Newsgroups: alt.ufo.report,alt.paranet.ufo
From: MrPostingRobot@kymhorsell.com
Subject: mars weather predicts ufos seen on earth (despite seeming tampering with dataset)

[uploaded 5 times; last 19/09/2024]

EXECUTIVE SUMMARY
- We look at a sequence of images of Mars and select certain pixels to
  predict UFO activity over N Am the next week.  The initial model
  performs well.
- But there's something unusual. It performs much better in the first
  1/2 of the image set then in the 2nd half. Why?
- Disregarding that, some AI programs make a model that uses
  combinations of pixels and predict UFO activity equally well in the
  2nd 1/2 of the images as the first.
- They spot characteristic fingerprints of air-brushing in the 2nd 1/2
  of the images.
- By reverse-engineering they discover 2 UFO types were most affected
  by the air-brushing. It seems the tampering was designed to cover
  evidence of certain types of things associated with those types of
  objects. Given the types they turn out to be, we might suspect the
  Martian images, when blown up, with show unusual things that
  "obviously" are not supposed to be there.
- They further point to the date of the transition between the "no
  air-brushing" to "air-brushing" happened just on the dates we
  previously found 2 Adobe photoshop images substituted for the regular
  .png images. The filenames were still marked .png, which is what
  caused some programs to crash and draw attention to the discrepancy.
  Something of a "botched job". Maybe arguing it was done by someone
  outside the companies involved in producing and curating the images.
  Evidence of MIB?
(See <kym.massbus.org/MARS-WEATHER> for related plots and images).
 
The AI's have been chugging away on a set of images taken by the Mars
Observer probe and assembled by Malin Space Science Systems of SD CA
US (MSSS).  The images cover the period 2007 to 2022 with some gaps.
The gaps are mostly related to tuning the pipeline from probe to final
product. There was also that Pandemic thing that seemed to cause some
absences from the offices during 2022. But there may be other
reasons, too. :)

What they've come up with kills a few birds with one stone.  They can
use the images -- even tho they are (mostly) taken 1 week apart of a
randomly-positioned Mars in the full color frame -- to predict various
types of unusual objects seen over N Am about 1 wk later (travel
time?). The simplest program they came up with predicted the total UFO
count for next week correctly 70-80% of the time.  They also came up
with a 2nd, much more refined, program that pushed predictions up to
80-90% correct. And there is a good reason for this as we will go into
in the "photo detective" section of the article (way below).

I've set up a display area on my website for a couple Mars images (the
full set runs to several 100s and they are much of a muchness to the
human eye) and some highlighted areas showing which regions the AI's
zeroed in on for their predictive magic. Some areas are so diffuse
they can't be illustrated by a patch of yellow highlight or a red box
drawn around them. But {shrug}.

The AIs ran through the set of images (which themselves were selected
by the good folks at MSSS as representative of movies they made each
week of the red planet in the prev week; said sample image was
attached to the weekly weather report the company put out each week
until end 2022). Looking at each color value in all the images they
selected a bunch that appeared in around 1/2 the images. Such pixels
have, theoretically, the highest information content. (In information
theory something that always happens or never happens has no info
content; c.f. news items about car crashes and some celebrity found dead).

It assembled a count of each selected color (the bands were widened
out so any pixel within +-5 units of the R, G and B components of the
wanted color were considered a match) and created a time series of how
many pixels in the images were present on each publication date of the
image. They then lined that up against NUFORC sightings reports for
the following week, having done other work just recently that
suggested this kind of time lag might be a good thing to try.
Automated science in action. :)

Putting the various created data through a simple learning network
produced high levels of predictive success for each UFO type.  Only a
couple of types could not be handled in this first run.  They turned
out to be new categories at the NUFORC that have not received much
data yet -- e.g. "Star" types.

