How good is the photometry in the Hubble Source Catalog, and how does it compare with the previous version of the HSC? Due to the diversity of the Hubble data, this is a hard question to answer. We have taken a three-pronged approach to address it.

We first examine a few specific datasets, comparing magnitudes directly for repeated measurements using the same camera and filter. The second approach is to compare repeat measurements in the full database. While this provides a better representation of the entire dataset, it can also be misleading since the tails of the distributions are generally dominated by the faintest sources in the catalog near the detection limit.  The third approach is to compare the HSC photometry to catalogs in the literature that have observations of the same objects.  The third approach measures the absolute photometric accuracy of the HSC magnitudes, while the first two explore the noise using relative photometry.

In summary:

• The WFC3/UVIS and ACS/WFC photometry is significantly improved in HSCv4 compared with HSCv3 (by as much as a factor of 2) when measured by the variation of repeated measurements of the same object in globular cluster M4 and the SWEEPS field in the Galactic halo.
• The WFC3/IR data is significantly poorer in HSCv4 than in HSCv3. We attribute this to the larger pixel size for the HAP images compared with the HLA images.  For the other cameras, the overall noise distribution looks similar in the two versions.
• In a single tested field, HSCv4 has smaller biases in absolute photometry compared with HSCv3, and both agree very well with the deep photometric catalog.  The HSCv4 offsets in two filters are 0.006 and 0.003 magnitudes. The absolute photometry was tested by comparing with the Brown et al. (2009) deep ACS/WFC catalog of the outer disk of M31.

Relative Photometry

WFC3/UVIS photometry in globular cluster M4

Globular Cluster M4 in WFC3/UVIS F467M image from HAPFor our first case we examine the repeat measurements in the globular cluster M4. For this study, as well as the next two, we use `MagAper2 values (i.e., aperture magnitudes), which are the default for the HSC. Globular cluster M4 has more than 100 repeated observations in the WFC3/UVIS filter F467M, so the median absolute deviation (MAD) values and mean magnitudes are very accurately measured.  The figure at right shows HSCv4 overlaid on a typical image; restricting the HSCv4 sample to objects with many measurements produces a high quality catalog.  This cluster has data both in HSCv3 and HSCv4, which facilitates a comparison of the photometry in the two catalogs.

The figure below shows the noise for objects that have at least 50 repeated measurements in the F467M filter.  Each dot represents a single star that appears in both HSCv3 and HSCv4.  The x-axis shows the magnitude in the WFC3/UVIS F467M filter, and the y-axis is the noise (median absolute deviation or MAD). The HSCv4 photometry (bottom) has noise that is a factor of 2 lower than the noise in HSCv3 (top).  The horizontal lines are at the same location in the two panels and indicate the median noise level for stars between magnitude 17 and 18.  The scatter in HSCv4 is only 0.007 mag (less than 1%).

Note that the HSCv4 magnitudes are better for brighter objects.  The objects with magnitudes brighter than 16 in HSCv4 do appear in HSCv3 but have magnitudes that are much fainter (typically ~17) and have magnitude error values larger than 0.1 magnitudes (so they do not appear on the HSCv3 plot).  These sources are all saturated, which reduces the accuracy of the HSCv4 magnitudes, but the scatter remains reasonably small even for objects several magnitudes above the saturation threshold. The properties of bright sources in the SWEEPS field (below) are similar.

The photometry in HSCv4 in this field is excellent, and it is distinctly improved compared with the photometry in HSCv3 (which was already very good).

ACS/WFC photometry in the SWEEPS field

SWEEPS field in HAP image from ACS/WFC F606W and F814W observationsThe SWEEPS field is a target in the galactic halo with hundreds of thousands of stars and hundreds of repeated observations using the ACS and other instruments.  The observations are spread over more than a decade, so this field is a good test of the long term stability of the photometry.  The SWEEPS data has been extensively studied in HSCv3 and is a good comparison field for HSCv4.

The SWEEPS field is extremely crowded and is a challenging for the catalog source detection and photometry.  The figure at right shows one ACS/WFC image using the F606W and F814W filters.  There are approximately 80 similar images, each with a catalog of approximately 100 thousand stars.

The figure below shows the photometric noise in repeated measurements using the F606W filter.  As for the M4 plot above, each dot represents a single star that appears in both HSCv3 and HSCv4.  The x-axis shows the magnitude in the ACS/WFC F606W filter, and the y-axis is the MAD noise. The lines on the plot show the median noise in magnitude bins. Again the HSCv4 photometry (bottom) has lower noise than the HSCv3 photometry.   And again the noise for sufficiently bright stars is less than 0.01 magnitude. 

