isabella232 / travel-impact-model

Travel Impact Model 1.5.0

(Implementation of the Travalyst Shared Framework by Google)

Table of contents

• Background
• Model overview
  • Flight level emission estimates
    • Flight level CO₂ estimates
    • Data sources
  • Breakdown from flight level to individual level
    • Data sources
    • Factors details
    • Outlier detection and basic correctness checking
• Example emission estimation
• Legal base for model data sharing
• Versioning
• Limitations
• Data quality
• Contact
• Glossary
• Appendix
  • Appendix A: Aircraft type support

Background

In this document we describe the modeling assumptions and input specifications behind the Travel Impact Model (TIM), a state of the art emission estimation model that Google’s Travel Sustainability team has compiled from several external data sources. The TIM aims at predicting carbon emissions for future flights to help travelers plan their travel.

Model overview

For each flight, the TIM considers several factors, such as the Great Circle distance between the origin and destination airports and the aircraft type being used for the route. Actual carbon emissions at flight time may vary depending on factors not known at modeling time, such as speed and altitude of the aircraft, the actual flight route, and weather conditions at the time of flight.

Flight level emission estimates

Flight level CO₂ estimates

The Travel Impact Model is based on the Tier 3 methodology for emission estimates from the Annex 1.A.3.a Aviation 2019 published by the European Environment Agency (EEA).

There are several resources about the EEA model available:

• the main documentation
• the data set
• further documentation on pre-work for the EEA model

The EEA model takes the efficiency of the aircraft into account. As shown in Figure 1, a typical flight is modeled in two stages: take off and landing (LTO, yellow) and cruise, climb, and descend (CCD, blue).

(Fig 1)

For each stage, there are aircraft-specific and distance-specific CO₂ emission estimates based on the fuel burn of the aircraft. Table 1 shows an example emissions forecast for a B789 aircraft:

Aircraft	Distance (nm)	LTO CO2 forecast (kg)	CCD CO2 forecast (kg)
B789	500	5'439	18'318
B789	1000	5'439	33'925
B789	...	...	...
B789	5000	5'439	164'982
B789	5500	5'439	180'903

(Table 1)

By using these numbers together with linear interpolation or extrapolation, it is possible to deduce the emission estimate for flights of any length on supported aircraft:

• Interpolation is used for flights that are in between two distance data points. As a theoretical example, a 5250 nautical miles flight on a Boeing 787-9 will emit 172778.5 kg of CO₂ during the CCD phase (where 172778.5 equals 164827 + (180730 - 164827)/2 and figures for 5000nm and 5500nm entries were taken from Table 1).
• Extrapolation is used for flights that are either shorter than the smallest supported distance, or longer than the longest supported distance for that aircraft type.

There is information for most commonly-used aircraft types in the EEA data, but some are missing. For missing aircraft types, one of the following alternatives is applied in ranked order:

• Supported using the Piano-X data set: If an aircraft type is supported in the Piano-X data set and a comparable type is supported both in the Piano-X and the EEA data set, a correction factor is derived by comparing the Piano-X output for both types across a range of missions. The correction factor will be applied to the LTO and CCD numbers of the comparable type in the EEA database.
• Supported by fallback to non-optimized aircraft type: If there are estimates in the EEA data set for an aircraft that is identical except for the lack of optimizations such as winglets or sharklets, the non-optimized counterpart is used for the estimate.
• Supported by fallback to previous generation aircraft type: If there are estimates in the EEA data set for a previous generation aircraft type in the same family, from the same manufacturer, the previous generation aircraft is used for the estimate.
• Supported by fallback to least efficient aircraft in the family: For umbrella codes that refer to a group of aircraft, the least efficient aircraft in the family will be assumed.
• Not supported: For aircraft types for which none of the cases above apply, there are no emissions estimates available.

See Appendix A for a table with detailed information about aircraft type support status.

