Large carbon dioxide emissions avoidance potential in improved commercial air transport efficiency

Efficiency model

The comparison of commercial passenger air transport efficiencies requires a comparable indicator. CO₂ intensity, i.e. Emissions per revenue passenger kilometer (g CO₂ per RPK) can be calculated for aircraft, airports, and city pairs. This indicator is directly proportional to fuel use per revenue passenger kilometer (kg fuel per RPK). It is influenced by many factors, including the aircraft design (airframe, engine type, winglets), operational aspects (distance, speed, flight route, flight altitude, weather, detours, holding patterns, green flying techniques), as well as on-the-ground conditions (taxiing, ground power supply). Energy use per passenger also depends on aircraft layout (seat numbers), load factors, and co-loaded cargo. Data is limited to aircraft with 30 seats and more, i.e., private aircraft and business jets are not part of the analysis. Data allows us to calculate average load factors for each city pair, from which global load factors and fuel use can be derived.

To factor these aspects into the calculation of emissions per RPK, an efficiency model is used that considers distance, aircraft type, and payload (passenger and cargo weight). 2023 data for the efficiency model is derived from:

1. 1)

   T100I (AirlineData): flight-specific load factors for flights to and from the USA and Canada, annual means.

2. 2)

   ICAO TFS (Traffic by Flight Stage): similar to T100I with global coverage, though not for all airlines.

3. 3)

   IATA WATS (World Air Transport Statistics): annually averaged, airline-specific load factors (i.e., not disaggregated by city pair or aircraft type), further split into international and domestic services, for about 219 airlines.

4. 4)

   FlightGlobal: similar to IATA WATS, with airline-level data (annual mean load factors) for 178 airlines.

Data consequently has three levels of granularity for load factors:

• Level 1: Airline average over one year (52% of all flights).

• Level 2: Airline–city pair average over one year (12.4%).

• Level 3: Airline–aircraft–city pair average over one year (15.2%).

T100I and ICAO TFS provide level 3 data. From these, level 2 and level 1 averages can be derived by averaging revenue passenger kilometers (RPK) and available seat kilometers (ASK) across aircraft serving a given city pair (level 2) and additionally across all city pairs (level 1, ICAO TFS only). By contrast, IATA WATS and FlightGlobal only provide level 1 data.

For each individual flight, the highest level of load factor is sought, following a fallback cascade across the data sources. When IATA WATS is established as the best data source, the corresponding value for either domestic or international flights is used. As a last fallback, a global load factor over all airlines is calculated for three distance categories: short-haul (≤1500 km), medium-haul (1501−3500 km) and long haul (>3500 km). This load factor is used when no other load factor (level 1–3) can be found.

The induced error can be calculated for using global means instead of level 3 values. When bootstrapping 100,000 samples, the per-flight absolute error mean is 0.0406, and the median is 0.0145. The bootstrapped 95% CI for the absolute error mean is [0.0389, 0.0423]. This corresponds to a relative error of 5.38% with a 95% CI of [5.15%, 5.62%]. Notably, this error affects 20.4% of the load factor, as a higher-level granularity for load factors is available for 79.6% of flights.

Total fuel use is a function of total load and aircraft-specific performance data. Seat-classes are weighted by multiplying average per-passenger emissions with a seat-class factor based on IATA’s cabin class method²¹. This method considers the number of seats per class in a given aircraft configuration. The formula for this calculation is provided in (Eq. (1)); the passenger share of fuel use in (Eq. (2)), also involving data from Flight Global, which provides passenger data. Fuel use is transformed into CO₂ in (Eq. (3)).

More specifically, fuel consumption is calculated for each individual flight, accounting for the specific aircraft model, cabin layout, and passenger/cargo load factors. Fuel consumption is derived from a physical aircraft performance model, which calculates fuel consumption for a given aircraft model, considering flight distance and takeoff weight. The total fuel consumption of the flight is then broken down into fuel use per passenger and converted to CO₂ emissions per passenger. All subsequent analyses are based on these individual-flight emission characteristics, allowing for more nuanced results than calculations based on airline-average or city pair-average load factors or performance data.

