Fuel consumption behaviour in aggregated air traffic metrics

Currently there exist a wide variety of models that can be used to assess the fuel consumption of a single flight, from conventional models based on physics and flight performance to more innovative ones based on avant-garde techniques such as artificial intelligence. However, the quality of the fuel consumption estimated by these models usually relies strongly on the quality of data available. As consumed fuel is impacted by a wide variety of features, such as aircraft type, engine family, meteorological conditions, flight path, etc, the more information available, the more accurate the estimations will be. However, having access to such granulated data is not always trivial and, moreover, the computational cost that could be derived from assembling data coming from different agents in the aviation field (airports, airlines, manufacturers, meteorological stations), plus the processing of the data and afterwards the computation of a refined fuel consumption model will be very high. The work presented here has been developed within the framework of the project E.R.A. (Environmentally Responsible Aviation) funded by Red.es, and it presents an extensive analysis on how consumed fuel and carbon dioxide emissions estimations could be made with limited access to information. Moreover, the aim is to be able to prove that for aggregated metrics, that being a set of flights and not a single flight, the consumed fuel can be easily estimated thus helping accounting for the carbon dioxide emissions that are produced at a global level.


1.
Introduction Aviation accounts for approximately 2% of global anthropogenic CO 2 emissions [1].This amount is even more significant due to the high growth forecasted for this sector, thus making decarbonisation a main topic for the aviation industry.The Kyoto Protocol [2] assigned the control and mitigation of international aviation emissions to the International Civil Aviation Organization (ICAO).
The need of applying all possible efforts to control and reduce air transport CO 2 emissions was introduced in ICAO's policy after the publication of the International Panel on Climate Change (IPCC) study [3], that analyses and quantifies different aviation emissions while forecasting the next 50 years.The adopted action line covered four groups of measures, colloquially baptised as four pillars, including new technologies for future aircraft, engines and systems, Air Navigation and operating procedure improvements, environmental optimization of aeronautical infrastructures, and finally the introduction of Market based Measures (MBM) in the case that the previous elements were not enough to mitigate CO 2 emissions.Four different MBM types were established and evaluated: Voluntary agreements, Environmental charges, Environmental taxes and Emissions Trading [4].
As no general agreement was achieved within ICAO in these four systems worldwide application, a compromise solution was approved in 2016, establishing a scheme to offset the yearly amount of CO 2 emissions that exceed the global level of the average 2019-2020 years by buying ICAO's approved Eligible Emissions Unit (EEU), corresponding to certified CO2 reduction activities.This CORSIA (Carbon Offsetting and Reduction Scheme for International Aviation) has already started in 2021, although the first three years are a testing period, and the number of participating States will be increasing in the subsequent period [5].The application of the other MBMs is left to the criteria of each State.For the time being, just the European Union has enforced the EU Emissions Trading System (ETS) to the flights between two points in its territory since 2013.Up to eight People's Republic of China provinces have decided to include air transportation in their local ETS.In addition, several other countries around the world have in study or subject to legal approval other MBM economic mechanisms to be applied in this field.
In order to be able to gain knowledge on emissions the first step is to be able to measure them.Focusing on CO 2 , which is the emission currently targeted by market-based measures tackling aviation emissions such as CORSIA, the common approach is to estimate it as a linear relationship with the consumed fuel.A correlation factor between 3.15 and 3.16 has been approved valid by the scientific community and the regulatory authorities.
Currently there exist a wide variety of models that can be used to assess the fuel consumption of a single flight, from conventional models based on physics and flight performance to more innovative ones based on electronic data capture [6] or on avant-garde techniques such as artificial intelligence [7].However, the quality of the fuel consumption estimated by these models usually relies strongly on the quality of data available.In other words, the more information available of a flight, the better estimation of the consumed fuel, and therein, carbon dioxide emissions.As consumed fuel is impacted by a wide variety of features, such as aircraft type, engine family, meteorological conditions, flight path, etc, the more information available, the more accurate the estimations will be.
However, having access to such granulated data is not always trivial and, moreover, the computational cost that could be derived from assembling data coming from different agents in the aviation field (airports, airlines, manufacturers, meteorological stations), plus the processing of the data and afterwards the computation of a refined fuel consumption model will be very high.In the work presented here an extensive analysis on how consumed fuel and carbon dioxide emissions estimations could be made with a limited access to information is performed.Moreover, the aim is to be able to prove that for aggregated metrics, that being a set of flights and not a single flight, the consumed fuel can be easily estimated thus helping accounting for the dioxide carbon emissions that are produced at a global level.

