H Photometric Error Due To Airglow

Conclusions
Introduction
Sample Selection
Analysis
Fraction of Scans With A Given H Photometric Error
Table Of All Plots

Airglow contributes extra noise at H band that causes the H fluxes of extended objects to have significant additional scatter. The flux error is directly correlated with the mean background-removed noise measured in the Atlas Images.

Below a background-removed noise of ~1.0 DN, any additional flux error caused by airglow is negligible. However, above a background-removed noise of ~1.1 DN, the additional scatter caused by airglow causes a violation of the Level 1 specification on galaxy photometry. The additional scatter reaches one sigma values of 0.27 mag for galaxies with H = 13.8 mag in scans with severe airglow, making the H photometry nearly useless for extended sources. Fortunately, only 1% of scans are so affected.

About 25% of all scans have enough extra noise to cause the H photometric error to exceed the Level 1 specification of 10% error. About 10% of all scans have an H photometric error of 15% for galaxies with H = 13.8 mag.

The background-removed noise can therefore be used to assign a "airglow quality score" to each scan for the H extended source photometry, which can then be factored into the overall quality score of a scan. One possible algorithm is to assign "airglow quality" = 0 or 1 for all scans above some threshold of acceptability, such as 0.15 - 0.20 mag error. The scores 1 or 2 through 10 could then simply be scaled with the background-removed noise, so that noise values corresponding to a 10% photometric error at H would receive "airglow quality" = 10. 75% of all scans would then receive "airglow quality score" = 10, with 25% of all scans getting a lesser score.

Thus, for example, if 0.15 mag error is picked as the threshold for "airglow quality" = 1, the "airglow quality score" would be:

airglow quality score = int( 7.75 - 5.6 * measured background-removed noise)

We would clearly have to somewhat relax the Level 1 specification on galaxy photometry in order to accept more than 75% of scans into the catalog. However, with this scoring scheme we could do that, and the user could choose whether to work with extended sources from scans with quality < 10 in order to increase sky coverage.

Analysis of Noise In The 2MASS Atlas Images showed that airglow contributed extra noise at H band with an average sigma of 0.20 DN over all scans. J and K bands were mostly immune from airglow except for a small number of scans with severe airglow. For some examples of Images heavily affected by airglow, see T. Jarrett's Severe Airglow Scan 097, 971005n, specifically an H Image before and after background removal.

Because the airglow contribution is constant over the small size of most 2MASS galaxies, a value of 0.20 DN would cause a systematic flux error at H of 4-13% for a source with H = 13.8 mag. This alone flirts with violating the Level 1 Specification for galaxy photometry of 10%. (Recall that Poisson errors themselves are already close to a 10% error, leaving little margin for additional error sources.) However, my previous analysis could not evaluate the actual effects of airglow on extended source photometry, since the distribution of extra airglow noise was not known.

Ideally, to understand the effect of airglow one would like to examine duplicate observations of the same galaxy from scan overlaps. Unfortunately, this is impractical due to the low numbers of galaxies. Fortunately, analysis of the J-H colors of galaxies can be used to derive the H band noise due to airglow.

This analysis therefore evaluates directly the effect of airglow on H galaxy fluxes by using the observed J-H colors of galaxies. Since airglow affects only H fluxes to first order, leaving the J fluxes unchanged most of the time (see Analysis of Noise In The 2MASS Atlas Images), examining the increase of the J-H scatter with airglow is equivalent to examining the increase of the H scatter with airglow. The J-K color could be used as well, but J is preferred to K since J has a higher SNR for galaxies, and thus J-H will have lower scatter than H-K.

The scatter of J-H colors of galaxies should depend only on the intrinsic dispersion of the J-H colors of galaxies and the measurement noise in the Images. The intrinsic dispersion of J-H colors of galaxies is of course independent of airglow and measurement noise, and is small enough so that the effects of airglow are easily measured. Although I will be able to measure only effects of airglow that exceed the sum of the intrinsic galaxy dispersion and the measurement error, those are the only effects that are important. Smaller airglow effects are within the measurement error, and thus are of no concern.

The Level 1 Specification on galaxy photometry is an upper limit of 10% error (one sigma) for a galaxy with H = 13.8 mag. Hence I have analyzed galaxies with H = 13.6 - 13.8 mag.

I have selected galaxies from the database for two samples, using stringent criteria to guarantee high reliability. The criteria are color_score > 0.15 and wedge-shape score > 10, which guarantee that the sample consists of galaxies to which the Level 1 Specification applies. (See 2MASS Galaxy Catalog: First Results.) The two samples are:

1. glat > 40°, with H = 13.7 - 13.8 mag, observed in the North before 01/16/98, producing 670 galaxies. The date cutoff was used simply to match my already available Image Atlas noise database used for my noise analysis.

2. galaxies from the nights of 971004n and 971005n, nights with very high airglow, with H = 13.6 - 13.8 mag, producing 237 galaxies. These sources are at glat ~ -25 to -44°, with source densities low enough so that confusion noise is not a problem. This sample reaches much worse airglow than the first sample, and hence the effects stand out most prominently here.

