Support the Arctic Sea Ice Forum and Blog

Author Topic: Reverse Engineering "Goddard's" Real (un)Scientific Method  (Read 14868 times)

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Reverse Engineering "Goddard's" Real (un)Scientific Method
« on: August 05, 2014, 08:07:13 PM »
Is there a Gimp expert in the house?

I've now politely asked Tony Heller on two occasions to provide enough information to enable an independent 3rd party to replicate his results. We're talking Arctic sea ice extent here, not temperature.  Needless to say such information has not been forthcoming. This will give you a flavour of the quality of the "debate".

https://archive.today/FcIlO

Are there any volunteers out there who can provide me with some helpful hints?
« Last Edit: August 12, 2014, 02:36:22 PM by Jim Hunt »
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

A-Team

  • Nilas ice
  • Posts: 2317
    • View Profile
  • Liked: 237
  • Likes Given: 14
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #1 on: August 07, 2014, 03:34:39 PM »
Try a PM to Wipneus here, this type of Arctic area change image is his specialty. In talking about areas, keep in mind that a polar stereographic projection is conformal (preserves angles), meaning that it cannot also preserve areas (otherwise it would successfully flatten the positive curvature of a sphere cap to the zero curvature of the plane -- an isometry preserves curvature). So it is a little problematic to simply count pixels of a certain color using the color picker tools of gimp ... these would have to be ramped up by a latitudinal trig adjustment to get accurate areas.

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #2 on: August 07, 2014, 11:10:12 PM »
Try a PM to Wipneus here, this type of Arctic area change image is his specialty.

Thanks for the suggestion A-Team - Will do.

Quote
In talking about areas, keep in mind that a polar stereographic projection is conformal (preserves angles), meaning that it cannot also preserve areas (otherwise it would successfully flatten the positive curvature of a sphere cap to the zero curvature of the plane -- an isometry preserves curvature). So it is a little problematic to simply count pixels of a certain color using the color picker tools of gimp ... these would have to be ramped up by a latitudinal trig adjustment to get accurate areas.

As an ageing cynic I rather doubt Mr. "Goddard" worries about such subtleties, but thanks for pointing them out. Obviously the next thing to do after replicating his results (if that proves to be possible) is to reveal their shortcomings. For the moment I'll give him the benefit of the doubt and assume his weapon of choice in this instance is Gimp rather than Photoshop!
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #3 on: August 09, 2014, 08:23:56 AM »
Wipneus assures me that "I do not consider myself an image expert. I am about the worst person to ask"

Whilst I feel sure this is mere false modesty, it looks like I shall now have to delve deep into the Gimp manual myself.

More in due course.
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

Patrick

  • New ice
  • Posts: 24
    • View Profile
  • Liked: 0
  • Likes Given: 0
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #4 on: August 09, 2014, 04:33:23 PM »
I don't know how far you're with your own experiments, but I'll just post here a step-by-step guide. This might not be the fastest possible way, but whatever... ;D

  • Download the two files you want to compare and open them in GIMP as layers. The order is not important, but I'll just assume the 2012 image is on top of the 2014 one.
  • Go to "Image" -> "Mode" and change it to RGB.
  • Toggle off the visibilty of the top layer and make sure you select the bottom layer to work on.
  • Choose the color selection tool (Shift+O) and select all the areas which are above 1m thickness via shift-clicking the associated colors on the scale.
  • Pick red as the foreground color and fill your selection via shift-clicking into the area with the "bucket fill" tool (Shift+B).
  • Invert your selection (Ctrl+I) and fill it with black.
  • Select the top layer (2012 one), toggle on visibility again and repeat steps 4-6, just instead of red use green.
  • Now you have to change the layer mode of the top layer to "Difference" et voilà...
  • As a last step you may want to get back the map as a background for orientation. So first choose "Layers" -> "Merge Down". Select the ice area with the color selection tool, right-click the layer and click on "Add Layer Mask". In the following pop-up window make sure you choose "Selection" and click "Ok". Now you can open again one of your downloaded files as a layer and lower it in the layer dialog. That should basically be it...
Hope it helps.  :)

