Beijing Capital (PEK) Observed Traffic '−92%' — a One-Airport Step That Never Recovered and Our Data Could Not Confirm, October 2026
- Our count at Beijing Capital fell from about 325 movements a day to 26 on 1 October, starting with a step at 13:00 Beijing time on 30 September — and stayed at 15–26 a day for every complete day through 11 October. This page closed on 13 October with no recovery to report.
- Everything around it read ordinary in our data the whole time: Tianjin, the wider Beijing region, our worldwide total, and the weather at Capital went from clear to hazy and back without the count moving. That pattern usually means our receivers lost the airport, not that the airport lost its flights.
- We never got an independent count to prove it either way: the second network we use for that sees nothing within 35 km of Capital, and our feed does not cover Beijing Daxing at all. Our notice set for the Beijing FIR holds no closure, but it was last refreshed on 4 October. Nothing on this page says anything about the operating condition of Capital — for that, check your airline and the airport.
Advisory timeline — how active is this episode
Our feed sees only part of Beijing Capital's traffic on an ordinary day, and its count there is noisy: between 148 and 402 observed movements a day across 2–29 September, a 325/day 30-day median, with four single-day dips (227 on 12 Sep, 199 on 15 Sep, 223 on 16 Sep, 148 on 19 Sep) that each recovered the next day while our worldwide table read normal. 30 September is different in shape: 132, of which 103 fall between 00:00 and 04:59 UTC and 29 in the nineteen hours after. 1 October reads 26 — the day that tripped the anomaly flag. Then nothing moves: 23, 15, 26, 20, 21, 23, 17, 15 across 2–9 October, all settled; 21 on 10 October is a recount from the underlying flight records on a day our whole feed ran short (the stored row holds 4, the worldwide table 28,657 against its usual 226,013–265,788) and is marked as a partial day; 22 on 11 October is provisional and can still settle a little. 12 October had not finished arriving at close-out and is not plotted. The 30 September bar is the recount we plotted on 3 October; the stored row, cut short by a partial refresh on our side at the time, has since settled to the same 132 — while 1 October settled at exactly the 26 we first published. The daily series over the window:
Fig. 1 — Observed movements per day at Beijing Capital International (PEK), 2 Sep – 11 Oct 2026 · * partial day · Data as of 13 Oct 2026, 22:05 UTC · closed · PEK series complete through 9 Oct; 10 Oct a feed-outage day plotted as a recount; 11 Oct provisional; 12 Oct partial, not plotted; NOTAM set refreshed to 4 Oct · Source: FlySafe monitoring (crowdsourced ADS-B receiver network).
Airports in the exposure zone
| Airport | Size | Observed signal | FIR |
|---|---|---|---|
| Beijing Capital International Airport (PEK) | large | — · 1 Oct: 26 observed movements (4 dep / 22 arr) vs a 325 baseline; 2–11 Oct: 15–26 a day, 229 movements in eleven days (33 dep / 196 arr), average 21; every hour 0–4 on 9–12 Oct; weather clear 1–7 Oct, hazy 8–12 Oct, clear 13 Oct; no closure notice in our set (refreshed to 4 Oct) | ZBPE |
| Tianjin Binhai International Airport (TSN) | large | 124 km · Same FIR, same feed: 503 on 30 Sep and 511 on 1 Oct after settlement, then 360–500 a day across 2–9 Oct and 323 on 11 Oct (provisional) — ordinary; low-altitude raw sample over Tianjin 18–32 distinct aircraft per six-hour block on 9–13 Oct, unchanged | ZBPE |
| Beijing Daxing International Airport (PKX) | large | 66 km · Not covered: our feed holds no movement counts for Daxing on any day, and the independent network has logged 2 there in total — we cannot say anything about traffic at Daxing | ZBPE |
| Guangzhou Baiyun International Airport (CAN) | large | — · 412 on 30 Sep and 343 on 1 Oct, then 477–555 a day across 2–9 Oct — the soft 1 Oct reading did not persist | ZGZU |
| Shenzhen Bao'an International Airport (SZX) | large | — · 888 on 30 Sep and 952 on 1 Oct, then 976–1,124 a day across 2–9 Oct — ordinary | ZGZU |
| Hong Kong International Airport (HKG) | large | — · 883 on 30 Sep and 1,028 on 1 Oct, then 843–987 a day across 2–9 Oct — ordinary | VHHK |
Distance alone does not determine impact — wind direction and the flight levels the ash reaches decide which airports and routes are affected on a given day. FIR = Flight Information Region.
