At long wavelengths, the sky is relatively insensitive to changes in the twilight sky, so one cannot use the method used for SW data. Instead, one observes a region of the sky with the telescope pointing to the Zenith, a region to the South with an airmass of 2 and a region that is even further South (an airmass of 2.4). The flat can then be created by subtracting the images that were taken at one of the higher airmasses from the image taken at Zenith. The resulting image is then normalised to 1. Five images are taken at each position. Since the position of these images is slightly different, stars can be removed before the frames are averaged.
In principle, one should apply the non-linearity correction to the flat field frames; however, in practice, correcting for the non-linearity in flat field frames will make very little difference to the final result as the pixel-to-pixel sensitivity variation in the Aladdin array is quite low, about 1%. As a comparison, the pixel-to-pixel sensitivity variation in the Hawaii array is much higher, about 7%. If a non-linearity corection is applied, a dark frame with the appropriate read out mode and DIT should be subtracted from all frames before the non-linearity correction is made.
If one divides one flat with another the pixel-to-pixel RMS is around 0.2%, so the S/N of the flats is high. However, there are other systematic effects which cause much larger differences. For flats that are taken close in time, these differences are caused by pupil ghosts (at the level of 2%), imperfections in the objective (at the level of 1-5%), and hot pixels (at the level 10-20%). For flats that are taken during different nights, the differences caused by imperfections in the objective are larger, and one starts to see differences which can only be described as smudges with a size of around 30 pixels.
Since the RMS of the flat is much lower than the typical deviation caused by hot pixels, one should replace these hot pixels with the local average.