README

The airbourne ADS-B reports are crucial as they are the only
computer-readable calibration data available for deducing
oscillator bias, drift and the Perth AFC error contributions.
Inmarsat obviously had access to a much larger training database.
 
Three public ADS-B datasets were identified and steps were taken to
obtain each in the highest quality possible, and to sanity-check the
data.


Flightradar24
=============

  http://www.flightradar24.com/data/pinned/mh370-2d81a27/#2d81a27
  16:42:30--17:21:03
  62 records: F-WMSA2 (35), F-WMKC1 (20), T-WMKP2 (5), T-WMKN1 (2)

Flightradar24 has more incoming ADS-B messages, but appears to have
consolidation, meaning that the entries shown possibly aren't quite
perfectly the original messages received, but what FR24 do have is
obtainable in a single unauthenicated compact call to:

  curl -X POST 'http://api.fr24.com/api/common/v1/json-rpc/' -X POST -d '{"jsonrpc":"2.0","method":"Aircraft.doGetPlaybackSingle","params":[47716903],"id":1}' -o fr24-pinned-47716903.json

I asked Flightradar by email in tracking number ID [#41330] about the
unconsolidated messages.  The reply I received appeared to be
uninformed and somewhat miss the point I was enquiring after.
Another follow-up does not appear to have illicited a reply.
(Nor subsequent follow-ups to Fredrik Lindahl/Mikael Robertsson
directly).

A second set of data appears to be served out via the minute-by-minute
updates; some of these appear to have valid rate-of-climb (RoC) in
multiples of 64ft/minute (as would have been transmitted).

A JPEG image was posted on the FR24 website press page, with the last
ten altitude/rate-of-climb records; the low-precision locations are
probably a result of FR24's multilateration ("MLAT") setups using
time-of-arrival differentials.

  http://www.flightradar24.com/press/
  https://dl.dropboxusercontent.com/u/256889957/FR24%20Website/Press/mh370.jpg


Flightaware
===========

  http://uk.flightaware.com/live/flight/MAS370/history/20140307/1635Z/WMKK/ZBAA/tracklog
  16:43:26--17:02:40
  31 records: WMSA (17), WMKF (7), WMKK (4), WMKP (2), [null] (1)
  
  [long gap] then 17:50:50 (exactly at IGARI), with no feeder ID...

Flightaware seems to have less feeders.  There does not appear to be
a single location to obtain everything in the FA database; eg. the
proper API call (which needs a paid account + authentication) is to:

  https://username:flightxml.flightaware.com/json/FlightXML2/GetHistoricalTrack?faFlightID=MAS370-1394087710-airline-0005

The JSON response does not return feeder ID, which can only be
screen-scraped from the inline HTML (not in JSON).  Notably,
Flightaware's coverage is less, and starts later than Flightradar24's.

There is an obscure single entry at 17:50:50 (precisely at
IGARI!)---possibly inserted by hand in order to cope with FA's
coverage overwise ending much sooner than FR24's; the insertion
of this record brings into question the reliability of the rest of
the data.

In regard of the accuracy of the date, 'GetHistoricalTrack' claims to
only return true data:

  http://discussions.flightaware.com/post135440.html?sid=4fb56f9afcf10ef5d3916827896ce081#p135440

  "FlightXML’s GetHistoricalTrack intentionally returns only actual
  data points, which means that approximated positions that have been
  synthesized based on extrapolations of movement are not
  included. This design choice was originally made in order to ensure
  that people making use of our FlightXML data for research or
  analysis purposes are not mislead by our synthetically generated
  data"

If indeed 'GetHistoricalTrack' only returns what's in the database,
then a conscious human decision was likely taken to add the (faux)
17:50:50 entry into the database _itself_; which brings into question
the overall reliability of the rest of the Flightaware data.

The basis of the 17:50:50 record was queried in ID [#26771] which
produced no satisfactory answers.  I'm grateful to Jeff Lawson who
was able to confirm the suspected manual adding of the 17:50:50 point:

  "I've researched this and that single position was actually manually
  inserted later based upon other established data sources in order to
  make the map seem more consistent with user expectations, however
  the rest of the positions in that flight are unaltered."

In the end Flightaware have been extremely helpful and provided a
database dump outside of the normal mechanisms: 'fa-database-dump.csv'


Planefinder
===========

  http://planefinder.net/flight/MAS370/time/2014-03-07T16:50:00%20UTC
  16:46:59--17:00:41
  14 records: [null] (14)

Planefinder proactively put out a blog post and CSV file, containing
nineteen records rounded to the nearest minute, the last five of which
are extrapolated predictions.

  http://planefinder.net/about/news/malaysian-boeing-777-missing-presumed-crashed/
  http://cdn-misc.pinkfroot.com/MH370.csv

Querying the following URL at timestamps corresponding to sixty-second periods allows
the extraction of up to one true record per minute:

  http://planefinder.net/endpoints/playback/update.php?utc=1394210760&loc=1394210760

These records also contains the +5 minutes predictions which
correspond to the extraneous extrapolated lines in the published CSV
file.  The altitude granularity is finer and has not suffered from
rounding.


Lib Home Radar
==============

  ..~17:02:30

Libhomeradar recorded last contact at 4.70732,102.52782 (unknown time).
It does not not appear possible to provide a permalink, as the page
number is counted back in time from the present, and new flights keep
on being added.  A binary search can be made using the "pg=NNNN"
parameter to locate 2014-03-07:

  http://www.libhomeradar.org/databasequery/details.php?sid=6673947949&pg=1800

The flyover of 9M-MRO in the opposite direction earlier in the same
day in the opposite direction was last recorded at 3.54640,102.70950
which is on the longitude; as such libhomeradar are probably likely to
only have a single feeder in the region.


Radarbox24
==========

  http://www.radarbox24.com/2014-03-07/16/53/MAS370
  16:52:04--16:58:44 (17:12:08?)
  6 records

Radarbox24 were proactive in putting up a post and screenshot of the
tracked flight, showing an apparent track up until ~17:12:08:

  https://plus.google.com/+Radarbox24/posts/Yigduhqzkep
  https://lh3.googleusercontent.com/-dO6doa68uqg/Uxt_aTSnoNI/AAAAAAAAACo/rWmeG5hXtBQ/s2560/MAS370-2.png

The return from the 'playback_history' used by the display in the
webpage shows the MAS370 flight as 'starting' at 16:52:04, which
is likely to be the time of first reception, and this aligns
to the data points returned.  Interestingly Radarbox24 appears to
retain heading from the dlat/dlon, but not speed over ground.


Summary
=======

Aside from the fake FA entry at 17:50:50, the datasets have reasonably
separate sourcing and can be correlated.