What our monitoring showed: a step in one hour, at one airport — and twelve days of flat line after it
FlySafe tracks observed departures and arrivals at 493 airports from crowdsourced ADS-B receiver data. Observed counts are a floor, not a total — receiver coverage varies — but an airport measured against its own baseline is a reliable shape. Beijing Capital International's last 30 days:
Fig. 2 — Observed daily movements at PEK, 2 Sept – 11 Oct (latest complete UTC day). Data as of 13 Oct 2026, 22:05 UTC · closed · PEK series complete through 9 Oct; 10 Oct a feed-outage day plotted as a recount; 11 Oct provisional; 12 Oct partial, not plotted; NOTAM set refreshed to 4 Oct · Source: FlySafe monitoring (ADS-B receiver network).
- 30-day baseline in our own feed: 325 observed movements/day (median of 2–29 Sep), range 148–402. Same-weekday check: the four previous Thursdays read 268–347, median 320.
- 1 Oct: 26 — 4 departures, 22 arrivals — 92% below both baselines. The detection layer fired at −89% against a 7-day average of 241.
- After the step, nothing came back. 1–11 October: 26, 23, 15, 26, 20, 21, 23, 17, 15, 21, 22 — 229 movements in eleven days, 33 departures and 196 arrivals, an average of 21 a day, 94% below the 325 median. Every hour of 9–12 October reads 0–4 movements. The 10 October figure is a recount on a day our whole feed ran short; 11 October is provisional.
- The fall was a step, not a slope. On 30 September the hours 00:00–04:59 UTC read 19, 19, 23, 28, 14 — in line with 30, 23, 19, 26, 21 on the 28th. From 05:00 UTC (13:00 Beijing time) the hourly count dropped to 5, 2, 2, 1, 1, 1, 1, 1 across 05:00–12:59, against 133 and 131 movements in the same eight hours on the 28th and 29th.
- Settlement did not rescue the number. Our table refines each day for several nights after it closes. Tianjin's 30 September and 1 October rose on settlement from the 402 and 453 we first published to 503 and 511; Capital's 30 September settled to the 132 we had already recounted, and its 1 October stayed at exactly 26.
- Why the flag went quiet. By 11 October Capital's own seven-day average had adapted to 21, so the 22 of that day reads +5% against it and the detection layer no longer flags the airport. Against its 30-day same-weekday median of 389 the same day is −94%. The silence of the flag is a property of a rolling average, not a recovery.
- Earlier dips were one-day and recovered. 12, 15, 16 and 19 September read 148–227 at Capital while our worldwide table read normal — a measure of how thin our sample at this airport is. None of them lasted into a second day; this one lasted the rest of the series.
Our counts are a floor, not a total: they are what our receivers saw and what our aggregation has finished writing, and at Beijing Capital that floor sat well below the airport's real traffic even on an ordinary day. A thinner sample prints a smaller number even when nothing at the airport changed. Live view: PEK airport brief.
What it means for flights
A crowdsourced count can fall without any traffic falling. Observed movements are reconstructed from ADS-B receivers run by volunteers, then aggregated on our side. When the receivers that see an airport's low-altitude arrival and departure legs go offline — power, internet, a failed antenna, a moved feeder — flights keep operating, but our reconstruction can no longer tie them to the runway. The number drops; the runway does not. At Beijing Capital our sample is thin even on a normal day, which is why its series already carried four one-day dips in September. We publish observed counts as a floor, not a total, and a −92% reading opens an investigation rather than a headline.
