IRAS Explanatory Supplement
VII. Analysis of Processing
E. Point Source Processing Considerations
Table of Contents | Index | Previous Section | Next Section
- The Nature of Rejected Sources
- Bright Source Problems
- Sources of Incompleteness
- Effects of Failed Detectors
- Setting the Seconds-Confirmation Threshold
E.1 The Nature of Rejected SourcesDetections had to survive a number of stringent confirmation tests to become an HCON. The nature of those sources that became HCONs but did not go on to become cataloged objects is of interest to those trying to understand the completeness and reliability of the IRAS survey (see Chapter VIII).
E.1.a Single HCONsAway from highly confused regions there were three main causes of single HCONs. Single HCONs can be due to inertially fixed sources faint enough to be below the completeness threshold, to moving sources such as asteroids and comets, and to wholly spurious sources generated by noise, radiation hits, diffraction spikes, and debris near the spacecraft.
The number of true, fixed sources which appear only as single HCONs can be estimated from the completeness of the catalog in each flux range. A crude estimate based on the preliminary figures for completeness given in Section VIII.D suggests that half of all single HCON sources in the sky with two HCON coverages are real but incomplete; only one third of the single HCONs in the region covered with 3 HCONs are predicted to be real. An estimate of the number of asteroids and comets is given in Section VII.F.
Figures VII.Ap.17-20 show the distribution of single HCON sources detected in a given band in Galactic coordinates. These plots look much like the plots shown earlier for catalog sources, with the obvious addition of asteroids and comets in the ecliptic plane.
E.1.b Rejected Weeks-Confirmed SourcesRoughly 10,000 weeks-confirmed sources were rejected because they did not have consistent sightings of acceptable quality in at least one wavelength band. Most of these sources were caused by the ubiquitous infrared cirrus.
Regions with high source density were specially processed to generate sources that were relatively isolated, had repeatable fluxes and stood out prominently as point sources against the local background. High source density rules were applied to measurements in those wavelength bands for which the density of sources in a 1 sq. deg bin exceeded the confusion limit threshold. Sources were rejected from the catalog only if the high source density processor rejected all the bands that it processed and if the bands that it did not process failed to meet the normal rules for inclusion in the catalog (Section V.H.5).
The number of 1°2 bins precessed according to high source density rules was 690, 631, 1382 and 6192 at 12, 25, 60 and 100 µm, respectively, or roughly one-seventh of the sky at the longest wavelength. In all, some 300,000 individual sources consisting of one or more hours-confirmed sightings were examined. Of these, approximately 170,000 were immediately rejected as having only one hours-confirmed sighting. Under no circumstances could any of these have reached the final catalog. About 40,000 were rejected for failing to have two hours-confirmed sightings with at least two "perfect" sightings (with FSTAT = 7) in at least one band.
Another 17,000 weeks-confirmed sources subsequently failed one or more of the tests described in Section V.H.6. Because these objects lacked high quality fluxes in any band they were thus excluded from the catalog.
Table VII.E.1 gives the fraction of sources for which a particular reason was the cause of the rejection of a measurement in a particular band. The tests were applied in the order listed in the table and represent an obstacle course that all bands being processed according to high source density rules had to survive. Since a band could be rejected at any point in the sequence, the reason that finally led to its rejection is counted in the table.
It can be seen from the table that the dominant reasons for rejection vary with wavelength. In all cases the first test which demanded at least two sightings with a correlation coefficient greater than 0.97 removed the bulk of the sources, particularly at 60 µm . The correlation coefficient test served to reject both low signal-to-noise sources and extended sources. The latter evidently dominate confused regions at 60 µm. The effects of neighboring sources accounted for most of the rest of the rejections.
Figure VII.E.1a-d The effects of the high density criteria are
shown in the four wavelength bands. Plotted is the number of bins
with a specified number of sources with a measured flux in a given
band. The open bars show the number of bins with a given number
of fluxes before high source density processing. The solid bars
show the number of high quality fluxes that remain after processing.
The striped bars show the number of high and medium quality fluxes
Figures VII.E.1a-d show the effects of high source density processing on the number of sources within a 1°2 bin in the four wavelength bands. In these histograms the open bars give the number of bins containing the specified number of sources, where in this case a source is counted in a band if it meets the criterion of having at least two hours-confirming sightings with FSTAT 3. The confusion limit corresponding to 25 beams per source is marked. The solid bars show the results after the high source density criteria are applied. In this case a source was counted only if it had a high quality flux in a given band according to the high source density criteria. The figure shows a drastic reduction in the average source density after the more stringent criteria were applied. There are relatively few 1 sq. deg bins with more than the confusion limited number of sources. The striped bars indicate that the number of sources per sq, deg increases when high and moderate quality fluxes are included suggesting that confusion effects may be important for moderate quality fluxes in highly populated areas.
