Data Variance

There is optional data variance creation for all instruments. By default only the polarimetry and mid-infra-red NOD_CHOP family of recipes have variance processing enabled. To switch on variance calculations for the other recipes is a simple edit of the recipe. See the notes on customising recipes for instructions. The initial variance is calculated as follows.

• First the readnoise variance for the instrument is applied. It is currently a constant for all pixels, which takes the gain, the number of exposures, and the number of array reads per exposure into account. For more than one read, n, the noise scales by the following factor,

  
  

  $$\sqrt{(n(n+1))/(12(n-1))}$$
  
  

  which is derived from linear-regression theory. The read noise reduces by the factor $1/\sqrt{\rm {number~of~exposures}}$.
  [_ADD_READNOISE_VARIANCE_]

  The raw read noise in units of electrons comes from the readnoise calibration (set within the ARRAY_TESTS) recipe, or should one not exist, ORAC-DR selects a representative default. The UFTI and IRCAM defaults come from the instrument Web pages such as this and depend on speed, gain, and read type. Typical values are as follows: Classic Cam 40e^-, INGRID 25e^-, IRCAM 47e^-, IRIS2 15e^-, ISAAC 18.5e^-, NACO 53e^-, NIRI 50e^-, UFTI 26e^-, and UIST 40e^-. Michelle reductions merely use an estimate of 1000e^-.
  [<instrument>/_GET_READNOISE_]

  The gain--the number of electrons per analogue-to-digital unit--comes from the data headers. If for some unusual reason, the header is absent, the recipe subsitutes suitable time-dependent defaults. The gain is 7.5e^- for Classic Cam, 4.1e^- for INGRID, $\sim$6e^- for UFTI and IRCAM, 5.2e^- for IRIS2, 4.6e^- for ISAAC, 500e^- for Michelle, 10e^- for NACO, 12.3e^- for NIRI, and 15e^- for UIST.
  [<instrument>/_GET_GAIN_]

• The second step is to add the Poisson variance. This simply adds the data array to the variance component taking into account the gain of the detector and the number of exposures.

  For read modes where the bias level is not removed, such as CHOP or STARE, a bias calibration frame must be subtracted first. If no suitable bias is found, the bias is deemed to be zero; ORAC-DR issues a warning that the variance is wrong, likely overestimated.

  The calculations also derive the ratio of the read noise to the Poisson noise, and reports the percentage of background-limited pixels, i.e. those where the Poisson noise is greater than the read noise.
  [_ADD_POISSON_VARIANCE_]