Statistical Studies on the Results Obtained

STATISTICAL STUDIES ON THE RESULTS OBTAINED

WITH THE GSC-II PIPELINE AT ESO/ECF AND STSCI

by

R. Pannunzio, R. Morbidelli and A. Spagna

Osservatorio Astronomico di Torino

Internal Report nr.47/99

Pino Torinese, 16 July 1999

ABSTRACT

In the present report, preliminary studies on the accuracy of the results obtained by means of the pipeline of the GSC-II project at the ESO/ECF Institute of Garhing (D), and at the STScI of Baltimore (U.S.A.), are given. For this study, only the first reduced 1,536 plates of the ESO/ECF Institute and the 1,776 plates of the STScI, have been taken into account, considering the most significative parameters collected in the DID file reports of the GSC-II pipeline.

This paper is the first of a series of reports that should give an indication of the accuracy in the results of the pipelines used both to the ESO/ECF Institute of Garching and at the STScI of Baltimore.

The final goals of these studies are the comparison of the results obtained in different Institutes with the same pipeline but using different parameters (different catalogs, thresholds and so on) in order to have some criteria to follow for the realization of the Q.A. on the plates of the CRA proposals reduced at Torino.

INTRODUCTION

At the beginning of this year the ESO/ECF Institute of Garching (Germany) had reduced 1,536 Schmidt plates belonging to different Surveys using the current version of the pipeline of that period, while the STScI, about in the same period, had processed 1,776 plates with the same pipeline but using different options. The DID file reports of the pipelines of the two Institutes have been sent to our Astronomical Observatory for checking the reliability of the results and at the same time to study the global quality of the data of 2 billion objects produced up to now.

From the DID file reports, the 56 most significative variables have been selected and converted in a XDR file easily-handled in IDL environment.

Table 1. Number and type of plates processed.

The set of images includes both plates scanned with a 25 m pixel (14,000x14,000 pixels, PIM files) and plates scanned with a 15 m pixel (23,040x23,040, pixels VIM files). Table 1 lists the number of images processed by STScI and ESO/ECF respectively.

These plates are taken from the all sky surveys of the Palomar Observatory (POSS-I, POSS-II, PAL QV, SERC-EJ) and from the Anglo-Australian Observatory surveys (SERC-J, SERC-ER, SES-R). Table 2 reports the main properties of these surveys and the number of plates reduced by the two Institutes.

Table 2. Surveys processed for GSC-II

The Catalogs used for the astrometric and photometric calibrations are the PPM and TYCHO Catalogs respectively for the plates processed by ESO/ECF, while the PPM and the TYCHO - GSPC-II are the Catalogs used by STScI.

The gnomonic projection for the astrometric plate-solution has been adopted by both the Institutes.

Finally, object classification (Star/Non Star/Defect) is based on the default criteria (i.e. same Decision-Trees).

These are practically the most important conventional parameters required by the s/w pipeline.

In the next section a detailed statistical analysis of the pipeline results performed in both Institutes, is shown.

PLATE DISTRIBUTION

The first statistics realized with the variables collected by the DID file reports of the ESO/ECF and STScI pipelines is the distribution of the plate centers in the sky.

In Fig. 1A are shown the plate centers of different Surveys in galactic coordinates of the ESO/ECF plates, while in Fig.1B the same plate centers are represented in equatorial coordinates. How it is possible to see the 1,536 plates are quasi uniformly distributed on the whole sky with the exception of a narrow belt close to the celestial equator (Fig 1B) while in galactic coordinates (Fig. 1A) the distribution of the plates along the galactic plane, at present, is low and not uniform.

In particular Fig. 1B shows the plates XP, corresponding to the POSS-II R Survey, correctly placed in the north emisphere, while the plates XS and ER corresponding to the Surveys AAO SES and SERC ER respectively are placed in the southern emisphere. Similar considerations for the plates XJ corresponding to the Survey POSS-II J that are placed in the north, while the plates S referred to the SERC J Survey are all placed in the south.

The distribution of the STScI plates both in galactic and equatorial coordinates of Fig. 1C and 1D show no appreciable differences with respect to the ESO/ECF plates.

The only difference is the presence of some XE plates of first-generation scan (PIM images) at very high declination not included in the previous ESO/ECF plate distributions.

1

1

OBJECT DISTRIBUTION

The set of Figs. 2 give the distribution of the total number of objects per plate in function of the galactic latitude and longitude for both Institutes. More precisely Fig. 2A shows the total number of objects obtained by the sum of the single stars and of the children (given by the deblending of multiple objects) present on each plate in function of the galactic latitude for the ESO/ECF plates with a galactic longitude included in +/- 45 with respect to the galactic center. As expected, the trend of the total objects increases simmetrically in the two galactic emispheres with the decreasing of the absolute value of the galactic latitude. However, in the same time, it is evident that the two exponential distributions of points are not simmetric to the line of galactic latitude 0 but the center of simmetry of the two point distributions is shifted in the southern galactic emisphere of few degrees. This is probably due to the fact that the Sun is located about 20 parsec above the galactic plane. Then, the stellar density observed at a given northern galactic latitude +|b| is lesser than the density seen at -|b| in the southern emisphere.