E.2 Bright Source ProblemsThe point-spread function of the telescope caused bright point sources to illuminate many more detectors in the focal plane than just those over which their image centers passed. These extremely bright sources caused special problems and received special handling during the data processing. During final source selection the WSDB was searched for spurious sources due to optical cross-talk from bright sources. As discussed below, 22 of these were deleted.
Often, the extra detections of a bright source could not be combined into a single seconds-confirmed sighting and created a variety of problems: four detections from one source in one band were sometimes processed as two separate seconds-confirmations; sources with "too many" triple detections could fail to band-merge; leftover unconfirmed detections could take priority over the seconds-confirmed source to produce an incorrect measurement of the source in one or more bands; with enough extra detections both a primary and one or more false HCONs could be produced. The spurious sources could have either similar or much weaker fluxes than the primary source.
Many of these problems were solved in the normal course of the data processing. Optical cross-talk processing eliminated weaker neighbors of bright sources (Section V.D.2.c). Slot extensions for edge detections, priority for adjacent wavelength bands and priority for seconds-confirmed sources over non-seconds-confirmed objects produced more complete band-merging. Second-confirmed sources were given priority over non-seconds-confirmed sources in hours-confirmation. All of these changes had the effect of pulling more detections into a given confirmed source, leaving fewer odds and ends to confirm and to cause unreliable sources.
Figure VII.E.2 Strip-chart tracings of the detector outputs
at 60 µm during a passage of IRC+10216 over the focal plane.
The detector timing has been adjusted so that the detector samples
correspond to the same in-scan positions.|
Figure VII.E.3 The vicinity of IRC+10216 is shown before
and after the weeks-confirmation requirement was imposed on sources
in the Working Survey Data Base. Spurious sources due to cross-talk
are shown as triangles. Those that survived weeks-confirmation
were subsequently deleted.|
Figure VII.E.3 shows the area around IRC+10216 in the WSDB. This figure shows the optical cross-talk problem at its worst. IRC+10216 generated spurious HCONs in three bands. Spurious sources line up in the typical spider arm pattern ±60 and ±120° from the in-scan direction. Because this source lies in the ecliptic plane, the scan angle was always along the local meridian so that there was no HCON-to-HCON variation in twist angle to prevent weeks-confirmation of the cross-talk sources. for For sources at higher latitudes, however, the scan direction usually varied enough to prevent the weeks-confirmation of spurious HCONs. Eleven spurious week-confirmed sources surround IRC+10216, of which seven were good enough to pass initial catalog selection rules. These seven were deleted by hand from the final catalog, along with 15 other such sources.
The final catalog data base was checked for bright sources and their neighbors above Galactic latitude 5°. Catalog sources brighter than 450 Jy at 25 µm or brighter than 100 Jy at 12, 60, or 100 µm were examined. These thresholds were selected after analysis of several hundred bright sources and are the levels at which cross-talk confirmations begin to survive final catalog screening. Any bright source neighbor, i.e., having a position within a 1000" window and a high quality flux that was band-compatible with the bright source, was identified and evaluated for evidence of optical cross-talk. Analysis of the raw data in strip chart form, e.g., Fig VII.E.2, gave the surest evidence of cross-talk. Neighbors judged to be independent sources had detections that were free from influence of the nearby bright source: the detector plots showed local minima separating the two sources. Cross-talk neighbors, on the other hand, were not separated by a local minimum but were found on a plateau of emission that was a function of the intensity, direction, and distance from the central source.
In addition to being found next to an extremely bright source, cross-talk neighbors also tended to have the following characteristics: detection in only a single band; only 2 HCONs even if more were expected; a location in one of the preferred spider-arm directions; and a brightness proportional to its distance from the primary object.
Twenty-two neighbors survived final catalog screening but were found to be due to cross-talk. These are summarized in Table VII.E.2. Notice that only the cross-talk source near Sco was confirmed in two wavelength bands. All others are single band sources. Galactic latitudes within 5° of the plane were not examined. In addition, the Orion region (especially around Mon R2, OMC 1, NGC 2024, and NGC 2071) had too many neighbors to examine. In such crowded regions the high source density processor suppressed most remaining cross-talk sources, but users should be aware that significant numbers of cross-talk sources may remain.
