Newsgroups: alt.paranet.ufo
From: kymhorsell@gmail.com
Subject: predicting UFO sightings from satellite data and planetary positions

[uploaded 64 times; last 31/10/2024]

EXECUTIVE SUMMARY
- A simple neural net (NN) is created to predict in advance California
  monthly UFO sightings as per NUFORC database.  Some adjustments were
  made to allow for the intro of a web report form in approx March 2006.
- The input data are planetary positions as per an ephemeris nominally
  tuned to 2000-2050 but supposedly of reasonable accuracy back to
  1950. Other data include surface temperature measures from ground
  stations and satellites (the JMA gridded monthly dataset).
- The model is able to predict monthly California UFO sightings 12m in
  advance to an accuracy of around +-2. In particular, the NN is
  trained to predict "outliers" or UFO sighting clusters in preference
  to the mundane seasonal month-to-month average counts. But it seems to
  perform adequately for all months in the data.
- Similar models built to predict sightings in other US states were
  not always as successful although many were. It seems some US states
  generate sightings similar to "white noise". This does not mean
  those data are "junk" and all California sightings are "something".
  The model is not built to decide whether a particular predicted
  sighting is the result of "weather", "alcoholic haze", or "unusual
  physical object".
- Further work is in train to produce an annual or monthly UFO
  prediction s/w that might be useful to pre-position or at least
  anticipate UFO reports from certain regions up to 12m in advance.

  
This article is a outline of a very simple neural network that
predicts UFO sightings from sat-based surface temperature data and an
ephemeris of various planetary positions.

These data were not chosen because they are the best correlates of UFO
sightings data, but because they are reliably curated and updated by
govt agencies around the world and will provide a continuing quality
data stream into the future. The ephemeris data is generated by a
program distilled from tabulated data that can also be updated into
the future and also can be zoomed into hour by hour resolution if
necessary. Supposedly the Dec and RA for the planets in the dataset
are accurate to within a couple of arc minutes. Adequate for our
purposes here.

The s/w -- simulating a simple "brain" about the size of a sea squirt
(~200 neurons) -- takes a large set of monthly time series from the
online datasets and is trained to predict UFO sightings for California
from 1950-present as given by the NUFORC database. The present set-up
predicts sightings 12 months into the future. I.e. given data upto the
end of 2020 we can use the s/w to predict UFO activity in California
upto the end of 2021.

It turns out these sightings can be predicted to high precision given
they are only month by month resolution. The training procedure finds
the final error is within +-2 sightings for each month.  In particular
the error band applies to many of the "flaps" or sightings clusters
that have been seen in Cal from 1950.

We can not decide just here whether the "weather data" or the "planet
data" has a larger influence on the accuracy of the predictions.  In
any case, simply building a "good predictor" can't tell you who is
operating the UFO objects (or whatever) in question or whether some
large fraction of the sightings are e.g. weather-induced
hallucinations of some kind.

But it is quite interesting the input data chosen can predict the
"outliers" so well.

Here's a sample of the "observed" versus "predicted" California
sightings since 2000:

