IPP Software Navigation Tools IPP Links Communication Pan-STARRS Links

Changeset 39840


Ignore:
Timestamp:
Dec 8, 2016, 5:30:05 PM (10 years ago)
Author:
eugene
Message:

add updated text to calibration

File:
1 edited

Legend:

Unmodified
Added
Removed
  • trunk/doc/release.2015/ps1.calibration/calibration.tex

    r39838 r39840  
    1 \documentclass[iop,floatfix]{emulateapj}
     1% \documentclass[iop,floatfix]{emulateapj}
    22% \pdfoutput=1
    33
    44% see latex.readme.txt for notes on using the PS1 template
    5 %\documentclass[12pt,preprint]{aastex}
     5\documentclass[12pt,preprint]{aastex}
    66%\documentclass[manuscript]{aastex}
    77%\documentclass[preprint2]{aastex}
    88%\documentclass[preprint2,longabstract]{aastex}
    99\RequirePackage{color}
    10 % \input{astro.sty}
     10\RequirePackage{code}
     11\input{astro.sty}
    1112
    1213% online version may use color, but print version needs b/w
     
    8687
    8788% insert additional keywords as appropriate:
    88 \keywords{Surveys:\PSONE }
     89%\keywords{Surveys:\PSONE }
    8990
    9091\section{Introduction}\label{sec:intro}
     
    141142images are also corrected for fringing: a master fringe pattern is
    142143scaled to match the observed fringing and subtracted.  Mask and
    143 variance image arrays are generated with the \changed{detrend
    144   analysis} and carried forward at each stage of the IPP processing.
    145 Source detection and photometry are performed for each chip
    146 independently.  As discussed below, preliminary astrometric and
    147 photometric calibrations are performed for all chips in a single
    148 exposure in a single analysis. 
     144variance image arrays are generated with the detrend analysis and
     145carried forward at each stage of the IPP processing.  Source detection
     146and photometry are performed for each chip independently.  As
     147discussed below, preliminary astrometric and photometric calibrations
     148are performed for all chips in a single exposure in a single analysis.
    149149
    150150Chip images are geometrically transformed based on the astrometric
     
    266266A third form of the astrometry model is used in the context of the
    267267calibration determined within the DVO database system.  We retain the
    268 two levels of transformations (chip $\rtarrow$ focal plane $\rtarrow$
     268two levels of transformations (chip $\rightarrow$ focal plane $\rightarrow$
    269269tangent plane), but the relationship between the chip and focal plane
    270270is represented with only the linear terms in the polynomial,
     
    359359stars.  For all possible pairs between the two lists, the values of
    360360\[
    361 $\Delta X = X^{\rm ref}_{\rm chip} - X^{\rm obs}_{\rm chip}\\
    362 $\Delta Y = Y^{\rm ref}_{\rm chip} - Y^{\rm obs}_{\rm chip}
     361\Delta X = X^{\rm ref}_{\rm chip} - X^{\rm obs}_{\rm chip}\\
     362\Delta Y = Y^{\rm ref}_{\rm chip} - Y^{\rm obs}_{\rm chip}
    363363\]
    364364are generated.  The collection of $\Delta X, \Delta Y$ values are
     
    404404The astrometry solutions from the independent chip fits are used to
    405405generate a single model for the camera-wide distortion terms.  The
    406 goal is to determine the two stage fit (chip $\rtarrow$ focal plane
    407 $\rtarrow$ tangent plane).  There are a number of degenerate terms
     406goal is to determine the two stage fit (chip $\rightarrow$ focal plane
     407$\rightarrow$ tangent plane).  There are a number of degenerate terms
    408408between these two levels of transformation, most obviously between the
    409409parameters which define the constant offset from chip to focal plane
     
    424424
    425425\note{describe the output smf file?}
     426
     427\note{discuss the real-time photometric calibration}
    426428
    427429\section{DVO Description}
     
    522524contains 9 times as many rows as the Average table.  The order of the
    523525table is fixed in relation to the Average table: row $i$ of Average
    524 defines the object with photometry contained in rows $9i \rtarrow 9i +
     526defines the object with photometry contained in rows $9i \rightarrow 9i +
    5255278$ ($i$ is zero counting). 
    526528
     
    716718such as the instrumental variations and atmospheric attenuation. 
    717719\[
    718 M_{rel} & = & m_{inst} + ZP + M_{cal} \\
     720M_{rel} = m_{inst} + ZP + M_{cal}
    719721\]
    720722
     
