Modeling Civil Aviation Emissions with Actual Flight Trajectories and Enhanced Aircraft Performance Model

Abstract

Aviation emissions are continuously increasing along with the rapid development of air transportation, and results in the deterioration in regional air quality and the global climate. Accurate emission estimation is of great importance for relevant policies promotion and the sustainable development of the environment. Previous studies focused on the total emissions of a flight and lacked high precision in both spatial and temporal resolutions, especially aviation activities near ground. In this research, we propose an open-sourced emission calculation framework based on actual flight trajectories (TrajEmission), which calculates both the ground and airborne emissions simultaneously according to the configuration parameters, trajectory characteristics, and ambient conditions. We compare the emission results with five emission inventory methods. The results indicate that pollutant (nitrogen oxides, carbon monoxide, and unburned hydrocarbons) emissions in the landing and takeoff (LTO) cycle might usually be underestimated due to a lack of trajectory-based methods. In addition, in the overall results, the method based on the great circle route leads to an overestimation of 56.8% of pollutant emissions compared to the method based on actual routes. We also investigate the extent to which other factors could influence the emission results. To summarize, the TrajEmission framework can build inventories for the whole process of flight movements with high spatial–temporal resolutions and provide solid data support for environmental science and other related fields.

1. Introduction

Economic development leads to growing aviation transportation, which contributes to increasing pollutant emissions including nitrogen oxides (NO_x), carbon monoxide (CO), unburned hydrocarbons (HC), sulfur oxides (SO_x), particulate matter (PM), etc., all of which have negative impacts on both air quality and climate change [1,2,3]. It is noted in previous studies that aviation emissions can affect the environment at local and global scales. For example, CO may undergo a chemical reaction in the atmosphere that converts it to carbon dioxide, adding more greenhouse gases, increasing positive radiative forcing, and contributing to climate change. NO_x can cause acid rain and chemical smog, and it also reacts with ozone, causing damage to the ozone layer in the lower stratospheric layer. Aircraft engines emanate more pollutants, such as hydrocarbons, soot, and carbon monoxide, and further promote the formation of contrails [4]. Aircraft contrails alter the radiative balance of the atmosphere and have a net warming effect on the climate by reflecting sunlight and absorbing heat from the Earth’s surface. Aircraft contrails also increase cloudiness in the atmosphere, potentially altering local temperature and precipitation patterns [5].

Several policies have set a goal to reduce emissions and mitigate environmental impacts [6], such as the Intergovernmental Panel on Climate Change (IPCC) reports [7], Carbon Offsetting and Reduction Scheme for International Aviation (CORSIA) [8,9], International Air Transport Association (IATA)’s resolution to reach net zero carbon emissions by 2050 [10], and the European Union Emissions Trading System (EU ETS) [11].

Aviation emission calculation comprises top-down and bottom-up methods. The top-down methods prefer to use statistical data like annual aviation energy consumption and LTO numbers of airports in an area, but they lack fine results. Meanwhile, the bottom-up methods can fulfill the demand but need more accurate data, like operational procedure parameters, trajectory data, historical scheduled flight databases and so on. In practice, the International Civil Aviation Organization (ICAO) set forward three approaches (simple/advanced/sophisticated) to calculate aviation emissions [12]. Similarly, the European Environment Agency/European Monitoring and Evaluation Program (EEA/EMEP) proposed Tier 1, Tier 2, and Tier 3 methods to estimate aviation emissions with different model complexity and accuracy [12,13].

In general, emission models are built for the landing and takeoff (LTO cycle) phase and CCD (climb–cruise–descent) phase. LTO encompasses the takeoff, climb, approach, and idle phases (altitude ≤ 3000 ft), while CCD specifically means the flight trajectories at higher altitude (>3000 ft) [13]. The simple/Tier 1 approach relies on the number of inbound and outbound flights to calculate LTO emissions at designated airports. The advanced/Tier 2 approach takes into account the operational time (time-in-mode), engine type, and fuel consumption, utilizing empirical databases, like the ICAO Aircraft Engine Emissions Databank (EEDB) [14]. The sophisticated/Tier 3A approach accurately represents real aircraft emissions utilizing proprietary data and models, which highly depend on actual data, such as aircraft engine thrust, aircraft configuration, real-time speed, and dynamic emission indices derived from the Boeing Fuel Flow Methods 2 (BFFM2) [15] or Deutsches Zentrum für Luft- und Raumfahrt Method (DLR) [16,17]. Tier 3A adds departure and destination airports as model inputs. Some Tier 3B methods use the great circle route as the trajectory data to obtain dynamic parameters [13]. The simple/Tier 1 and advanced/Tier 2 approaches typically use statistical and empirical data to estimate, which may lack a description of the actual flight status and contain model bias. Their results mostly update annually, and resolution is city-scale and route-scale emissions. The use of flight trajectory can solve the problems to some extent. Some integrated tools based on the above methods are built to calculate emission inventory, like AEIC [18] and SAGE [19]. But at the same time, it also requires high fidelity of input data, so it could not be applied widely until actual trajectory data were easily accessible.

