Ninety days over the Baltic, from publicly broadcast aircraft data.
A measurement, openly derived and open to challenge.
Loading…
Point at the map for cell detail.
Day by day, across the region
Transport-category aircraft, pooled across every well-observed cell, so
each point rests on hundreds of thousands of reports rather than the
handful any single cell contributes. The dashed
line is the background rate for that same population. Ticks below the axis mark days the
detector produced at least one candidate — candidates for review,
not findings, and 98% of them fall in control airspace where a
firing is a false positive by construction.
There is no per-cell time slider, and that is a data limit rather
than a missing feature. Across the backfill only 4.9% of
cell-weeks carry the thousand integrity reports a rate needs. At any
position of such a slider 95% of the map would be blank, and the rest would
move mostly on sampling noise. Animating that would look like information
and be nothing of the kind. Pooled region-wide the same days are measurable
many times over, so the time axis is regional.
Light aircraft are excluded, and the difference is the whole
point. Their receivers report poor integrity almost everywhere,
for reasons that have nothing to do with the sky.
At 54.74°N 9.39°E, one of the best-observed cells in the
region, 51.5% of all integrity reports are degraded — and 0.6% of
the transport-category ones. Fifteen Cessna 172s report 68.8% and
two gliders report 100%, while 264 Boeing 737s, 159 A320neos and 105
Boeing 787s in the same cell report between 0.0% and 0.5%. Pooled together
that cell looks alarming and is measuring avionics.
The filter is not a mute button. At 59.78°N 29.79°E, where
63 Airbus A350s report degraded integrity, the cell reads
99.8% before the restriction and 99.8% after. The artifact
disappears; the signal does not move. Interference does not politely skip
airliners.
Aircraft types are classified by the altitude they fly at, never by what
they report — a filter defined by the quantity it is meant to clean would
prove whatever it was built to prove. The cut is 14,000 ft at the 90th
percentile, which keeps regional turboprops such as the ATR 72 and Dash 8
and excludes piston singles.
These are still measurements, not findings.
What has been measured
Ninety days are loaded and continuous — 10 May to 7 August
2026, 141 million position reports and 35 million navigation integrity
reports from 12,734 aircraft, with no gaps. Across the region
6.02% of all integrity reports are degraded — but only
0.93% of transport-category ones. The lower figure is the
background a finding has to rise above, because it is measured over the same
population the map draws. The gap between the two is light aviation, and
mistaking it for interference is the single easiest error to make here.
The primary detector has been measured against control airspace: over a
31-day window across 106 cells where nothing was reported, it fires on
0.07% of the cell-hours it evaluates — about five firings
in a month. That is a tenfold improvement on the previous version, and it
came from a rule rather than a threshold: a firing now requires several
different transport-category aircraft types to show the
degradation, because light aircraft avionics report it almost everywhere.
How much it would catch is not merely unknown — it is
undefined. The one effect in this region large enough to serve as a
reference sits in airspace so thinly flown that the detector never evaluates
it at all. Of the 6,057 cell-hours near it in the control window, sixteen
carry enough integrity reports, seven also carry enough aircraft, and none
of those has a baseline to compare against. A detection rate needs a
denominator and there is none, in every configuration tested — including
with the coverage restriction removed entirely.
An earlier estimate of 81% came from a different configuration than the
false-positive figure beside it, and has been withdrawn rather than
restated. Both halves are now measured by one script through the detector's
own code path, so a configuration cannot produce one number without the
other. Knowing what a method misses matters as much as knowing when
it is wrong, and that half is open.
No findings are published here. Half of the bar has been
cleared — the control-airspace test — and half has not: no independently
documented, dated event has yet been found in airspace this method can
actually see. Until one is, what appears on this site is measurement, and it
is labelled as such.
Region
Baltic — 53.5°N to 66.0°N, 9.0°E to 30.0°E
Cell size
Hexagons roughly 20 km across, on 15-minute windows
Cadence
Daily batch processing of published archives. Nothing is real-time
Expansion
Further regions only after the method is validated in this one
About
What this is, how it works, and the rules it is built to.
Satellite navigation signals reach an aircraft receiver extremely weak —
roughly the power of a car headlight seen from twenty thousand kilometres
away. They are correspondingly easy to disrupt.