1.1
Framework This work is developed in the framework of the E.R.A project, funded by Red.es.The aim of this research project is the development of a digital prototype tool that allows users to allocate air traffic in the future and know their environmental cost.This project tackles two challenges in the aviation field, on one hand the traffic distribution, and, on the other hand the environmental cost of the different itineraries.Let an example serve to illustrate the scope of the project: picture trip between two airports, also known as Origin Destination pair (O&D).There can be many itineraries that cover this trip, from one flight trip, to trips involving one or more flying connections between that origin-destination combination.Depending on different elements, such as total travelling time, ticket pricing, flight availability and many other factors, the passenger flow between those two points will be distributed along all different itineraries, all of those having their associated environmental cost that needs to be estimated.The project aims at forecasting this passenger flow and attaching the carbon footprint for each itinerary.To do so the project is composed by three main modules: • Demand forecast module: This first module is devoted to estimate the demand evolution in air transport considering the elasticity of the main macroeconomic and demographic and social factors affecting the air traffic behaviour.To do so, a model based on econometrics gravitational equations fed with thousands of historical passengers movements data has been developed.The output of this model, which is the expected traffic demand for different aviation itineraries, is then fed to the second module of the project.
• Traffic allocation module: the second module consists of an artificial intelligence based model whose aim is to predict the passengers flow share along different itineraries.To do so, the model is fed with the demand forecast along with different data regarding historical operations and passengers' flow.The mathematical requirements of this module are specific (the passenger flow cannot be negative and the sum of the allocation among all itineraries needs to be 100%), so a novel artificial intelligence algorithm has been designed and developed to ensure the robustness of the solution.
• Finally, the environmental cost module is coupled with the rest of modules of the project.The objective of this module is to provide an environmental cost, in terms of emissions to the forecasted itineraries and flow share obtained by means of the traffic allocation module.This paper's work is focused on the activities regarding the environmental module on the project.One of the main challenges of the ensemble of the modules arises in the connection between them in terms of data.Each module relies on different data sources to be built, so, in the end, consistency between all data must be ensured.This may seem trivial, however, it has been proved to be challenging, as the granularity of data in each module is different.For instance, the first module (the demand one) relies on passenger data and socio economic data retrieved from data sources which are defined in a monthly basis, that is, for a given route, the number of flights and the number of passengers will be known, but the day and time in which the flights took place is unknown.With such limitation, the estimation of the burnt fuel becomes challenging, as the most common methods [8], [9], [10] for its estimation rely on trajectory data (more than one instance per minute), which is much more granular than monthly data.For that reason it becomes necessary to develop and validate fuel burnt emissions models that are able to provide trustworthy information by using a limited amount of data or aggregated data.
Given the above mentioned constraints, the objective is to be able to perform analysis at network level, meaning for the total number of flights flying a given itinerary during a large period of time.By doing so, we aim at reducing the dimensionality of the data required and thus, the computational cost.A new model has been developed and validated using both different aircraft performance data and actual fuel consumption in a wide range of routes provided by airlines.
The fuel consumption of a commercial aircraft flight depends on multiple performance, network and environmental factors, most particularly the wind en route.Regarding the parameters that depend on the operations of airline companies, probably the most relevant is the takeoff mass of the plane, which greatly depends on the mass of fuel loaded (function of the distance to travel), and the mass of the payload.In passenger transport, the mass of payload depends on the size of the aircraft and the load factor.Thus, it seems of interest to determine, for a given size of aircraft and a route with a characteristic distance, what is the impact of the payload, for normal intervals of the load factor in the usual operations of the airlines 2.
