Questions and answers
What this site does and does not say
The questions a project like this invites, answered before they are asked. If yours is not here, write and it will be — the list is meant to grow.
How much does aviation actually matter?
About 2.5% of the world's CO₂ — and that figure is used to end the argument as often as to start it. It is small beside electricity, industry or heating, and it has grown roughly eightfold as a share since 1940 while almost every other sector shrank in relative terms. Counting contrails and nitrogen oxides, which tonnes of CO₂ do not capture, aviation accounts for about 3.5% of human-caused radiative forcing. The two percentages are of different quantities and must not be added or swapped. The context page sets out the series behind all of this, with the sources.
Are you saying airlines and airports are wasting fuel?
No, and the distinction is the whole point of this site. We measure the distance between what a flight actually burnt and what the same aircraft would have burnt flying the direct route at the most efficient profile, in the same real wind. That ideal is a theoretical limit: separation between aircraft, route structure, closed airspace and arrival queues put it out of reach of every real flight. Published estimates of what is genuinely recoverable are far smaller — EUROCONTROL puts about 39 kg of fuel per flight on continuous climb and descent procedures. Of the 5.1 points of vertical gap the median flight carries — the headline 4.6 is the fleet aggregate, which weighs long flights more heavily — 2.2 remain even for a flight going direct through an empty night sky. We read that as the baseline being unreachable rather than anybody's inefficiency — a reading, not a second measurement.
That is not the only figure on this site. Alongside the distance from the theoretical optimum, the front page reports what closing the spread between comparable flights would be worth: 1.1 Mt of CO₂ a year if the flights above the median flew like that median, and 0.5 Mt on the most cautious assumption. EUROCONTROL's own estimate for continuous climb and descent procedures is 1.1 Mt. Those are the figures to set beside published estimates of avoidable emissions; the 12.1% is not one of them.
What does an airport's number actually mean?
It describes the flights that touch that airport, not the conduct of the airport. Each flight is counted at both ends and its gap is measured over the whole flight, so the columns on dep. and on arr. split the same figure by the role the airport played. The phase split goes further: for arrivals, a median of 70% of an airport's deviation was produced within 40 NM of the airport itself. That is a location, not a cause. It says where the fuel was burnt; it does not say whether the profile was chosen by the operator or imposed by the traffic, and nothing here distinguishes the two.
Your figures divide to 11%, not 12%. Which is right?
Both, and the difference is the denominator. Dividing 2.50 Mt of gap by the 23.4 Mt actually emitted gives 11% — the gap as a share of what was burnt. The headline 12.1% is the gap as a share of what the ideal flight would have burnt, which is the smaller number, so the percentage is larger. Both are ratios of sums, not averages of per-flight percentages. The remaining fraction of a point between them is the type calibration, and the honest way to say it is that the two figures do not share a basis: the tonnages above are calibrated against the ICAO reference, the percentages are computed on the uncalibrated fuel. It is not that the factor cancels — inside a ratio of sums it does not, because it varies by aircraft type and the types have different gaps. On the calibrated basis the headline would be 12.0% instead of 12.1%, a difference of 0.14 points. Which basis is the right one for the headline is an open question, and it is recorded as one rather than settled quietly. Neither figure is wrong; they answer different questions, and this site uses the second because every comparison here — route against route, airport against airport — is made against the ideal, not against the actual. Per flight that gap is about 431 kg of fuel, of which 163 kg is the vertical component: the part continuous climb and descent procedures address.
Can I trust the ranking order?
Only its tails. About half the routes sit within a few points of the norm, inside the uncertainty of the method, and their ordering carries no information. At the head of the airport table ten positions can be separated by as little as 1.7 points within a single month. What is stable is the distance from the norm, not the position: read "well above comparable flights", never "third worst in Europe".
Is this peer reviewed?
No. It is an independent open-data project, not an institutional or academic publication. What it offers instead is verifiability: the source data is public, the method is documented in full, the code is open, and every figure can be recomputed from scratch. In July 2026, four organisations were given advance notice of figures that a later correction changed substantially; those figures are not in this release. Any reply from anyone named here is published in full and unconditionally.
Who pays for this?
Nobody. There is no funder, client, sponsor or advertising, and no organisation has had sight of the figures before publication beyond material provided on request, on terms open to anyone named here. The rules that keep it that way — including what happens if that ever changes — are written down under independence.
What does it leave out?
CO₂ is not the whole climate effect of flying. Contrails and nitrogen oxides contribute a large share of aviation's total warming effect — by published assessments, the majority of it — and these figures contain none of them. It also excludes ground operations, and ADS-B coverage does not include oceanic sectors. Read the figures here as what they are: fuel burnt in the air over Europe, turned into CO₂.
Why only 2026, and why no comparison with last year?
Because 197 days of 2026 is what has been processed so far, and a year-on-year comparison built on a single period would be an invitation to read weather as a trend. Releases come twice a year from now on, each covering twelve months, so the first honest comparison becomes possible once two of those windows exist.
Does this track individual flights, aircraft or people?
No. Nothing is published below an aggregate of at least 10 flights, and rankings need at least 100. The pipeline keeps no registration and no callsign — only the aircraft type and the airports — and no figure on this site describes an identifiable flight, operator crew or passenger.
Is this an emissions inventory?
No, and the difference is the point. An inventory answers
how much, and where. The most complete one built from the same raw material
as this site — GAIA,
published in Atmospheric Chemistry and Physics in 2024 —
reconstructs 103.7 million ADS-B trajectories worldwide and
gives CO₂, nitrogen oxides and particulate on a grid, for atmospheric
research. It is the reference for what aviation emits.