    877879values for each GPC1 chip (40x40 pixels per point).  \note{show the
    878880  flat-field residual images, discuss the features?}.  We then used
    879 \node{setphot} to apply this new flat-field correction, as well as the
     881\code{setphot} to apply this new flat-field correction, as well as the
    880882ubercal flat-field corrections, to the data in the database.  At this
    881883point, we re-ran the entire relphot analysis to determine zero points
     
    896898the bright end.  \note{recommendation}
    897899
     900\section{PV3 DVO Master Database}
     901
     902Data from the GPC1 chip images, the stack images, and the warp images
     903are loaded into DVO using the real-time analysis astrometric
     904calibration to guide the association of detections into objects.
     905After the full PV3 DVO database was constructed, including all of the
     906chip, stack, and warp detections, several external catalogs were
     907merged into the database.  First, the complete 2MASS PSC was loaded
     908into a stand-alone DVO database, which was then merged into the PV3
     909master database.  Next the DVO database of synthetic photometry in
     910the PS1 bands (see Section~\ref{sec:synthdb}) was merged in.  Next,
     911the full Tycho database was added, followed by the AllWISE database.
     912After the Gaia release in August 2016, we generated a DVO database of
     913the Gaia positional and photometric information and merged that into
     914the master DVO database.
     915
     916\note{need to describe the assignment of flags, etc, for the external
     917  data sources}.
     918
    898919\section{Astrometry Analysis}
    899 \begin{verbatim}
    900 * initial astrometry based on real-time calibration
    901 * relative astrometry calibration of images
    902   * bright objects, images
    903 * first pass to deter
    904 \end{verbatim}
    905 
    906 \section{Systematic Residuals}
    907 
    908 \subsection{Camera-Scale Trends}
     920
     921Once the full PV3 dataset loaded into the master PV3 DVO database,
     922along with supporting databases, and the photometric calibrations were
     923performed, relative astrometry could be performed on the database to
     924improve the overall astrometric calibration.
     925
     926In many respects the relative astrometric analysis is similar to the
     927relative photometric analysis: the repeated measurements of the same
     928object in different images are used to determine a high quality
     929average position for the object.  The new average positions are then
     930used to determine improved astrometric calibrations for each of the
     931images.  These improved calibrations are used to set the observed
     932coordinates of the measurements from those images, which are in turn
     933used to improve the average positions of the objects.  The whole
     934process is repeated for several iterations.  Like the photometric
     935analysis, the astrometric analysis is performed in a parallel fashion
     936with the same concept that specific machines are responsible for
     937exposures and objects which land within their regions of
     938responsibility, defined on the basis of lines of constant RA and DEC.
     939Between iteration steps, the astrometric calibrations are shared
     940between the parallel machines as are the improved positions for
     941objects controlled by one machine but detect in images controlled by
     942another machine.  Like the photometric analysis, the entire sky is
     943processed in one pass.  However, there are some important differences
     944in the details.
     945
     946\subsection{Systematic Effects}
     947
     948First, the astrometric calibration has a larger number of systematic
     949effects which must be performed.  These consist of: 1) the
     950Koppenh\"offer Effect, 2) Differential Chromatic Refraction, 3) Static
     951deviations in the camera.  We discuss each of these in turn below.
     952
     953\subsubsection{Koppenh\"offer Effect}
     954
     955The Koppenh\"offer Effect was first identified (DATE) by Johannes
     956Koppenh\"offer (MPE) as part of the effort to search for planet
     957transists in the Stellar Transit Survey data.  He noticed that the
     958astromety of bright stars and faint stars disagreed on overlapping
     959chips at the boundary between the STS fields.  After some exploration,
     960it was determined that the X coordinate of the brightest stars was
     961offset from the expected location based on the faint stars for a
     962subset of the GPC1 chips.  The essence of the effect was that the
     963bright stars were advanced along the serial register more quickly than
     964they should have been.  The brighter the star, the more the charge
     965cloud was pushed ahead on the serial register.  The amplitude of the
     966effect was at most \note{XXX}.  Only the \note{2-phase} chips suffered
     967from this effect.  By adjusting the \note{which?} voltages on the
     968camera, the effect was prevented in exposures after \note{DATE}.
     969However, this left \note{XXX,XXX exposures (XX\%)} already
     970contaminated by the effect. 
     971
     972We measured the Koppenh\"offer Effect by accumulating the residual
     973astrometry statistics for \note{how many} stars.  For each chip, we
