I’m going to shift a bit here, though the next two posts on this topic are already planned (though Sharon over at Doubtful News just pre-empted me tonight on the Crrow777 stuff that’s hit Newsweek). Instead of discussing pseudoscience that I’ve seen elsewhere, I’m going to discuss my own. Not pseudoscience, per se, but where science can go wrong when you have little sleep and are under extreme pressure to do things quickly.
But before I get specifically to this, I want to emphasize: News reports that there are “no craters on Pluto” are wrong. There are clearly impact craters. It’s that there are no unambiguously yet observed impact craters on Sputnik Planum. That out of the way:
I made a boo-boo. But, science is ultimately self-correcting because if it’s wrong, then when people try to duplicate it, they will get different results …
I generally study impact craters (among other things). One of my primary science areas of research for the Pluto-Charon system is to understand their crater populations to tease out what the impacts are like out there 40AU from home and what the geologic history of the bodies are. To do that, you have to map craters. I’m going to be focusing on that in the coming months (and currently) and I’m also going to be focusing on how our mapping changes as we start to get lossless data and higher pixel-scale data (not higher “resolution,” for “resolution” means number of pixels, while “pixel scale” refers to the length per pixel). This latter focus has been something I’ve been publishing on in the last year.
As I’ve mentioned before on this blog, images right now are being sent down lossy compressed. Meaning they are full of JPEG artifacts that wash out a lot of small features … like impact craters. So when mapping, I’m assigning a subjective confidence level that indicates how certain I am that a feature is a crater or not. Since we have repeat imagery, already, I’m going over each area multiple times, blindly, with the different images.
One area that’s hit the news is Sputnik Planum, on the “left” side of the bright albedo feature Tombaugh Regio. It’s bright, and it’s young, and we know it’s relatively young because it has no unambiguous impact craters in the images that we have so far. I’m very careful with that phrasing: unambiguous impact craters in the images that we have so far.
Except, I thought I found one. A rather large one. But I didn’t.
When I initially mapped it in the image that came down a week ago (the full-frame image that was unveiled the morning of the encounter), I gave it a confidence level of 4 out of 5. We had the lossy-compressed JPEG version of the image, and after we had attempted to remove some of the JPEG artifacts through Fourier Transform truncation and then deconvolved it with the point-spread function of the camera (the camera inherently blurs things a teeny bit), it looked like a crater, and I was pretty certain it was a crater. Since it was many pixels wide and the image had a pixel scale of 3.8 km/px, that is a significantly sized crater, at least 30 km in diameter.
Except, it wasn’t. We have since gotten a mosaic at 2.2 km/px of the planet, and we have gotten higher pixel scale images at 400 m/px that have not yet been released. In none of these is that very large, very obvious crater present.
What happened?
We made a tiny artifact bigger by image processing. It was a simple cosmic ray hit.
Here’s what happened:
- Cosmic ray hit the detector, meaning there was a very bright pixel with a lot of electrons in it.
- This detector has the annoying property that if you have a bright spot, a dark streak forms behind it. You can see this in all of the over-exposed hazards search images. So the bright pixel now had a dark streak behind it.
- This was lossy JPG compressed on the spacecraft by a severe amount. Heavy JPG compression can make things “ring” because it represents the data as a series of cosine waves.
- One of our basic image processors took that image and first deconvolved it, sharpening the ringing JPEG noise.
- He then looked at the image in frequency space and made a series of clips that when brought back into spatial space (what we’re used to) will dampen a lot of the obvious JPG blockiness and make for an image that is more aesthetic and helps to make out a lot more features because you don’t have the 8×8 grid of JPG blocks dominating.
This is perfectly reasonable to do, and so long as you understand the kinds of artifacts that it can introduce and don’t over-interpret it, you’re fine.
Unfortunately, it makes this particular kind of cosmic ray hit on this particular detector look like a very clear, very obvious impact crater. Despite my best efforts at not over-interpreting early images that clearly showed artifacts from the image processing, I over-interpreted this feature.
Fortunately, it never made it into a press release or a paper (though I will be talking about it in a paper I’ll be writing as a cautionary tale), but when doing stuff like this, I’m always reminded of how (and this is going to sound arrogant) I’m different from a pseudoscientist, and how working on skepticism for the past (nearly) decade has helped me to become a better scientist. Someone like Richard Hoagland, Mike Bara, Keith Laney, or the guy I talked about in the last blog post probably would not hesitate to make a big deal out of these kinds of features.
To be blunt, I’m a crater expert. I am considered to be an expert in mapping impact craters due to my experience at mapping over 1 million impact craters across 7 solar system bodies (so far). Yet, I made this significant mistake. What separates me from the pseudoscientist, though, is that when I was presenting this to people, I said that this looks very much like a certain crater, but we need to wait to see the uncompressed version of the image, and we need to wait for the higher-resolution maps before saying it’s certain. And if it isn’t, “it will be very interesting to figure out why it isn’t a crater.” I specifically said that in a team meeting on Sunday.
Many things right now are provisional simply because of the very lossy image compression. Features like craters are particularly difficult to tease out, unless they are very large and very obvious (as are many). Contrast that with the people trumpeting “geometric structures” on Pluto and Charon in these images. Of course there are “geometric structures” that were “artificially created” … all in the lossy JPG compression algorithm! I keep thinking I’m repeating myself with this — and I am — but people still keep making this claim.
