GPS interference over the southern North Sea, 9 November 2026
From 17:08Z to 17:27Z on 9 November 2026, about 80 aircraft over the Netherlands, the southern North Sea, eastern England and north-west Germany lost GPS at the same time.
The pattern fits a single radio transmitter at or near sea level over the North Sea, probably about 80 to 110 NM north of Texel. It does not fit a source over the Dutch mainland.
It is the largest GPS-interference event in this area in the past year. The only similar event, on 23 July 2026, came from a different place, off the English coast.
The event was first reported by gpswise.aero and discussed in a Bluesky thread by auonsson. gpswise reported 50 flights affected for 20 minutes over the Netherlands. This report checks that claim against ADS-B data and estimates where the interference came from.
Key findings
These are measurements from ADS-B data:
- degradation started within 6 seconds, at 17:08:35Z to 17:08:41Z, on aircraft spread across about 510 km, from 55.7N 0.2W to 51.9N 4.2E
- it ended at 17:26Z to 17:27Z, so it lasted about 19 minutes
- in the 17Z hour, 68 aircraft over 50.5N to 55.5N, 1E to 9E reported NACp below 6, against about 4 in the same hour on 6, 7, 8 and 10 November
- in ADS-B Exchange snapshots, 70 to 88 aircraft were affected at any one moment, out of 386 to 497 comparison aircraft: those above 8,000 ft within 47N to 58N, 4W to 14E whose GPS was healthy from 16:55Z to 17:07Z
- aircraft at cruise altitude were affected, while aircraft below FL150 were not
- at 17:09Z and 17:18Z, no aircraft beyond the radio horizon of the fitted source was affected (0 of 372 and 0 of 319); across all snapshots, one aircraft far to the south-east was briefly marked as affected at 1.7 times its horizon distance
- 39 affected aircraft were west of the Amsterdam FIR, broadly in UK-controlled airspace, while affected; some were over eastern England (an approximate classification from the Amsterdam FIR boundary)
- we found no GPS-interference notice for this area and time in HEXBIRD's NOTAM archive (from the FAA's US NOTAM System, which includes Dutch civil and military notices) or in Dutch, German and UK navigational warnings to shipping
- from 1 November 2025 to 10 November 2026, no other hour in this area came close: the next largest had 58 affected aircraft, and a typical hour has 3
- affected aircraft show no shared false position, so there is no sign of coherent spoofing
- GPSJAM's daily map stayed green because it averages over a whole day: in each hex, the aircraft affected during the burst were at most 7 of the 250 to 700 seen that day. Weak degradation was not the reason, as 92% of reports from affected aircraft had NACp 0
Key judgments
HEXBIRD assesses that it is almost certain that a radio-frequency transmitter caused the GPS loss, rather than space weather or a data-processing fault. Confidence is high. The simultaneous start and stop, the sharp line-of-sight boundary and agreement between independent data sources all point the same way.
HEXBIRD assesses that it is likely that the source was at or near sea level over the southern North Sea, roughly 80 to 110 NM north of Texel. Our best estimate is 54.65N 5.05E. The model's 95% region (54.55N to 54.75N, 4.95E to 5.15E) does not mean a 95% chance that the transmitter is there: it comes from a simple model that makes many assumptions and may be wrong. Confidence in that position is low. We made this estimate after a convoy was reported nearby, and it depends on how sharply the jamming stopped at the radio horizon: if it reached all the way to the horizon, the best fit moves about 18 NM north. The location comes from a propagation model, not a direct measurement.
The data cannot show who operated the transmitter or why. This report makes no judgment on that.
Timing and scale of the event
default.traces. These counts understate the event, because the trace archive records a field only when it changes.default.traces.Only aircraft at cruise altitude were affected
The share of affected aircraft depended mainly on altitude. Aircraft below FL150 were almost never affected, even close to the centre of the affected area. This is what happens when a transmitter near the surface is hidden from low aircraft by the curve of the Earth.