The summary table for this first run looks like:

UFO type	Percent correctly predicted 
		the week after the image
colorwhite	0.72723
ALL		0.72023
colorbright	0.71917
Unknown		0.71540
Disk		0.71169
Triangle	0.70934
Cross		0.69769
colorblack	0.69763
Other		0.69713
Rectangle	0.69704
colordark	0.69646
Oval		0.69583
Light		0.69232
Changing	0.68727
Diamond		0.68720
Flash		0.68057
Circle		0.67323
Cylinder	0.67185
Egg		0.67099
colorred	0.65620
Cone		0.65385
Teardrop	0.64992
coloryellow	0.64215
Sphere		0.63969
Chevron		0.62814
Formation	0.61680
Cigar		0.60307
Fireball	0.54430
colororange	0.53172
Star 		NO MODEL FOUND
faint 		NO MODEL FOUND

This far and away beats anything we've produced in the past 3-4 years
of looking at the various datasets. See the Archive for examples. :)
The last record -- some tweaks to what I call the "nav model"
predicted UFO sightings about 45% correct. One of the advantages of
that model (that assumes object travel around the solar system in
straight lines at fixed speeds, partly at random and partly because
they selected a destination that was the closest one to their current
position) was it gave a path the object was supposed to have come in
order to line up with some known UFO sighting. Since it plotted the
path of the relevant simulated objects it could also show in the sky
where they would appear (at long distance) and how they would move
related to known stars and constellations as they got closer and
closer to Planet Dirt. And, nicely, some of the movies they made
seemed to match up with what some people have reported seeing in terms
of "long distance" UFO's out there at relatively large distances
(compared to the trip down to the local burger joint).

The models are are a bit more subtle. They use the distribution of
certain colored pixels to calculate what objects are expected to
appear and when. They don't really prove the pixels "are" the objects
but just something nebulously related to them.  E.g. you can see from
the sample images some things like cloud and dust storms *may* be
related to UFO travel.  These things have characteristic colors. Maybe
the AI's picked up on just that. Maybe a dust storm makes Mars appear
brighter on Earth and some kid looks up in the sky and sees it and
thinks it's moving across the sky and is a UFO. But we don't really
think so.

And here is one reason to think these images mean something more.
Around 1/2 way through the project to make all these weather reports
there was a change. The reason I came across this was the AI programs
fell over trying to process 2 images from the set we're talking about.
My s/w can handle various format images. But it turned out none of
them can handle Adobe photoshop images. Some of them crash. And,
hence, it came to my attention that for some reason 2 images had been
altered for reason or reason unknown.

But in a previous post we found that the dates of the 2 altered
images correlated with the week after being a UFO flap -- in one case
a very large flap. So we thought we have narrowed down why those 2
images might have been altered.

But now for the sequel. In the models the AI's are creating this time
from the total sequence of images, they noticed that images in the 2nd
half of the set predicted things OK, but were less "fulsome" in their
predictions. In face, plotting out some details (see the web page) you
can typically see the 2nd half of the plot mostly consists of a
straight line with some bumps. The bumps still correspond to some UFO
sightings. But compared with the 1st 1/2 of the plot they are nothing.
In the first 1/2 the prediction goes up and down wildly. But very much
in sink with the way sightings (next wk) were going up and down.  In
the 2nd 1/2 everything was muted. And it seemed to get more muted as
things went on.

The AI's -- with some experience of looking at images these past few
years -- diagnosed a case of "air brushing". No matter what (whoever)
calls it, the technique involves scrambling around the pixels in
localised regions of an image to remove unwanted features. New pixels
are generally introduced but not too many -- mostly the pixels
already in the region and swirled around so as to make the
alternation merge in seamlessly (it is hope) with the rest of the
image already there. So all this leave a strong statistical
fingerprint that you can spot. If you're an AI.

So now things get interesting. The AI's ran off and tried to figure
out WHAT was being covered up. We can see from the table above all
types of UFO's are predicted about the same. More than 1/2. Less than
80%.  Kinda hard to tell if any particular type is better predicted
than the average type.

But there is a way to find out. If you can produce a much better model
that gets around the air brushing, you can maybe figure out which UFO
type is getting most hammered by the airbrushing that's going on.

(And, as before, I don't want to impune anyone's reputation here.  The
mere fact the photoshopped images were left on public display kinda
hints as being done outside the relevant companies.  The images being
used were selected for the weather reports by taking one frame from
the related movie of Mars probe images over the prev week. The company
surely would just have gone to another image if the sample one was not
suitable for some reason.  If they had to change it then it would have
still been a real PNG file instead of an Adobe image with the name
just sloppily changed to .png).

Does one particular UFO type just happen to get more effected by the
tampering than any other? You need a better model to compare with.

And since the model, above, is "grade school level" for model
building, there is a big upside we can explore. So the AI's went off
to build a simple "highschool level model".