This site answers a narrower question: how far from a reference. It does
not try to count Europe's aviation emissions — it compares each flight with an
ideal version of itself, and reports the difference by route and by airport. An
inventory tells you what was emitted; this tells you how much comparable flights
differ from one another. The two are complementary, and where they overlap the
inventory is the better source.
How accurate is the fuel model?
Fuel burn comes from OpenAP, an open performance model from TU Delft, anchored per aircraft type to the ICAO Carbon Emissions Calculator methodology. The check that matters is on the types the model was not corrected for: the most common airliners in the sample land within about 5% of a reference the model never saw. Types that needed correction, and why, are listed in the methodology — the correction compensates a documented limitation, not an unexplained discrepancy.
A route here cannot fly its direct path. Is that counted as inefficiency?
It is measured, and it is flagged. 208 ranked routes have a great circle crossing closed or systematically avoided airspace, and they carry a ⚑ wherever they appear. That detour is not recoverable while the closures hold. Note also that an overflight ban binds European carriers and not third-country ones, so a figure for such a route averages operators that must divert with operators that need not.
Can I reuse these figures?
Yes, and the terms differ by what you reuse. The text and the charts are additionally offered under CC BY 4.0 — this project's own expression. The figures as data derive from adsb.lol trajectories under ODbL, so a dataset built from them is a derivative database and carries the same share-alike obligation. In both cases attribution is required. The detail is under licence and reuse.
I am named here and I disagree with the figure. What happens?
Write to hello@co2gap.org. A reply is published on this site in full, unconditionally and next to the figure it concerns — not summarised, not answered selectively. If the disagreement is about the method rather than the reading, the pipeline is open and a reproducible counter-example is the fastest way to change what is published.
Where this method is weak
Not a disclaimer. These are the places where I already know the method is soft, written down so that someone who works in this field can tell me how wrong I am. Criticism of the method is what is being asked for — not endorsement. Whatever comes back is published here, including the parts that do not suit the conclusions.
1. The cruise baseline is not as optimal as it claims
Measured over the cruise alone, our gap comes out slightly negative: the real aircraft burns marginally less than the profile we call optimal. That is not a result about aviation, it is a defect in our reference: the optimal cruise altitude we compute is not the fuel-optimal one. Published work on cruise efficiency finds a clear positive gap on the same perimeter, so the disagreement is ours to explain.
We now know where it comes from. Compared with the altitude aircraft actually reach, our baseline cruises about 1,000 ft below them on the longest sectors — and lower, up there, is not better. The reference is burning more than it should exactly where cruise dominates, which is enough to swallow the gap and turn it negative.
It affects the level of the vertical term, not the attribution: cruise contributes very little to what separates one airport from another. Correcting it would most likely make the headline figure larger, since a genuinely optimal reference burns less.
2. CO₂ is not the whole climate effect
Contrails and nitrogen oxides account for a large share of aviation's warming effect — by published assessments the majority of it — and these figures contain neither. A route flown at a level that avoids contrail formation could be worse by this site's figures and better for the climate.
Both have now been measured, for European traffic across 201 days of data, to 24 July 2026 — a slightly wider window than the CO₂ figures above, whose 197 days end on 20 July because that is where the wind data ends. The method, the figures and what they cannot support will be published with the January 2027 release, when the window covers a full calendar year: contrail forcing per flight differs by a factor of about four and a half between January and July, so seven months is not a year, and publishing a figure now would mean publishing one that then moves. The January release will also be explicit about which parts of it can be broken down and which cannot — some of what a per-airport contrail ranking appears to show follows from modelling choices rather than from the air above the airport.
Anyone who works on non-CO₂ effects and can say how badly that changes the reading of these rankings would be doing this project a service.
3. Fuel modelling for the types the model does not cover
The performance model carries calibrated fuel curves for a limited set of aircraft types; the rest fall back on a generic model rescaled from a static take-off figure. Our per-type correction compensates that, and the check is that the types needing no correction land within about 5% of an independent reference. But a correction is still a correction.
The diagnosis, including which types are affected and why, is written up in the methodology and has been put to the model's authors publicly.
4. The uncertainty of the figures is not quantified
Aircraft mass is estimated, not known, and it is the largest physical uncertainty in the model. We say so, but we do not say how much it moves the result — there is no ± on any number here.
And the honest version of that is harder than it sounds. This metric is a difference between two model runs, the real flight and the ideal one. An error that is systematic cancels out in the subtraction; one that varies with altitude, weight or phase of flight does not, and lands squarely on the gap. Showing that the model reproduces published fuel burn to within a few per cent says nothing about either case. Until that sensitivity is measured, only the extremes of these rankings should be read as meaning anything, which is why that caution appears wherever a ranking does.
And one we found ourselves
The quality gate is geometric: it checks that a flight covered the distance it should have, and never looks at the fuel. A handful of flights in the published period therefore carry burn figures that are physically impossible, the residue of degenerate trajectories: the worst of them is an A320 whose track carries no airspeed at all, and which therefore integrates to a burn 96% below the ideal. They are far too few to move any published statistic — removing the 37 worst shifts the headline by 0.002 of a point, and leaves the airport and route rankings identical — but they are there, and the defect pushes the gap down, never up. A gate on fuel plausibility is due in the next release.
How to send something useful
Corrections are welcome and are published. These arrive in a form that can actually be acted on:
For a figure you think is wrong: the airport or route, the period, which number you are disputing, and what you believe it should be. If you hold traffic or fuel data of your own, saying how far ours is from yours is more useful than saying that it is wrong.
For the method: the step you disagree with, and — if you can — the case that breaks it. The pipeline is open, so a reproducible counter-example changes what is published faster than any argument.
If you are named here and want to reply: say so, and the reply is published in full, next to the figure it concerns, without editing.
Still unanswered?
hello@co2gap.org. Corrections and replies are published on this site, in full and unconditionally.