As aviation trajectory data have become available to the public recently, some studies demonstrated the feasibility of using them to meet the requirements of dynamic updating and fine-grained modeling for aviation emissions [20]. One typical case of those datasets is Automatic Dependent Surveillance–Broadcast (ADS-B) data. ADS-B records flight trajectory waypoints and can be accessed on open websites such as OpenSky [21,22], Flightaware [23], or Flightradar24 [24]. Zhang used ADS-B data to identify real routes and distances instead of the great circle routes, which effectively reduces the uncertainty of the emission inventories [25]. Quadros used airline schedules and ADS-B data from Flightradar24 and OpenSky to establish global aviation emission inventory monthly from 2017 to 2020 [26]. Filippone used ADS-B data and Mode-S data from OpenSky to calculate non-LTO emissions and verified the advantages of real-time flight data on emission modeling [27,28]. Teoh used flight trajectories derived from ADS-B telemetry and reanalysis weather data to develop the global aviation emissions inventory, and captured the spatiotemporal distribution of aviation activity and emissions during 2019–2021 [29]. Considering the significant spatiotemporal variation in aviation emissions, utilizing actual trajectory data, on the one hand, enables the capturing of variations in routes, fuel costs, and emissions due to weather and air traffic management. On the other hand, it replaces the great circle trajectory and obtains spatial–temporal emission results with higher accuracy. The open-sourced ADS-B dataset also makes it possible to be widely applied in aviation emission evaluations, like the sophisticated approach/Tier 3B method.

Nevertheless, there are still some disadvantages in the previous trajectory-based methods. Firstly, there are few calculations using ground trajectories directly. Actually, real LTO trajectories will have significant impacts on overall emission results, especially for short-haul flights, due to the higher proportion the LTO phase takes in whole flights. Secondly, the potential of aviation trajectories has not been fully exploited. ADS-B trajectories can record altitude and speed, which can help to calculate real-time aircraft mass, vertical climb rate, heading angle, dynamic engine parameters, and route inefficiency in each flight segment nearly in real time. Those factors will affect emission results in different ways [30]. Furthermore, fuel flow reflects the magnitude of thrust with a non-linear relationship. However, in some aircraft performance models (e.g., BADA) [31], it is usually treated as a linear one. It inevitably results in deviations in fuel consumption, for thrust is far higher at takeoff and quite lower when cruising [32]. Additionally, a higher heading angle of an airplane will result in a higher fuel flow. As such, these factors have not been fully considered and may increase the uncertainties in aviation emission inventories.

To solve the above problems, this paper proposes a technical framework named TrajEmission for civil aviation emission modeling, with real three-dimensional flight trajectories and a fully open-sourced aircraft performance model, openAP [33]. TrajEmission considers the spatiotemporal characteristics for the whole aircraft procedure from ground to the air and aggregates the fuel costs and pollutant emissions in flight segments with regard to the aircraft type, aircraft mass, and trajectory characteristics. It is argued that the proposed framework can not only correct the actual LTO emissions using ground tracks, but also support the timely calculation of the fuel consumption and emissions, so as to improve the model accuracy and spatial–temporal resolution of the results.

This paper is organized as follows. Section 2 describes the data and methodology. The experiments are conducted with real flight data and the results are analyzed in Section 3 compared with some baseline methods. Section 4 investigates the factors that may affect the emission results. Finally, Section 5 draws the conclusion and carries out discussion on the implications and future work.

2. Materials and Methods

2.1. Data

The ADS-B system is widely used in aeronautical monitoring. It makes aircrafts automatically send flight-related information to the ground stations, from which the status of aircrafts can be effectively monitored [34]. In this study, ADS-B trajectory data are from Flightradar24, and used to obtain actual flight operation process and trajectory records for both the LTO and CCD phases. Each aircraft continuously generates location records by global navigation satellite systems (GNSS) every 30–45 s during flight. Every record contains the ICAO code, real-time coordinates of the aircraft, timestamp, altitude, ground speed, etc. Flight information is also derived from flight schedules in Flightradar24 including the ICAO code, flight time, departure and landing airport code, operating aircraft type, engine manufacturer, engine number, etc. The ICAO code is used to join the two databases and generate a complete dataset to facilitate the emission calculation. Data samples are shown in

Table 1. Flight schedule information data samples. (Org = origin, Dst = destination).

Flight	ICAO	Airline	Org	Dst	OTime	Dtime
3U110	CSC110	Sichuan Airlines	TFU	HGH	01:10:00	03:15:00
MU2101	CES2101	China Eastern	XIY	PKX	07:30:00	08:49:00
CZ2735	CSN2735	China Southern	CKG	NGB	13:00:00	15:08:00
SC1153	CDG1153	Shandong Airlines	XMN	PEK	09:25:00	11:53:00
3U110	CSC110	Sichuan Airlines	TFU	HGH	01:10:00	03:15:00

and

Table 2. ADS-B trajectory data samples.

Timestamp	Coordinate	Ground Speed/kt	Altitude/ft
1634465087	[121.332, 31.2058]	147	0
1634465097	[121.3315, 31.2121]	146	4
1634465112	[121.3294, 31.2217]	146	10
1634465118	[121.3275, 31.2259]	147	13

and some ADS-B waypoints’ spatial distribution is shown in Figure 1. Table 1 presents part of the flight schedule information in the data samples. Table 2 shows the attributes in the ADS-B trajectory, where each flight corresponds to one set of trajectories.