     974measured the mean X and Y displacements of the astrometric residuals
     975as function of the instrumental magnitude of the star divided by the
     976FWHM$^2$.  \note{was there is significant difference using a surface
     977  brightness version?}  We measured the trend for all chips in a
     978number of different time ranges and found the effect to be quite
     979stable, in the period where it was present.  The effect only appeared
     980in the serial direction.  Figure~\ref{fig:koppenhoefer} shows the KE
     981trend for a typical affected chip both before and after the
     982correction.  For the PV3 dataset, we re-measured the KE trends using
     983stars in the Galactic pole regions after an initial relative
     984astrometry calibration pass: the Galactic pole is necessary because
     985the real-time astrometric calibration relies largely on the fainter
     986stars which are not affected by the KE.  The trend is then stored in a
     987form which can be applied to the database measurements.
     988
     989\subsubsection{Differential Chromatic Refraction}
     990
     991Differential Chromatic Refraction (DCR) affects astrometry because the
     992reference stars used the calibrate the images are not the same color
     993(SED) as the rest of the stars in the image.  For a given star of a
     994color different from the reference stars, as exposures are taken at
     995higher airmass, the apparent position of the star will be biased along
     996the parallactic angle.  While it is possible to build a model for the
     997DCR impact based on the filter response functions and atmospheric
     998refraction, we have instead elected to use an empirical correction for
     999the DCR present in the PV3 database.  We have measured the DCR trend
     1000using the astrometric residuals of millions of stars after performing
     1001an initial relative astrometry calibration.  We define a blue DCR
     1002color ($g-i$) to be used when correcting the filters \gps,\rps,\ips, and a red
     1003DCR color ($z - y$) to be used when correcting the filters $zy$.  In
     1004the process of performing the relative astrometry calibration, we
     1005record the median red and blue colors of the reference stars used to
     1006measure the astrometry calibration for each image.  As we determine
     1007the astrometry parameters for each object in the database, we record
     1008the median red and blue reference star colors for all images used to
     1009determine the astrometry for a given object.  For each star in the
     1010database, we know both the color of the star and the typical color of
     1011the reference stars used to calibrate the astrometry for that star. 
     1012
     1013We measure the mean deviation of the residuals in the parallactic
     1014angle direction and the direction perpendicular to the parallactic
     1015angle.  For each filter, we determine the DCR trend as a function of
     1016the difference between the star color and the reference star color,
     1017using the red or blue color approriate to the particular filter, times
     1018the tangent of the zenith distance.  Figure~\ref{fig:DCR} shows the
     1019DCR trend for the 5 filters \grizy, as well as the measured
     1020displacement in the direction perpendicular to the parallactic angle.
     1021We represent the trend with a spline fitted to this dataset.  The DCR
     1022trend has an amplitude of \note{XXX - XXX} in the five filters. 
     1023
     1024\note{write down the DCR formalae for reference}.
     1025
     1026\subsubsection{Astrometric Flat-field}
     1027
     1028After correction for both KE and DCR, we observe persistent residual
     1029astrometric deviations which depend on the position in the camera.  We
     1030construct an astrometric ``flat-field'' response by determining the
     1031mean residual displacement in the X and Y (chip) directions as a
     1032function of position in the focal plane.  We have measured the
     1033astrometric flat using a sampling resolution of 40x40 pixels, matching
     1034the photometric flat-field correction images.
     1035Figure~\ref{fig:astroflat} shows the astrometric flat-field images for
     1036the five filters \grizy\ in each of the two coordinate directions.
     1037These plots show several types of features.
     1038
     1039The dominant pattern in the astrometric residual is roughly a series
     1040of concentric rings. The pattern is similar to the pattern of the
     1041focal surface residuals measured by (REF), which also has a concentric
     1042series of rings with similar spacing.  The ``tent'' in the center of
     1043the focal surface reflected in these astrometry residual plots.  Our
     1044interpretation of the structure is that the deviations of the focal
     1045plane from the ideal focal surface introduces small-scale PSF changes,
     1046presumably coupled to the optical aberrations, which result in small
     1047changes in the centroid of the object relative to the PSF model at
     1048that location.  Since the PSF model shape parameters are only able to
     1049vary at the level of a 6x6 grid per chips, the finer structures are
     1050not included in the PSF model.  The PV2 analysis shows the ring
     1051structure more clearly, with a pattern much more closely following the
