Galaxy Zoo Starburst Talk

Emission line fluxes, SDSS vs. MPA-JHU pipelines

  • mlpeck by mlpeck

    I said here that I'd present a bit of a comparison of the SDSS "idlspec2d" flux estimates with the MPA-JHU estimates in the quench data table.

    Here is a comparison for the 4 lines that are used in the standard BPT diagram. The SDSS pipeline has a standardized way to warn that a flux estimate is missing or shouldn't be used, but it does occasionally produce a huge negative flux estimate. I've culled points that were obviously hugely in error. We've already discussed the problem of the occasionally hugely erroneous estimates in the MPA-JHU data, and I've culled those too.

    I've just plotted flux against flux here, with the SDSS estimates as the x-values. The color coding of the points is the same as in the BPT diagrams I posted earlier. There's a mistake in the legend at bottom right: AGN are blue points and LINERs cyan.

    Sometime later perhaps I'll post a comparison of the [O II] doublet at 3727-3729 Å. The situation there is much worse than for these lines.

    enter image description here

    Posted

  • jtmendel by jtmendel scientist, moderator in response to mlpeck's comment.

    Just a quick question: if I remember correctly, the MPA-JHU line fluxes are measured from spectra with a slightly different flux calibration than the default SDSS pipelines (at least, I think this is the case for DR7). Do the comparisons above take in to account the spec_to_fiber scaling in the MPA-JHU data?

    Posted

  • mlpeck by mlpeck

    Hmm, according to both the DR8 release paper and the MPA top level page (http://www.mpa-garching.mpg.de/SDSS/DR7/) the spectra are scaled to (r band) fiber magnitudes, but I can't find a discussion of the significance of that. There is an entry labelled "spectofiber" in the galSpecLine table. Can that be used to convert between flux calibrations?

    Also, the MPA fluxes are corrected for galactic extinction and I think the SDSS fluxes are not. I took the SDSS fluxes from the SPZLINE table in the spectra files and did not adjust them.

    The fluxes in the Quench data table appear to have been taken unmodified from the galSpecLine tables.

    Posted

  • mlpeck by mlpeck

    OK, I just did weighted1 least squares fits of the SDSS fluxes against the MPA ones for these 4 lines and except for H beta the relations all have slopes between 1.17 and 1.22 with fairly small zero point offsets. The H beta slope is 1 with a larger zero point offset.

    The DR8 release paper (http://arxiv.org/abs/1101.1559) has a set of tiny graphs on page 14 of the arxiv version comparing these 4 lines that look not much at all like mine even allowing for the fact that different quantities are plotted. But the caption says

    In performing
    this comparison, we have put all measurements on a common scale
    by removing the MilkyWay reddening and and spectrophotometric
    zeropoint corrections from the MPA-JHU line measurements.

    The extinction corrections should be small, but the "spectrophotometric zeropoint corrections" might account for the slopes differing from 1.

    The more sophisticated stellar contribution estimates in the MPA-JHU measurements could also account for the different behavior of H beta where stellar absorption is much more important than in the other lines.

    1Weighted least squares isn't quite the right technique since both sets of data are measured with error, but I don't have error estimates for the MPA data. Also a proper treatment would best be done with Bayesian methods which I don't feel like taking the time to do right now.

    Posted

  • JeanTate by JeanTate in response to mlpeck's comment.

    Very interesting plots! 😃

    Can you share with us the IDs of the objects you culled?

    All but the [OIII] plot seems to contain some clear outliers; how many of these outliers are common to at least two plots (i.e. same object, > 1 outlier)?

    There's a big train-wreck of black (no classification) objects in the H-beta plot; any idea why?

    Note to those who want to see the two right-hand plots (one of which includes the legend): you can right-click on the image, and then copy/paste the URL into a new tab/window of a browser (this may vary somewhat, depending on what browser/OS combo you're using). To check that you're on the right track, here is that URL: http://wildlife-pix.com/quench/emlinecomp.jpg

    Posted

  • mlpeck by mlpeck

    The only objects I've culled so far from my personal quench sample are the possible QSO/star blend (per DR9 classification -- ID forgotten) and the one isolated star (AGS00001z3). Decisions on what else to remove are above my pay grade.

    One thing I noticed since I made that post is that the MPA pipeline does have a way to indicate missingness, namely a flux estimate of 0 with an error estimate also 0. That accounts for at least one outlier in the H alpha plot.