Here, to demonstrate our method, we use all the flights between Beijing Capital International Airport (PEK) and Shanghai Hongqiao International Airport (SHA) from 16 to 22 October 2021 to conduct experiments. Figure 2 illustrates the number of flights for each aircraft type on the PEK-SHA route. A333 accounts for 29%, while B773 represents 22%, together comprising 51% of total flights. Additionally, A359 flights make up 12%, and both B77W and B789 each account for 8%.

Data preprocessing is conducted to deal with the inevitable trajectory record mistakes or missing by gross error detection and imputation.

2.2. Methodology

The technical framework of TrajEmission is shown in Figure 3. Firstly, flight schedule information and ADS-B trajectories are integrated. The trajectory dataset is filtered by missing field values and preprocessed to interpolate the missing parts and reconstruct trajectories. Secondly, overall fuel consumption is calculated with parameters depending on the flight status, by aggregating the flight segment fuel cost. Thirdly, real-time emissions will be estimated based on fuel consumption, with emission indices affected by aviation pollutant types, air speed, and altitude. Subsequently, we examine the impact of the model parameters on the emission results.

2.2.1. Three-Dimensional Flight Trajectory Imputation

The recorded waypoints suffer from discontinuity due to the incomplete station coverage. To ensure trajectory integrity, we interpolate the missing trajectory segments utilizing an open-source toolkit openAP [33]. The kinematic aircraft performance model (WRAP) in openAP is used for operational parameter estimation and simulated trajectory generation. It proposes distinct methods for modeling each flight phase, and shows better performance in estimating real-time engine thrust, air speed, vertical rate, altitude, and travel distance [35]. Before data processing, we use linear regression to fill blank field values. If the total flight track has a climb–cruise–approach phase but is missing some parts, we will use linear regression to interpolate missing coordinates and other fields, to reduce computational costs. If the flight phase is not complete, WRAP can help to simulate it if the trajectory segments miss a lot and linear interpolation does not work well. The example of trajectory imputation is shown in Figure 4. Specific steps are as follows:

(1)

Identify flight phase integrity and determine whether to use linear regression or WRAP.

(2)

For the idle and takeoff phase missing part (altitude ≤ 200 m), assign the start point of the runway in the departure airport, and the end point of the runway in the landing airport as the start and end points of the flight trajectory. If there is a gap between the first/last observation point and the start/end point, missing waypoints will be filled by simulated takeoff/landing trajectory vertically and great circle trajectory horizontally.

(3)

For the idle phase, add the airport average idle time as a variable parameter because some flights lack the internal trajectory in the airport from the runway to the apron. Thus, we can adjust the idle time of the specific airport to supplement the ground emission part. The PEK and SHA average is set as 900 s.

(4)

If the flight phase type is less than 3, which means the missing waypoints are too many, these flight records will be replaced by the simulation trajectory on the whole, according to the maximum cruise altitude. Since the maximum cruise altitude varies between aircraft types, it is set to the empirical altitude from actual ADS-B data. Simulation time interval is set as 30 s.

2.2.2. Fuel Consumption on Flight Trajectories

Fuel flow indicates the amount of fuel consumed by the engine per unit time, in kg/s. It is related to the aircraft type, engine parameters, flight operational time (time-in-mode), flight time. The calculations of fuel consumption and fuel flow are shown as Formulas (1) and (2) [25,26]. Fuel flow varies in terms of the aircraft mass, true air speed, and flight mode. An aircraft will likely have maximum fuel flow at takeoff and less at cruise. The physical meaning of each parameter is shown in Figure 5.

$$FC=\sum {FF}_{fp,eng,alt}\times N_{eng}\times t_{fp}$$

(1)

$${FF}_{fp,eng,alt}=f\left(m,tas,alt,pa\right)$$

(2)

where $FC$ denotes the fuel consumption of all phases for a flight. It is the aggregated results of each flight segment. ${FF}_{fp,eng,alt}$ represents fuel flow in the flight phase $fp$, with engine type eng, and altitude $alt$. $N_{eng}$ is the engine number for type eng. Time-in-mode $t_{fp}$ is up to the operation procedure. And $m$ means aircraft mass, $tas$ means true air speed, and $pa$ means trajectory path angle.

In this study, considering we only use domestic flights as the experiment dataset, $tas$ is replaced by ground speed for computational efficiency and default engine type of different aircraft types as type eng in openAP. Aircraft mass also has great impacts on fuel consumption. It involves passenger occupancy, belly cargo, and aviation kerosene mass. We employ 100%, 70%, and 50% max takeoff mass (MTOW) as inputs to compare results. Meanwhile, aviation kerosene mass will be subtracted sequentially by fuel burn at each flight segment. The path angle also determines fuel consumption. During the climb phase, the greater the path angle, the more fuel is burned. We use it as the input to correct the fuel consumption calculation.

2.2.3. Emission Model

Aviation emission is a function of fuel consumption and emission indices ($EI$). Meanwhile, in air pollution and emission inventory fields, emission factors (EF) are used more often. To be clear, we use $EI$ in this study.