The shape: one clock hour, then flat. Operational disruptions — weather, a runway closure, a restriction — tend to build through a schedule, leave knock-on delays, and recover in stages. Our count at Capital did something else: it read normally from midnight UTC on 30 September, stepped from 14 movements in the 04:00 hour to 5 in the 05:00 hour, and sat at 0–5 an hour for the rest of the series — every hour of 9–12 October reads 0–4, day and night. That is the shape of a source switching off. It is suggestive, not proof — a real stoppage at a fixed time would also start abruptly — which is why the checks below matter.
The geography test: the hole is Capital's alone. If something were wrong with our collection across northern China, the neighbours would fall with Capital. They did not. Tianjin Binhai, in the same FIR and 124 km away, read 402 on 30 September and 453 on 1 October when this page was first published, inside its 386–457 range for the previous six days; both later settled higher, to 503 and 511. Further afield, Shenzhen read 952, Hong Kong 1,028, Seoul Incheon 1,207 and Tokyo Haneda 1,361 on 1 October, each in its usual range; Guangzhou ran soft at 343. Our worldwide table read 230,642 movements on 30 September and 245,335 on 1 October as first published, inside its 222,287–264,575 range for 22–29 September. A collection problem in our pipeline would not pick out one airport this cleanly; a receiver problem near that airport would.
The raw positions: the low-altitude sample around Capital halved; Tianjin's did not. Below the movement count sits our raw position sample, taken in regular snapshots. Counting distinct aircraft below 10,000 ft over the northern Beijing terminal area around Capital, 06:00–17:59 UTC: 69 and 71 in the two six-hour blocks on 29 September, then 33 and 29 on 30 September and 41 and 34 on 1 October; the overnight 00:00–05:59 block, 61–74 from 27 to 30 September, fell to 31 on 1 October and 23 on 2 October. Over Tianjin the same blocks read 72 and 61, 75 and 62, 71 and 50 — unchanged. Over the wider Beijing region at all altitudes the sample held at 943–1,049 in each six-hour block between 00:00 and 11:59 UTC on all three days. So aircraft were still being seen in Beijing's airspace, just far fewer of them close to the ground near Capital — enough to break our runway reconstruction, which needs exactly those low legs. This narrows the cause to the area around one airport. It does not, on its own, tell receiver loss apart from a real reduction in flights there.
What we could not check. Our strongest single test for a coverage artifact is a second, independent ADS-B network built from different receivers: if it keeps counting normally while ours collapses, the gap is ours. That network has no coverage at Beijing Capital — it recorded between 0 and 3 movements a day there through September — so the test was unavailable from the start. Our feed also holds no counts at all for Beijing Daxing, the city's second hub 66 km to the south, so any change in how traffic is split between Beijing's two airports would be invisible to us. 1 October was the first day of China's National Day holiday; the step in our count came the afternoon before, at a single hour, which is not how a holiday schedule looks — but we hold no schedule data to rule anything out. These are the limits of what this page can say, stated before any conclusion.
What the airspace record and the weather said — and what they did not. Our active NOTAM set for the Beijing FIR (ZBPE) held 17 notices overlapping 28 September – 3 October: navaid maintenance at two VOR stations, works areas, approach-aid limitations and missed-approach procedure changes at Capital, and one parking stand closed for works since June. None closes the aerodrome, a runway or airspace. Our closure-event layer did open a low-confidence record for ZBPE covering 29 September – 1 October; its inputs were that parking-stand notice, a news item about airspace in western China outside the Beijing FIR, and FIR-level traffic readings from the same feed this page examines — none of which describes a closure at Capital, so we do not treat it as evidence of one. Capital's weather office kept issuing routine reports every half hour (49 on 30 September, 49 on 1 October, 48 on 2 October); conditions were CAVOK at the hour of the step and every one of the 146 reports in those three days was in visual conditions. These are FIR-level and aerodrome-weather observations; they do not describe operations at the airport, and we publish no operational verdict on Beijing Capital — that belongs to the airport, the airlines and the Chinese authorities.