| Altitude | 0 to 75 NM | 75 to 150 NM | 150 to 225 NM | 225 to 300 NM |
|---|---|---|---|---|
| Below 5,000 ft | 0 of 10 | 1 of 16 | 0 of 15 | 1 of 37 |
| 5,000 to 15,000 ft | 0 of 6 | 0 of 12 | 0 of 15 | 0 of 38 |
| 15,000 to 25,000 ft | 1 of 4 | 1 of 11 | 0 of 26 | 0 of 37 |
| 25,000 ft and above | 21 of 30 | 27 of 73 | 5 of 98 | 6 of 139 |
default.traces.Two aircraft near the fitted source show the same pattern:
- a Dutch Navy NH90 helicopter,
NEPTN61(48104f), flew at about 900 ft near Den Helder, 88 to 105 NM from the fitted point and well beyond its radio horizon, and its GPS stayed healthy throughout - a Phenom 300,
OK-PHN(49d62d), at FL430 lost GPS when the event started, about 65 NM from the fitted point, and recovered at 17:28Z when it ended
Where the source probably was
We fitted a model of a transmitter at sea level to ADS-B Exchange snapshots taken about a minute apart from 17:09Z to 17:26Z. Inside the transmitter's reach, a fixed share of aircraft lose GPS; beyond it, almost none do. The reach is set near each aircraft's radio horizon, and one range of bearings may be jammed more weakly. Positions come from ADS-B only: an affected aircraft is placed on the straight line between its last good position before losing GPS and its first after regaining it.
The best fit is at 54.65N 5.05E. Inside the transmitter's reach, 86% of aircraft lost GPS, except to the north-east (bearings 0° to 90° from the source), where 18% did. Reach ended at 0.94 of the radio horizon. If it had ended exactly at the horizon, the best fit would move to 54.95N 5.1E, but the data fit that less well. At the best fit, 14 of the 15 aircraft that lost GPS while climbing, or regained it while descending, did so between 0.90 and 1.00 of their radio horizon. The 15th was in the weakly jammed north-east sector.
We simulated events from a known source, keeping each aircraft's tendency to stay jammed or healthy, and refitted them to set the region's size. The model's 95% region (shaded) covers about 54.55N to 54.75N, 4.95E to 5.15E, roughly 50 square NM. In 200 test simulations from the fitted model it contained the true source 97% of the time.
This is not a 95% chance that the transmitter is in the region. The region only measures how much random noise in the data could move the estimate if the model were exactly right, and the model is simple. It assumes a single transmitter at sea level, a smooth Earth with standard refraction, a sharp edge to the jamming, the same jammed share everywhere inside it except one weaker sector, and aircraft positions interpolated in straight lines. If any of these is wrong, the true position could lie well outside the region. For example, a transmitter whose jamming reached all the way to the radio horizon would put the best fit about 18 NM further north.
The convoy
On the morning of 9 November, the ship spotter vesselspotter reported three Russian-flagged cargo ships, MAIA-1, LADY D and LADY R, heading south through the North Sea towards the English Channel at about 10 knots. MarineTraffic shows MAIA-1 (138 m) and LADY R (123 m) last calling at Baltiysk, Russia, and lists both under EU and US sanctions, and LADY R under UK sanctions too. On 13 November auonsson, crediting vesselspotter, reported that the convoy, possibly with a warship escort, passed through the area during the event and attributed the jamming to it.
We made this estimate after that report and knew where the convoy was when we chose the model. The convoy passed 13 to 17 NM north-east of the best fit, outside the model's 95% region. If the jamming reached all the way to the radio horizon, the best fit would move about 18 NM north, closer to the convoy, so this estimate does not rule the convoy out. Sentinel-1 shows the three ships at 17:17Z (below). In ADS-B data from 11:00Z to midnight we saw no maritime patrol aircraft over the southern North Sea; the closest aircraft of any kind below 15,000 ft passed 28 NM from the convoy's track, an offshore helicopter at 14:28Z.
Sentinel-1 radar pass during the burst
Sentinel-1C passed over at 17:16:36Z to 17:17:26Z, during the burst, and imaged about 70% of the model's 95% region, including the best estimate, at about 17:16:58Z; a strip 1 to 4 km wide along the region's western edge falls in the image's no-data border. It shows no ship in the imaged part. The nearest detection is a 265 m vessel candidate (length includes smear) 4.4 NM east of the best fit. The detector finds almost all ships of 100 m or more, but fewer small ones. Three ships in a north-east to south-west column at 54.83N to 54.87N, 5.24E to 5.28E match the convoy's MarineTraffic positions: LADY D and MAIA-1, and probably LADY R. Their radar extents (170 to 203 m) are longer than the ships (123 and 138 m where known) because of smear and wakes. A fourth target, 118 m and unidentified, lies 0.4 km from LADY D; it could be an escort, but the radar cannot tell.
The past year in this area
We counted, for every hour from 1 November 2025 to 10 November 2026, aircraft above FL200 over 51N to 56N, 1E to 9E that reported NIC below 5 or NACp below 6. That is about 9,000 hours.