As shown in Formula (3), ${Emiss}_{pollu}$ represents the total amount of aviation emission of specific pollutant $pollu$ for all the flight trajectories. ${FC}_i$ means the fuel consumption for trajectory $i$. ${EI}_{pollu,i}$ means the emission indices of pollutant $pollu$ with trajectory $i$. NO_x, CO, and HC are involved as the typical pollutants for emission modeling.

$${Emiss}_{pollu}=\sum {FC}_i\times {EI}_{p,i}$$

(3)

$EI$ depicts the emission intensity under certain conditions. They are considered as functions affected by temperature, humidity, and pressure. The dynamic emission indices are evaluated referring to the DLR correction model in openAP-Emission model [36]. Firstly, the corresponding atmospheric condition for flight trajectories is transformed to the International Standard Atmospheric condition (ISA) at sea level. Then, a parabolic model is used as the fuel flow–emission indices reference function. Finally, $EI$ is multiple by reference factors and is converted to $EI$ at a certain altitude [17]. $M$ is mach number. $\theta$ denotes the ratio of the actual temperature and standard temperature (288.15 K). $FF$ is fuel flow, $\beta ,\delta ,\omega ,ratio$ are the parameters to help correct $EI$ under the ISA condition. $E{I}_0$ represents emission indices at the sea level and $EI$ is the correction result. The formulas are as follows:

$$\beta =e^{0.2\times \left(M^2\right)}$$

(4)

$$\theta ={\displaystyle \frac{T\left(alt\right)}{288.15\beta }}$$

(5)

$$\delta ={\displaystyle \frac{\left(1-0.0019812\ast alt\right)\ast \frac{5.255786}{288.15}}{{\beta }^{3.5}}}$$

(6)

$$ratio={\displaystyle \frac{{\theta }^{3.3}}{{\delta }^{1.02}}}$$

(7)

$$\omega ={\displaystyle \frac{e^{-0.0001426\ast \left(alt-12900\right)}}{{10}^3}}$$

(8)

$$EI\_NOx=E{I}_{0}NOx\ast {\displaystyle \frac{e^{-19\ast \left(\omega -0.00634\right)}}{\sqrt{ratio}}}$$

(9)

$$EI\_CO=E{I}_{0}CO\ast ratio$$

(10)

$$EI\_HC=E{I}_{0}HC\ast ratio$$

(11)

TrajEmission accounts for dynamic emission index (EI) values across various temperatures and pressures at multiple altitudes. We apply the DLR method to correct the static $E{I}_0$ values in ISA. For example, A333 has the highest number of flights on the PEK-SHA route.

Table 3. ICAO advanced method: LTO NO_x\CO\HC EI (Aircraft: A333, Engine: Trent 772).

Flight Phase	Fuel Flow (kg/s)	EI NOx (g/kg)	EI CO (g/kg)	EI HC (g/kg)
Takeoff	3.139	35.56	0.21	0.01
Climb	2.53	26.82	0.49	0.01
Approaching	0.821	10.42	1.56	0.04
Idle	0.27	4.66	23.97	2.46

presents the EI values for the ICAO advanced approach sourced from the ICAO Emission Estimation Database (EEDB). It only has the $EI$ values for the LTO cycle. To accurately adjust the EI values during cruise, we correct the $EI$ for each flight segment based on the actual flight status, as illustrated in Figure 6. Here, in Figure 7, as the altitude and flight speed increase, $EI\_NOx$ gradually decreases. In contrast to $EI\_NOx$, the $EI\_HC$ and $EI\_CO$ exhibit distinct trends. Under low-altitude and low-speed conditions, the HC and CO emission indices are higher, corresponding to a greater proportion of emissions that may occur at the approach and idle phase.

3. Results

3.1. Experimental Results

In total, the PEK-SHA route emitted 64,300 kg of pollutant emissions during the LTO phase and 365,200 kg of pollutant emissions during the CCD phase. As shown in Figure 8a, the average daily emissions during the weekend (Date: 16 and 17 October) are lower than the emissions on working days (Date: 18–22 October). LTO emissions account for 10–20% of the total emissions. The total amount of NO_x emissions is the 40.7t, followed by 21.5 t of CO and 2.1 t of HC. In terms of the phase share in the LTO cycle from Figure 8b, NO_x is emitted most at takeoff, then the climb phase, and emitted less at the approach and idle phase. This is due to the fact that at the beginning of takeoff, the aircraft requires more thrust. The engine power efficiency is higher at this time and ultra-high temperatures promote NO_x conversion. When the aircraft starts to land and idle, less thrust is required, and the NO_x emission share reduces. In contrast, CO and HC emissions are higher while descending and taxiing. It is probably because when the engine power reduces as the aircraft lands, aviation kerosene incomplete combustion leads to more CO and HC [5].

3.2. Uncertainty Analysis

Typically, the 95% confidence interval is used to assess the uncertainty of emission inventories, but it will have a bias given the non-normal distribution. We use Bootstrap to analyze the uncertainty in the estimation of the aviation emission results here. The effectiveness of this method for uncertainty analysis of carbon emissions has been demonstrated [37]. The advantage of Bootstrap is that it does not depend on the distributional assumptions of the data, but rather generates a large number of bootstrap samples from the target dataset, and then calculates the statistic for each bootstrap sample to construct a distribution of the statistics [38,39]. We take 1000 samples of the fuel and emission results of each flight, select the statistics as the mean, and calculate the 95% confidence interval of the results. Figure 9 represents the bootstrap samples’ probability density distribution for one flight. On the PEK-SHA route, the fuel consumption is between 11,959.3 kg and 12,319.2 kg/per flight, the NO_x emission is between 402.0 kg and 420.3 kg/per flight, the CO emission is between 110.7 kg and 115.7 kg/per flight, and the HC emission is between 8.8 kg and 9.3 kg/per flight within the 95% confidence interval. Therefore, the uncertainty of the estimation of fuel ranges ±185 kg/per flight on the PEK-SHA route using TrajEmission, and NO_x uncertainty ranges ±8.9 kg/per flight. CO uncertainty ranges ±2.6 kg/per flight. HC uncertainty ranges ±0.25 kg/per flight.

