5.2 Notes on Uncertainty Propagation in the BCD Pipeline
The modules that propagate uncertainties are SNEST_IRS, DNTOFLUX, LINEARIZ, SLOPE_FINDER, and FLATAP, as described above. No other modules propagate uncertainties, and for them the output *unc.fits are identical to the input ones. The uncertainties are correlated among pixels in the operations of DROOPOP, ROWDROOP, DARKBASE, DARKDRIFT, DROOPRES, and REMOVSTRAYCROSS. The correlated uncertainties in these modules are unrealistically large. Since they are the same for all pixels in the array, or in entire read-out channels, or in entire rows, they do not reflect the actual statistical noise from pixel to pixel. Uncertainties are not propagated in CUBESUB, because the darks include significant zodiacal background signal; the uncertainties from the latter are instead propagated during the background subtraction (“bksub”) pipeline for low-resolution IRS data.