Data sources

Used for flight level emissions:

• EEA Report No 13/2019 1.A.3.a Aviation 1 Master emissions calculator 2019 (link)
• Piano-X aircraft database (link)

Breakdown from flight level to individual level

In addition to predicting a flight’s emissions, it is possible to estimate the emissions for an individual seat on that flight. To perform this estimate, it’s necessary to perform an individual breakdown based on three relevant factors:

1. Number of total seats on the plane in each seating class (first, business, premium economy, economy)
2. Number of occupied seats on the plane
3. Amount of cargo being carried

The emission estimates are higher for premium economy, business and first seating classes because the seats in these sections take up more space. As a result, those seats account for a larger share of the flight's total emissions. Different space allocations on narrow and wide-body aircraft are considered using separate weighing factors.

Data sources

Used to determine which aircraft type was used for a given flight:

• Aircraft type from published flight schedules

Used to determine seating configuration and calculate emissions per available seat:

• Aircraft Configuration/Version (ACV) from published flight schedules

Factors details

Seating class factors

Seating parameters follow IATA RP 1726. An analysis of seat pitch and width in each seating class in typical plane configurations confirmed the accuracy of these factors.

• Narrow-body aircraft
  • Economy and Premium Economy 1
  • Business and First 1.5
• Wide-body aircraft
  • Economy 1
  • Premium Economy 1.5
  • Business 4
  • First 5

Load factors

Load factors are derived from a projection of past passenger statistics from 2019 U.S. data average (source):

• Passenger load follows a seasonal pattern with low in Jan (~79.3%) and high in June (~89.8%)
  • Aggregated overall average applied in the model is 84.5%
• Cargo load not included

Outlier detection and basic correctness checking

If there are no individual seat configuration numbers for a flight available from the published flight schedules, or if they are incorrectly formatted or implausible, the TIM uses aircraft-specific medians derived from the overall dataset instead. Basic correctness checks based on reference seat configurations for the aircraft are performed, specifically:

• The calculated total seat area for a flight is the total available seating area. This is calculated based on seating data and seating class factors. For example, the total seat area for a wide-body aircraft would be:
  • 1.0 * num_economy_class_seats +
1.5 * num_premium_economy_class_seats +
4.0 * num_business_class_seats +
5.0 * num_first_class_seats
• The reference total seat area for an aircraft is roughly the median total seat area.
• During a comparison step: If the calculated total seat area for a given flight is within certain boundaries of the reference for that aircraft, the filed seating data from published flight schedules is used. Otherwise the reference total seat area is used.

Example emission estimation

Let’s consider the following flight parameters:

• Origin: Zurich ZRH
• Destination: San Francisco SFO
• Aircraft: Boeing 787-9
  • Economy seats: 188
  • Premium Economy seats: 21
  • Business seats: 48
  • First seats: 0

To get the total emissions for a flight, let’s follow the process below:

1. Calculate great circle distance between ZRH and SFO: 9369 km (= 5058.855 nautical miles)
2. Look up the static LTO numbers and the distance based CCD number from aircraft performance data (see Table 1), and interpolate CO₂ for a 9369 km long flight:
   • LTO 5.43 metric tons of CO₂
   • CCD 166.87 metric tons of CO₂ calculated
     • 165.0 + (5058.9 - 5000) * (180.9 -165.0) / (5500 - 5000)
3. Sum LTO and CCD number for total flight level result:
   • 166.87 + 5.43 = 172.3 metric tons of CO₂

Once the total flight emissions are computed, let’s compute the per passenger break down:

1. Determine which seating class factors to use for the given flight. In the ZRH-SFO example, we will use the wide-body factors (Boeing 787-9).
2. Calculate the equivalent capacity of the aircraft according to the following
   C = first_class_seats * first_class_multiplier + business_class_seats * business_class_multiplier + …
   • In this specific example, the estimated area is:
     0 * 5 + 48 * 4 + 1.5 * 21 + 188 * 1 = 411.5
3. Divide the total CO₂ emissions by the equivalent capacity calculated above to get the of CO₂ emissions per-economy passenger: 172.3 t CO₂/411.5 = 418.71 kg CO₂
4. Emissions per-passenger for other cabins can be derived by multiplying for the corresponding cabin factor.
   • Business: 418.71 * 4 = 1674.85 kg CO₂
   • Premium Economy: 418.71 * 1.5 = 628.06 kg CO₂
   • Economy = 418.71kg CO₂
5. Scale to estimated load factor 0.845 by apportioning emissions to occupied seats:
   • Business: 1674.85 / 0.845 = 1982.067 kg CO₂
   • Premium Economy: 628.06 / 0.845 = 743.28 kg CO₂
   • Economy = 418.71 / 0.845 = 495.52 kg CO₂

Legal base for model data sharing

The carbon emission estimate data will be available via API under the Creative Commons Attribution-ShareAlike CC BY-SA 4.0 open source license (legal code).