I used the default magnitudes j_m, h_m and k_m from the database, which are isophotal fiducial magnitudes, with the isophote usually fixed for all bands to be the K isophote. If k_m is fainter than 13.5 mag, the fiducial band switches to J. In this analysis, it doesn't matter which band is fiducial - only that the apertures are indeed the same so that the J-H color is not affected by aperture differences between J and H.

I used the date, scan and coadd information for every galaxy to look up the the H background-removed noise (called background_sigma) and the H background value for that coadd. The background-removed noise is measured using the histogram of Image Atlas pixels after bright sources and their artifacts are masked, and a four parameter fit to the background is removed. Specifically, the noise is half the distance in DN (counts) from the 16% (-1 sigma Gaussian) to the 84% point (+1 sigma Gaussian). The four parameter fit removes background structure with scales longer than ~4-5'. This is the parameter "bkgnd_sig" in the "Survey Coadd Parameter File".

I calculated the excess background sigma by subtracting the fit:

0.25 + 0.043 * sqrt(background)

from the background-removed noise. This fit is derived in Analysis of Noise In The 2MASS Atlas Images.

A plot of J-H color vs. H background sigma for 971004n and 971005n shows clearly that there is increased noise in the H fluxes of galaxies as the background sigma increases. While galaxies certainly have some intrinsic spread in J-H color, that spread ought to be independent of background sigma.

The J-H colors are tightly distributed for H background sigma < 1.0 DN, and the scatter grows much larger at higher values of H background sigma. In fact, the scatter grows so large that it is possible that this plot suffers from truncation error, since J-H = 1.7 implies J = 15.5 mag for H = 13.8 mag. For comparison, the completeness limit of the Extended Source Catalog is J ~ 15 mag. However, because the photometric error must increase as the background increases, further analysis is needed to show that airglow contributes this extra scatter, and that the error growth is not that expected with increased background levels.

The theoretical expected noises for the J and H fluxes are calculated for each galaxy from the number of pixels in the aperture used to calculate the galaxy fluxes and from the measured background sigma in the coadd. Previous analyses have shown that this noise value is roughly correct for galaxies, except that often there is additional photometric error which reflects the uncertainty in the aperture size. By using colors, this additional photometric error is virtually canceled, since the same aperture is used at all wavelengths. Hence the theoretical expected noise for the J-H color should be very close to the actual measured noise.

A plot of the observed J-H sigma vs. the theoretical expected sigma for 971004n and 971005n shows that for H background sigma < ~1 DN, the measured sigma is roughly equal to the expected sigma, validating the above expectations. However, above 1 DN, the observed scatter grows increasing larger than the expected scatter, to the point where the H measurement is practically meaningless, with a sigma of 0.28 mag.

To make the comparison to the Level 1 Specification for H photometry, for H background sigma < 1 DN, the average theoretical J and H magnitude errors are 0.064 and 0.073 mag, making the average theoretical J-H error 0.097 mag. Keeping the J magnitude error fixed, an H magnitude error of 10% corresponding to a mag error of 0.108 makes the J-H error 0.126. Hence all data with H background sigma > ~1.1 DN is definitely in violation of the Level 1 specification. The value of 0.28 mag reached for the J-H scatter is equivalent to an H one sigma error of 0.27 mag, far in excess of the Level 1 specification.

This increase of H photometric error with background sigma is not peculiar to these two nights. The same plot for all galaxies with glat above 40° shows general but not exact consistency with the previous plot, although no night covering high galactic latitudes reached the same high H background sigmas as on 971004n and 971005n so far. Overlaying the two curves shows this comparison more clearly.

The last point in the glat > 40° plot may possibly indicate that the relationship changes from night to night, which would be consistent with intra-night variations plotted in Analysis of Noise In The 2MASS Atlas Images. However, the distribution of points from which the sigma was derived for glat > 40° looks a bit peculiar at its highest values of background sigma and perhaps something else is affecting that bin. Also, when one takes into account the slightly lower theoretical expected noise for that point and calculates the extra H photometric error (see next paragraph), that point looks less anomalous.

The extra H photometric error can be derived for each bin and plotted vs. background sigma for 971004n and 971005n, for glat > 40°, and for both. The error bars for these points probably make the values of 5% found below 1.0 DN consistent with zero extra noise. (See the scatter of observed vs. theoretical noise used to calculate these values in glat > 40° and 971004n and 971005n.)

The mean J-H color versus background sigma for 971004n and 971005n and for glat > 40° both show the disturbing trend for a bias in the H flux with background sigma. Overlaying the two curves shows that above the same background sigma of ~1.0 DN, the mean H flux becomes increasingly brighter, as one would expect from the imposition of a magnitude threshold. The bias becomes ~5% at a background sigma of ~1.2 DN. However disturbing this may be, it is not in danger of violating the Level 1 specification of 10% spatial uniformity.

Finally, these increases with background sigma are consistent with the increase of excess background sigma with background sigma: 971004n and 971005n, glat > 40°, and both.

The following table gives the fraction of scans that have a given H photometric error for a galaxy with H = 13.8 mag. The error is given in magnitudes. The total H photometric error is also plotted vs. fraction of scans.

http://spider.ipac.caltech.edu/staff/tchester/2mass/analysis/galaxies/photometry/j_h_sigma/h_photometric_error.html
Comments and feedback: Tom Chester
Last update: 23 October 1998.