spacer link to MAST page spacer logo image spacer
 
link to STScI page


Most STScI network links to services and websites will be UNAVAILABLE Thursday May 3, 2018 7:00 PM - 11:00 PM.

Processing OM Mosaics

This text describing the processing used to create the OM mosaics distributed by MAST is excerpted from Kuntz et.al. (in prep) with permission of the author.

The paper has now been published:

Kuntz et. al. OMCat: Catalog of Serendipitous Sources Detected with the XMM-Newton Optical Monitor PASP, 120:740-758


Image mode processing: For the most part, we have used the standard omichain processing with the default setting; exceptions are detailed below. The standard omichain processing (processing from images) handles the images produced by each science window separately. For each science window image omichain applies a flatfield. The photon splash centroiding algorithm calculates the centroid to 1/8 of a pixel, but due to the algorithm, not all values are equally likely. This problem results in "modulo-8" fixed pattern noise. The omichain processing applies a redistribution to correct for this effect.

For every science window image omichain runs a source detections algorithm, measures the count rates for the sources, and applies a calibration to convert to instrumental magnitudes. Once all of the science windows are processed, omichain produces a "master" source list by combining the source list for each science window, matching sources in common between the lists, and determining the mean (α,δ) for each source. The standard processing has the option to use an external catalogue to correct the coordinates of the master source list; we have used this option with the USNO-B1 catalogue. It should be noted that the coordinate correction using the USNO-B1 catalogue will fail if there are too few matching sources, in which case no significant solution can be found. The omichain algorithm requires at least 10 matches in order to produce a significant coordinate correction. If there are too many sources in the field coordinate correction will also fail, presumably because some fraction of matches are spurious and the solution will not converge. We have found that the coordinate correction can fail for almost any density of sources, though we have not yet determined the cause of that failure.

In addition to combining the source lists for all the filters, omichain mosaics the low-resolution science window images (but not the high-resolution science window images) for each filter. The world coordinates system (WCS) of the first science window for a given filter sets the coordinate system of the entire image mosaic. Note that the standard omichain processing can correct the master source list, but not the images. Further, the correction using an external catalogue will only be applied if there are at least ten sources. One does have the option of doing the same correction to the individual science windows, but there are often not enough sources in a single science window to perform such a correction. We have applied the correction derived from the external catalogue to the low-resolution mosaics. Since the mosaic images use the WCS from the first science window in the mosaic, we compared the (α,δ) in the source list for the first science window to the (α,δ) in the USNO corrected master list. We then determined the (Δα,Δδ) which must be added to the (α,δ) of the first source list in order to obtain the (α,δ) in the USNO corrected master list. We then applied that correction to the header keywords of the mosaicked images. Two points must be noted. 1) Since the telescope can drift by 1-2 arcseconds between exposures, this correction is done separately for each filter. 2) There are often not enough sources in a single science window to attempt to determine the offset and rotation that would best align the first source list with the master list. Therefore, we have assumed that the rotation between the master list and the external catalogue can be applied to all of the science windows.

The standard omichain processing does not combine the images from the high-resolution science windows. We have combined the high resolution images, though not with ommosaic, the standard SAS tool. The ommosaic used the WCS keywords to determine the offsets needed to align the WCS frames of the individual science windows before summing. We allow the WCS of the summed image to be set by the first science window. For each successive high resolution image, we compare the source list to the source list from the first image, determine the (Δα,Δδ) required to match the source lists, and apply that offset to the image before adding it to the mosaic. Not all sources are used for determining the offsets, only sources appearing in at least half of the images; this selection removes sources with poorly determined positions. The offsets are rounded to the nearest integer pixel; subpixelization did not seem to produce a significant improvement in the resultant PSF, and so was not used for this processing. Compared to the direct sum of the images made by ommosaic, our processing does improve the PSF of the summed image, sometimes improving the FWHM by as much as a pixel (see Figure 2). We compare the source list from the first image with the master source list to correct the summed image in the same manner used for the low-resolution mosaics.

hires demo
Fig. 2. -- Image created with the 25 high-resolution science windows using ommosaic (left panel) compared with the same image created with our processing (right panel). Note that the central point source is rounder after our processing, and a number of knots of faint emission stand out more.
hires profile
Comparison of the profile of the central point source red: ommosaic black: our processing. The FWHM improves by almost a pixel in this case and the peak intensity increases by 15%.

Further coordinate correction: After our initial processing of the public archive we found that the pipeline coordinate correction using the USNO catalogue failed for ~38% of the fields. Further, the failure was not limited to extremely high or extremely low source densities (see Figure 3). We thus found it worthwhile to create our own coordinate correction routine using the USNO catalogue. Although the bulk of fields need only a small correction, some fields need substantial corrections (~2"). Thus, although we attempt to find high precision corrections for all fields, it is still worthwhile to find lower accuracy corrections for those fields that do not have a large enough number of matches with the USNO catalogue to attempt a high precision solution.

histogram of number of sources in each field
Fig. 3 -- Histogram of the number of sources detected in each field. The dashed red line shows the number of fields for which the pipeline coordinate correction failed as a function of the number of sources. The thick blue line shows the number of fields for which the post-pipeline coordinate correction failed as a function of the number of sources

We used a fairly simple and robust algorithm for matching the OM sources to the USNO sources. If there were > 10 matches we iteratively solved for the offset in the (α,δ) that minimized the offset between the OM source list with the USNO source list. By iterating the solution we could eliminate some portion of the false matches. We have not attempted to solve for a rotation for two reasons: 1) adding a rotation to the fit did not significantly improve the solution, and 2) given that there are systematic offsets from one science window to another, the rotation could be strongly biased by the offset of a single science window. If there were 3 < n < 10 matches we merely calculated the mean offset between the OM and USNO sources, and used that offset as the coordinate correction. If there were < 3 sources we did not attempt a correction. We applied the same correction to the individual images that we applied to the source lists.