3.3. Comparison with Other Emission Inventory Methods

We compare five different emission calculation methods for the PEK-SHA route using the same dataset. These methods are designed to quantify emissions based on various data types and levels of granularity.

The ICAO simple approach/EMEP Tier 1 utilizes the number of flights in the LTO at specific airports to estimate emissions. This method can allow for a quick result, but only estimates LTO emissions at the airport level, and does not distinguish between aircraft types. As a realization of this approach, we select reference average emission parameters from the Technical Guidelines for the Preparation of Emission Inventories for Non-Road Mobile Sources (Trial) [40] and calculate the PEK and SHA airport emissions, separately. The emission parameters for NO_x, CO, and HC are 16.29, 9.14, and 2.68 kg/per LTO, respectively.

The ICAO advanced approach estimates emissions in each phase of a flight, with specific aircraft types. The LTO cycle is divided into four phases, takeoff, climb, approach, and idle. The time-in-mode of them is 0.7 min, 2.2 min, 4 min, and 26 min. The thrust setting is 100%, 85%, 30%, and 7%. The fuel flow and EI values are related to aircraft and engine types, which can be found in the EEDB based on empirical statistical data. In this way, the advanced approach takes full account of the flight phase and aircraft type differences and improves the emission accuracy. The EI value examples are shown in Table 3.

The EMEP Tier 2 method is similar to the ICAO advanced approach; in addition, it also distinguishes between domestic flights and international flights. If it does not do so, the LTO will be overestimated. The EMEP methods are for use in the European area, so for emission background information consistency, we choose a similar airport, Heathrow Airport, as the background setting, which is also a 4F airport like PEK, and the average emission statistics of several aircrafts are utilized as inputs.

Table 4. (a). Emission values in EMEP Tier 2 method (LTO). (b). Emission values in EMEP Tier 2 method (CCD).

(a)
Aircraft	Mass of Fuel Burnt (kg)	Mass of CO Emitted (kg)	Mass of HC Emitted (kg)	Mass of NOx Emitted (kg)
A320	996.290	7.91565	1.57578	15.8733
A321	1290.70	4.89888	0.0751630	22.9054
A332	2657.84	20.0979	1.96336	49.7137
A333	2657.84	20.0979	1.96336	49.7137
B77W	3605.49	35.7075	3.68612	95.9020
B789	1999.62	13.2788	0.430738	24.1298
(b)
Aircraft	Fuel Burn (kg)	Mass of CO Emitted (kg)	Mass of HC Emitted (kg)	Mass of NOx Emitted (kg)
A320	2068.55	6.26	1.30	35.16
A321	2568.96	3.97	0.11	47.32
A332	4790.40	15.00	1.34	88.43
A333	4729.68	12.90	1.09	85.97
B77W	6521.82	22.74	2.29	165.69
B789	4370.69	9.01	0.33	74.01

From EMEP/EEA air pollution emission inventory guidebook 2023.

present the emissions per aircraft in the EMEP Tier 2 method.

Table 5. Emission values in EMEP Tier 3A method (Aircraft type: A333).

Flight Phase	Duration	Fuel Burn (kg)	NOx (kg)	CO (kg)	HC (kg)
ICAO	00:32:54	2168.08	35.32	21.19	2.10
CCD	1:15:28	6677.37	111.19	15.90	1.26
Total ICAO LTO + CCD 539 nm.	1:48:22	8845.45	146.51	37.09	3.35

From EMEP/EEA air pollution emission inventory guidebook 2023.

shows the emissions per aircraft in the EMEP Tier 3A method. In these methods, pollutant emissions per aircraft are provided using average emission factors.

In addition to exploiting the reference parameters, the real flight condition can also be used in the emission model. The EMEP Tier 3A method takes into account the origin and destination (OD), and the distance of the CCD phases. In this study, the distance for the PEK-SHA route is 998 km.

The EMEP Tier 3B method uses full flight trajectory information, a real one such as Quick Access Record (QAR) and ADS-B, or a simulated one from the aircraft performance model. To compare with the actual trajectory results, in this study, the great circle distance is employed as the flight distance in conjunction with the simulation vertical profiles from openAP for the implementation of this method.

To summarize, the ICAO simple approach/EMEP Tier 1 uses numbers from the LTO. The ICAO advanced approach is able to distinguish between aircraft types in each flight but uses empirical parameters in EEDB. Both of those methods can only estimate LTO emissions. The EMEP Tier 2 method and EMEP Tier 3A method can estimate emissions using flight schedule information. All of the methods described depend on statistics data or flight schedule information. As for the EMEP Tier 3B method and TrajEmission, these two methods are both based on trajectories; the difference is one is based on the simulated trajectory, and another is based on the actual trajectory. We aim to compare TrajEmission with both the statistical methods and trajectory methods and analyze the differences.

