After finishing the intial 2008 LuTop BSG matrix I corrected some MOBY LuTOp data and saw an large upturn at the UV end of the BSG right near the edge (first 12 ish pixels). Once I finished the 2011 BSG matrix I SLCed the same M245 MOBY file with the 2008 and 2011 BSG matrix and amazingly the UV upturn is gone in the data corrected using the 2011 matrix. So I compared the two matrices to figure out what was wrong with the 2008 data set. The page below shows the comparison. See the NIST2011_03 webpage and page num 3.20 for the version of this page before the 2008 data set was fixed.
Original webpage about the 2008 UV upturn problem

After comparing the two matrices I went into the smoothed BSG LuTop data from 2008 and fixed the laser scans which looked like they were causing the problem. You can see the changed laser scans at the bottom of this page. I then recreated the matrix and the huge UV upturn is done. The 2008 data is still higher than the 2011 data but the large upturn is gone. If you look at the 2008 data you will see it is much noiser than the 2011 data and there were a number of problems with the data set. Making the matrix craeted from this data set more uncertain than the 2011 data set.

MOBY245 LuTop data is shown below the blue is corrected with the 2008 matrix and the green line with the 2011 matrix.. The magenta line is the unfixed 2008 corrected data. Notice the upturn is gone in the 2008 and 2011 corrected data. The 2008 data is still higher than 2011.

Figure 1

Zoomed in view of the first 30 rows of the 2011 and 2008 SLC matrix. Notice were the 2008 data set lower and smoother.

Figure 2

In-water LuTop ADU/pix/sec data SLCed with the 2011, 2008 and old algo and uncorrected. You can see this looks fine for both 2011 and 2008.

Figure 3

Lamp Signal from LuTop ADU/pix/sec data SLCed with the 2011, 2008 and old algo and uncorrected. You can see the 2008 SLC data is much better (blue line compared to the magenta line). The UV is not pulled down too much. It looks alot more like the data corrected with 2011 SLC matrix looks better.

Figure 4

Zoomed in view of the first 30 rows of the 2011 and 2008 SLC weighted matrix. Weighted matrix with the diagonal (laser peak pixels) set to zero.
Lo = LuTop signal, SPR = correction matrix. inv takes the inverse of the matrix. eye is the n by n idenity matrix and adds a one up the diagonal.
A = inv((eye(512,512) + SPR));
for k = 1:512
  SPRw(:,k) = A(:,k)*Lo(k,1);
end
The 2008 matrix is now less noisy and the values are lower.

Figure 5

UV rows in the 2011 and 2008 weighted matrix (Fig 5) with the diagonal (laser peak pixels) set to zero. The large valley at pixels 300-370 is from the large bump. These are the 2, 4, 6 ... rows from figure 5. If you sum these up you get the second value in the legend.

Notice the larger negative numbers in the 2008 data set are much closers to the 2011 data set at pixels 200-250 and > 480

Figure 6

This shows the cumulative sum of the FIgure 6 lines. Where the 2008 and 2011 line diverge are where the differences between the matrices are. You can see the spread is much better than before.

Figure 7

This is the cumsum 2011 - 2008. You can see two change points, around pixel 200 and after 480 are less than before.

Figure 8

Below are the laser data set which were "fixed" which reduced the UV upturn in the 2008 SLC. The green line is the 2011 data set, you can see it is lower at the edge for most of these laser data set The red line shows the original data which caused the upturn and the blue line is the new fixed data which reduced the problem a ton. The little upturns at the end might have been caused by two processes. 1) there is a little "noise ball" at the end of some of these scans which could have caused this data to be artifically high. 2) the smoothing algo might be having a hard time with these at the edge. Seemed like the smoothed data was higher than the real data in these cases.

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15