Changeset 40714 for trunk/doc/release.2015/ps1.calibration/calibration.tex

Timestamp:

May 5, 2019, 8:58:31 PM (7 years ago)

Author:

eugene

Message:

adding polar issue test from stefano

File:

• trunk/doc/release.2015/ps1.calibration/calibration.tex (modified) (1 diff)

trunk/doc/release.2015/ps1.calibration/calibration.tex

@@ -2817 +2817 @@
 proper motions will obviate the need to correct for the Galactic rotation.
 
+\section{Polar Astrometry Issues}
+
+Internal consistency testing of the PV3 stacks measurements indicated
+potential problems with the astrometric registration of the exposures
+in small areas near the North Pole.  These issues were originally
+suggested by a few high-latitude sources with significant differences
+in morphology or position across bands, including strong (and
+anomalous) apparent color gradients.  Direct investigation of a few of
+these anomalous sources demonstrated the presence of significant
+misalignments between exposures; one of the worst cases is shown in
+Figure~\ref{fig:pole.issue.exampe}.  While such sources appeared to be
+rare, astrometric registration errors have the potential to affect
+several different source properties: morphology and photometry in
+addition to astrometry.  Therefore we carried out an astrometric
+regsitration test for all skycells North of $ \delta=+70\deg$.
+
+\begin{figure*}[htbp]
+  \begin{center}
+  \includegraphics[width=\hsize,clip]{{pics/A1}.pdf}
+  \caption{\label{fig:pole.issue.example} Example of a stack source badly affected by polar astrometry failures.  Source from multiple detections from skycell 2643.093.}
+  \end{center}
+\end{figure*}
+
+This test was based primarily on the ``original detection positions'',
+\ie, the positions of sources (detections) found in individual
+exposures as measured after each exposure's astrometric calibration,
+but before recalibration of the combined values to the Gaia reference
+frame (described in Section 7.3).  We started by collecting the
+original detection positions (as defined above) for each skycell.  To
+ensure good signal-to-noise ratios and minimize potential spurious
+detections, we used only the top quartile (in flux) of detections
+within each chip.  We grouped these detections on a filter-by-filter
+basis within a radius of $ 2\farcs5 $ (10 pixels), ensuring that each
+group contained only one source per exposure, and retaining only
+groups with at least five detections; we then recorded the 2-D
+position dispersion for each group.  The mean positions for each group
+were cross-correlated against the Gaia DR2 sources, showing that these
+were real sources and providing information on their absolute
+astrometry.
+
+\begin{figure*}[htbp]
+  \begin{center}
+%  \includegraphics[width=\hsize,clip]{{pics/A2}.pdf}
+  \caption{\label{fig:pole.issue.example} Example of a stack source badly affected by polar astrometry failures.}
+  \end{center}
+\end{figure*}
+
+Overall, the vast majority of the detection groups thus defined have
+good consistency between source positions, resulting in an astrometric
+dispersion of 1 pixel or less.  A few ``bad'' groups, defined as
+having an internal dispersion $ > 1 $ pixel, can result from spurious
+sources or other anomalies, and are generally rare (fewer than a few
+percent of al groups).  However, some skycells have a significant
+fraction ($ > 10\%$) of bad groups.  Direct inspection demonstrates
+that the incidence of bad groups is related to astrometric
+registration failures.  Figure~\ref{fig:pole.astrom.failures} shows an
+example of a good and of a bad group.
+
+%% [Note: the rest of this
+%%   paragraph, and Figure A3, may be too much information for this
+%%   paper.]  It also appears that registration problems, when present,
+%% are not uniform within a skycell; Figure (A3) shows the difference
+%% between mean group position and the position of individual detections
+%% for all G band exposures overlapping skycell 2637.088, which has one
+%% of the worst-case mismatches in the g band.
+
+% caption: Map of astrometric displacement for all g-band exposures
+%    overlapping skycell 2637.088, with one of the worst astrometric
+%    registration issues. [Optional]
+
+\begin{figure*}[htbp]
+  \begin{center}
+  \includegraphics[width=\hsize,clip]{{pics/A4}.pdf}
+  \caption{\label{fig:pole.bad.histogram} Histogram of the fraction of bad groups for each skycell (red line).}
+  \end{center}
+\end{figure*}
+
+Bad skycells, defined as those with more than 10\% bad groups, are
+essentially limited to the North polar cap ($ \delta > +80^{\degree}$).
+Of the 2500 skycells in this region, 164, or 6.6\%, have more than 10\% 
+bad groups; 64 of these have more than 20\% bad groups.  By comparison,
+essentially no skycells between $ +70^\degree $ and $ +80^\degree $ have
+more than 10\% bad groups.  Figure~\ref{fig:pole.bad.histogram} shows a histogram
+of the fraction of bad groups for each skycell.
+
+In order to have an independent validation of the impact of this
+astrometric alignment issue, we also carried out a photometric test
+based on a comparison of stack to mean object photometry.  In the
+presence of modest registration errors, mean object photometry would
+not be affected, as individual detection woulds have the correct
+signal, and averaging their flux in catalog space would yield the
+correct total magnitude.  On the other hand, imperfect stacking would
+result in a dilution of the total signal on a pixel-by-pixel basis,
+and result in potentially larger estimated sizes and smaller total
+flux for stack sources.  Indeed, mean magnitudes are brighter than
+stack magnitudes for a significant fraction of the sources in the same
+skycells that are identified as bad by the relative astrometry test.
+Therefore we confirm that the astrometric registration issues result
+in poor stack photometry for the affected skycells.
+
+\note{discuss the cause of the failure due to the duplicates in the reference catalog, and the original polar astrometry failures}
+
+As a result of these tests, we decided to 1) exclude from the main DR2
+catalogs all sources in the skycells with more than 10\% bad groups,
+and 2) to reprocess all such skycells with an improved procedure.  The
+reprocessing was carried out in late 2018, and the astrometric
+registration test was repeated on the reprocessed exposures.  The
+reprocessing greatly ameliorated the registration issue, as shown
+Figure (A4).  Here the red line shows the histogram of the fraction of
+bad groups for each skycell {\sl before reprocessing}, while the black
+line refers to the results {\sl after reprocessing}.  The improvement
+is apparent.  After reprocessing, only 23 cells, instead of the
+original 164, exceed 10\% of bad groups, and even for these the
+fraction of bad groups is substantially reduced.  Sources in the
+previously bad, now fixed skycells will be included in an upcoming
+partial release.
+
+\note{the above is not quite accurate -- a test reprocess demonstrated
+  partial improvement, but did not use a totally repaired ref catalog.
+  we are running a new analysis based on a DR2-tied catalog with
+  pristine source set.}
+
 \section{Conclusion}