What we said would decide this — and what moved. On 3 October we named three things that would close this page. Recovery in our own count: it did not come. Capital's hourly movements never returned to their usual 10–30 in the 00:00–12:59 UTC hours; every hour of 9–12 October reads 0–4. Settlement: it happened, and it changed nothing in Capital's favour. The cut-short stored row for 30 September rose to the 132 we had already recounted; 1 October stayed at 26 to the movement, while Tianjin's provisional 402 and 453 for the same two days settled up to 503 and 511. The airspace record: no closure or restriction notice for Capital entered our set — with one caveat we owe the reader: our notice collection has not refreshed since 17:03 UTC on 4 October, for any FIR, so the record on this page answers for 28 September – 4 October and is silent on the days after. The closure-event layer's low-confidence record for ZBPE ran to 3 October, and two further one-day records at even lower confidence (0.18, cause unattributed) opened and closed on 6 and 11 October; none describes the aerodrome, and nothing has been open since the end of 11 October. With no recovery to wait for and the one decisive test still unavailable, keeping the page open would be waiting for a number that has shown no sign of moving in twelve days. So we close it as what it is: a documented steady state, cause unconfirmed.
What the data showed — close-out, 13 October. The number. From 1 to 11 October our feed recorded 229 observed movements at Beijing Capital — 33 departures and 196 arrivals — an average of 21 a day against a 325/day September median, −94%, with no complete day above 26 and none below 15. The 10 October value (21) is a recount on a day our whole feed ran short; 11 October (22) is provisional. The neighbourhood. Tianjin read 360–500 a day across 2–9 October; Shenzhen 976–1,124; Hong Kong 843–987; Seoul Incheon 1,189–1,294; Tokyo Haneda 1,340–1,404; Guangzhou 477–555, its soft 1 October reading gone the next day. Our worldwide table ran 226,013–265,788 on its complete days, with one feed-outage day on 10 October (28,657) that we mark as partial everywhere. The raw positions at close-out. Within a box of about 35 km around Capital, below 10,000 ft, we counted 48, 48 and 46 distinct aircraft in the three daytime six-hour blocks of 29 September; on 11–13 October the same blocks read 1–9. In a tighter box of about 17 km the contrast is starker: 31–38 per block at Capital on 28–29 September, a single aircraft in a single block across 9–13 October — while Tianjin, in a box of the same size, read 18–32 per block on the same October days, and the wider Beijing region at all altitudes still carried 558–655 distinct aircraft per six-hour block. The low-altitude sample near Capital did not halve this time; by the end of the series it was close to zero, with everything around it intact. The weather. Capital's reports were CAVOK on all 47–49 half-hourly observations a day from 1 to 7 October, hazy from 8 October (37 of 48 reports in marginal conditions, 2 in instrument conditions), down to 33 of 48 in instrument conditions on 12 October, and CAVOK again on all 43 reports to 21:00 UTC on 13 October. The count read 15–26 on the clear days and 15–26 on the hazy ones. A weather-driven disruption shows up in a count; a count that cannot see the runway shows nothing either way. The independent network. It logged 3 departures at Capital on 28 September and 5 on 6 October and nothing else since 25 September; in the same 35 km box it recorded no aircraft at any altitude across 6–13 October while its feed ran normally elsewhere, latest position at 21:43 UTC on 13 October. That is a network with no receivers in reach of Capital, not a network counting zero flights — and it means the test that would settle this page never became available. What this adds up to. Every check we could run came back the way a coverage loss at one airport comes back, and none came back the way a disruption does. We still do not call it: a coverage reading is an inference until a second source confirms it or the count returns the way it left. What would reopen this page. Our count at Capital returning to its 150–400 band in a single step — which would confirm the coverage reading in hindsight, and we would say so — or a notice closing the aerodrome or a runway at Capital appearing in our set for the days this page covers. Until then, the record stands as published, and we publish no verdict on operations at Beijing Capital.