Versioning

The model will be developed further over time, e.g. with improved load factors methodology or more fine grained seat area ratios calculation. New versions will be published.

A full model version will have four components: MAJOR.MINOR.PATCH.DATE, e.g. 1.3.1.20230101. The four tiers of change tracking are handled differently:

• Major versions: Changes to the model that would break existing client implementations if not addressed (e.g. changes in data types or schema) or major methodology changes (e.g. adding new data sources to the model that lead to major output changes). We expect these to be infrequent but they need to be managed with special care.
• Minor versions: Changes to the model that, while being consistent across schema versions, change the model parameters or implementation.
• Patch versions: Implementation changes meant to address bugs or inaccuracies in the model implementation.
• Dated versions: Model datasets are recreated with refreshed input data but no change to the algorithms regularly.

Limitations

The model described in this document produces estimates of carbon emissions. Emission estimates aim to be representative of what the typical emissions for a flight matching the model inputs would be. Estimates might differ from actual emissions based on a number of factors.

Actual flight distances: When modeling the distance between a given origin and destination, the Great Circle Distance between the origin and destination airport is used, as opposed to the actual distance flown.

This simplifying assumption enables the model to be used even when precise flight path information is not available, such as when computing emission estimates for future flights.

Aircraft types: The emissions model accounts for the equipment type as published in the flight schedules. The majority of aircraft types in use are covered. See Appendix A for a list of supported aircraft types.

Some aircraft types are supported by falling back to a related model thought to have comparable emissions. See Flight level emission estimates for more details.

If no reasonable approximation is available for a given aircraft, the model will not produce estimates for it.

Cargo load factors: Cargo load is not yet supported in the model.

Engine information: Beyond the aircraft type, there are other aircraft characteristics that can have an effect on the flight emissions (e.g. engine type, engine age, etc.) that are not currently included when computing emission estimates.

Fuel type: The emissions model assumes that all flights operate on 100% conventional fuel. Alternative fuel types (e.g. Sustainable Aviation Fuel) are not supported.

Greenhouse gases: Greenhouse gases other than CO₂ are not included.

Passenger load factors: The same default load factor is used for every flight instead of including seasonal, regional, and airline level factors.

This simplifying assumption was introduced during Covid-19 pandemic times.

Seat configurations: If there are no seat configurations individual numbers for a flight available from published flight schedules, or if they are incorrectly formatted or implausible, aircraft specific medians derived from the overall dataset are employed.

Data quality

The CO₂ estimates were validated by comparing against a limited amount of real-world fuel burn data. The finding was that the TIM is underestimating by 7% on average.

The EEA guidebook (chapter 4) cites sources from ICAO that estimate the uncertainty of the LTO factors between 5 and 10%. The CCD factor uncertainty is estimated between 15 and 40%.

Contact

We are welcoming feedback and enquiries. Please get in touch using this form.

Glossary

CCD: The flight phases Climb, Cruise, and Descend occur above a flight altitude of 3,000 feet.

CO₂: Carbon dioxide is the most significant long-lived greenhouse gas in Earth's atmosphere. Since the Industrial Revolution anthropogenic emissions – primarily from use of fossil fuels and deforestation – have rapidly increased its concentration in the atmosphere, leading to global warming.

Contrail-induced cirrus clouds: Cirrus clouds are atmospheric clouds that look like thin strands. There are natural cirrus clouds, and also contrail induced cirrus clouds that under certain conditions occur as the result of a contrail formation from aircraft engine exhaust.

Radiative Forcing (RF): Radiative Forcing is the instantaneous difference in radiative energy flux stemming from a climate perturbation, measured at the top of the atmosphere.