Figure 2B shows the same effect of Fig. 2A but for plates selected in direction of the galactic anti-center and covering a range of +/- 45 from the galactic meridian at l=180. In this figure we note that the maximum number of objects (about 1 million) is more than 3 times less than the counts toward the galactic center. Moreover in this case the asymmetric shift towards southern latitudes is not evident, maybe because of the lower stellar density and because of the contamination of extra-galactic sources.

Figure 2C and 2D show the object distributions as a function of the galactic latitude for the STScI plates toward the center and anti-center of the Galaxy, with the same constraints used for the ESO/ECF plates. Looking at the Figs. 2C and 2D and comparing them with the corresponding figures (Figs. 2A and 2B) of the ESO/ECF plates we don’t see particular differences in the trend of the point distributions for which the same considerations made for the ESO/ECF plates are valid for the STScI plates. A global comparison of the Figs. 2 put in evidence that, in general the Surveys in Blu Band (XJ and S-Plates) shown for constant galactic latitudes less objects than the corresponding surveys in Red Band (GR/ER/XS and XP-plates) independently from the magnitude limits of the different surveys.

1

1

DISTRIBUTION OF STARS, NON STARS AND DEFECTS

As far as the objects counts in all the plates are concerned, the Tables 3 at the end of the present report give in second column the whole number of objects with different classifications both for the ESO/ECF and for the STScI. It is possible to see that the total number of objects till now processed (single+children) is about 1 billion for both Institutes.

Furthermore the Tables 3 show that the typical number of objects per plate (median) is centered around to 350,000 reaching the maximum values of 4,000,000 - 4,500,000 of objects in some few plates taken on the galactic plane (see also the Figs. 2).

The frequency distribution of the objects classified as Stars, Non Stars and Defects are presented in Fig. 3A (ESO/ECF) and in Fig. 3B (STScI).

The analysis of these figures offers a first check on the reliability of the pipeline classifier. Indeed a small and almost constant number of objects per plate classified as Defects (about 20,000 as deduced from the column of the median of the two Tables 3) is realistic because the generation of Defects should be independent from the number of the objects in the plate.

On the other hand, the broad distributions of the stellar and non-stellar objects depends clearly on the latitude effect shown in Figs. 2A-2D. Note also that the mean number of Non Stars per plate is slightly higher than the mean number of Stars (cfr. Tables 3).

Figures 4 show how many objects are classified as Stars as a function of the total number of detected objects. The various Surveys at High and Low Galactic Latitude for the ESO/ECF and STScI respectively are reported with different symbols. Here, we assume as high galactic latitude (HGL) plates those having a number of objects lesser than 500,000 while low galactic latitude (LGL) plates those with more than 500,000 objects. According to Figs. 2A-2D, such a threshold corresponds to a galactic latitude b=20-30.

The number of Stars is very correlated to the total number of objects along two straight branches which are more evident in the HGL plates and which correspond to two plate populations having different probability of “finding” stellar objects.

In particular the upper branch presents a higher probability of classifying objects as Stars. It includes the plates which have been scanned with the larger 25 micron-pixel (PIM images) and consequently also many extended objects (blended stars, galaxies, etc.) on these plates have been erroneously classified as Stars. This situation is present also at LGL but is less evident probably because galactic plane fiels were processed later when their images have been replaced with the higher resolution (15 micron) scans.

Finally, in the lower branch which corresponds to the reliable 15 micron scans (VIM images), the number of Stars does not particularly depend on the type of Survey (having different colors and magnitude limits). In fact the spread

1

1

1

of the points is very narrow until to HGL plates with 300,000 objects, and after that the dispersion does not exceed 10%.

The situation is practically equivalent for the STScI plates of Fig. 4C and 4D. Also in this case there are two branches corresponding to the PIM images (upper branch) and to the VIM images (lower branch). In fact in the upper branch of both figures are reported the Stars classified in the S and XE-plates scanned to 25 micron-pixel, while in the lower branch there are all the surveys scanned to 15 micron.

The set of Figs. 5 have the same structure of the previous Figs.4 but are referred to the distribution of the Non Stars with respect to the total number of objects for HGL and LGL of the ESO/ECF and STScI plates. As expected, we see two distribution of points, but in a complementary situation, where the lower branch is referred to the PIM images having a lower efficiency of resolve extended objects, while the upper one corresponds to the VIM images with a higher percentage of detection.

Finally, comparing the counts of Star and Non Stars in Fig. 4 and 5 of the more reliable VIM images, we remark that the number of Nonstars is larger than the number of Stars in the HGL fields because galaxies dominates the counts at these magnitude limits (see Table 2). Apparently it is surprising to note that nonstellar objects dominates the classification also for LGL fields. The reason derives from the fact that in such crowded fields many blended stars are not resolved properly by the pipeline and are considered as extended nonstellar objects by the Classifier.