We consider this an excellent result for HSCv4. In these plots, the noise increases for fainter stars due to counting noise in the image (which is both unavoidable and expected).  The noise distribution is visibly much tighter for HSCv4.  Note that the photometry is also better for the brightest stars; those stars are simply missing in HSCv3 (or more correctly, they appear with magnitudes that are too faint).  Increasing saturation effects cause the noise levels to increase as the stars get brighter, but the data is still usable well above the brightness limit of the HSCv3 data.

Poor photometry for WFC3/IR

We tested the photometry for WFC3/IR using the galaxy cluster ACT-CLJ0205.7-5829, where there are many repeated observations in the filters F105W and F140W.  We knew before we did this study that the quality of photometry in the WFC3/IR HAP source lists used for HSCv4 is not as good as the HLA source lists used for HSCv3.  This test confirmed the problems.

The figure below compares the HSCv4 and HSCv3 catalogs overlaid on an image of the field.  We have chosen an image from the HLA due to its higher resolution.

The blending of close sources in the HAP source lists will lead to some outliers in the magnitudes (where the HSCv4 magnitude is considerably brighter than the corresponding HSCv3 magnitude).  It will also lead to astrometric offsets between the catalogs for individual sources.

We cross-matched the HSCv3 and HSCv4 catalogs in this field to compare the properties of other sources.  The sources that are visibly split in the figure above are mostly not included in this cross-match because their positions disagree.  The results are mainly for the more common sources that are isolated.  The figure below compares the catalog median absolute deviation (MAD) values for the matched sources.

All objects in this figure have at least 25 measurements in both HSCv3 and HSCv4, so their magnitudes and MAD values are accurately determined. The noise is much higher in HSCv4: note the points are almost all above the diagonal line where the noise would be the same in the two catalogs.  The typical MAD noise for these objects in HSCv3 ranges from 0.005 to 0.06 mag; for HSCv4, the best case noise is nearly 0.03 mag and the worst is nearly 0.15 mag. 

The high noise level for WFC3/IR measurements in HSCv4 is due to the larger pixel size used for drizzling the HAP images. We discuss this issue further in the next section.

Distribution of flux differences across all instruments and sources in HSCv4

We now turn to our second approach; looking at repeat measurements for the entire HSC database. The following figure shows the distribution of comparisons among independent photometric measurements of  sources that belong to the same match and have the same filter in the HSC for version 3 (right) and version 4 (left). For each match with 2 or more measurements in a filter, we compute the median MagAper2 flux and the median absolute deviation (MAD) among the flux values. The x-axis is the ratio of these values, MAD/Med .  This ratio is similar to the MAD magnitude value given in the catalog for matches. The y-axis is the number of sources per bin (with a bin width of 0.0025) that is normalized to unity at a MAD/Med value of zero. The distributions are shown for the different cameras included in the HSC.

The HSCv4 flux difference distributions are similar to HSCv3 with the notable exception of the WFC3/IR camera (green line), which has much larger errors than for HSCv3.

Direct comparison of these distributions is complicated because they are affected by factors such as the fields that are included and the depth of the source lists (deeper catalogs include more faint sources, which naturally have higher noise in their photometry).  These plots include all sources regardless of magnitude and signal-to-noise, so the differences are dominated by faint sources near the detection limits of the images. Note that is different from the detailed comparisons reported above, where we compared observations of the same sources in HSCv3 and HSCv4. The main point of this figure is to demonstrate that typical photometric uncertainties in the HSC are better than 0.10 magnitude for a majority of the data.

To do a more direct comparison of the HSCv4 and HSCv3 photometry, we have also computed these distributions using only matches that appear in both HSCv3 and HSCv4.  We restrict the sample to objects that match within 0.1 arcsec using the Matchv3  table in the HSCv4 database.  The figure below separates the distributions by camera and plots the HSCv3 and HSCv4 curves on the same plot.

These distributions have been normalized by the integral under the curve rather than by the value at x=0.  For ACS/WFC and WFC3/UVIS, these distributions were somewhat better in HSCv3 than HSCv4. But in the case of WFC3/IR the distribution was much better in HSCv3. The values are almost identical for WFPC2 (unsurprising since the WFPC2 source catalogs are the same for HSCv4 as for HSCv3). Note that despite starting with the same 23 million matches, the number of sources for each instrument varies because the contributions to a match by different instrument are often not the same in the two versions.

The obvious difference here is that WFC3/IR (bottom left) has much larger photometric errors in HSCv4.  Note from the legend that there are also fewer WFC3/IR sources in these matches for HSCv4.  The HAP WFC3/IR source lists used for HSCv4 are both shallower and have poorer photometry than the HLA source lists used for HSCv3. 

Direct comparisons of magnitudes for matched sources in HSCv3 and HSCv4

Another approach is to directly compare the magnitude measurements in HSCv3 and HSCv4.  We have constructed a sample of sources with reliable matches between the two catalogs that have measurements in the same filter/detector combinations in the two catalogs.  These measurements still may have different numbers of source detections in the filters, but this is the most direct comparison of the magnitudes in the two versions of the catalog. The plot below shows the results.

