Tinetti et al have a paper in Nature, July 12th, claiming infrared spectroscopic detection of water on an extrasolar planet.

Just appeared on the exoplanet.eu mailing list/web site

Paper is there Water vapour in the atmosphere of a transiting extrasolar planet (pdf)
(here is Nature link)

Detection is by comparing broadband IRAC mid-infrared data from Spitzer with detailed atmospheric models.

Object is transiting hot Jovian HD189733b – Jupiter mass planet in 2 day orbit around a K dwarf about 60 lightyears away.

Claims is the absorption seen in mid-IR is only consistent with water vapour at about 0.05% in the upper atmosphere – they also note they get a slightly different radius in the IR than in the optical during transits – possibly due to high altitude clouds (transparent to IR but opaque in optical).
Claim robust detection of molecular species through spectroscopy.

Looks pretty good, I’m sure some people will work hard to see if there are alternative explanations.

There have been previous contested claims of water detection, this one pushes things further a bit, two separate issues: one is proof of concept that we can in fact do chemistry through absorption spectroscopy of extrasolar planets, rest is just engineering; and, there is water out there, as there should be.

Good solid science, pushing what is doable with current instruments that we should remember were not designed to do this science, it is serendipitous that this sort of data can be squeezed out of them since they were spec’d to do different but related science.

Comments

  1. #1 Peter Erwin
    July 13, 2007

    I like the highly specific title The New York Times came up with for their story:

    “Scientists Find Evidence of Water on Planet”

  2. #2 ic348
    July 13, 2007

    I’m concerned about the paper in two respects.

    - IRAC observations are taken two at a time (3.6 & 5.8; 4.5 & 8), yet they do not show the 4.5 micron data. The observations in each filter not done sequentially. Unless this has somehow changed without my knowledge, there’s simply no way that they could have just not observed the source at 4.5 microns. The exclusion of that filter’s data seems rather suspicious. For all we know, the 4.5 micron data point lies well above or below the model predictions.

    And (in case they make this rebuttal) I don’t think it’s good enough to say that the data wasn’t included because of possible contamination with rovibrational CO. The authors are basing their argument on model fits of H20 spectral features to three data points. Perhaps I’m missing something, but I don’t see how modeling the CO lines (and thus ‘subtracting them out’) is an insurmountable problem. CO is one of the most frequently observed molecules in astronomy. There could be other lines overlapping the other three IRAC bandpasses, yet possible contamination of these lines does not deter the authors from modeling the emisssion.

    - The errors shown in the figure are only 1 sigma error bars. The photometry cannot possibly be anything close to 10-20 sigma. And if they had shown 3 sigma errors it would have been rather clear that the data is also consistent with no spectral features at all.

    At some point someone will provide a compelling case for water in the atmosphere of an exoplanet. For this paper, count me as unconvinced.

  3. #3 Lab Lemming
    July 17, 2007

    “And if they had shown 3 sigma errors it would have been rather clear that the data is also consistent with no spectral features at all.”

    You mean, that there was a 0.4% chance that the data was consistent with no spectral features?

  4. #4 ic348
    July 18, 2007

    “You mean, that there was a 0.4% chance that the data was consistent with no spectral features? ”

    Not quite the same thing. I’m saying that if you included three sigma error bars you could fit a straight line through all three data points that stays within the error bars. I wonder what the chi^2 value is for a straight line using these data.

    If there were a much larger number of data points then you could quantitatively estimate the likelihood that the data and a test set with no spectral feature are the same and also estimate the significance of the difference. As I remember, the authors of the paper did nothing other than say ‘see, this looks like a feature’. My response is ‘see, it’s also consistent with no feature at all’.