Effective Radiative Forcing (ERF): Radiative forcing effects can create rapid responses in the troposphere, which can either enhance or reduce the flux over time, and makes RF a difficult proxy for calculating long-term climate effects. ERF attempts to capture long-term climate forcing, and represents the change in net radiative flux after allowing for short-term responses in atmospheric temperatures, water vapor and clouds.

European Environment Agency (EEA): An agency of the European Union whose task is to provide sound, independent information on the environment.

Google’s Travel Sustainability team: A team at Google focusing on travel sustainability, based in Zurich (Switzerland) and Cambridge (U.S.), with the goal to enable users to make more sustainable travel choices.

Great circle distance: Defined as the shortest distance between two points on the surface of a sphere when measured along the surface of the sphere.

ICAO: The International Civil Aviation Organization, a specialized agency of the United Nations.

LTO: The flight phases Take Off and Landing occur below a flight altitude of 3000 feet at the beginning and the end of a flight. They include the following phases: taxi-out, taxi-in (idle), take-off, climb-out, approach and landing.

TIM: The Travel Impact Model described in this document.

Short Lived Climate Pollutants (SLCPs): Pollutants that stay in the atmosphere for a short time (e.g. weeks) in comparison to Long Lived Climate Pollutants such as CO₂ that stay in the atmosphere for hundreds of years.