But, I’m perfectly willing to be corrected. In fact, I have now written 1000 words about how and why I was wrong, and the exact reasons and process that led me to that erroneous conclusion: Based on better data, I can re-examine things and see what went on and if it’s real. Contrast that with what I listened to earlier today which was a discussion between Richard Hoagland, Keith Laney, and the host of Skywatchers Radio. This quote involves all three men, talking about the Norgay Montes image released last week, and where one stops and the other starts doesn’t really matter, for all three were complicit in this train of thought:
“Look around in that image. You will be amazed. The more you look, the more you’ll see. It’s pretty incredible. Blow the image up as much as possible and look at every little part of that image. There’s so much artificial stuff in there! Again, as denoted by the geometry.”
QED
Yep. You have to initially look at the point spread function (PSF) in the raw image and then toss out, based on examining the PSF, what are obvious cosmic ray hits. That is kind of hard to do since a cosmic ray hit totally destroys whatever image information was being recorded in the few image pixels which were hit by the cosmic ray during the exposure. One is left with the only option of mapping “into and onto” the pixels hit, by the cosmic ray, data from the surrounding pixels in order to make a “guess” of what would have been recorded by those pixels if the cosmic ray had not struck those pixels within the CCD sensor.
Here is a way for the layman to imagine the effect of a cosmic ray hit on a CCD sensor: Take a 35mm slide film photograph. Lay it on a table. Poke the slide film with a needle, just enough to create an extremely tiny hole in the film. The end result is similar to a cosmic ray hit on a CCD sensor. On the slide film, you have now created a tiny hole. You have no idea what image information in the form of exposed silver film grains existed in the tiny spot where the hole is. You can only make a guess, based on what is seen in the film around the hole.
During my years of hunting for supernovae (six to my credit), we ruled out hundreds of potential candidates by examining the PSF. In nearly every case, a PSF which didn’t match the PSF of nearby stars of similar magnitude always turned out to be a cosmic ray hit. This is pretty obvious to do when one is looking at stars on a nearly black background. But with New Horizon’s images of a planetary surface, we don’t have the really high contrast associated with a star against a dark background. On top of this issue is the fact that the images which New Horizons has delivered so far, during its close flyby of Pluto, are in a lossy JPEG compressed image format. The JPEG format, depending on the chosen compression ratio, creates blocks within the image of a given size. The issue is that not only has loss of some (or possibly a lot) of the original image information occurred (depending on the chosen JPEG compression ratio), but also that the JPEG compression algorithm inherently creates high contrast lines or points along the edge of each JPEG block within the image.
There are third party software programs which attempt to reverse the artifacts which are created by JPEG compression, but of course these programs can not fully “reverse engineer” the JPEG compression scheme since, by definition, JPEG compression results in the loss of some of the original image information. Thus such third party software, while being able to mostly remove the JPEG compression artifacts, can NEVER restore the original raw image information which is missing from the original uncompressed image. What is lost by JPEG compression is detailed information in the horizontal and vertical image planes, along with dynamic range of the original image colors. JPEG compression also results in a slight yet obvious shift in hue around the pure red color since magenta wraps around to red within the JPEG compression algorithm.
The upshot is that we will have to wait a few months for the New Horizons raw image data to be downloaded at a tricklingly small transmission rate due to the distance and extremely faint radio signal from New Horizons.
Comment by Michael Marcus — July 21, 2015 @ 10:18 pm |
Thank you for sharing this with us, and pointing out how a real scientist’s work differs significantly from that of pseudoscientists. It takes a very strong personality to admit to a mistake, then explain in detail what went wrong.
Something I’ve noticed from your last few entries, is how quickly the latter tend to jump all over anything that seems to confirm their biases. They appear almost in competition with one another, to see who can be the first to claim they found ruins of a civilization on whatever world has recently been photographed, no matter the quality of the photo. I suppose this is so whoever had the most convincing story, or maybe even whoever has the biggest platform on which to trumpet their findings, can then tour their idea around conventions and maybe even write a book, to make enough money to pay the bills for a while. Did you notice in the explanatory video from Alien Art TV, that the narrator mentioned his earlier clip, “Pluto Has Buildings,” “might go viral”?
I can’t help but do a facepalm, because these guys come across as so pathetic when compared to the real thing (you).
Comment by Rick K. — July 21, 2015 @ 10:44 pm |
Yup, I noticed that he said it “might go viral.” I think I rolled my eyes when he said it.
I’ve also noticed that they are very quick to say that the latest confirms everything they said before. That’s why I was able to say in my last post (#4) that I was able to predict exactly what they would say. It’s predictable.
By the opposite token, for the real scientists, what we’re seeing geologically is some part predicted by some part not. I can’t discuss unreleased stuff, but I’ll say to stay tuned to the press conference and associated releases on Friday. Stuff we had never predicted. And what’s more, also contrary to what pseudoscientists would have you believe, scientists are excited by that! Because the predictions were wrong – or incomplete – that means that there’s still more to learn, still more to find out, still more to discover!
Comment by Stuart Robbins — July 21, 2015 @ 10:50 pm |