E.3 Sources of IncompletenessThe causes of lost HCONs are various and include: (i) missing detections due to radiation hits or noise spikes (which cause correlation coefficients to fall below threshold); (ii) band-merge failure caused by noise detection in another band; and (iii) missing detections because the detections fell below the signal-to-noise ratio correlation coefficient thresholds. Causes (i) and (ii) usually are significant only because a failed detector also removed a detection as well. There exist two additional causes of incompleteness at 100 µm. First, the instability of the noise estimation caused by the presence of cirrus sometimes causes the noise to be erroneously estimated high by a factor of two. Second, cirrus itself often creates confusion and cross-scan position shifts, resulting in a failure to hour-confirm. Finally, as discussed in more detail in Section VII.F, asteroids can also cause lost HCONs. One bright source lost the HCON because of a coincidence with asteroid Valentine.
E.4 Effects of Failed DetectorsThe data processing allowed for failed or degraded detectors by giving the status of non-seconds-confirmed due to a failed detector (NSCF) to any detection whose failure to seconds-confirm might have been caused by the potentially confirming detector being failed or excessively noisy. NSCFs enjoyed the same status as seconds-confirmed band-merged detections and had the effect of significantly increasing completeness at the cost of hurting reliability.
Because detectors 17 and 20 in the 25 µm band were dead, their seconds-confirming partners 40, 44, and 41 often produced NSCF detections as did detectors 9 and 13, opposite the dead detector 36 at 60 µm. Additional detectors declared dead due to their degraded performance were: 28, 25, and 26 at 12 µm and 42 at 25 µm. The designation of so many "dead" detectors produced a flood of NSCF detections with NSCFs outnumbering seconds-confirmations by about 5 to 1. The purpose of declaring active, but degraded, detectors as failed was to avoid penalizing the more sensitive, confirming detector. Bright sources could still confirm on both detectors, since seconds-confirmation was attempted for detections produced by the degraded detectors. Weaker sources had a chance of confirming in the nominal way on subsequent sightings with a more favorable combination of detectors.
An unfortunate side effect of allowing NSCFs was that weak radiation hits which passed the correlation coefficient test could masquerade as valid detections. When this occurred on the twelve detectors subject to NSCF status the radiation hit could band-merge and provide an erroneous flux or position for a true source. The requirement of weeks-conformation in each wavelength band prevented this from becoming a serious source of error for the catalog.
Seconds-confirmation was still possible despite passage over a failed detector. Consider a bright source passing over detectors 20 (failed), and 44 and 40 (fig. II.C.6). Although detections from 44 and 40 could seconds-confirm, they generally did not because any displacement from the overlap region resulted in a failure of the flux test. More than half the time the weaker detection band-merged into the main source while the stronger detection became a separate source. The weaker detection was more successful at band merging because it had a greater in-scan uncertainty than the stronger detection. Thus, at hours-confirmation one could often find two versions of the source, e.g., one with a good 25 µm flux and nothing else, the other with good fluxes at 12, 60, and 100 µm but a low flux at 25 µm. This problem was corrected in the reprocessing of SOPs 29-446 by suppressing the weaker detection in these nearly overlapping cases. An error in the computation of the cross-scan uncertainty prevented band merging of the stronger detection in about one-third of these special cases.
E.5 Setting the Seconds-Confirmation ThresholdWhen a confirmation threshold is set optimally, practically all true matches should be accepted, and practically all false matches should be rejected. One way to obtain a feeling for whether an acceptable threshold setting is possible and has been obtained is to process the same data with different thresholds. If one begins with a very high value, then lowering the threshold should result in a significant increase in the number of confirmation. Continuing to lower the threshold further should eventually yield little increase in the number of confirmations. This should happen as the completeness approaches unity. If the threshold is lowered still more, the number of confirmations will begin to rise again at some point as more spurious sources are accepted, unless there is no noise in the process.
If a reasonable range of threshold values is found in which the number of confirmations is essentially constant, then that range is assumed to contain the optimal value. Such a range should occur at thresholds which seem consistent with the properties of the measurement error. Assuming that gross blunders in estimating the measurement errors have not been made, then if no such range is found, even for threshold values which are clearly unreasonably large, then the noise contamination has probably blended significantly into the signal well before the onset of completeness for the matches.
The position test thresholds in the seconds-confirmation processing were varied as described above. The optimal range was found at 12, 60, and 100 µm. The search was not as successful in the 25 µm band. While the number of additional confirmations rolled off as the threshold was raised, the total number never reached a plateau such as those of the other bands. Time constraints prohibited a thorough study of this anomaly, and an attempt to identify the spatial distribution of the excess events was inconclusive. The flux distribution appeared to be more strongly concentrated to the fainter end than the overall distribution of objects detected.
A cursory analysis of the rate at which these sightings survived hours-confirmation on indicated that only very few succeeded. It is likely that none at all passed through weeks-confirmation, so that no impact on the completeness or reliability of the catalog is expected.
Table of Contents | Index | Previous Section | Next Section