Appendix

Appendix A: Aircraft type support

IATA aircraft code	Aircraft full name	Mapping (ICAO aircraft code)	Support status
100	Fokker 100	F100	Direct match in EEA
146	British Aerospace 146	BAE146	Direct match in EEA
221	Airbus A220-100		Supported via correction factor derived from Piano data
223	Airbus A220-300		Supported via correction factor derived from Piano data
290	Embraer 190 E2	E190	Mapped onto older model
295	Embraer 195 E2	E195	Mapped onto older model
310	Airbus A310	A310	Direct match in EEA
313	Airbus A310-300	A310	Direct match in EEA
318	Airbus A318	A318	Direct match in EEA
319	Airbus A319	A319	Direct match in EEA
320	Airbus A320-100/200	A320	Direct match in EEA
321	Airbus A321	A321	Direct match in EEA
330	Airbus A330	A332	Mapped to least efficient in family
332	Airbus A330-200	A332	Direct match in EEA
333	Airbus A330-300	A333	Direct match in EEA
339	Airbus A330-900neo	A333	Supported via correction factor derived from Piano data
340	Airbus A340	A345	Mapped to least efficient in family
343	Airbus A340-300	A343	Direct match in EEA
345	Airbus A340-500	A345	Direct match in EEA
346	Airbus A340-600	A346	Direct match in EEA
350	Airbus A350	A350	Mapped to least efficient in family
351	Airbus Industrie A350-1000	A350	Supported via correction factor derived from Piano data
359	Airbus A350-900	A350	Direct match in EEA
380	Airbus A380	A380	Mapped to least efficient in family
388	Airbus A380-800	A380	Direct match in EEA
717	Boeing 717-200	B717	Direct match in EEA
732	Boeing 737-200	B732	Direct match in EEA
733	Boeing 737-300	B733	Direct match in EEA
734	Boeing 737-400	B734	Direct match in EEA
735	Boeing 737-500	B735	Direct match in EEA
736	Boeing 737-600	B736	Direct match in EEA
737	Boeing 737	B734	Mapped to least efficient in family
738	Boeing 737-800	B738	Direct match in EEA
739	Boeing 737-900	B739	Direct match in EEA
744	Boeing 747-400	B744	Direct match in EEA
747	Boeing 747	B744	Mapped to least efficient in family
752	Boeing 757-200	B752	Direct match in EEA
753	Boeing 757-300	B753	Direct match in EEA
757	Boeing 757	B753	Mapped to least efficient in family
762	Boeing 767-200	B762	Direct match in EEA
763	Boeing 767-300	B763	Direct match in EEA
764	Boeing 767-400	B764	Direct match in EEA
767	Boeing 767	B764	Mapped to least efficient in family
772	Boeing 777-200/200ER	B772	Direct match in EEA
773	Boeing 777-300	B773	Direct match in EEA
777	Boeing 777	B773	Mapped to least efficient in family
781	Boeing 787-10		Supported via correction factor derived from Piano data
787	Boeing 787	B789	Mapped to least efficient in family
788	Boeing 787-8	B788	Direct match in EEA
789	Boeing 787-9	B789	Direct match in EEA
32A	Airbus Industrie A320 (Sharklets)		Supported via correction factor derived from Piano data
32B	Airbus Industrie A321 (Sharklets)		Supported via correction factor derived from Piano data
32N	Airbus A320neo		Supported via correction factor derived from Piano data
32Q	Airbus A321neo		Supported via correction factor derived from Piano data
32S	Airbus A318/A319/A320/A321	A321	Mapped to least efficient in family
73C	Boeing 737-300 (winglets)	B733	Mapped to non-optimized aircraft
73E	Boeing 737-500 (winglets)	B735	Mapped to non-optimized aircraft
73F	Boeing 737 Freighter	B734	Mapped to least efficient in family
73G	Boeing 737-700	B737	Direct match in EEA
73H	Boeing 737-800 (winglets)		Supported via correction factor derived from Piano data
73J	Boeing 737-900 (winglets)	B739	Mapped to non-optimized aircraft
73M	Boeing 737-200	B732	Direct match in EEA
73N	Boeing 737-300	B733	Direct match in EEA
73Q	Boeing 737-400	B734	Direct match in EEA
73S	Boeing 737-200/200 Advanced	B732	Direct match in EEA
73W	Boeing 737-700 (winglets)		Supported via correction factor derived from Piano data
74E	Boeing 747-400 Mixed	B744	Direct match in EEA
74F	Boeing 747 Freighter	B744	Mapped to least efficient in family
74H	Boeing 747-8I	B744	Mapped onto older model
74N	Boeing 747-8F (Freighter)	B744	Mapped onto older model
74Y	Boeing 747-400F Freighter	B744	Direct match in EEA
75T	Boeing 757-300 (winglets)	B753	Mapped to non-optimized aircraft
75W	Boeing 757-200 (winglets)		Supported via correction factor derived from Piano data
76W	Boeing 767-300 (winglets)		Supported via correction factor derived from Piano data
77F	Boeing 777 Freighter	B773	Mapped to least efficient in family
77L	Boeing 777-200LR	B772	Mapped to similar model