The 17Z hour on 9 November had 111 affected aircraft. A typical hour has 3. In 99.9% of hours there were 15 or fewer.
default.traces. Short bursts are diluted in hourly counts, so smaller events can be hidden.Only one other hour stands out, 19Z on 23 July 2026, with 58 affected aircraft. It looks like the same kind of event:
- it started sharply at 19:00Z and ended sharply at 19:18Z, so it also lasted about 19 minutes
- about 60 aircraft a minute were affected at its peak
- no NOTAM in the archive covers it
It came from a different place, though:
- the fit puts the 23 July source near 53.4N to 53.6N, 1.2E to 1.4E, about 40 NM off the Lincolnshire and Norfolk coast
- that is about 150 NM west-south-west of the 9 November fit, and the 9 November position fits the 23 July data much worse (log-likelihood 51 to 60 units lower)
- on 23 July, no aircraft beyond half its radio horizon was affected, so the footprint was limited by transmitter power rather than by the horizon
- because of this, the model cannot estimate the 23 July source height
The 23 July fit lies inside the area of UK NOTAM B1139/26, which warned of GNSS jamming for Exercise Storm Warrior off the English east coast. But that notice covered 09:00Z to 11:00Z on 21, 23, 28 and 30 July, from the surface to 20,000 ft. Those windows show no effect on aircraft above FL200. The 19:00Z event falls outside every notified window.
Hourly counts can hide a short burst, so we also scanned every minute of every day for sudden jumps in the number of affected aircraft above 8,000 ft. Only 3 bursts in the year reached 12 or more aircraft at once:
- 9 November 2026, 17:08Z to 17:26Z, peak 70 aircraft
- 23 July 2026, 19:01Z to 19:18Z, peak 37 aircraft
- 14 October 2026, 09:08Z to 09:10Z, peak 14 aircraft, centred near 54.4N 6.2E
We have not examined the 3-minute burst on 14 October. The scan also flags many small bursts around dawn, between 03:30Z and 05:30Z. These involve 5 to 10 aircraft while traffic is light, and they look like normal low-accuracy aircraft rather than interference.
Two other hours were slightly raised, 08Z on 22 February (29 aircraft) and 08Z to 09Z on 14 October (26 aircraft). Both rise and fall gradually over an hour or more. They do not look like the same kind of event.
Why the GPSJAM map shows almost nothing
GPSJAM's map for 9 November leaves the Netherlands green, with 8 yellow hexes over the sea. The main reason is that it averages over a whole day. For each H3 hex (about 45 km across) it counts aircraft once per UTC day, as bad if ADS-B Exchange marked them as having lost GPS. It then colours the hex by (bad − 1) ÷ all aircraft: yellow from 2%, red from 10%.
A 19-minute burst is a small share of a day's traffic. Busy hexes over the Netherlands see 250 to 700 aircraft a day, and the burst added at most 7 bad ones. The worst hex scored (7 − 1) ÷ 426 = 1.4%.
| Counting method, 66 hexes hit by the burst | Median share | Highest | Yellow / red |
|---|---|---|---|
| 16:55Z to 17:30Z only | 12.5% | 50% | 20 / 45 |
| Whole day, also counting MLAT positions | 2.0% | 6.9% | 32 / 0 |
| Whole day, as published | 0.7% | 3.6% | 8 / 0 |
2 smaller factors add to this. The minus one removes much of a signal that is only 4 to 7 aircraft strong. And GPSJAM skips MLAT positions, which were 78% of reports from affected aircraft, so most were counted bad in only 1 or 2 hexes.
The 8 yellow hexes are quiet offshore cells with 26 to 47 aircraft all day, where the burst's 2 bad aircraft were enough. gpswise also shows small daily numbers for its Dutch hexes (6 of 1,625 and 11 of 2,604 flights); it shows the event through hourly detail.