     1052focal surface deviations.  In the PV2 analysis, the PSF model used at
     1053most a 3x3 grid per chip to follow the shape variations, so any
     1054changes caused by the optical aberrations would be less well modeled in
     1055the PV2 analysis, as we observe.
     1056
     1057A second pattern which is weakly seen in several chips consists of
     1058consistent displacements in the X (serial) direction for certain
     1059cells.  This effect can be seen most clearly in chips XY45 and XY46.
     1060In the PV2 analysis, this pattern is also more clearly seen.  In this
     1061case, the fact that the astrometric model used polynomials with a
     1062maximum of 3rd order per chip means the deviation of individual cells
     1063cannot be followed by the astrometric model. 
     1064
     1065A third effect is seen at the edge of the chips, where there appears
     1066to be a tendency for the residual to follow the chip edge.  The origin
     1067of this is unclear, but likely caused by the astrometry model failing
     1068to follow the underlying variations because of the need to extrapolate
     1069to the edge pixels.  Finally, we also identify an interesting effect
     1070{\em not} visible at the resolution of these astrometric flat-field
     1071images.  Fine structures are observed at the \approx 10 pixel scale
     1072similar to the ``tree rings'' reported by the DES team and others
     1073(G. Berstein REF \& REFS).  We explore these tree rings in detail in
     1074\note{SECTION or REF?}.
     1075
     1076After the initial analysis to measure the KE corrections, DCR
     1077corrections, and astrometric flat-field corrections, we applied these
     1078corrections to the entire database.  Within the schema of the
     1079database, each measurement has the raw chip coordinates
     1080(\code{Measure.Xccd,Yccd}) as well as the offset for that object based on each of
     1081these three corrections: \code{Measure.XoffKH,YoffKH,
     1082  Measure.XoffDCR,YoffDCR, Measure.XoffCAM,YoffCAM}.  The offsets are
     1083calculated for each measurement based on the observed instrumental
     1084chip magnitudes and FWHM for the Koppenhoffer Effect, on the average
     1085chip colors and the altitude \& azimuth of each measurement for the
     1086DCR correction, and on the chip coordinates for the astrometric
     1087flat-field corrections.  The corrections are combined and applied to
     1088the raw chip coordinates and saved back in the database in the fields
     1089\code{Measure.Xfix,Yfix}.  At this point, we are ready to run the
     1090full astrometric calibration.
     1091
     1092\subsection{Galactic Rotation and Solar Motion}
     1093
     1094The initial analysis of the PV2 astrometry used the 2MASS positions as
     1095an inertial constraint: the 2MASS coordiates were included in the
     1096calculation of the mean positions for the objects in the database,
     1097with weight corresponding to the reported astrometric errors.  In this
     1098analysis, the object positions used to determine the calibrations of
     1099the image parameters ignored proper motion and parallax.  After the
     1100image calibrations were determined, then individual objects were
     1101fitted for proper motion and possibly parallax, as discussed in detail
     1102below.
     1103
     1104Using the PV2 analysis of the astrometry calibration, we discovered
     1105large-scale systematic trends in the reported proper motions of
     1106background quasars.  This motion had an amplitude of 10 - 15
     1107milliarcseconds per year and clear trends with Galactic longitude.  We
     1108also observed systematic errors of the mean positions with respect to
     1109the ICRF milliarcsecond radio quasar positions, with an amplitude of
     1110\approx 60 milliarcseconds, again with trends associated with Galactic
     1111longitude.  Since the 2MASS data were believed to have minimal average
     1112deviations relative to the ICRF quasars, this latter seemed to be a
     1113real effect. 
     1114
     1115We realized that both the proper motion and the mean position biases
     1116could be caused by a single common effect: the proper motion of the
     1117stars used as reference stars between the 2MASS epoch (\approx 2000)
     1118and PS1 epoch (\approx 2012).  Since we are fitting the image
     1119calibrations without fitting for the proper motions of the stars, we
     1120are in essencence forcing those stars to have proper motions of 0.0.
     1121The background quasars would then be observed to have proper motions
     1122corresponding to the proper motions of the reference stars, but in the
     1123opposite direction.  We demonstrated that the observed quasar proper
     1124motions agreed well with the distribution expected if the median
     1125distance to our reference stars was \approx 500 pc. 
     1126
     1127For PV3, we desired to address this bias by including our knowledge
     1128about the distances to the reference stars and the expected typical
     1129proper motions for stars at those distances.  With some constraint on
     1130the distance to each star, we can determine the expected proper motion
     1131based on a model of the Galactic rotation and solar motions.  We can
     1132then calculate the mean positions for the objects keeping the assumed
