Block Adjustments Aided by Distortion & Scale-X Corrections
H.L. McCallon 06-06-01
Martinizing tests currently being performed on Dec band 17 (-12 to -6 deg.) show that the overlap sigmas from v2 processing can be reduced by almost a factor of four by combining scan-segment block adjustments, distortion and x-scan scale factor corrections. The effect of the x-scan scale factor corrections is far more significant than I had expected. It's not just a matter of pulling in a few wayward scans from the tails of the distribution. For this case at least, the entire distribution of x-scan differences is narrowed significantly, reducing the population sigma by a factor of two. This brings the final x-scan overlap sigmas to about the same value as the in-scan sigmas. The relative improvement due to the various corrections will likely vary with Dec band. Also keep in mind that these are mean differences between overlapping scan-segments. Even if they could all magically be driven to zero, other errors sources (such as extraction errors) would remain.
I. Background:
As a verification of the Martinizing results, the UCAC has been used to independently derive block adjustments (by scan-segment) and x-scan scale corrections (by scan). These are then compared to the adjustments and x-scan scale corrections determined by minimizing mean scan overlaps differences and Tycho-2 residuals. The latter process will be referred to as "Martinizing". The x-scan scale determination feature is still a few days away from being available for use in the Martinizing algorithm, In order to have a consistent set of numbers which include the x-scan scale effect, the scan-segment adjustments determined using UCAC were fed into the OLST (mean overlap differences) file to compute residual overlaps. Histograms of the resulting residual differences will be presented showing the improvement with each correction. Finally a comparison of difference sigmas when using UCAC as the source of adjustments as opposed to the Martinizing algorithm will be presented.
II. Improvements by Type of Correction:
Figure 1 presents a comparison of overlap difference histograms (normalized) before and after the basic scan-segment adjustments. "Before" is in dashed red lines and "after" in solid black lines. X-scan histograms are in the left panel and in-scan histograms in the right. The x-scan sigma drops from 206 to 117 mas and the in-scan sigma from 195 to 78. Figure 2 histograms scan-segment residual differences "before" and "after" the distortion corrections. The in-scan sigma shows another drop from 78 to 58 mas, but x-scan drops only from 117 to 110, leaving it almost twice as large as in-scan. The problem remaining is x-scan scale factor error, as can be seen from Figure 3 with histograms before and after the x-scan scale correction. X-scan sigma drops by a factor of two down to 54 mas, bringing it in line with the unaffected in-scan sigma of 58 mas. Finally, Figure 4 shows rather dramatically the full improvement over v2 processing. Mean scan-segment differences after v2 processing are histogramed in red and residual differences after block adjustment, distortion correction and scale-x correction are histogramed in black.
III. Comparison with Martinized Results:
The same analysis was performed using scan-segment adjustments determined via the Martinizing
algorithm, minus the x-scan scale adjustments. Results from the two methods are compared in
Table 1. All numbers in the table are standard deviations of the
residual scan-segment overlaps expressed in mas. Note that they provide about the same improvement,
with Martinizing falling off by only a few mas in-scan. Since the Martinizing was done on the single
Dec band and with preliminary weighting factors, further improvements should be possible.
http://spider.ipac.caltech.edu/staff/hlm/2mass/PB17/PB17.html
Comments to: Howard McCallon
Last update: 07 June 2001