Attempt at some science: removing SHO emission from RGB

[Placidus]

We wanted to produce a Gabriela Mistral in SHO with RGB stars. We took about 22 hours of SHO, and 1 hour total of RGB.

We noticed that in the RGB image, when stretched enough for the stars to be brighter and fatter than the SHO stars, there was a lot of nebulosity, with Gabriela's face still very recognizable, but in the "wrong" colours for SHO.

If we stretched less, then when they were combined, there was still a conspicuous annular ring from the very highly stretched SHO stars. What to do?

I used multiple linear regression to statistically remove ("correct for") the emission nebulosity in the RGB image, given the SHO image.

If I used the whole image, it just didn't work.

But if I ignored the brightest purple nebulosity close around the bright star under Gabriela's chin (the one with the purple lightning bolts coming out of it), it worked beyond splendidly: 99.99% of the variance in the background of the RGB image was accounted for by SHO emission nebulosity.

If you scroll through the five attached images, you will see:

(a) The straight SHO image, with big purple stars
(b) The straight RGB image, with Gabriela's face clearly visible (big one here (https://photos.smugmug.com/Category/Star-Forming-Regions/i-rKJgnJT/0/234076e4/X2/Gabriela%20Mistral%20RGB%20stretche d%20100-X2.jpg))
(c) RGB given SHO (shorthand: RGB|SHO): the RGB image "corrected for" the SHO image. (big one here (https://photos.smugmug.com/Category/Star-Forming-Regions/i-qR9s9LZ/0/044c07b7/X4/Gabriela%20Mistral%20RGB%20correcte d%20for%20SHO-X4.jpg)). There should be less than 0.01% of the SHO emission nebulosity left in the image. Any nebulosity that is still there is not SHO, but something else. This is the important result.
(d) A starless, deconvolved, wavelet sharpened SHO image.
(e) Just for fun, the SHO image with the "corrected" RGB stars dropped in (result = brighter of RGB|SHO and starless SHO.
The final image is here (https://photos.smugmug.com/Category/Star-Forming-Regions/i-4HXghm4/0/cbf0d012/X5/Z%20brighter%20of%28%20RGB%20correc ted%20for%20SHO%2C%20starless%20sha rpened%29-X5.jpg).


So what is the remaining nebulosity in the RGB|SHO image? What is the nature of those purple "whiskers" or lightning bolts?

The most obvious answer is reflection nebulosity. Gas well behind the big bright star, with the ropy whiskers nothing to do with the star, but to do with shock fronts in the distant gas.

Another answer is that it is something exotic. One possibility is another emission line, for example N2. But one would expect N2 to be statistically highly correlated with one or more of SII, H-alpha, and OIII. There are other weird kinds of radiation, such as synchrotron radiation (the blue glow in the Crab nebula) due to accelerating charged particles in a magnetic field, but the blue glow in the crab nebula is uniform, not ropy.

Conclusion: the bright excess nebulosity close to the bright star under the chin, after exclusion of SII, H-alpha, and OIII, would most prosaically be reflection nebulosity from gas and dust well behind the star. It could just possibly be something more exotic. A spectrometer could distinguish between continuum light (such as reflected starlight) and exotic narrowband emission.

Best,
Mike

[Merlin66]

The use of a spectrograph would allow you to analyse the actual emission wavelengths of the nebula.
The colour “representation” could then be matched to the actual spectrum....

[Atmos]

That’s a really interesting methodology MnT, not something I’d considered doing but it’s worked well!
What makes it interesting is that it’s purple in the SHO image and OIII in colour in RGB which is suggestive of both SII and OIII emissions. If it was purely reflection nebulosity then it shouldn’t show up in SHO at all like other reflection nebula.

It’s a bit difficult to see in these images but I’m assuming it still has the “squiggles” in Version C?

The overall look around that star appears to be reflection nebula after the SHO has been removed so I think we can say that it’s a given. It’s a bright energetic star so it’s not surprising to see reflection nebulosity around it.
It also cannot be purely reflection otherwise it wouldn’t appear so strong in OII and SII. As to what could cause that process, it’s still racking my brain a bit. It’s definitely suggestive of high energy processes in the past with SII but a decent NII emission isn’t suggested by the blue colour. In pure RGB it still looks like a very strong OIII emission.

