I have fiddled around with the spectra files, even last night a bunch. The background doesn't remove very neatly at all. I tried time periods that were roughly equal to 7 and ones twice as long as background.
I selected and compared ROIs between the…
The different traces have different times. Are you normalising each to counts/hour and taking average of background then subtracting value multiplies by length of trace 7 (about 4 hours I think)?
When determining how "extreme" the signal is it is worth dividing total number of counts/hour by background number of counts/hour.