The ACS/WFC and WFC3/UVIS distributions are tight and well-behaved, with at most small shifts in the photometry. (In both cases the HSCv4 magnitudes are slightly brighter, which probably is an indication of better exposure alignment in the HAP images.)  The WFPC2 plot is not shown because the magnitudes are actually identical due to both catalogs using the HLA source lists for WFPC2.

On the other hand, the WFC3/IR distribution is clearly poor.  It is both shifted significantly (by nearly 0.05 mag) and shows much larger scatter (approximately 0.26 mag FWHM).

There is little doubt that the degradation for WFC3/IR is due mainly to the change in pixel scale for the HAP images, which used 0.1275 arcsec pixels instead of the 0.09 arcsec pixels used by the HLA.  The area of the HAP pixels is twice as large as the HLA pixels.  That increases the size of the point-spread function (making aperture photometry noisier).  It also makes it harder to detect objects both in crowded regions (where there is more blending) and near the background noise level (because there are more background counts due to larger pixels).  

We are considering the possibility of reducing the HAP pixel size to match the HLA pixels for the next round of HAP processing.  If implemented, that will significantly improve the quality of the WFC3/IR data in the HAP source lists and the HSC.  Implementation of that change is to be determined based on the availability of the necessary resources.

Absolute Photometry

Comparison with the Brown et al. (2009) M31 catalog

M31 disk field from HAP-SVM image of ACS/WFC F814W with HSCv4 sourcesNext we compare HSC data with an external catalog. The case shown below is a comparison between the HSC and the Brown et al. (2009) deep ACS/WFC observations of the outer disk of M31 (proposal = 10265). The observing plan for this proposal resulted in 60 separate one-orbit visits (not typical of most HST observations), hence provide an excellent opportunity for determining the internal uncertainties by examining repeat measurements. A sample image with the HSCv4 catalog sources over-plotted is shown at the right.

This field has been used in HSC Use Case #1 and in Whitmore et al. (2016) to test the HSC photometry. Here we compare with the photometry in HSCv4 and HSCv3. If you would like to make detailed comparisons yourself you will need to download the Brown et al. (2009) catalog via the MAST High Level Science Product (HLSP) webpage. Below we provide a brief summary of the results, including some of the use case discussion; more details are available in Whitmore et al. (2016).

The HSC uses the ABmag system while Brown et al. (2009) use STmag. The conversions, as provided in Brown et al. (2009), are:

ABmag_F606W = STmag_F606W - 0.169
ABmag_F814W = STmag_F814W - 0.840

The HSC MagAper2 values use an aperture with radius R = 3 pixels while Brown et al. correct to total magnitudes. To compare, we add additional corrections:

TotMag_F606W = ABmag_F606W - 0.248
TotMag_F814W = ABmag_F814W - 0.292

where these corrections are 2.5*log(Encircled Energy) using the values for Encircled Energy from Sirianni et al. 2005, Table 3.

Cross-matching the HSCv4 and the Brown et al. (2009) catalog for the disk field reveals an astrometric shift in the Brown catalog of 0.5 arcsec in declination. (This is certainly in the Brown et al. catalog rather than the HSC since the same shift is seen in a comparison with Gaia DR3.)  There are also other small distortions in the Brown catalog coordinate system, but those are small enough to ignore for our purposes (+- 0.06 arcsec in RA).  After correcting for those offsets, we find about 8,500 reliable matches (omitting for simplicity objects that have multiple close matches in the HSC).  There are a similar number of matches to HSCv3.  The figure below compares the photometry for the two catalogs after applying the constant offsets to convert to total AB magnitudes.

Clearly there is good agreement between the magnitudes.  The Brown et al. catalog is roughly 3 magnitudes deeper than the HAP catalogs because it is based on combined data from all the individual visits, while the HAP catalogs are each based on a single visit.  The Brown et al. catalog was truncated at a F814W magnitude of 28.0 (ABMag) to speed up the cross-match.

The second plot shows the same data but plots the difference between the HSC magnitudes and the Brown et al. magnitudes.

There is little if any bias in the HSCv4 magnitudes: the median offsets for the two filters are 0.003 mag and -0.009 mag.  HSCv3 does show a small offset in the F606W filter (0.022 mag).  The noise in the two HSC versions is similar in this field.

Note that the HSCv4 catalog is deeper than HSCv3 in this field.  That is most visible in the F814W filter (right column) where the magnitude cutoff for HSCv3 clearly leads to a Malmquist bias for the faintest sources.  To remove the catalog depth as a variable, we cross-matched the HSCv3 and HSCv4 catalogs in this field.  The plot below compares the magnitude differences from the Brown et al. (2009) catalog.

The final results clearly favor HSCv4, which has a smaller median offset in the magnitude and a smaller MAD scatter than HSCv3.