    Some of the apparent outliers would be seen not to be outliers if I had plotted error bars along with the points, that is the estimated statistical errors were very large.

    What that clump of black points in the H beta plot means is that the vast majority of non-classified objects (in the BPT diagnostic) had weak or negative H beta fluxes. Of those 4 lines that's the hardest to get a good estimate of emission line flux because the correction for stellar absorption is large, especially for this sample.

    Posted

  • JeanTate by JeanTate in response to mlpeck's comment.

    the possible QSO/star blend (per DR9 classification -- ID forgotten)

    I think that's AGS00002ak (DR7 ObjId 758877274402460510).

    What that clump of black points in the H beta plot means is that the vast majority of non-classified objects (in the BPT diagnostic) had weak or negative H beta fluxes.

    I wonder how many have error bars large enough to make them indistinguishable from zero H-beta flux?

    because the correction for stellar absorption is large

    Would you mind explaining this a bit please?

    My - possibly quite wrong - understanding of what this mean is that around 486nm (rest frame) there are lots of metal lines, in the spectra of stars of type ~B right through to ~M, so estimating what's combined metals absorption and what's H-beta is very difficult, not least because the former depends sensitively on assumptions about the mix of spectral types and their metallicity. An example is AGS00000iy; zooming in on the interactive spectrum, near 580nm, you can see that H-beta is in a broad, complex absorption trough, yet at 5806.307Å (H-beta at z=0.19379) there's a tiny, single pixel, emission peak.

    Posted

  • mlpeck by mlpeck

    I wonder how many have error bars large enough to make them indistinguishable from zero H-beta flux?

    Basically all of them. I haven't explicitly checked but I doubt there were many "not classified" objects that had a detection in H-beta and no detection in another line. Any exceptions would probably be due to missing data -- for some reason the region around H-alpha is masked relatively often, so there could be a few cases where H-alpha or [N II] flux estimates are missing.

    Would you mind explaining this a bit please?

    Oh, my reasoning is much more simplistic and has nothing directly to do with metal content. The line strengths of the Balmer lines are at least roughly equal in absorption, whereas there's a strong Balmer decrement in emission. For a fairly wide range of conditions in the ionized gas the ratios H-alpha:H-beta:H-gamma are about 2.9:1:0.45. If you add in dust the effective Balmer decrement is even steeper.

    So H-beta in emission is intrinsically weaker than H-alpha and relatively speaking the correction for stellar absorption is greater too.

    The claim in the DR8 release paper is that the MPA-JHU pipeline produced more accurate and better physically motivated stellar contributions to the spectra, which if true would be a good reason to use it instead of the "native" estimates.

    One thing I had hoped was that it would be possible to substitute SDSS pipeline flux estimates in cases where the MPA fluxes were missing or hugely in error, but at best that would require transforming from one calibration to the other as well as galactic extinction correction (which is no problem) and H-beta would still be problematic.

    Posted

  • JeanTate by JeanTate

    Thanks! 😃

    Just in case this thread has readers other than us two, and in case any may be wondering about the terms, here's my off-the-top-of-my-head explanations (they contain omissions and errors; I hope a professional astronomer will correct them):

    • Balmer: atomic transitions in the hydrogen atom, to the n=2 level; they are named H-alpha (n=3 to n=2), H-beta (4 -> 2), H-gamma (5 -> 2), etc (except that the 8 -> 2 transition is named "H8", rather than "H-zeta"; etc) WP
    • line strength: flux in a line
    • Balmer decrement: change in relative line strength, going from H-alpha to H-beta to ...
    • effective Balmer decrement: the actual, observed, Balmer decrement, to which many physical processes contribute (not just the atomic physics to do with relative populations in excited states in a uniform radiation field). As dust almost always 'reddens' an object - reduces the intensity of shorter wavelengths coming from the object relative to longer ones - it makes the ratio of H-alpha to H-beta (both in emission) even greater than what it would be if there were no dust
    • galactic extinction: dust (and gas) in our own, Milky Way, galaxy both absorbs and reddens the light we receive from external objects (e.g. galaxies). Astronomers call this 'extinction'. A lot of work has gone into making maps of this galactic extinction, so you can 'correct for' it - estimate the flux assuming there was no dust (or gas) in our own galaxy.

    One thing I think follows: errors aside, objects with H-alpha emission flux < ~2.9 times the H-beta emission flux are (likely) outliers.

    Posted

  • jules by jules moderator

    I'm reading! Thanks for the glossary - it helps! 😃

    Posted