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #5 on: August 09, 2014, 10:18:00 PM »
Thanks Patrick,

That looks to be precisely what I need! Fortunately in all the circumstances I was distracted from Gimp by the imminent arrival of ex Hurricane Bertha in my own back yard:

http://econnexus.org/bertha-brings-flood-warnings-to-sw-england/

I'll let you know the results of my experiments, but probably not until tomorrow, since it's late here now (UTC). Always assuming we don't get washed or blown away before then!
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #6 on: August 10, 2014, 03:05:49 PM »
Time to roll my sleeves up, but Steve/Tony keeps on moving the goalposts!

https://archive.today/JmxXJ#selection-1079.0-1091.27

Quote
Let me see if I have got this straight. Your 56% headline number is based on a methodology of your own devising? If so where is this novel methodology described?

I’m afraid “About the same as DMI’s 30% measurements” just doesn’t cut the mustard.
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #7 on: August 10, 2014, 11:05:39 PM »
Thanks to Patrick, this is what I've come up with. It compares ACNFS thickness on July 29th 2012 with 2014. Perhaps unsurprisingly, it doesn't look a whole lot like the (un)Real Science version.

I wonder why that is?
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

SteveMDFP

  • Nilas ice
  • Posts: 1389
    • View Profile
  • Liked: 166
  • Likes Given: 15
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #8 on: August 11, 2014, 01:03:37 AM »
Thanks to Patrick, this is what I've come up with. It compares ACNFS thickness on July 29th 2012 with 2014. Perhaps unsurprisingly, it doesn't look a whole lot like the (un)Real Science version.

I wonder why that is?
I've been eagerly awaiting your results.  Yes, the yours and his differ pretty strikingly in the Beaufort especially.  I was wondering if you'd done your own pixel counts, to compare with "Goddard"'s claim of 18 percent increase?  I also asked over there, but I don't know if my question will get through moderation/censorship.

ChasingIce

  • New ice
  • Posts: 85
    • View Profile
  • Liked: 0
  • Likes Given: 0
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #9 on: August 11, 2014, 02:31:08 AM »
Jim, Goddard's image purportedly is limited to ice over 1m thick.  You would have to crop out the thin ice in both images 2012 & 2014 before starting the comparison if I'm understanding the whole thing correctly.

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #10 on: August 11, 2014, 06:17:09 AM »
Steve - Pixel counts, plus A-Team's "latitudinal trig adjustment" to follow eventually. Since I'm a Gimp noobie, how do you persuade it to count pixels?

ChasingIce - Patrick's method does that, so in the image above ice < 1 m (according to ACNFS!)  is already cropped.
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #11 on: August 11, 2014, 08:15:37 AM »
With the aid of the Great God Google I've discovered the histogram. I've also reverse engineered (approximately at least) "Steve's" JAXA methodology. Basically selecting everything within a threshold of 100 of pure white.

Shock News! 45.8% more ice than this date in 2012, based on pixel counting!
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

Wipneus

  • Citizen scientist
  • Young ice
  • Posts: 3973
    • View Profile
    • Arctische Pinguin
  • Liked: 545
  • Likes Given: 0
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #12 on: August 11, 2014, 08:36:53 AM »
Impressive Jim.

I suspect the Jaxa maps use an EASE grid, which is an equal area projection -> no worries about "trig latitudal adjustments"

Patrick

  • New ice
  • Posts: 24
    • View Profile
  • Liked: 0
  • Likes Given: 0
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #13 on: August 11, 2014, 07:13:22 PM »
I wondered why your ACNFS-example is looking so rough, Jim, and I think I've found the issue. When using the color selection tool make sure you untick "Antialiasing" on the left-hand side under tool options. That should give you a nice and clean result like this.

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #14 on: August 11, 2014, 07:27:21 PM »
Thanks again Patrick. I had in fact worked that out by the time I produced the JAXA version above.

However I sadly neglected to document it properly!
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

ChasingIce

  • New ice
  • Posts: 85
    • View Profile
  • Liked: 0
  • Likes Given: 0
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #15 on: August 12, 2014, 04:16:57 AM »

ChasingIce - Patrick's method does that, so in the image above ice < 1 m (according to ACNFS!)  is already cropped.