Not only did the new model score generally 10 points better than the
first model on all categories, the plots of the performance suffered
much less from the degradation seen in the first one. There was still
a but of a straight line -- just in the past 12 m of the data inside
of several years with model #1 -- and the ups and downs of the
prediction were now similar to the fantastic swings in the early part
of the dataset. This new model was "seeing through" the airbrushing.
As B Ross would say "I knew it would".

So, finally, we can compare the first model with the
2nd. Specifically, we are looking to see if any particular type of UFO
was more affected by the apparent airbrushing than any other. If so,
we can figure out what is being hidden and -- since we know what color
pixels are involved -- exactly where on Mars these things are being seen.

The summary table of the relevant differences are thus:

UFO Type	Diff1	Diff2    absdiffofDiffs  ratioofDiffs
Chevron		12.2842 11.9734 diff=0.3108 ratio=0.974699
colorblack	2.6522 3.6893 diff=1.0371 ratio=1.39103
Changing	18.784 16.2309 diff=2.5531 ratio=0.864081
Cone		7.1061 3.6254 diff=3.4807 ratio=0.510181
Oval		29.397 32.973 diff=3.576 ratio=1.12165
Teardrop	1.574 5.3165 diff=3.7425 ratio=3.3777
Egg		7.7851 12.2155 diff=4.4304 ratio=1.56909
Cross		3.549 9.1951 diff=5.6461 ratio=2.5909
Formation	6.202 14.29 diff=8.088 ratio=2.3041
Flash		8.2378 17.6767 diff=9.4389 ratio=2.1458
Cigar		50.366 60.2836 diff=9.9176 ratio=1.19691
colordark	15.1701 2.9989 diff=12.1712 ratio=0.197685
Other		19.18 4.439 diff=14.741 ratio=0.231439
Diamond		11.715 27.7992 diff=16.0842 ratio=2.37296
Triangle	3.539 26.511 diff=22.972 ratio=7.4911
Disk		19.796 43.533 diff=23.737 ratio=2.19908
Fireball	208.579 180.931 diff=27.648 ratio=0.867446
Cylinder	7.3732 40.7922 diff=33.419 ratio=5.5325
Unknown		43.54 5.1 diff=38.44 ratio=0.117134
Rectangle	2.9233 42.8395 diff=39.9162 ratio=14.6545
Sphere		66.789 26.23 diff=40.559 ratio=0.392729
Circle		32.342 74.013 diff=41.671 ratio=2.28845
ALL		329.01 374.61 diff=45.6 ratio=1.1386
colorbright	11.917 147.456 diff=135.539 ratio=12.3736
Light		21.901 477.446 diff=455.545 ratio=21.8002

The first column is the obvious UFO type we're looking at.  Diff1 is
the first models average predictive success between the first and 2nd
halves of the list of images.  The Diff2 column is the same for Model
#2.  The "absdiffofDiffs" is the difference between Diff1 and Diff2.
The "ratio" is the ratio of the 2 Diffs.  When the "absdiff" or
"ratio" get big -- it means the 2nd model was doing WAY BETTER than
the first model and predicting in the 2nd half of the images.

I've sorted the table by the "diff=" column. At the bottom we see what
UFO types seem to be worst affected by the suspected
tampering. "Light" and "bright". Well.

Show of hands. Does anyone find that objects someone might report as
"light in the sky" or "bright light in the sky" just *might* be
visible as the same kind of thing on Mars images and maybe someone
doesn't want anyone to see that?

Of course the debunkers might argue the whole thing is a beat-up.  The
process was all above board. As scientists fiddled with the data they
found better ways to represent it and the colors maybe drifted. (Part
of the problem my own AI's found in the beginning was PNG and JPG type
images that got mixed together in early parts of the data had
somewhat different color models and some of the pixels found in PNG
images to be predictive were not present or not predictive when found
in JPG images; but they quickly found a way to handle that -- making
the color bands less selective -- allowing a +- 5 range for each R, G
and B value to match the colors wanted for the models).

But the final nail in the coffin I think is the date where this change
over from un-airbrushed to airbrushed images happened.  It was in the
middle of the data. The date of the relevant parts of the plots is
around middle 2013.

And what were the dates of those 2 photoshopped images?

