature" in the scattered-light patterns.
    b. We will not do radial averaging to look for pupil ghosts - instead we will expect the generic fitting to recover pupil ghosts. If not, we can add expected pupil-ghost forms to the fitting functions being used.
    c. This will probably be done first during Commissioning, but we will repeat during SV to check for time stability.
    d. We need to choose a target field. Criteria would be to have high stellar density to make the process more precise, but not so high that crowding makes aperture photometry unreliable. Also want to have a coarse network of astrometric & photometric standards to remove the degeneracy at linear level which exists in these internal solutions.
    e. Need to decide how pixel-area variations should be treated - they're easy to calculate, but need to be clear on where the correction is applied. Do star-flattened images represent flux or surface brightness in a given pixel?
    f. Will we need to be making initial stellar color corrections during the star flat derivation process? This would be true if CCDs have sufficiently different color terms that we need to make a color tweak to get most accurate inter-CCD comparisons of stellar mags.
  5. Validate star flat: With star flat in hand, we use it to flatten a series of sky exposures of any field. The sky exposures should be dithered so that stars move around the focal plane. We then compare relative mags in exposure A to relative mags in exposure B and expect agreement if the star flat is accurate. Cal-R1 and Cal-G4 are req/goal level of agreement. Notes:
    a. We are not specific about the field or the dither pattern used in the test. But it makes sense to use something like expected DES tiling pattern, and there are other reasons to have SV include a small survey with expected DES full-depth tiling pattern.
    b. Test should be conducted in photometric conditions, but not necessarily all on the same night.
    c. We can also use the collection of SV sky exposures to refine or re-derive the star flat as a test of the "ubercal" software that we have used to derive it.
    d. Comparison of DECam mags to SDSS/UKIDS mags in some Stripe 82 field will also give some information on quality of star flat, albeit not as precise as internal comparisons.
  6. Clouds: We want to also examine relative photometry residuals on exposures taken on non-photometric nights of some field for which we have several dithered photometric exposures. First purpose is Cal-G5 goal of seeing whether RASICam can tell us when there are enough clouds that extinction varies across the DECam array. Second, we want to exercise the DESDM module that assigns rough mags to non-photometric data (what code does this, how?).
  7. Linearity: We will assume that shutter-timing and linearity tests are done as part of DECam commissioning. However we will check linearity two ways during SV:
    a. Seeing if bright and faint stars yield same star flat solutions.
    b. Taking at least one set of observations of the validation field with halved exposure time, then see if relative magnitudes (compared to full exposure time) have a correlation with brightness of sources.

Absolute Calibration SV Steps

With a star flat and sky subtraction method that yield consistent stellar mags across the array, we can begin to place the system response on an absolute scale.

  1. Color-term sanity check: Initial observations of some Stripe 82 will yield DECam instrumental mags to regress against SDSS and UKIDS grizY mags.
    a. Derived color terms can be compared against those expected from synthetic photometry of estimated DECam spectral response, and should meet the Cal-R2, Cal-R3 sanity checks. Color terms that seriously depart from expectations could be signs of some defective element of the optical system.
    b. Cal-G5 is the desire that we understand our system response by having observed color terms agree more closely with synthetic photometry. For this we will want full sweep of DECam filters with DECal flats to map expected color terms across the field.
  2. Absolute response stability checks: We will return to the same Stripe 82 photometric reference field at the twilights and sometime in the middle of every clear night, and again calculate zero points and color terms for each CCD.
    a. Cal-R6 tests that the resultant zeropoints should not change any more than we expect the clear-sky transmission coefficients to change (assuming fixed airmass - we'll need extinction coefficients for each night??
    b. During one photometric night, we will return to the photometric reference field once per hour to check shorter-term stability.
    c. From these data we can decide whether the observed inter- and intra-night zeropoint stability mean that taking standards at morning/evening twilight will be sufficient to insure that photometric nights meet goals for Year 1 survey calibration (Cal-G10)
    d. We can collect database of zeropoints in these exposures (and any others taken in fields with photometric standard stars) and regress against seeing (Cal-G9) as test of the stellar magnitude algorithm. Also regress zeropoints / color terms against the various sky monitors (Cal-G12) to begin understanding how to use them as absolute calibration tools. And regress against focal-plane temperature (Cal-G11) to look for changes in CCD amplifier gain and/or red spectral response with device temperature. Do we want to force a change in the DECam focal plane operating temperature during one clear SV night?
    e. Take DECal flats every cloudy night of SV so we can check stability.
  3. Primary calibration: Take one look at BD+17 - probably will be done during Commissioning, but we may wish to repeat this on a photometric night when we've also looked at our Stripe 82 reference field - maybe the night when we're doing hourly Stripe-82 observations. Main job here is to see that BD+17 instrumental mag is stable (shutter time is repeatable) and get some initial data to exercise the full absolute calibration chain. Also desired by SN group.

Other stuff

We considered it useful to have some stellar-locus software available to run as a test of photometric accuracy and repeatability on the Stripe-82 and the DES-tiling fields.

Questions of sky level vs moon phase, distance, etc., are in the Signals/Noise category, and scattered light from nearby bright stars will be treated as "Anomaly" category, not Calibrations.

Tricky issue is how to look for relative gain changes between CCDs (more specifically, between amplifiers).