     1133proper motion fixed.  When calibrating a specific image, the reference
     1134star mean position is then translated to the expected position at the
     1135epoch of that image.  The image calibration is then performed relative
     1136to these predicted postions.  This process naturally accounts for the
     1137proper motion of the reference stars.  In order to make the
     1138calibrations consistent with the observed coordinates of an external
     1139inertial reference, we perform the iterative fits using the technique
     1140as described, but assign very high weights in the initial iterations
     1141to the inertial reference, and reduce the weights as the astrometric
     1142calibration iterations proceed.
     1143
     1144In order to perform this analysis, we need estimated distances for
     1145every reference star used in the analysis.  Green et al (REF)
     1146performed SED fitting for 800M stars in the 3$\pi$ region using PV2
     1147data.  The goal of this work was to determine the 3D structure of the
     1148dust in the galaxy.  By fitting model SEDs to \note{all?} stars
     1149meeting a basic data quality cut \note{(describe)}, they determined
     1150the best spectral type, and thus $T_{\rm eff}$, absolute $r$-band
     1151magnitude, distance modulus, and extinction $A_V$ (the desired output
     1152and used to determine the dust extinction as a function of distance
     1153throughout the galaxy).  We use the distance modulus determined in
     1154this analysis to predict the proper motions. 
     1155
     1156To convert the distances to proper motions, we use the Galactic
     1157rotation parameters ($A,B$) = (14.82,-12.37) km sec$^{-1}$ pc$^{-1}$
     1158and Solar motion parameters ($U_{\rm sol}, V_{\rm sol}, W_{\rm sol}$)
     1159= (9.32, 11.18, 7.61) km sec$^{-1}$ as determined by Feast \&
     1160Whitelock (REF) using Hipparchos data.  Proper motions are determined
     1161from the following:
     1162\begin{eqnarray}
     1163\mu^{\rm gal}_{l} & = & (A \cos (2 l) + B) \cos (b) \\
     1164\mu^{\rm gal}_{b} & = & \frac{-A \sin (2 l) \sin (2 b)}{2} \\
     1165\mu^{\rm sol}_{l} & = & \frac{U \sin(l) - V \cos(l)}{d} \\
     1166\mu^{\rm sol}_{b} & = & \frac{(U \cos(l) + V \sin(l)) \sin(b) - W \cos(b)}{d}
     1167\end{eqnarray}
     1168where $d$ is the distance and $l,b$ are the Galactic coordintes of the
     1169star. \note{some reference?}  Note that the proper motion induced by
     1170the Galactic rotation is independent of distance while the reflex
     1171motion induced by the solar motion decreases with increasing
     1172distance.  Also note that this model assumes a flat rotation curve for
     1173objects in the thin disk; any reference stars which are part of
     1174the halo population will have proper motions which are not
     1175described by this model; the mostly random nature of the halo motions
     1176should act to increase the noise in the measurement, but should not
     1177introduce detectable motion biases.  Also, if the distance modulus is
     1178not well determined, we can assume the object is simply following the
     1179Galactic rotation curve and set a fixed proper motion.  If we do not
     1180have a distance modulus from the Green et al analysis, we assume a
     1181value of 500pc. 
     1182
     1183\note{plots to show how well this worked for PV3 pre Gaia}
     1184
     1185\subsection{Gaia Constraint}
     1186
     1187After the full relative astrometry analysis was performed for the PV3
     1188database, the Gaia Data Release 1 became available.  This afforded us
     1189the opportunity to constrain the astrometry on the basis of the Gaia
     1190observations.  Gaia DR1 objects which are bright enough to have proper
     1191motion and parallax solutions are in general saturated in the PS1
     1192observations.  Thus, we are limited to using the Gaia mean positions
     1193reported for the fainter stars.  We extracted all Gaia sources
     1194\note{not marked as a duplicate} from \note{where?} and generated a
     1195DVO database from this dataset.  We then merged the Gaia DVO into the
     1196PV3 master DVO database.  We re-ran the complete relative astrometry
     1197analysis using Gaia as an additional measurement.  We applied the
     1198analysis described above, applying the estimated distances to
     1199determine preliminary proper motions.  The Gaia mean epoch is reported
     1200as 2015.0, so all Gaia measurements were assigned this epoch.  We
     1201wanted to ensure the Gaia measurements dominated the astrometric
     1202solutions, so we made the weight very high for the Gaia points:
     12031000$\times$ the nominal weight in the initial fits (to lock down the
     1204reference frame), decreasing to 100$\times$ the nominal weight for the
     1205last fits.  We also retained the 2MASS measurements in the analysis,
     1206but gave them somewhat lower weights than Gaia: while the 2MASS data
     1207does not have the accuracy of Gaia, the coverage is known to be quite
     1208complete, while the Gaia DR1 has clear gaps and holes.  Having 2MASS,
     1209even at a lower weight, helps to tile over those gaps.
     1210
     1211\note{Figures showing the Gaia residuals}
    9091212
    9101213\section{Discussion}
Note: See TracChangeset for help on using the changeset viewer.