Allocation of total fuel burn of a given flight to individual passengers and subsequent conversion to CO₂ emissions follows these steps:

(a) Calculation of fuel use for payload in ton kilometres (tkm)

(b) Calculation of fuel per passenger (pax)

Here, passenger emissions and 50% of the emissions from freight transport are allocated to passengers, assuming a passenger (including baggage) weighs 100 kg³⁹.

(c) Calculation of CO₂ per Passenger (pax)

The resulting per-passenger emission intensity enables comparisons of flight efficiency. Figure 8 illustrates the inherent dependence of emission intensity on flight distance, with longer flights generally exhibiting lower emission intensities. This is because takeoff and ascent to cruising altitude – the most fuel-intensive phase of a flight – account for a larger proportion of total fuel burn on short-haul routes.

These data are used to assess all flights; where no data for a specific flight is available, global mean values are used for three distance classes (short-/medium-/long-haul). Flight distances are based on great circle distances, which are converted to flown distances by using ICAO’s correction factor: +50 km for flights up to 550 km, + 100 km for flights between 550 and 5500 km, and +125 km for flights >5500 km⁴⁰. Data is available for 26,155 city pairs and 61,607 ‘flights’, with one flight being defined as a unique combination of departure airport, arrival airport, airline, and aircraft type. Non-operational code-sharing flights have been removed from the dataset; only the airline carrying out the flight is considered. A weakness of the dataset is that it includes mostly scheduled flights.

Validation checks

The dataset comprises 27,451,887 flights in 2023, carrying a total of 3,554,769,475 passengers over a distance of 6,813,991,167,301 RPK, and causing emissions of 577,968,750 t CO₂. This is less than the 35.3 million flights, 4.319 billion passengers, 8232 billion RPK, and the 530 Mt CO₂ reported through the CORSIA Central Registry (said to cover 99% of total 2023 emissions on scheduled services reported for 2023 by ICAO⁴¹)^42,43. Our data consequently covers 83% of RPK and 82% of passengers in comparison to ICAO data. A potential explanation is that there are data gaps related to the reporting of charter operators.

Consistency analysis of our database shows that the number of seats in aircraft is always higher than the number of passengers. Airline load factor comparison with three airlines, Ryanair, Lufthansa, and American Airlines, for which data is publicly available, also suggests that variation is within 10%. For example, in 2023, Ryanair reported a 93% load factor, whilst the model suggests 93.6%; Lufthansa reported 82%, whilst the dataset is 78.9%; and American Airlines reported 83.5%, whilst the model suggests 81%. It is difficult to validate emission intensities for airlines due to variable operating conditions and inconsistent reporting standards. Annual reports may not include RPK numbers, refer to financial years rather than calendar years, combine data on operational fuel use for cargo and passengers, combine different accounting scopes, deduct sustainable aviation fuel emission “savings”, or calculate CO₂e rather than CO₂. Airlines also have subsidiaries that may have been included/excluded in the accounting. However, the communication teams of all airlines mentioned in this article (LATAM, Ryanair, EasyJet, Qatar Airways, Emirates, China Southern, British Airways, United Airlines, Delta Air Lines, Air Algérie) were provided with the data sources used in our analysis, as well as the airline-specific emission efficiencies (g CO₂ per RPK) calculated, and were asked to comment. LATAM responded by saying that it does not wish to comment. EasyJet provided a link to their own calculations, suggesting slightly lower emission efficiencies (66.64 g CO₂ per RPK) than calculated using our model (74.2 CO₂ per RPK).

Emission attribution

Emissions are assessed at different scales of analysis, for airports, airlines, or countries. For airports, emissions are calculated by aggregating half the emissions from all incoming and outgoing flights. This ensures that half of the emissions for each city pair are attributed to the respective departure and arrival airports. The approach addresses that fuel use can vary between incoming and outgoing flights due to differences in flight paths or weather. Airline emissions consider all emissions from flights operated by an airline, with intensities calculated by dividing the total amount of CO₂ by the number of RPK. Country emissions are calculated by aggregating emissions from all airports within national borders.