Other explanations considered
| Explanation | What it predicts | What we see |
|---|---|---|
| Space weather (ionospheric scintillation) | a gradual, patchy onset over a wide region, with no altitude dependence | a start within 6 seconds over about 510 km and a sharp boundary at the radio horizon. Rejected. |
| Data or processing fault | effects on one feed, one receiver network or one avionics type | more than 20 operators and many aircraft types, seen in ADS-B Exchange, airplanes.live (through GPSJAM) and gpswise. Rejected. |
| Source over the Dutch mainland | low aircraft near Schiphol and Lelystad affected | no low aircraft affected. The model fit is much worse. Rejected. |
| Airborne transmitter | low aircraft near the source affected, with a fitted height well above sea level | best fit at sea level, but height is poorly constrained. No candidate aircraft identified. Not ruled out. |
| Several transmitters | a footprint that is not one circle, or starts at different times | one start time and one footprint. Two nearby sources cannot be told apart from one. Not ruled out. |
| Spoofing rather than jamming | affected aircraft reporting a shared false position or a coherent shift | only small scatter and isolated glitches that jump back and forth. No sign of spoofing. |
Limitations
- the location is a model result, not a measurement, and it depends on the line-of-sight assumption
- the 95% region reflects noise in the data under the model, not the chance that the model is right; it is not a 95% probability that the transmitter is inside it
- the fit assumes a source at sea level; a source a few hundred feet up could not be told apart in our tests
- the region pools snapshots about a minute apart, so the same aircraft appear many times; the simulations allow for this, but not for a wrong model
- the position depends on where the transmitter's reach ended: each 0.02 of the radio horizon moves the best fit about 5 to 9 NM north or south
- jamming was weak to the north-east of the source, which a directional or partly blocked transmitter could explain; terrain cannot, as those paths cross open sea
- the trace archive records only changed fields, so counts based on NACp alone are undercounts (for example,
OK-PHNshows up only through its NIC) - in the snapshots fitted, 59 to 107 aircraft each time were left out because their data was stale or lacked quality fields, and 3 to 12 affected aircraft were left out because their position was unknown; missing affected aircraft could shift the edge of the footprint. Affected aircraft are placed by straight-line interpolation, which is wrong for aircraft that turned while jammed
- the NOTAM archive starts in June 2026 and may miss notices; we did not check Dutch Notices to Mariners or any UK warning issued and cancelled on the day
- we did not have ship-tracking data for the fitted area; the convoy's positions come from MarineTraffic screenshots published by auonsson
Next checks
These checks would most narrow down the source:
- ship-tracking (AIS) data for 54.4N to 55.0N, 4.8E to 5.5E from 17:00Z to 17:30Z, to find vessels or installations near the fitted area and confirm the convoy's track
- Dutch defence or coastguard information about activity at sea that evening (a navy NH90 was flying nearby)
- pilot reports or air traffic control records from the Dutch, German and UK air navigation service providers for the burst window
- whether gpswise's flight list matches the aircraft listed here
Method and sources
- ADS-B trace points from HEXBIRD
default.traces(ADS-B Exchange and airplanes.live archive), queried 12 and 13 November 2026 - ADS-B Exchange
aircraft.jsonsnapshots, about every 15 seconds, 16:55Z to 17:30Z, from a private archive. These provide the receiver-side GPS-lost marker (gpsOkBefore) - degraded means NACp below 6 or NIC below 5, counted only for aircraft that reported NACp 9 or higher at some point that day
- for the fit, the comparison group is aircraft seen at least 3 times from 16:55Z to 17:07Z with healthy GPS throughout
- the fit uses 17 snapshots from 17:08:52Z to 17:26:03Z; for each position on a 0.05° grid it searches reach from 0.86 to 1.06 of the radio horizon and a weakly jammed sector (start every 15°, width 60°, 90° or 120°); the region uses 300 calibration and 200 validation simulations, and 200 more with the source at 54.95N 5.1E
- positions are ADS-B only; MLAT positions are not used
- patrol-aircraft search: HEXBIRD
default.traces, aircraft below 15,000 ft within 60 NM of the convoy's projected track (10 kn on 215°) from 11:00Z to midnight, and patrol, surveillance and military types over 52N to 58.5N, 0E to 10.5E - horizon crossings are switches between lost and healthy GPS that held for at least 3 snapshots (about 45 seconds) either side, from 17:09:40Z to 17:25:40Z, where the aircraft was climbing or descending at the switch and the healthy side was at least 3,000 ft lower
- NOTAMs from HEXBIRD
default.notams, searched for GPS, GNSS, jamming, interference and exercise notices active from 17:00Z to 18:00Z - maritime warnings: UK Hydrographic Office NAVAREA I and coastal warnings, Admiralty Weekly Notices to Mariners 46 and 47, German BSH nautical warnings, and NAVTEX broadcasts for 7 to 10 November from the navtex.net archive
- gpswise video and screenshots from the Bluesky thread linked above
- analysis started and directed by John Wiseman; the data queries, model fitting and drafting were done with HEXBIRD, an AI-assisted aviation analysis tool, and reviewed by him