The influence of the load factor The evolution of the specific consumption of an aircraft throughout its flight path, depending on its weight variation due to the fuel burnt, has been studied in detail [11].Using this system for various payloads and flight distances, curves such as the one in Figure 1 can be obtained [12].It can be seen that the difference in specific consumption between the two extreme load factors (60% and 100%) ranges between 10 and 20%, being approximately half for intermediate load factor variations.Some authors [13] have studied energy efficiency in real flights in a long-haul market in detail.In this particular case, the efficiency was calculated for the 20 largest companies in the North Atlantic between 2014 and 2017.The factors that explained the differences from one company to another, in descending order of importance, were: the type of aircraft, the seat density of the configuration, the proportion of cargo in the total payload and, lastly, the passenger load factor, responsible for 11% of the differences observed between companies.Other studies [14] have calculated global efficiencies of all commercial flights in a country, in this case the United States from 1990 to 2007.The study analyses the efficiency of several single-aisle aircraft and two twin-aisle categories (conveniently anonymous) in a variety of routes.The results indicate that fuel consumption increases between 3 and 4% of the additional payload per 1000 km in smaller aircraft and between 2 and 4% in twin-aisle aircraft.Less detailed calculations use approximate values to estimate the effects of increased load factors on fuel consumption [15].The results, for the current technology, with load factors of the order of 80 -85% suggest an increase of between 0.02 and 0.03 kg of fuel per 1000 km for each kg of added weight.The cruise phase is the most affected by the influence of the payload.Some specific analysis of this flight phase [16], [17] study the effects of payload and initial fuel load on cruise fuel consumption.In general, using the parameter of increase in fuel consumption per unit of increase in payload, these works confirm the margins noted above, with a range of values of 2 to 3% in long-haul flights and 3 to 4% in short and medium range.This is more or less constant with the different jet aircraft models in service in the last 50 years [18].
To check the degree of approximation of these estimates, two exercises have been carried out, using the BADA v3.16 program.The simulation of a medium-distance flight has been performed, corresponding to the average stage of flights in EUROCONTROL airspace, which is around 2 hours.An aircraft that is also common in this type of flights, the A320, has been used for the simulation, equipped with a V2500 engine.To calculate the effect of the load factor, a typical configuration of this aircraft with 180 seats has been selected.For the purposes of the simulation, and following the BADA methodology, the flight has been discretized into three segments: climb, cruise and descent.In each of these three segments, a sensitivity analysis has been carried out on the take-off mass, varying this mass in increments corresponding to variations in the load factor, from 85% to 93%, typical interval in the operation of airlines.
For the purposes of the flight parameters in each phase, the usual options that BADA offers have been used, which are the following: • Climb: including take-off, up to cruising altitude, is simulated with the option "climb at given CAS/Mach" and "calculated CAS (departure/arrival) speed profile" • Cruise at 33000 ft • Descent from cruising altitude to the end of the flight, including landing, with the same speed assumptions as for climb.
The fuel consumption calculations have been carried out by varying the take-off mass in intervals of 100 kg (1 passenger with his luggage).The results show that the difference in consumption of the same flight with 85% and 93% passenger load factor is 92 kg (5643 versus 5551 kg), 1.65% of the total.The difference would be 14 passengers (1400 kg), corresponding to a fuel increase of 6.5%, which coincides with the high values of the intervals indicated above, if it is assumed that In the two hours about 1700 km would be covered.The fuel variation of the different flight segments, depending on the load factor, is represented in Figure 3.The variation in all cases is very small and linear.It can be seen how the consumption during takeoff and the climb phase represents a relevant fraction of the total fuel consumption in flights of this duration, around 27%, while the cruise represents 68% of the total.To better understand the difference of 92 kg between fuel consumption with a load factor of 93% compared to consumption with a load factor of 85%, the breakdown by flight segment of this variation is shown in Table 1.The relevance of the take-off and climb phase for the fuel consumption of this type of flights can be appreciated again.The simulation of a long-distance flight has been carried out, corresponding to 6 hours, which is a characteristic average distance in this type of flight.An aircraft also common in this type of flights has been used for the simulation, the A350-900, equipped with a Trent XWP-84 engine.To calculate the effect of the load factor, a typical configuration of this aircraft with 314 seats has been selected.Using the same hypotheses regarding flight segments as in the previous case, the results show that the difference in consumption with 85% and 94% passenger load factor is 301 kg (42889 kg versus 43189 kg), a 0.70% of the total.The difference would be 28 passengers (2800 kg), which would increase fuel consumption by 11%.If it is considered that 5000 km are covered in six hours, the values are in the lower part of those previously mentioned.It can be seen in Figure 2 how the consumption during the cruise phase represents a majority fraction of the total consumption, 89%, which represents a substantial difference with respect to what has been previously shown for medium-distance flights (the case of the 2-hour flight ), where the contribution of takeoff was much more significant than on these longer flights.The breakdown by flight segments of this variation is shown in Table 2.