[codemonkey]

Super cool, M&T! Like Colin, not something I'd ever considered.

What did you use to produce D?

[multiweb]

Fascinating. And a great result to boot. :thumbsup:

[Placidus]

Thanks Colin! Totally agree. The one slight gloss is that if you hold an OIII filter up to continuum light, then some tiny percentage of it does get through. Hence the some ridiculously tiny percentage of the light from a reflection nebula (say 3%) does get through an OIII filter.



Yes. Big version here (https://photos.smugmug.com/Category/Star-Forming-Regions/i-qR9s9LZ/0/044c07b7/X4/Gabriela%20Mistral%20RGB%20correcte d%20for%20SHO-X4.jpg). I've edited the original post to include this.



Again, a well-reasoned argument, and we thank you for taking the time to think about it. We reckon astrophotography is all the richer if one stops and ponders what one is actually looking at. Titanic forces at play.

We still can't quite see what the physics of the "lightning bolts" might be, but they do seem to involve both emission nebulosity and reflection nebulosity. A guess might be that there is foreground emission nebulosity with background reflection nebulosity. We need a resident super-guru, to give the answer in the back of the book.



Thanks Lee. We removed the stars using our own in-house software package "GoodLook". A few mouse clicks and you find all the stars (bandpass filter, find the local peaks, select those with a high correlation with a stellar template). Then you use bilinear polynomial regression to try to guess what would have been "under" the star if it weren't there, given the immediate background. It works superbly except for the biggest, baddest, burned out stars, of which there were two in this image.

We tried leaving out this star-removing step, and adding the RGB|SHO directly into the HSO image, but it didn't work because the OIII and SII fringes around the SHO stars are so much bigger than the RGB stars.



Thanks muchly Marc! I guess we achieved our original goal of making an NB image with halfway decent RGB stars, by using multiple linear regression to remove SHO "contamination" from the RGB image, and on the way, realized that Gabriela Mistral must include a lot of reflection nebulosity. That was a bit of an adventure, almost as good as growing potatoes.

Best,
Mike

[Merlin66]

All you need is a Spectral image of the nebula to fully show the emission lines and any absorption lines.
You could then map these to you filter cut off wavelengths.......

[rustigsmed]

really cool stuff M&T (and an interesting investigation!) - thanks for sharing.

Russ

[Placidus]

Thanks Russ, that's kind.

[Peter Ward]

I have seen this in some of my data.

The whiskers you mention are however easily explained.

Gabriella is an old Greek lady after all....:D

[Placidus]

:D I am myself one sixteenth Greek. Great Great Grand-Dad found gold not so far from the farm. I am sure that you are at least 1/16 correct. Mike. :D

[Paul Haese]

I really like the E version Mike. RGB stars and NB detail. :thumbsup:

[Peter Ward]

OK..well...the less said about Catfish, the better. :D

[marc4darkskies]

Very interesting analysis M&T! A nice image too! :thumbsup:

[Placidus]

Thanks very much Paul and Marcus.

It will be interesting to see how it goes on other nebulas.

If there is no, or negligible, reflection nebulosity, the technique is easy. A great weakness of the technique is that if there is reflection nebulosity, like in this case, you have to use experience and judgement to find representative parts of the image that you guess are pure emission nebula. The maths will then tell you unequivocally (to an amazing number of decimal places) whether you are right or wrong. Even including one single patch of reflection nebulosity hugely changed the multiple regression goodness of fit, from say 99.99% to say 80%, so there is no doubt about it. (A bit like testing for a gas leak with a lighted match).

But automating it would require some slow process like a genetic algorithm, where you randomly select a bunch of say 20 patches of the image, call that set A, then another random set called B, etc, etc, etc, and you pit them against each other in a survival of the fittest, where (:D) fitness is the multiple regression goodness of fit. You choose the best 5 contenders, get them to mix and match their patches, and try again. That would keep me off the streets.

Nah, doing it manually doesn't take so very long.