Date of data	Adjusted(*) NUFORC sightings for California
		for 1 year later
YYYY.MM		OBS	PRED
2000.04		44	44
2000.12		55	55
2000.21		44	44
2000.29		55	55
2000.38		44	44
2000.46		110	110
2000.54		110	109	<- prediction too low by 1
2000.62		55	55
2000.71		131	131
2000.79		66	65	<- 1
2000.88		55	54	<- 2
2000.96		55	55
2001.04		55	54
2001.12		44	44
2001.21		44	44
2001.29		22	22
2001.38		55	56
2001.46		88	88
2001.54		44	43
2001.62		88	88
2001.71		55	55
2001.79		44	44
2001.88		55	55
2001.96		22	21
2002.04		44	43
2002.12		22	21
2002.21		22	22
2002.29		55	55
2002.38		44	44
2002.46		66	66
2002.54		88	88
2002.62		66	66
2002.71		44	44
2002.79		44	43
2002.88		55	55
2002.96		55	55
2003.04		55	56
2003.12		22	21
2003.21		22	22
2003.29		110	109	<-- predicts cluster
2003.38		66	66
2003.46		153	153	<-- predicts cluster
2003.54		66	66
2003.62		22	22
2003.71		88	88
2003.79		88	88
2003.88		22	22
2003.96		22	22
2004.04		44	44
2004.12		22	21
2004.21		44	45
2004.29		88	88
2004.38		44	44
2004.46		131	131	<-- predicts cluster
2004.54		153	153	<-- predicts cluster
2004.62		66	66
2004.71		22	22
2004.79		66	65
2004.88		66	65
2004.96		55	55
2005.04		44	44
2005.12		131	131	<-- predicts cluster
2005.21		11	11
2005.29		33	33
2005.38		36	36
2005.46		34	34
2005.54		62	62
2005.62		52	52
2005.71		44	44
2005.79		44	44
2005.88		54	54
2005.96		66	66
2006.04		55	55
2006.12		31	31
2006.21		59	59
2006.29		36	36
2006.38		30	30
2006.46		46	47
2006.54		57	57
2006.62		50	50
2006.71		42	42
2006.79		48	48
2006.88		70	70
2006.96		58	58
2007.04		72	71
2007.12		52	52
2007.21		49	49
2007.29		59	59
2007.38		44	44
2007.46		65	65
2007.54		38	38
2007.62		54	54
2007.71		57	57
2007.79		77	77
2007.88		83	83
2007.96		50	50
2008.04		98	97
2008.12		61	60
2008.21		47	47
2008.29		30	30
2008.38		34	34
2008.46		55	55
2008.54		65	65
2008.62		66	66
2008.71		54	54
2008.79		42	42
2008.88		43	44
2008.96		44	44
2009.04		52	52
2009.12		30	30
2009.21		33	33
2009.29		38	38
2009.38		42	43
2009.46		42	42
2009.54		78	78
2009.62		63	63
2009.71		67	67
2009.79		52	51
2009.88		57	57
2009.96		41	41
2010.04		46	46
2010.12		54	53
2010.21		42	42
2010.29		52	51
2010.38		39	38
2010.46		37	37
2010.54		60	60
2010.62		70	70
2010.71		47	47
2010.79		55	55
2010.88		50	50
2010.96		65	66
2011.04		63	63
2011.12		35	34
2011.21		48	48
2011.29		57	57
2011.38		51	51
2011.46		79	79
2011.54		69	69
2011.62		75	75
2011.71		70	70
2011.79		73	73
2011.88		71	71
2011.96		60	59
2012.04		40	40
2012.12		45	45
2012.21		52	52
2012.29		53	53
2012.38		56	56
2012.46		58	58
2012.54		67	67
2012.62		64	64
2012.71		61	61
2012.79		57	57
2012.88		64	64
2012.96		102	102	<-- predicts cluster
2013.04		90	89
2013.12		73	72
2013.21		61	61
2013.29		77	77
2013.38		60	60
2013.46		80	80
2013.54		72	72
2013.62		79	79
2013.71		66	66
2013.79		68	67
2013.88		55	55
2013.96		67	67
2014.04		60	60
2014.12		39	40
2014.21		45	46
2014.29		34	34
2014.38		36	36
2014.46		34	34
2014.54		53	53
2014.62		48	48
2014.71		73	74
2014.79		72	73
2014.88		250	251	<-- predicts cluster
2014.96		50	51
2015.04		38	39
2015.12		64	65
2015.21		38	39
2015.29		34	34
2015.38		23	23
2015.46		49	50
2015.54		91	91
2015.62		51	51
2015.71		63	64
2015.79		40	41
2015.88		42	42
2015.96		36	36
2016.04		31	31
2016.12		34	34
2016.21		33	33
2016.29		51	51
2016.38		47	47
2016.46		37	37
2016.54		48	48
2016.62		46	46
2016.71		48	48
2016.79		58	58
2016.88		46	47
2016.96		95	95
2017.04		34	34
2017.12		42	42
2017.21		20	21
2017.29		16	16
2017.38		23	23
2017.46		10	11
2017.54		36	36
2017.62		32	32
2017.71		25	25
2017.79		44	44
2017.88		15	16
2017.96		17	18
2018.04		18	19
2018.12		23	23
2018.21		23	23
2018.29		27	27
2018.38		25	25
2018.46		37	37
2018.54		48	48
2018.62		39	39
2018.71		60	60
2018.79		68	68
2018.88		77	77
2018.96		48	48
2019.04		52	52
2019.12		71	70
2019.21		74	74
2019.29		98	99
2019.38		42	41
2019.46		37	37
2019.54		53	53
2019.62		47	47
2019.71		27	27
2019.79		30	29
2019.88		46	46
2019.96		42	42

(*) Data prior to around March 2006 was scaled by 9.8 to allow for the
    creation of a web report form around then. The adjustment
    minimizes the relevant F statistic to around 1.02 i.e.  the
    variance of the pre-2006 and post-2006 data is about equal.


--
Upcoming events:
9 Jul 2021		NOAA bn Dollar Disasters Q2

Experts Weigh In on Pentagon UFO Report
Scientific American, 08 Jun 2021 11:10Z
The vast majority of examined incidents were not caused by US advanced 
technology programs, the forthcoming report concludes. So what's going on?

Nasa has just rejected missions to moons of Jupiter and Neptune - here's 
what we would have found out
The Conversation UK, 07 Jun 2021
It's been 30 years since Nasa last visited Venus, with the Magellan orbiter 
in 1990. Now, 2 new missions have been selected to explore the deadly ...

[More TBs!]
Graphene could allow hard drives to hold 12 times more data
Engadget, 08 Jun 2021 0:55Z
Solid-state drives may be the go-to for phones and most PCs and laptops, but 
the humble mechanical hard drive still has a place in the computing world.

Early American Astronauts No Strangers to UFO Sightings
Townhal, 07 Jun 2021 18:23Z
The Pentagon is set to release a special report for Congress on recent 
sightings, essentially detailing what is known about ...