Efficiency calculations

The efficiency of air transport is calculated for aircraft, airlines, airports, city pairs, and countries, using the same indicator (g CO₂ per RPK). Airport emissions are calculated for 3218 airports by summing up the CO₂ for all departures and arrivals from/at each airport, and dividing the value by two to assign half the emissions from every flight to this airport. In an identical approach, half of all RPKs for every departure and arrival are calculated. The mean CO₂ intensity (g CO₂ per RPK) is then derived by dividing an airport’s CO₂ emissions by its RPK. Aircraft comparisons are based on city pairs, for which data is available for all aspects of the flight. For these, fuel consumption per RPK can be calculated for each aircraft based on fuel use, emissions, and RPK. On this basis, aircraft can be ranked by observed efficiency.

Optimum efficiencies on city pairs can be calculated for all routes served by more than one flight configuration (13,666 out of 26,156 city pairs, or 52.2%). Avoided CO₂ can be determined by dividing the most efficient average (g CO₂ per RPK) by the average efficiency on the city pair, multiplied with CO₂ emissions on this city pair. Values can then be aggregated for all city pairs, and total emission be compared to current emissions in the system. This can be expressed as (Eq. (4)):

E_best = Best (minimum) efficiency for a given route (e.g., fuel per tkm)

E_i = Efficiency of airline i on the same route

CO2_i = Current CO₂ emissions of airline i

CO2_best,i = CO₂ emissions at best efficiency for airline i

Then, the formula for CO₂ emissions at the best efficiency is:

If there are multiple airlines on a city‑pair, the total emissions improvement can be calculated as (Eq. (5)):

Assuming that only the most efficient aircraft were operated on all city pairs served by more than one carrier, emissions would be 10.66% lower than current emissions (517 Mt CO₂, rather than 578 Mt CO₂). This is the lower theoretical threshold for emission reductions based on observed operational efficiency.

Theoretical maximum efficiency. There are three ways to increase the efficiency of air transport: using only the most efficient aircraft, switching to one flight class (economy); and increasing load factors. The maximum efficiency for the year 2023 can be calculated in three steps.

Step 1: Global fleet of most efficient aircraft models. The effect of replacing the entire 2023 fleet with the two most efficient models – the Boeing 787-9 (55.4 g CO₂ per RPK) and the Airbus A321neo (61.4 g CO₂ per RPK) would lead to emission reductions in the range of 26.6% (all Airbus) to 33.7% (all Boeing), compared to the current fleet average of 83.6 g CO₂ per RPK, with (Eq. (6)):

I_avg = average CO₂ intensity (gCO₂ per RPK)

I_opt = CO₂ intensity of the Boeing 787‑9 (gCO₂ per RPK)

S = relative savings

Step 2: Economy class only configuration. To determine the difference between an air transport system consisting only very efficient aircraft models, we calculate frequency weighted averages for average seat numbers in economy, business, and first class for the five configurations of the most efficient aircraft models (A350-1000; A350-900; Boeing 787-1000; Boeing 787-800; and Boeing 787-900). As Table 1 illustrates, first class seating is uncommon in these aircraft, partially explaining their efficiency.

The difference in space use between all-economy vs. business and first class aircraft layout can be calculated using current economy to business to first ratios, with each first class seat equalling 5 economy seats, and every business class seat 4 economy class seats²¹. The percentage increase in seat numbers can be calculated based on data from Official Airline Guide (OAG) and FleetsAnalyzer (FA) for models A350 and B787, which represent the range of most efficient aircraft. In contrast to FA, OAG does not distinguish economy and economy-premium class; this might explain some of the difference in observed maximum seat numbers. OAG data refers to the 2023 flight schedule that lists the actual cabin layout for every flight; FA data considers actual aircraft configurations in fleets of various airlines (2023 data).

Table 1 shows the observed maximum seat number for the different models for both databases, and the average seat numbers for each model. The theoretical maximum number of seats can be calculated based on the following formula (Eq. (7)):

where S is the number of seats per class, and the factors are the IATA seat class factors, and economy (eco), premium-economy (prem.eco), business (bus) and first (first).

The increase in seat numbers in an all-economy layout is then determined based on an “assumed maximum”, representing either the calculated theoretical maximum seat number or the certified maximum seat number (whichever is lower). Certified maximums refer to type certification that are based on emergency evacuation demands¹³. This means that the number of seats in the Airbus A350-1000 is capped at 440, and at 420 seats for Boeing 787-1000 and 787-900.