The results based on the above methodologies are shown in

Table 6. Comparison of emission inventory methodologies (PEK-SHA route).

Methods	Description	Phase	Fuel/t	NOx/t	CO/t	HC/t
ICAO simple approach/EMEP Tier 1	Numbers of LTO cycle and average emission factors.	LTO	-	13.2	7.4	2.2
		CCD	-	-	-	-
ICAO advanced approach	Based on ICAO Aircraft Engine Emissions Databank.	LTO	1735.7	30.6	17.2	1.7
		CCD	-	-	-	-
EMEP Tier 2 method	Use of aircraft-specific LTO EIs and average EIs for CCD.	LTO	2116.8	41	16.4	1.6
		CCD	3625.8	69.5	10.3	0.9
EMEP Tier 3A method	Use of specific aircraft type/engine data, OD flight information.	LTO	1725.2	29.2	17.4	1.7
		CCD	5450.3	95.7	13.1	1.1
EMEP Tier 3B method	LTO part: ICAO advanced approach. CCD part: using great circle distance and simulated vertical profile.	LTO	1735.7	30.6	17.2	1.7
		CCD	15,454.6	632	76.3	4.0
TrajEmission (This study)	Using actual trajectory and dynamic parameter configurations both at LTO and CCD phase.	LTO	1230.3	40.7	21.5	2.1
		CCD	8612.1	292.7	70.2	5.3

. The ICAO simple approach has overall lower pollutant emissions than the other approaches. In the CCD phase, the results show that bottom-up methods, such as EMEP Tier 3B and TrajEmission, usually give higher results than top-down methods, such as EMEP Tier 2. Moreover, emissions using the actual trajectories are lower than the simulated ones. This is because, in reality, airlines are more willing to use fuel-efficient routes to reduce costs. Although the simulated trajectory is optimal in the idealistic condition, it is not necessarily the most fuel-efficient route.

Possible reasons for the methodological deviations are as follows: There is a discrepancy between the actual flight status and ideal situation. Top-down approaches are inadequate in accounting for the variations in the actual flight process, and obviously, fine-scale bottom-up emission methods exhibiting a biased fit to the actual circumstances are superior to them. For example, the real flight time may change at different airports in a given year. The PEK average idle time is 15 min, but the ICAO advanced methods consider 26 min for all airports. Furthermore, the actual trajectory exhibits non-typical acceleration or deceleration behavior, as well as other temporary detours, which directly lead to fuel consumption fluctuation. As a comparison, simulated trajectories are generated in the ideal condition and lack representation of the real circumstances, like speed and emission intensity variation.

3.4. Emission Impact Factors

Many factors can lead to deviations in emission evaluations. We will quantify the impacts of aircraft types, flight trajectory characteristics, and aircraft mass, those which are inputs in TrajEmission, to better understand the extent to which these factors affect the emission results.

3.4.1. Aircraft Types

TrajEmission entails aircraft types in emission modeling, and we posit that the contribution of different aircraft engines and specifications varies considerably. On the PEK-SHA route, aircraft A333, B773, and A359 cover more than 50% of the entire flights, as shown in Figure 2.

Taking the small- and medium-sized aircraft A320, A321, medium-sized A333, and large-sized B77W as examples, Figure 10 describes the emission distinctions of these aircrafts. We conduct bootstrap sampling on the results. After sampling 1000 times, with a confidence interval of 95%, the fuel consumption of a small- and medium-sized flight is 4.5~5.3 t/per flight, that of a medium-sized flight is 11.4~13.2 t/per flight, and that of a large aircraft is 14.3~17.0 t/per flight. The fuel consumption and emission distribution for small- and medium-sized aircraft is more concentrated, with less variation in the emission results per flight compared to large aircraft. Large-sized aircraft consume between 12.5 t and 20 t fuel per flight, which is approximately 3~4 times higher than that of small- and medium-sized aircraft.

Regarding NO_x, small- and medium-sized aircraft emit less than 200 kg, medium-sized aircraft emit between 300 and 500 kg, and large aircraft emit between 600 and 800 kg. For CO, large aircraft have an average emission of around 200 kg, whereas smaller aircraft emit less than 150 kg. Similarly, large aircrafts’ average HC emissions are 18 kg, while smaller aircrafts’ are less than 12.5 kg.

3.4.2. Flight Trajectory Characteristics

TrajEmission can use trajectory characteristics to improve emission model accuracy. In this part, we will discuss the impact of flight routes, dynamic engine parameters, and path angles when evaluating pollutant emissions.

Differences in flight routes can also result in variations in emission calculations. Many studies have taken the great circle trajectories between the OD airports as the flight paths to calculate aviation emissions, whereas in reality, airlines tend to follow fuel-efficient or safer routes. Moreover, if temporary air traffic control is encountered, a flight path could be diverted accordingly. The different assumptions underlying the emission calculation methods can lead to biased estimates of real flight emissions.