Passenger questions, answered from the data
Our data cannot confirm that, and at close-out on 13 October it still could not. Our own observed count at Capital fell from a 325/day median to 26 on 1 October after a step at 13:00 Beijing time on 30 September and stayed at 15–26 a day through 11 October, but nearby Tianjin, the wider Beijing region and our worldwide total read ordinary throughout, the weather changed from clear to hazy without moving the count, and no notice in our NOTAM set for the Beijing FIR (last refreshed 4 October) closes the aerodrome or a runway. That pattern usually means lost receiver coverage rather than lost flights — but the independent network we would normally use to confirm it does not cover PEK. For the status of a specific flight, check your airline and the airport.
Because that is what our receivers and our aggregation produced, and we publish what we see. Our detection layer flags any tracked airport that falls 20% or more below its own rolling average; Capital read 26 movements against a 7-day average of 241 on 1 October, so the flag fired at −89%, and against the 30-day median of 325 the day is −92%. The investigation behind the flag — hourly shape, neighbouring airports, raw positions, the airspace record and weather — decides whether a drop is operational or a coverage artifact. Here it pointed to coverage but never proved it, and this page says exactly that.
No. Every complete day from 1 to 11 October read 15–26 observed movements — 229 in eleven days, an average of 21, 94% below the 325/day September median — and every hour of 9–12 October read 0–4. The detection flag went quiet because Capital's own seven-day average adapted to the new level (21), not because the count came back; against its 30-day same-weekday median of 389 the 11 October reading is still −94%. We closed the page on 13 October as a documented steady state with the cause unconfirmed; a return of the count to its 150–400 band, or a closure notice for Capital in our set, would reopen it.
We cannot say. Our feed holds no movement counts for Daxing on any day, and the independent network we cross-check against has logged 2 movements there in total, so there is nothing in our data to compare. What we can say is that Tianjin Binhai, the other large airport in the same FIR, read ordinary throughout: 503 and 511 on 30 September and 1 October after settlement, and 360–500 a day across 2–9 October. For Daxing, check your airline and the airport directly.
For media — the full kit
Everything on this page — charts, CSV, quotes — is published under CC BY 4.0: free to reuse with attribution to FlySafe.zone. Charts come in dark (social/web) and light (print/light layouts) versions; attribution and license are embedded in every file.
“Observed traffic at Beijing Capital International fell to 26 movements on 1 October 2026 in FlySafe's own crowdsourced ADS-B feed — 92% below its 325-a-day median — after a single-hour step at 13:00 Beijing time on 30 September, and stayed at 15–26 a day on every complete day through 11 October, an average of 21, while Tianjin Binhai and the wider Beijing region read ordinary, the weather at Capital went from clear to hazy without moving the count, and no published notice in FlySafe's set closed the aerodrome; the pattern points to a loss of receiver coverage rather than a disruption to flights, but no independent count at the airport ever became available to confirm it, and the page was closed on 13 October with the cause unconfirmed, according to flight-monitoring service FlySafe.zone.”
Boilerplate — about the source (for your article footer)
FlySafe.zone monitors airspace risk and observed airport activity worldwide: 334 flight-information regions scored daily, GPS-interference signals, volcanic-ash exposure and daily observed movements at 493 airports from the crowdsourced ADS-B receiver network. Methodology: flysafe.zone/methodology.
Need a different cut of the data, a comment, or numbers for another airport? Leave an email — we answer fast during active events.