Results suggest that an all-economy layout would allow airlines to carry 26% to 57% more passengers, which is equivalent to the theoretical reduction in fuel use of an all-economy layout.

Step 3: 95% load factor. Limiting capacity growth in global air transport will lead to higher load factors. In particular low-cost carriers have reported load factors of up to 94%⁴⁴. Here we assume a hypothetical maximum load factor of 95% that would reflect on an air transport system with considerably reduced capacity²⁵. Given the current load factor of 78.9%, aircraft could carry 16.1% more passengers if available capacity was utilized at 95%.

Fuel penalty. Calculations in Step 2 and 3 do not consider that optimization for efficiency will incur a fuel penalty due to the additional weight carried. Analysis for the two aircraft models characterizing the range of layout gains, i.e., A350-1000 and Boeing 787-800 (26.1–56.7% more seats), suggests that this fuel penalty is small. This is illustrated in Fig. 9 for the A350-1000: depending on distance, fuel consumption increases by 5.8% to 13.9%, implying a small fuel-penalty even in comparison of a near-empty (20% load factor) to a close to full (80% load factor) aircraft. We conclude that increasing seat numbers or load factor will marginally increase fuel use. The potential reduction in fuel consumption in an optimized air transport system is thus in the range of −55% to −75%, not considering the fuel penalty.

Fuel use increases as a function of load, here illustrated for the A350-1000. The difference in fuel consumption between a near-empty (20% load factor, lower green curve) and near-full (80% load factor, upper blue curve) ranges between 5.8% and 13.9%, suggestion that the fuel penalty for additional weight carried is small.

CO₂ intensity cap model

A CO₂ intensity cap model illustrates how much emissions from global aviation would fall if flights were required to operate at a mandated maximum CO₂ intensity. However, considering such a cap as constant is largely unrealistic as it is strongly dependent on flight distance and load factors. Ignoring changes to load factors or flight distance, we isolate the variation independent of those. Other scenarios investigate increasing load factors (in exchange for fewer aircraft, see above).

The CO₂ intensity c is therefore modelled as a function of load factor l and flight distance d, linear in both predictors (Eq. (8)):

Inverse proportionality is assumed as increasing load decreases the intensity, so does a longer flight (larger aircraft, less resistance at higher altitude, smaller contribution of fuel-intensive takeoff). This model is fit via least-squares, estimating the constants to (rounded) a = 54 g CO₂ per RPK, b = 35,000 g CO₂ per pax, d₀ = 180 km. The offset in load factor l₀ is manually set to 0.01 to prevent intensities going to infinity for very low load factors which is not well constrained with the data given how few flights operate that empty. This model is considered to be the medium cap model representing an average intensity. The low c_low and high cap model c_high are defined as c_low = 4c/3, c_high = 2c/3, representing 33.3% lower/higher efficiency, respectively. Theoretically, for very long flight distances fuel consumption increases due to the additional weight of large amounts of fuel. This motivates to add a term proportional to distance d, which however, does not yield a positive proportionality constant, contradicting this theory with data. In that sense, higher intensities for very long flight distances are not supported with the data here.

Limitations

Findings are characterized by the following limitations: Our data comprises only a share of global commercial passenger air transport (83% of RPK and 82% of passengers in comparison to ICAO^42,43). It is unclear how the difference affects the representativeness of our findings for commercial passenger air transport. We also acknowledge that ICAO’s distance correction is crude, as the ratio of actual flight distance to great-circle distance depends on multiple factors such as weather, airspace restrictions, or airport congestion. Load factors are calculated based on three levels of granularity, and a fallback option that is the global average. Analysis suggests that the error affects in particular small, remote, or politically isolated territories (Nauru, St. Pierre and Miquelon, Wallis and Futuna, Bhutan, or Greenland), countries where reporting is more limited (China, Russian Federation, India), or specific airlines and airline subsidiaries (e.g., Sichuan Airlines, TUI Airways Ltd., Interglobe Aviation Ltd.). Last, our proposed strategies to increase efficiencies in global aviation will in practice be met with various economic and political challenges. Airlines operate within economic constraints and a business environment shaped by subsidies, limited ambition for climate change mitigation, and expectations of continued growth. Efforts to enhance air transport efficiency will have to be addressed within this multifaceted context.