Taking the PEK-SHA route as an example, we used both the simulated great circle trajectory and the real flight trajectories to conduct the experiments. The route comparison is shown in Figure 11. A total of 761.6 t of pollutant emissions were generated following the great circle trajectory. By contrast, a total of 432.4 t of pollutant emissions were generated on the real flight trajectories, which are only 56.8% of the great circle trajectory. We argue that the methods based on the great circle trajectory will generally overestimate the aviation emissions.

As far as different phases are concerned, the LTO emission for the PEK-SHA route is 49.5 t on the great circle trajectory, which is lower than the 64.2 t on the real trajectory. The difference is mainly caused by the extended time-in-mode in practice under the requirements from air traffic management. Meanwhile, the CCD emission for the PEK-SHA route is 712.2 t on the great circle trajectory, contrasted with 368.2 t on the actual trajectories. The deviation is caused by the tendency of the flights to favor the fuel-efficient trajectories that take advantage of prevailing winds. The great circle trajectory usually gives the shortest flight distance between a city pair, but it does not mean it is the most fuel-efficient.

Also, inconsistencies in the actual paths lead to differences in emissions. The discrepancy resulting from real circumstances cannot be conveyed through great circle trajectories. Pollutant emissions are generally overestimated, as shown in Figure 12a. Median emissions per flight in the great circle trajectory (585.5 kg) are 14.8% higher than the actual trajectory results (510.0 kg). Nevertheless, the maximum and minimum values differ considerably, which are 3252 kg vs. 1104 kg and 152 kg vs. 112 kg, respectively. Outliers are fewer when using actual data. Figure 12b indicates that the round-trip trajectory emission results are biased in reality, but the simulated trajectory results will not change in the same city pairs. It is possible that the choice of route, or the traffic status at airports before landing, leads to bias all together. Actual flights will have more trajectory variation due to severe weather, air traffic congestion, or temporary air control. Busier airports have more cases of holding patterns before landing. A multitude of factors contribute to the inherent uncertainty of emission results.

In addition to the discrepancies in trajectories, the engine parameters along with flight status are also influenced. Figure 13 shows summary statistics that the average emission parameters change with altitudes. The EI can indicate the emission per fuel consumption unit.

The non-linear fuel consumption model is employed for the calculation of fuel consumption pertaining to the variation with altitude and speed. As illustrated in Figure 14, A333, the aircraft with the highest flight numbers on the PEK-SHA route, is taken as an example. In the initial low-altitude phase, below 5000 ft, the fuel flow demonstrates a gradual increase with altitude, followed by a subsequent gradual decrease. This indicates that during the takeoff and climb stages, the corresponding engine fuel consumption is elevated due to the larger thrust requirements. As the altitude rises, the engine transitions to the smooth, high-speed flight phase, and the thrust diminishes, resulting in a corresponding decline in the fuel flow. Additionally, a distinct phenomenon arises during the low-altitude and low-speed phase, wherein the fuel flow exhibits a markedly elevated level compared to that observed in the idle phase. At this juncture, the aircraft appears to be in a stop-and-go condition at the airport, with the highest fuel flow. It can be seen that the implementation of a well-designed landing approach and departure procedure, which minimizes the aircraft’s idling time, can effectively reduce the aircraft’s fuel consumption and emissions.

Changes in the path angle can also result in alterations to emissions. The discussion of the path angle’s impact on emissions is rarely mentioned in previous studies. In general, the higher the path angle, the higher the fuel flow when ascending. Pollutant emissions are 0.081 t less than in scenarios using the path angle in the trajectory.

3.4.3. Aircraft Mass

Aircraft mass can also affect fuel consumption and emissions. Some research points out that different initial payloads can affect the final emissions results [41]. These factors are not fully taken into account by the above methods. Some methods assume the 100% takeoff weight (MTOW) as the aircraft weight. Domestic flights did not exceed 80% seat occupancy rate in 2021 [42]. In order to measure the impact of weight on flight emissions, we use 100%, 70%, and 50% MTOW as the initial aircraft mass. Indeed, aircraft mass also decreases gradually over the course of the flight. In order to avoid mass bias, TrajEmission subtracts the weight of the fuel consumed on the previous flight segment in the process of calculating the next segment, updating the aircraft mass in real time.

Overall fuel consumption using 100% mass, 70%, and 50% mass is 9842 t, 8741 t, and 8211 t. The 50% mass fuel is 16.6% lower than the 100% one and the 70% result is 11.2% lower. As for NO_x emissions, the 50% mass result is 26.8% lower than the 100% one, and the 70% mass result is 19.0% lower than the 100% mass result. The frequency distribution discrepancy is shown in Figure 15, taking A320 as an example. Emissions will be exaggerated if the aircraft mass bias is not taken into account.

4. Discussion

Table 6 presents various estimation methods. The ICAO simple method relies on statistical data and does not differentiate between aircraft types. Consequently, it yields lower LTO estimates compared to other methods, primarily due to average emission factors. As indicated in Figure 10, the fuel consumption and pollutant emissions from large aircraft are significantly higher than those from smaller and medium-sized aircraft. On the PEK-SHA route in this study, which is predominantly operated by medium–small-sized aircrafts, the actual average emission factor should be lower than the empirical parameters provided in the manual, leading to the observed lower results. In other methods of LTO estimation, the results are consistently lower than those obtained using TrajEmission.