3.
The influence of wind Weather plays an important role in fuel consumption.Wind can have a relevant impact.In aggregated metrics it is not possible to compute atmospheric conditions for each flight.To decide how to incorporate this very important effect in the fuel consumption estimation, it has been decided therefore to compare the GCD (Great circle Distance) and ESAD (equivalent Air Distance) in the two way MAD-JFK route (Figure 3).A new model has been developed and validated using both, different aircraft performance data, and actual fuel consumption in a wide range of routes provided by airlines.The result is a linear correlation of fuel burnt vs ESAD (Equivalent Air Distance, which takes the effect of winds into consideration) for each aircraft model.An example of this type of function is shown in Figure 5. Proven the fact that load factor can be assumed constant, using actual aircraft fuel consumption provided by the airline, we propose linear correlations based on ESAD for each aircraft model.By linearly correlating the ESAD vs the fuel burnt we can obtain fuel burnt with origin, destination, month and aircraft type as only inputs.The average ESAD for a given route at a given month will be obtained from historical data.This approximation enables us to estimate the fuel burnt for large amounts of flights in aggregated metrics.The consumed fuel is then translated into CO2 emissions using a linear relationship and finally, it is translated into cost for the airline.Conclusions It can be concluded in view of the previous results that, with the current average load factors above 80%, the increases in fuel consumption due to variations in occupancy are small, between 1.5 and 0.5%, and are less influential than other factors, such as winds, temperatures and slots on airways.In aggregate calculations, normal aircraft occupations do not represent large deviations in the results.Therefore, for aggregated metrics it has been proven that it is feasible to assume a constant load factor.
Also, for aggregated metrics we can assume a linear correlation between the consumed fuel and the distance.In order to account for wind effects, we propose to use the ESAD and not the GCD as distance.
In conclusion then, for aggregated metrics it is acceptable to estimate fuel as dependent from origin, destination and aircraft type.

Figure 1 .
Figure 1.Example of the influence of range and payload on fuel consumption per nautical mile of a modern commercial aircraft.(Benito, 2006)

Figure 2 .
Figure 2. Variation of fuel consumption with the load factor of a 2-hour flight operated by an A320-200, and a 6-hour flight operated by an A350-900, broken down by the different flight segments (BADA simulation).

Figure 3 .
Figure 3. Histograms showing the statistical distribution of the fuel consumption for the MAD-JFK and JFK-MAD flights as a function of ESAD and GCD.

Figure 4 .
Figure 4. Fuel consumption as a function of ESAD for a typical narrow body and a typical wide body aircraft.(Source: own elaboration)

Table 1 .
Breakdown by flight segment of the variation in fuel consumption on the 2-hour flight with a load factor of 93% compared to 85%

Table 2 .
Breakdown by flight segment of the variation in fuel consumption on the 6-hour flight with a load factor of 93% compared to 85%