I did the eyeball test while being color blind...lol.   

keep up the good work.

A-Team

  • Nilas ice
  • Posts: 2317
    • View Profile
  • Liked: 237
  • Likes Given: 14
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #16 on: August 13, 2014, 07:22:56 PM »
There's no need to speculate. This is the azimuthal Ease grid (so ok to count pixels for area); the areal distortion of the Polar Stereographic map true at 70° N would have been -6% at the north pole.

A-Team

  • Nilas ice
  • Posts: 2317
    • View Profile
  • Liked: 237
  • Likes Given: 14
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #17 on: August 13, 2014, 10:24:53 PM »
How exactly does an azimuthal equal area projection differ from a polar stereographic conformal? They're easy to tell apart when both go all the way down to the equator -- the conformal map preserves shapes. However, restricted to the Arctic Ocean they are mighty hard to tell apart (animation below). Especially when in postage stamp maps with blurry jpeg land boundaries.

When a map is properly done, shift-clicking on all the colors in the color key selects *all* the colors deployed on the map and *nothing* else. Too often someone has done lossy compression along the way, dithering the histogram beyond repair by posterization etc.

I see lots published maps where clicking on key colors does not select *any* color on the map -- that arises from a failure to embed the key in the graphic. Of course they were supposed to embed metadata like the projection and scale.

All this is taught in US high school geography class, for sixteen year olds.

I only see this done right at a few sites such as NASA (which also provides graticules and text overlays as separate files from the scientific data).

Climate change graphics are often done badly in journals. The reason: the graphic arose first as a ppt talk at a meeting. Graphic quality doesn't matter in the back of an auditorium. The authors didn't feel like revising, so they submit the same crummy graphic to the journal, saying page layout there would degrade it anyway. Journals are not set up to host layer stacks. And let's not even talk about what legacy print would do to the image. And last but not least, a scientifically correct graphic might aid competitors. So they try to get away with it by calling it  'presentation quality'.

On your project, I think overall that comparisons are not useful unless grounded in a discussion of cumulative measurement error. The satellite has grossly inadequate resolution for the job it is supposed to do, the algorithm wouldn't work right even if it did, many processing steps at inadequate bit depth (precision) introduce error, as do re-projection to a grid, lossy compression, and screen display hardware. About all you can do error-free is count pixels.

A-Team

  • Nilas ice
  • Posts: 2317
    • View Profile
  • Liked: 237
  • Likes Given: 14
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #18 on: August 14, 2014, 04:01:45 AM »
This is hilarious. After first telling readers how ignorant they are, NSIDC proceeds to post an utterly ignorant reference map themselves -- an indexed gif using 126 colors to illustrate their equal area EASE.

Here, 3 colors were needed: land, water, backdrop. Never, ever use more colors than needed. Never, ever dither a digital scientific map project. Let downstream users do the degradation if they want.

"Discussion of map projections is often unnecessarily lengthy and sidetracked by ignorance or disregard for the fact that there is no one best map projection. Each projection has different properties and thus different best uses. Sometimes the question is raised as to why we chose equal-area projections over the other possibilities for the EASE-Grids, and the answer relies on a basic understanding of projection characteristics."

http://nsidc.org/data/ease/ease_grid.html#whyEqualArea

It is very very common these days to encounter maps utterly dislocated from their metadata. We wish to determine the projection by best fit to common projections. That involves rescaling reference projections. Since the ocean is featureless, we want a one-click land mask for pixel-perfect rescaling of a translucent layer. That cannot be done here with feathering, anti-aliasing, or color-picker radius setting.

I set out to provide us with the 4 reference projections used in Arctic maps and satellite imagery. That has morphed into a time-wasting search for a valid instance of this equal area map at a decent scale. Meanwhile JAXA prominently features all their land masks; Google maps also do a good job of embedding projection and scale.

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #19 on: August 14, 2014, 10:34:57 AM »
Thanks very much for the in depth analysis of assorted projections A-Team.

On your project, I think overall that comparisons are not useful unless grounded in a discussion of cumulative measurement error.