For each OM source in the source list with a significance in any filter > 3, the matching algorithm finds the closest USNO source. It then creates the distribution of the distances between the OM sources and their closest USNO counterparts, then this distribution would be a Gaussian whose width is the coordinate uncertainties of the two catalogues and whose peak is the offset between the two catalogues. If there were no true matches between the OM sources and the USNO catalogue, then the distribution would be given roughly by the probability distribution of sources:

P(r) = e[ρπr2]ρπ[(r+δr)2-r2]e[-ρπ[(r+δr)2-r2]]

where ρ is the surface density of sources and δR is the binsize of one's histogram of distances. For this distribution both the peak of the distribution and the width of the distribution scale as ρ-0.5. We expect that some fraction of the OM sources have true USNO matches and that the remainder will not. As a result, the observed distribution of sources will have a sharp peak with a width of ~0.3" due to matches and a low, broad distribution for the spurious matches. This algorithm has problems with high density regions; for source densities of 5000 sources/image (0.005 sources archsec-2), the distribution of spurious matches will peak at 6.3" with the lower half-maximum at 2". Although the peak of the matching sources typically has r 3", the true match rate is likely to be small compared to the spurious match rate, and so it is difficult, if not impossible, to find the true match peak in this distribution. However, at these sources densities source confusion is a serious problem as well, so even if the matching algorithm worked, the coordinate solution would remain problematic.

For the matching algorithm, we simply fit that distribution with a Gaussian. If the width of the Gaussian is smaller than 0".7, then algorithm takes all of the sources within 3σ of the peak of the distribution as real matches. An initial solution is determined from those matches, and a fit is made in the image coordinate frame to find the (α,δ) offset that minimizes the distance between the OM sources and their USNO matches. The source matching is redone with the new offset, and the process is iterated until it converges. After application of our coordinate correction routines, only ~ 14% of the fields remained without any coordinate correction. Besides fields with very few objects, the fields without coordinate corrections were characterized by very broad distributions matches suggesting a combination of large pointing error and large source density, and thus a large number of spurious identifications.

For fields where the pipeline processing found a good coordinate solution using the USNO catalogue our coordinate correction was not significantly different. However, our coordinate coorection did improve the RMS difference between OM and USNO coordinates for sim23% of fields.

Fast mode processing: The bulk of the fast mode processing is concerned with the production of light curves of the source. the fast mode images consist of 10".5 ×10".5 regions containing, typically, a single source. We combine all of the images for each filter using the same method applied to the high-resolution images. We do not currently attempt a coordinate correction. [However, since these sources are typically fairly bright, they should show up in the USNO-B1 catalogue, and a correction should be feasibile.]

Further processing: Further processing is required to provide a more useful source list to be incorporated into the OMCat. To the standard image catalogue (a binary fits table) we add images of each source from each filter. Each "postage stamp" image is 19 × 19 pixels in size, extracted from the low-resolution image mosaics (thus the pixel size is 0".95 and the image is 18".1 × 18".1 in size). Since the sources were derived from all of the science windows, some sources can fall in high-resolution sicence windows without low-resolution counterparts. In that case the postage stamp is extracted from the high-resolution image and binned to the same resolution and size as the other postage stamps. Sources that appear only in "fast" science windows are treated similarly. Note that postage stamps are extracted from all of the available filters, not just the filters for which the source was detected; many postage stamps will thus appear to be empty.

Processing summary: For each obsid our processing produces a coordinate corrected source list, a coordinate corrected low-resolution mosaicked image for each filter, a coordinate corrected high-resolution mosaicked image for each filter (if possible), or a summed fast mode image.

Caveats: 1) Individual science windows may be significantly (1-2") offset from the remainder of the mosaic. In this case the correction by use of the USNO-B1 catalogue will not be wholely satisfactory, and sources will appear to be offset in the postage stamps. The extent of this problem for any individual source can be determined by looking at the "RMS-RESID" column which contains the RMS residual from the fit of source list to the USNO-B1 catalogue. 2) The source lists will contain spurious sources, source due to ghost images, diffraction spikes, readout streaks, saturation around bright sources, and other effects. Some of these sources can be removed by consulting the Q_FLAG parameter. Similarly, confused sources are flagged by the C_FLAG parameter. However, we have found that filtering out sources with significances < 3 was a more efficient means of removing spurious sources than reference to the quality flags. 3) Although we provide the mosaicked low-resolution and mosaicked high-resolution images through the archive, these images should not be used for photometry. 4) Although there may be substantial exposure for a given image, the detection limit will not be substantially better; the source detection is done on the individual science windows, most of which have a mean exposure of ~2200 s rather than the mosaicked images.