77W	Boeing 777-300ER	B77W	Direct match in EEA
77X	Boeing 777-200F Freighter	B772	Direct match in EEA
7M8	Boeing 737MAX 8		Supported via correction factor derived from Piano data
7M9	Boeing 737MAX 9		Supported via correction factor derived from Piano data
7S8	Boeing 737-800 (Scimitar Winglets)		Supported via correction factor derived from Piano data
A32	Antonov AN-32	AN32	Direct match in EEA
A81	Antonov AN-148-100	AN148	Direct match in EEA
AB4	Airbus A300B2/B4/C4	A30B	Direct match in EEA
AB6	Airbus A300-600/600C	A306	Direct match in EEA
ABY	Airbus A300-600 Freighter	A306	Direct match in EEA
AN4	Antonov AN-24	AN24	Direct match in EEA
AN6	Antonov AN-26/30/32	AN32	Mapped to least efficient in family
AR1	Avro Regional Jet RJ100 Avroliner		Not supported
AR8	Avro Regional Jet RJ85 Avroliner		Not supported
ARJ	Avro Regional Jet Avroliner		Not supported
AT4	ATR 42-300/320	ATR42	Mapped to similar model
AT5	ATR 42-500	ATR42	Direct match in EEA
AT7	ATR 72	ATR72	Direct match in EEA
ATR	ATR 42/ATR 72	ATR72	Mapped to least efficient in family
BE1	Beechcraft 1900		Not supported
BE9	Beechcraft C99 Airliner		Not supported
BEH	Beechcraft 1900D		Not supported
BES	Beechcraft 1900/1900C		Not supported
BET	Beechcraft Light Aircraft twin engine		Not supported
BNI	Pilatus Brit-Norm BN-2A/B ISL/BN-2T		Not supported
C27	Comac ARJ21-700		Not supported
CNA	Cessna (Light Aircraft)	C208	Direct match in EEA
CNC	Cessna (Light Aircraft - single engine)	C208	Direct match in EEA
CNF	Cessna 208B Freighter	C208	Direct match in EEA
CNJ	Cessna Citation	C500	Direct match in EEA
CR1	Canadair Regional Jet 100		Not supported
CR2	Canadair Regional Jet 200		Not supported
CR5	Canadair Regional Jet 550		Supported via correction factor derived from Piano data
CR7	Canadair Regional Jet 700	CS700RJ	Direct match in EEA
CR9	Canadair Regional Jet 900	CS900RJ	Direct match in EEA
CRJ	Canadair Regional Jet	CS900RJ	Mapped to least efficient in family
CRK	Canadair Regional Jet 1000		Not supported
CS1	Bombardier CS100		Not supported
CS3	Bombardier CS300		Not supported
CVF	Convair 440/580/600/640 Freighter		Not supported
DH2	De Havilland-Bombardier DHC-8 Dash 8 Series 200	DHC8	Mapped to non-optimized aircraft
DH3	De Havilland-Bombardier DHC-8 Dash 8 Series 300	DHC8	Mapped to non-optimized aircraft
DH4	De Havilland-Bombardier DHC-8 Dash 8 Series 400	DHC8	Mapped to non-optimized aircraft
DH8	De Havilland-Bombardier DHC-8 Dash 8	DHC8	Direct match in EEA
DHC	De Havilland-Bombardier DHC-4 Caribou		Not supported
DHT	De Havilland-Bombardier DHC-6 Twin Otter	DHC6	Direct match in EEA
E70	Embraer 170 Regional Jet	E170	Direct match in EEA
E75	Embraer 175 Regional Jet	E175	Direct match in EEA
E7W	Embraer 175 (Enhanced Winglets)		Supported via correction factor derived from Piano data
E90	Embraer 190 Regional Jet	E190	Direct match in EEA
E95	Embraer 195 Regional Jet	E195	Direct match in EEA
EM2	Embraer EMB-120 Brasilia	E120	Direct match in EEA
EMB	Embraer EMB-110 Bandeirante	E110	Direct match in EEA
EMJ	Embraer RJ-170/175/190/195 Regional Jet		Mapped to least efficient in family
ER3	Embraer ERJ-135 Regional Jet	E135	Direct match in EEA
ER4	Embraer ERJ-145 Regional Jet	E145	Direct match in EEA
ERD	Embraer ERJ-140 Regional Jet	E145	Direct match in EEA
ERJ	Embraer ERJ-135/140/145 Regional Jet		Mapped to least efficient in family
F50	Fokker 50	F50	Direct match in EEA
F70	Fokker 70	F70	Direct match in EEA
FRJ	Fairchild Dornier 328JET		Not supported
IL7	Ilyushin IL-76	IL76	Direct match in EEA
IL9	Ilyushin IL-96-300	IL96	Direct match in EEA
J32	British Aerospace Jetstream 32		Not supported
J41	British Aerospace Jetstream 41		Not supported
JST	British Aerospace Jetstream 31/32/41		Not supported
L4T	LET L410 Turbolet	L410	Direct match in EEA
M1F	McDonnell Douglas MD-11 Freighter	MD11	Direct match in EEA
M80	McDonnell Douglas MD-80		Not supported
M83	McDonnell Douglas MD-83		Not supported
M87	McDonnell Douglas MD-87		Not supported
M88	McDonnell Douglas MD-88		Not supported
M90	McDonnell Douglas MD-90	MD90	Direct match in EEA
MA6	Xian Yunshuji MA-60		Not supported
S20	SAAB 2000		Not supported
SF3	SAAB SF 340		Not supported
SFB	Saab 340B		Not supported
SU9	Sukhoi Superjet 100-95		Not supported
SWM	Fairchild (Swearingen) Metro/Merlin		Not supported
TU5	Tupolev TU-154		Not supported
YK2	Yakovlev YAK-42		Not supported
YK4	Yakovlev YAK-40		Not supported