Re: Data sets over time

[tom <tdroege2@earthlink.net>]

One of the things that makes selecting data sets interesting is the search for 
what makes data "bad".   Over the last week I made such a search.  I looked 
at the area of Michael Sallman's star and extracted images around it for the 
this month and last month.  This produced 180 images from 13 different days.

This is enough for fair statistics on the error.  I first processed all the 
images and made mag vs sigma plots for V and I.  The errors were 
disappointedly large.  

I then looked at every I image with DS9.  As a result I threw out 22 V and I 
images.  I then processed the remaining 158 "good" images.  I could not 
detect any difference in the mag vs sigma plots.  Possibly means might differ 
in the second decimal place, but there was no obvious improvement.

Now what did I throw out?  "bad.jpg" is an example.  There were a half dozen 
examples of this.  The rest were clouds.  Apparently the TOM2 Dec drive has a 
mechanical problem around +40 degrees.  For some reason it decides to move 
somewhere in the middle of the exposure.  I have no idea why, and have 
searched for tubing in the way, slipping pulleys, and the like to no avail.  
What you see in "bad.jpg" makes your head fuzzy.  When you look at the full 
image one almost gets sick.  The brain really does not like to look at such 
things because it recognizes that something is moving and tries to correct.  
But it can't because the image is fixed.  So it works overtime to try to sort 
things out.  

Note that the pipeline happily processes these files.  I assume the triangle 
fit locks on to the brightest stars and finds a fit.  This means that there 
will be a lot of ghost stars in the data base.  That is life I think.  
Looking at the result, it gets a pretty good answer for the stars it 
processes.  Boggle!

My conclusion:

If throwing out a bunch of files that look like bad.jpg and a bunch of obvious 
cloud files does not change the result, then something else is going on that 
produces a spread in the data.  I really suspect that it is not the images, 
but something else.  My nominations (in order of suspicion):

1) Camera tilt to the optical axis
2) The alignment of the optics, lens spacing, tilt etc..  
3)  The lens design.  The best design is a ring part way out from the center.  
The design is worse in the center and corners.  
4) Night to night focus variations.  The tilt means some parts of the image 
have a better focus than others.

I have done what I can for these things.  They are not going to get better.

Possibly one can work 1), 2), 3) into some processing scheme.  I have looked 
at Michael's tech notes that are searching for a pattern and I can't imagine 
any of the above producing the kind of pattern that Michael sees.  

I figure we will have to live with it. Possibly the ensemble photometry scheme 
will gain some.  

Tom Droege

bad.jpg