readout_pattern

(Not sure if this is a reasonable name.)

The raw images show horizontal pattern, both in H and K band. The patterns are different for H and K, and different patterns are observed for each readout.

Here is an example of such a pattern shown in the FLAT_OFF image.

Here are comparison of patterns from different exposure (all are from FLAT_OFF images). Each panel shows a median values along a row (as a function of y position) for each image.

The pattern is repeated for every 128 pixel.
In fact, they are repeated for every 64 pixel with a mirrored pattern (improve the wording).

The easiest way to remove this is to subtract the median value for each row. This approach generally works well. However, it seems that the amplitude of the pattern varies with column, and some residuals remain.

A better approach is to estimate the pattern for each column. This can be done reasonably well since the pattern is repeated fro every 64 pixel. And subtract that pattern from that column.

The image below compares the two approaches.

left : original image
center : median along the row
right : pattern from a column

In general, the second approach works better than the first one.

link to original image

This works good for H-band FLAT_OFF images. However, the pattern should be estimated only using the inter-order pixels for actual data. This can be done by utilizing a suitable mask (the mask will be generated by a pipeline), and this method will be utilized in the pipeline.

Example mask for H

Example mask for K

Here are example when a mask is used.