At this juncture my project consists of learning a bit of Gimp whilst now being able to swiftly debunk "Goddard" et al. when they match a typical "Shock News!" headline with an apparently random graphic. Thanks to all for your assistance.
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

A-Team

  • Nilas ice
  • Posts: 2317
    • View Profile
  • Liked: 237
  • Likes Given: 14
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #20 on: August 14, 2014, 04:28:39 PM »
Jim,

Worthwhile project, that rubbish should not go down unrebutted.

For the record, I chased down the IBCAO map which we have seen used frequently on this site. It is a polar stereographic. While done to high standards as a shaded relief, that makes it unsuitable for an straight land mask.

The color key does not work at all but I am not sure if they have released an unshaded version where you could click on a point in the Arctic Ocean and determine its depth.

The Gimp measuring tools provide the data needed to rotate/rescale it to an unknown projection. The tool should be applied at 2x or 3x to position it accurately. Here I went from Big Diomede to the Sweden/Finland coastal border -- the longer the line, the less the measuring error due to placement of the ends. You may have to use a different pair of reference points, depending on what the unknown map covers, but again use two extremes.

Usually the rotation is an exact integer, like 60º. That should be used if measurement on the two maps suggests a rotation of 59.95º or some such, even if the unknown map ends up being in a different projection.

If you drill around in the Gimp menu, you'll find a tool that rotates and rescales in one motion. That gives a slightly better result than a two-step.

Once the unknown projection is identified, the north pole can be located and a lat,lon grid applied (better: held out as a separate translucent layer with all but the grid lines in a fully transparent alpha channel).

http://www.ngdc.noaa.gov/mgg/bathymetry/arctic/2012GL052219.pdf

A-Team

  • Nilas ice
  • Posts: 2317
    • View Profile
  • Liked: 237
  • Likes Given: 14
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #21 on: August 14, 2014, 05:22:35 PM »
An ordinary google map uses mercator projection and so does not cover the poles. It is also slightly unsuitable for Greenland because Landsat and scientific imagery use a localized mercator grid called 22 N with less distortion.

For Greenland, about half the scientific maps use this mercator and the rest polar stereographic (plus unknowns). These are easily distinguished by eye if any lat,lon is provided as the latter projection does not give a rectangular cell. Mercator is really more convenient because a rectangular grid layer in meters is easily made in Gimp.

Meanwhle, google earth does something else entirely, projection onto a tangent plane from a bird's eye point of view over the center of the browser window. They want this for fly-ins but it causes the projection to change slightly with height (which however is provided).

The ice-penetrating radar tracks are provided relative to this coordinate system, as are the precious ultra-high resolution Digital Globe images which, if you dig in far enough, are provided as a time series. As a single shot runs to ~ $3,000, this is the only practical source of such imagery (which is essential to determining glacial motion from movement of small features).

Mouse-over gives lat,lon as well as the DEM map elevation. Wipneus might know if there is a convenient way of implementing mouse-over in forum software or in images opened from it. I've had it going in wikimedia but that is a ways off.

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #22 on: August 14, 2014, 07:28:26 PM »
Worthwhile project, that rubbish should not go down unrebutted.

Case in point: https://archive.today/GRjr3
« Last Edit: August 16, 2014, 12:43:35 PM by Jim Hunt »
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

A-Team

  • Nilas ice
  • Posts: 2317
    • View Profile
  • Liked: 237
  • Likes Given: 14
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #23 on: August 14, 2014, 07:45:59 PM »
A couple more things while I'm in idle mode with Landsat under maintaince.

If you are counting pixels for areas of each color, then the graticule (lat,lon lines) covering these pixels should be removed. A cheap way of doing that is to simply bump the count in each bin proportionately so that total pixels add up to the Arctic Basin.

The graticule should never have been flattened onto the map in the first place: the gif89a format as modified in netscape 2.0 took care of this twenty years ago, not that many mapmakers noticed. That is, make the map in gimp as 3 layers, graticle, map, graticule on map via 'new layer from visible'. Gif is limited to 8-bit indexed colors, 256 of them, which can be limiting (amounts to compression) for high resolution color photography.