Episodes of degraded satellite navigation affecting commercial aviation over
parts of Europe have been reported publicly by airlines, aviation authorities
and the trade press. Individual accounts exist. What does not is a continuous,
independent, reproducible record of where and when aircraft
reported degraded navigation integrity — one built to survive being
challenged, and held by someone with no stake in the answer.
GNSS Observatory is being built to be that record.
How it works
Aircraft broadcast their position continuously over ADS-B, derived from
onboard GNSS receivers. Those broadcasts also carry navigation
integrity fields — the receiver's own assessment of how far its reported
position can be trusted.
When reception degrades, that self-assessment falls, and the aircraft
broadcasts the degradation to anyone listening. The effect publishes its own
evidence. We read those public broadcasts and look for populations of aircraft
whose navigation integrity falls together, in the same airspace, at the same
time — then record the result in a form that can be re-derived and audited
later.
What this is
A project, not a service. It is built in the open, one region
at a time, and it publishes what it can defend. There is no product, no
subscription and no customer. Where the method is uncertain the site says so,
and the uncertain parts are currently the interesting ones.
The public map is the whole of it today. Everything drawn there is derived
from archives anyone can download, using code whose parameters travel with
every number it produces, so a disagreement can be settled by rerunning it
rather than by argument.
Method and principles
These are design rules rather than aspirations. They exist because the failure
modes are worse than the missed detections — and because they are what allows
a finding to survive being challenged.
No conclusion is drawn from a single aircraft. One
malfunctioning receiver is indistinguishable from localised interference.
Any finding requires a population of independent aircraft, across more than
one aircraft type and operator.
Effects are observed; transmitters are not located.
Measuring the effect on aviation is data analysis. Locating a transmitter is
radio direction finding, which carries an entirely different legal and
regulatory posture. GNSS Observatory does not do it, and operates no radio
receiving equipment.
No attribution of cause or actor. The data can show that
navigation performance degraded over an area. It cannot show who or what
caused it, and no such claim will be made.
Every record carries the software version that produced it,
with its full parameter set, so any published statement can be reproduced,
re-derived or challenged afterwards.
Findings are superseded, not silently revised. When the
method improves, earlier findings are marked superseded rather than quietly
rewritten — the audit trail is part of the work. Where an aircraft is
excluded on request, its identifiers are removed regardless.
The method is measured against quiet airspace, not only against
known events. Any detector can be tuned until it finds the
incidents you already know about. What matters is how often it fires over
control airspace where nothing was reported, and that rate is measured and
published alongside the method rather than left unstated.
Human review before publication. Nothing reaches the public
automatically.
Data and licensing
Source
Publicly released daily archives of aircraft transponder broadcasts, from the
adsb.lol
community project
Licence
Used under the upstream project's published terms (ODbL 1.0 and CC0), with attribution
Coastline
Natural Earth, public domain — clipped and served by this site, not fetched from anyone
Collection
No radio equipment is operated. Only archives already published by others are processed
Get in touch
Corrections and challenges are the most useful thing anyone can send. If you
work in aviation, avionics or air traffic management and something here looks
wrong, please say so — especially about the map.
The data processed consists of transponder broadcasts made openly by
aircraft: position, altitude, velocity, and navigation integrity values,
together with the transponder's assigned address. It contains no
passenger or crew information, and none is derived.
Analysis is carried out at the level of airspace and populations of aircraft,
not individual flights, and individual aircraft are not identified in
the public map or in public findings.
A transponder address can, in some cases, relate to an identifiable person —
for example where an aircraft is privately owned. Where it does, processing
is carried out on the basis of legitimate interests under Article
6(1)(f) GDPR: the independent documentation of navigation
interference affecting aviation safety, using data already broadcast openly
and published by others.
You may object to that processing at any time. An owner or
operator may request exclusion, and the aircraft will be removed from
processing and from any published output. Requests go to the contact address
above. A full privacy notice will be published before any findings are.
This site sets no cookies, loads no third-party resources, and performs no tracking.
Important limitations
GNSS Observatory is not an aviation information service.
Nothing published here is certified, validated for operational use, or
suitable for navigation, flight planning, dispatch, or any safety-of-flight
decision. Authoritative information about GNSS availability is issued by the
relevant national aviation authorities and air navigation service providers,
through NOTAMs and equivalent channels. Always use those.
GNSS Observatory is independent and is not affiliated with, endorsed by, or
acting on behalf of any government, aviation authority, air navigation
service provider, or aircraft operator.