Let me just check: 2013.03.27.png  2013.07.03.png.
(They claim to be .png files but are Adobe format -- a totally different
thing, hence the falling-over stuff).

So it looks like someone probably altered some images with Photoshop
just where the image sequence started to change. And the change seemed
to "mostly" affect predicting of "bright" and "Light" objects.  Where
the simple model seemed to predict them well before 2013, it suddenly
went blind with images after 2013.  The 2nd model -- which actually
uses combinations of 2 colors at a time instead of just the one --
gets over that and predicts almost as well *with* the photoshopped
images as previously.  And even this is a characteristic of
photoshop -- it swirls pixels around -- they don't all get
deleted. Their density is decreased.  So provided you have a backup
pixel color to look for at the same time, you can still use the
count of the "mixed up pixels" to make the predictions you wanted. The
AI's can X-ray photoshopped images easy as pi.

--
[Secret UFO recovery program blown open:]
I hope this revelation serves as an ontological shock sociologically
and provides a generally uniting issue for nations of the world to
re-assess their priorities.
-- David Grusch, 05 Jun 2023
[Talking to Les Kean et al for The Debrief, Grusch called for an end to
nearly a century of global UFO secrecy and warned that humanity needed to
prepare itself for "an unexpected, non-human intelligence contact scenario"].

The US government portrays itself as the world's preeminent
superpower, so to acknowledge that there are things in their
airspace, whatever they are, that are faster and more manoeuvrable
and run rings around fast jets doesn't play very well.
So there's the embarrassment factor, and maybe a little bit of
fear that either an adversary has made a quantum leap in
development, which has left the US in a poor second place, or, as
some believe, this really is extra terrestrial, in which case we're
not at the top of the food chain anymore.
-- Nick Pope, 02 May 2023

"[F]or the few cases in all domains--space, air, and sea--that do
demonstrate potentially anomalous characteristics, AARO exists to help the
DOD, IC, and interagency resolve those anomalous cases. In doing so, AARO is
approaching these cases with the highest level of objectivity and analytic
rigor. This includes physically testing and employing modeling and
simulation to validate our analyses and underlying theories, and
peer-reviewing those results within the U.S. government, industry partners,
and appropriately cleared academic institutions before reaching any
conclusions."
-- Dr Sean Kirkpatrick, Senate Hearings on UFOs, 19 Apr 2023.

Palmdale UFO Scare Leads To Revelations About Mystery Drone Incursions Over
Secretive Plant 42
YAHOO!News, 20 Aug 2024 04:06Z
The U.S. Air Force's Plant 42 in California, America's premier hub for
advanced aerospace development work, especially highly ...

[Flap!]
Multiple UFO Sightings in California: What We Know
Newsweek on MSN, 19 Aug 2024 12:26Z
An apparent spike in UFO sightings has captured the public's imagination
after a spate of reports and footage coming out of California over the past
week.A series of videos are doing the rounds across ...

How the World Sleepwalked Into the Global Mpox Emergency
Bloomberg, 19 Aug 2024 05:40Z

Flu virus that killed millions of poultry found in dairy cows. Is it
in California?
Fresno Bee, 15 Aug 2024

Former head of secret government UFO program reveals why he is certain Pentagon has material from crashed ...
Daily Mail, 18 Aug 2024 15:24Z
Luis Elizondo, 52, helped run the Advanced Aerospace Threat Identification
Program (AATIP) from 2009 to 2017, investigating UFOs that harassed Navy...

Why China's fifth Antarctic base has the world's spies watching. Should
Australia be worried?
ABC News, 18 Aug 2024 05:32Z
It might look like an ice-bound resort but China's latest Antarctic
station, Qinling, has Australia's intelligence community concerned.

UFO crashed in Arizona and 'was stolen by officials in cover-up', leaked
government documents claim
Irish Star on MSN, 08 Aug 2024 17:55Z
A UFO crashed in Arizona, leaked documents claim. The US seized a UFO in May
1953 that crashed outside Kingman, near Route 66, according to a
conversation between government officials. The visitors ...

Are Underwater UFOs an Imminent Threat? The U.S. Government Sure Thinks So--
And Here's the Proof
Popular Mechanics, 08 Aug 2024 21:54Z
Legislators went so far as to formally change the way they refer to UFO
sightings over and under bodies of water.