We quantified the flight characteristics of the actual trajectories individually, aiming to accurately assess the flight conditions and thereby enhance the accuracy of the emission estimates. We considered the aircraft type, flight speed, rate of climb, trajectory distance, and flight angle as input variables. We modeled dynamic fuel consumption using the flight speed and altitude to reflect changes in the flight status and modeled the EI using the temperature and pressure to represent the impact of environmental changes. The experimental results indicate that LTO emissions under TrajEmission account for 10–20% of the total emissions, which is a little bit higher than in previous studies [12,43]. This leads us to question whether using empirical emission factors as parameters might underestimate pollutant emissions, reflecting the potential long-term underestimation of emissions in the LTO cycle.

As for CCD emissions, approaches based on trajectory (EMEP Tier 3B method and TrajEmission) are higher than those without using actual waypoints. It indicates that incorporating trajectories can add actual information and improve the accuracy of CCD emission estimation, but it may also lead to overestimations. For example, the EMEP Tier 3B method uses great circle routes and simulated vertical profiles. Although these trajectories represent the shortest distance route in an ideal condition, they are not necessarily the most fuel-efficient [44]. In reality, airspace planning and monsoon belts are considered cost factors in airline routing [45]. Therefore, we use real routes to partially correct for the overestimation of CCD emissions. In Section 3.2, we also highlight the uncertainties analysis of TrajEmission, providing a reference for researchers. We should also include a comparison with the BADA-related methods, but, due to license application, this comparison was not included.

In addition, TrajEmission also has some parts to improve. Firstly, a more complicated model may improve the accuracy of the missing attribute complementation, but this will also result in computational burden. It is necessary to identify an appropriate balance between the computational cost and accuracy.

Secondly, the trajectory from runway to tarmac is absent in some cases. It represents the genuine response to the efficiency of an airport. We utilize the airport latitude and longitude as the coordinates for the apron, calculating the shortest distance as the ground track. If airport internal data are available, the ground trajectory can be reconstructed.

Thirdly, the aircraft performance model also has its limitations, such as being unable to explain the flight characteristics of new aircrafts, such as the C919. As a result, those have to be replaced by similar aircraft models. In fact, some methods, like deep learning, can also be utilized to extract features of the actual trajectory, combined with the flight parameters to reconstruct the 3D trajectory. They try to solve dependency with empirical aircraft data. This can be further studied in 3D trajectory interpolation and prediction.

Wind effects should be taken into consideration in calculating the flight emissions. It is estimated that wind may generate an impact of approximately 20% on the fuel consumption of flights [41]. Actually, airline companies prefer to design fuel-efficient routes considering the influence of wind so as to obtain fuel-saving routes typically longer than great circle routes. Furthermore, true air speed should be employed as an input to the aircraft performance model. However, only ground speed is used in this study as input to the WRAP model. It may result in accuracy bias in the emission calculation.

5. Conclusions

In this study, we set forward a technical framework TrajEmission for a fine-scale aviation emission calculation using both ground and airborne trajectories. ADS-B waypoints, flight schedule information, and dynamic engine parameters are integrated to estimate pollutant emissions.

The PEK-SHA route generates approximately 64,300 kg of emissions during LTO in a week, and 365,200 kg during CCD. LTO emissions account for 10–20% of the total emissions. In terms of flight phases, NO_x emissions are higher during the takeoff and climb phases compared to the approach and idle phases, while CO and HC emissions exhibit the opposite trend: lower during the takeoff and climb phases and higher during the descent and idle phases. This difference arises from the increased thrust required during ascent, which leads to higher temperatures and subsequently greater NO_x emissions. In contrast, when descending, incomplete combustion of fuel leads to higher CO and HC emissions.

Analysis of the uncertainty of TrajEmission. On the PEK-SHA route, the uncertainties of fuel consumption, NO_x, CO, and HC are ±185 kg per flight, ±8.9 kg per flight, 2.6 kg per flight, and ±0.25 kg per flight, respectively.

We compared the results of five other emission inventory methods with TrajEmission. In this study, we used trajectory-based estimation for the LTO phase, and all other inventory results were lower than those of this study, indicating a potential underestimation of ground emissions. For the CCD estimation, the results from non-trajectory methods are lower than those from trajectory estimation methods. In trajectory calculations, estimates based on great circle routes are significantly higher than those based on actual routes by 56.8%. This is because actual routes tend to be fuel-efficient routes. Therefore, the estimation results of this study reduce the uncertainty of other inventory methods and are closer to real flight conditions.

Additionally, we analyze the effects of features extracted from the trajectory data on the emission results. For example, large-sized aircraft have 3–4 times more emissions than small–medium-sized aircrafts. Different route choices can also impact on the emission results. Round trips actually have different emissions, but in listed emission inventory methods, they are always considered the same. Dynamic EIs can vary under distinct altitudes. Fuel flow has a non-linear relation with altitude and speed. In terms of aircraft mass, a 50% mass reduction can reduce fuel consumption by 16.6% and a 30% mass reduction by 11.2%.

TrajEmission can be widely applied to the calculation of aviation emissions, providing an open-source framework. It provides more possibilities for further study such as the following: (1) emission model for regional and global aviation estimation, (2) emission results at multiple scales for numerical analysis, (3) analysis of 4D pollutant dispersion in the vicinity of airports, and (4) analysis of emission patterns in the regional area.