Then provide the map as an animation that finishes on the combined last layer but does not loop. That way, end users sees the final flattened product but can recover individual map layers simply by saving the image to disk and opening in Preview, ImageJ, Gimp etc.

This is better scientific practice. After all, we're headed into GIS and a map is just one component of that. For instance, there might be dozens of other layers such as ocean depth, surface temperature, ice thickness, wind velocity field, time series and so on.

The idea of GIS is correlations and math right on the image. Note ImageJ provides 19 operations under Process --> Math and Gimp 20 -- and that's before a kazillion free plugins. So you can do everything from divide to boolean logic on pairs of layers. And it all can be dumped into numbers if you just have to work temporarily in Excel stacks or R etc.

In collaborative GIS -- and nowhere is the need for that greater than in climate change -- scientists share layers. Right now we have a lot of idiots flattening down their layers. To where they can't be re-purposed.

In theory you can obliterate an embedded graticule by replacing black pixels with nearest neighbor colored pixels. If you can lift the graticule with the color-picker settings, the temptation in these radial coordinate projections is to replace exactly that selection with what is there after a few degrees of rotation about the north pole. However that computation will produce pixels not in the color key.

Since pixels always form a rectangular array, the alternative respecting that geometry amounts to replacing the graticules with pixels displaced, say to the north and then to the west of the remaining vertical line. This keeps you within the color key. The north pole should be marked in a separate layer to maybe put it back in later.

In practice, this gives so-so results. Often you cannot even lift the graticule because it has been so dithered out making diagonal lat,lon lines and circles. This could have been avoided had the graticule been properly posterized and thresholded to begin with (so it makes dotted lines).

I have not located online graticule vector graphics for our projections that can be rescaled to fit the given map scale. One-size-fits-all raster graticules will need cleanup as described above after rescaling and rotation.

I see quite a few sea ice maps with the North Pole not indicated. If the map is centered (not all are), it will be halfway over and up in terms of pixel coordinates. Otherwise, it has to be located using crossed lines between pairs of well-separated land points or by fitting a land mask with known pole. I am of the opinion the NP should be marked up; the lat,lon are usually a standard integral rotation from Greenwich and it is not too invasive to mark a few crosshairs where meridians and latitude meet.


Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #25 on: August 14, 2014, 10:29:06 PM »
That's reminiscent of this one

Or indeed this one: https://archive.today/g95uO

An unwritten rule at the ASIF, for a variety of reasons. Only ever link to archives of "skeptical" articles, never to the articles themselves!
« Last Edit: August 15, 2014, 10:37:04 AM by Jim Hunt »
Reality is merely an illusion, albeit a very persistent one - Albert Einstein

A-Team

  • Nilas ice
  • Posts: 2317
    • View Profile
  • Liked: 237
  • Likes Given: 14
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #26 on: August 15, 2014, 02:19:11 AM »
Wrapping up, if you ever need a lat, lon grid layer for one of the common Arctic projections, they can be made in a few seconds in Gimp. Filter --> Pattern --> Grid followed by Filter --> Distorts --> Polar Coordinates, set according to the scale of your map. Then posterize/threshold/color and dim via opacity controller according to taste.

Jim Hunt

  • Young ice
  • Posts: 4208
    • View Profile
    • The Arctic sea ice Great White Con
  • Liked: 246
  • Likes Given: 26
Re: Reverse Engineering "Goddard's" Real (un)Scientific Method
« Reply #27 on: September 03, 2014, 04:02:15 PM »
David Rose is now propagating "Goddardian" nonsense in the Mail on Sunday once again, so I've started GIMPing the Scandinavian data from http://osisaf.met.no/p/ice/nh/conc/conc.shtml.

Here's my first attempt at a "DMI 30% clone".  I quickly discovered that .JPGs are a pain in the posterior! Maybe I'll have to delve into .HDFs and .GRBs? In the meantime note in particular the broad "Unclassified" bands around the edges. Whichever way I cut it, pixel counting reveals 2014 < 2013
« Last Edit: September 03, 2014, 04:09:02 PM by Jim Hunt »
Reality is merely an illusion, albeit a very persistent one - Albert Einstein