Negative values in just distortion corrected data

[mgalloy]

It appears that applying the distortion introduces the negative values:

20240508.163116 DEBUG: UCOMP_L1_DISTORTION: [pre] cam0 range: 3.1 - 3210.2
20240508.163116 DEBUG: UCOMP_L1_DISTORTION: [pre] cam1 range: 0.0 - 4095.0
20240508.163116 DEBUG: UCOMP_L1_DISTORTION: [post] cam0 range: -127.7 - 3012.2
20240508.163116 DEBUG: UCOMP_L1_DISTORTION: [post] cam1 range: -205.9 - 4030.1

where the reported values above are given just before and after the interpolation is applied to each extension/camera:

cam0 = reverse(data[*, *, p, 0, e], 1)
cam1 = data[*, *, p, 1, e]

mg_log, '[pre] cam0 range: %0.1f - %0.1f', mg_range(cam0), name=run.logger_name, /debug
mg_log, '[pre] cam1 range: %0.1f - %0.1f', mg_range(cam1), name=run.logger_name, /debug

if (apply_distortion) then begin
  cam0 = ucomp_apply_distortion(cam0, dx0_c, dy0_c)
  cam1 = ucomp_apply_distortion(cam1, dx1_c, dy1_c)
endif

mg_log, '[post] cam0 range: %0.1f - %0.1f', mg_range(cam0), name=run.logger_name, /debug
mg_log, '[post] cam1 range: %0.1f - %0.1f', mg_range(cam1), name=run.logger_name, /debug

ucomp_apply_distortion is just a basic interpolation:

function ucomp_apply_distortion, sub_image, dx_c, dy_c
  compile_opt strictarr

  dist_corrected = interpolate(sub_image, dx_c, dy_c, cubic=-0.5, missing=0.0)
  return, dist_corrected
end

[mgalloy]

There are only about on the order of 10 negative pixels in RCAM and 80 negative pixels in TCAM, but I'm not sure how the interpolation creates them. They do seem to be in the corners. For example, below is for 20240330.201621.ucomp.1074.dist.p9.fts extension 1, data[*, *, 3, 1]:

[mgalloy]

OK, think this possible if there are sudden changes to 0 in the image. For example, in this simple one-dimensional example, we can induce negative values from an interpolation, even if there are no negative values in the original data:

IDL> y = [10.0, 0.0, 0.7, 0.9, 1.3]
IDL> print, interpolate(y, [0.9, 1.1, 1.9, 3.1, 4.0], cubic=-0.5, missing=0.0)
     0.611650    -0.361100     0.602100     0.933700      1.30000

The interpolation needs to drop from 10.0 down to 0.0 and then head back up. It dips into negative values to do it.

For the UCoMP data, in the corners where these negative values are appearing, the data is basically white noise, where there would be a lot of very sudden changes in values.