Also the set of Figures 6 is very similar to the previous ones but here the Defects as a function of the total number of objects case are reported for the plates processed by ESO/ECF and STScI. In this case the Defects detected in the PIM images are always lower of the corresponding VIM images probably because the large dimensions of the PIM pixel-size detects as Stars also the irregular shape of the plate defects (spots in the film, satellite trails etc.).

A general conclusion on the set of Figs. 4, 5 and 6 is that for a given density of objects per plate, the number of Stars, Non Stars and Defect is quite independent from the kind of Surveys but is strongly dependent from the used scansion (15 or 25 micron).

A complementary set of Figures is given by the distribution of the Stars, Non Stars and Defects for the single Surveys at High and Low Galactic Latitude and for both Institutes. It is interesting to remark that in Fig.7A and 7B where the objects of the ESO/ECF S-plates are reported the number of stars is greater than the Non Stars until that the number of the total objects per plate is lesser than about 500,000 (Fig. 7A - HGL) while the Non Stars become higher with respect to the Stars up to this value (Fig. 7B –LGL).

A similar situation is presented in the Figs. 7C and 7D in which are reported the objects of the STScI S-plates.

Completely different is the situation of the objects distribution when the XJ, XP ,ER/GR/XS–plates are considered in both Institutes. Indeed if we look at the set of Figs. 8, 9 and 10 it is possible to note that the number of Non Stars detected in each region of the sky (at Low and High Galactic Latitude) is always greater than the number of the Stars.

As discussed before, the difference of the object distributions seen in Figs. 7 (S-plates) with respect to the set of Figs. 8, 9 and 10 (all other surveys) is due to the fact that of S-plates are based on PIM and VIM images for HGL and LGL fields respectively, while all the other surveys (apart few XE-plates of STScI) include only VIM images.

The effect of the different resolution of PIM and VIM images on object classification is more evident in Figs. 11 where the distribution of the objects detected by the classifier in the S-plates for the PIM and VIM images are shown. In fact Fig.11A and 11C show an evident abundance of Stars with respect the Non Stars for S-plates PIM scan having less than 500,000 objects while the opposite is true for the PIM and VIM images with objects at low galactic latitudes.

The same conclusions can be confirmed looking at the Fig.12 in which only for the STScI we have some XE-plates with PIM images. The two branches of points classified as Stars and Non Stars show an inversions of the abundances around to 500,000 objects per plate.

However in all the previous statistics the number of objects classified as Defects is always small and independent from the different kinds of surveys.

ASTROMETRIC AND PHOTOMETRIC CALIBRATIONS

As far as the astrometry of the GSC-II project is concerned, both the Institutes have adopted the PPM Catalog in order to fit a plate solution based on a third order polynomial which transforms the plate coordinates (X,Y) to the standard coordinates (,) and then to celestial coordinates (, - J2000). The histograms of Figs.13A and 13B indicates the number of PPM stars contained in each plate processed in ESO/ECF and STScI. In both the distributions the most frequent number of PPM objects per plate is about 300 with a minimum of 100 and a maximum of about 600 - 800.

If we take into account the photometry of the GSC-II, photometric standards are necessary for defining the photometric transformations which converts the non-linear instrumental magnitude, 2.5 Log D, where D is the integrated density, to the magnitudes in the natural passband of the plate (see Table 2).

In this regard, before January 1999 ESO/ECF used photometric standards only from the TYCHO Catalog while STScI adopts the TYCHO and/or GSPC-II catalogs. In both Institutes the most frequent number of TYCHO objects is centered around 500 (Figs. 13C and 13D) while only for the STScI the most frequent number of GSPC2 objects is usually less than 100 because of the small area of the CCD photometric sequences.

The Figs. 14A and 14B show the distribution of the total astrometric error

obtained by the combination of the fit RMS in  and .

For the ESO/ECF Institute (Fig. 14A) the best astrometric accuracy is attained by the ER/XS and S surveys which present a typical error of about 0”.4 – 0”.5, while the accuracy decreases to 0”.6 for the XP-plates and to 1”.2 for the XJ-plates.

In Fig. 14B the distribution of the total astrometric error of the plates processed by STScI, is given. Also in this case the astrometric accuracy of the plates depends on the type of survey. In fact the GR/ER/XS-plates show a peak value of 0”.4 while for the S, XP-plates the accuracy decreases to 0”.6. Finally a bimodal distribution with two peaks around 0”.6 and 1”.0 are visible for the XJ-plates.

Looking at the Figs. 15 it is possible to see how the global astrometric accuracy does not depend on the number of objects per plate for a given survey while is strongly dependent by the surveys as quoted above.

The set of Figs.16 shows the photometric error of the ESO/ECF and STScI Institutes obtained by a polynomial fit of order 1 as default for the ESO/ECF, while for STScI the errors have been obtained (by default) with different polynomial fits till to the third order.