Introduction

Civil aviation is a vital energy-consuming sector1. Global air transportation and its emissions have witnessed remarkable growth at a rate of over 5% per year (before COVID-19)2. Especially in China, air traffic has grown at an astounding rate of ~10%3 per year over the past decade. This has resulted in a dramatic increase in aviation fuel consumption and pollutant/carbon dioxide (CO2) emissions4,5,6. Aviation contributed 1.03 billion Mg of CO2 in 20197, responsible for approximately 3% of total energy-related CO2 emissions8,9. And this proportion is expected to grow, as other sectors (e.g., power and road transport) move towards low-emission and green technologies10,11. The aviation industry is more challenging to save energy, achieve zero emissions, and decarbonize, given that there are currently no effective policies or large-scale solutions to drastically reduce its heavy reliance on fossil fuels12,13.

However, one potential cost-effective pathway to reduce the aviation-induced energy consumption, pollution, and carbon is to manage aircraft activities during the landing-takeoff (LTO) cycles12,14, in particular the taxi process on the runway15. Furthermore, the LTO (comprising four operating modes: takeoff, climb, approach, and taxi) is of short distance over the entire flight range, fuel consumption and CO2 emissions can be as much as 5−10% of the whole journey16; and LTO emissions contribute up to 19%, 76% and 71%17 of nitrogen oxides (NOx), hydrocarbons (HC) and carbon monoxide (CO), respectively, due to differences in engine conditions. Non-negligible aviation-attributable impacts on ground-level annual fine particulate matter (PM2.5) and ozone (O3) concentrations in China can reach 0.4–1.5 and 10.8–14.5 μg·m−3, respectively18. Similarly, aircraft LTO activities contributed approximately 10%19 to the surrounding air quality (15 km20) NOx at the Zhengzhou Xinzheng International Airport. In the United States, LTO emissions have been shown to increase PM2.5 concentrations by about 8% at Hartsfield–Jackson Atlanta International Airport21. High-intensity LTO emissions that occur below the mixing layer height (MLH) have more direct impacts on the air quality near airports22 and are of significant concern.

It’s especially important to accurately estimate LTO fuel and emissions. This can provide quantitative references for energy management and pollutant/carbon reduction. It will also supply a necessary precise basis for physical-chemical reactions of air quality models for aviation environmental impact research. However, the widely used Chinese National Guideline (NG) and traditional ICAO methods have significant uncertainties that limit their effectiveness. The NG method based on the fixed emission factors (kg/LTO) resulting in maximum type-specific emission factor differences of up to five times23,24,25. The ICAO method uses a constant time-in-mode for all flights and defines the LTO occurrence scale as beginning from the ground surface to the top of the MLH at an altitude of 3000 ft26,27.

To further improve the modeling accuracy of LTO emissions estimation, physics-based aircraft trajectory methods have been proposed, using detailed flight track data. A representative example is the Aviation Environmental Design Tool (AEDT)28 developed by the United States Federal Aviation Administration’s (FAA), which incorporates radar-derived trajectory data. However, much of AEDT’s input data, sourced from the FAA, remains proprietary and inaccessible to the broader research community. Similarly, EUROCONTROL’s Advanced Emission Model (AEM) relies on aircraft performance data from the Base of Aircraft Data (BADA)29,30 to estimate emissions. Another tool in this category is the commercial Piano software31,32, which provides emissions modeling capabilities under a paid license. Recently, a combination of trajectory data and machine learning algorithms33 has been used to improve estimation accuracy, including the use of flight data records, quick access recorders (QAR)34,35,36, and automatic dependent surveillance-broadcasts (ADS-B)37,38,39. However, the estimation of emissions based on complete flight trajectory data for each individual aircraft, as well as comprehensive datasets for all aircraft types, remains limited by data availability constraints. In particular, modeling using the ADS-B dataset to cover long periods and numerous airports is still underexplored38, limiting the scalability and broader applicability of such methods40. As an alternative, the dynamic time-in-mode-based estimation is a potential solution, but no studies have been conducted yet.

In this study, we aim to address the trade-off between generalizability and accuracy in aircraft LTO emissions modeling. Specifically, we target the research gap in dynamically estimating the time-in-mode for each stage of the LTO cycle across numerous flights and diverse airport scenarios. To this end, we develop a practical aircraft landing-takeoff time, fuel, and emission model (ALTFEM) to accurately estimate the LTO fuel consumption and emissions by considering the dynamic time-in-mode for each flight. The Ns-based taxi in/out time is estimated for the first time by modeling the hourly air-to-ground traffic at airports using Ns (Number of arrival/departure aircraft on the schedule). The ALTFEM also integrates the MLH-derived climb and approach time calculating model and constructed based on over 11,000,000 pieces of actual data. Moreover, we recommend airport-specific/aircraft type-specific emission factors (kg/LTO) for multiple available data scenarios to revise the NG in China and promote the model. Our study provides a general framework for improving LTO emissions estimation and a basis for assessing aviation-related energy consumption, environmental and climate impacts, as well as a scientific rationale for developing airport emissions management policies.

Results

Dynamic time-in-mode estimated by the ALTFEM and driving factors

The parameters of the taxi time calculation model have interpretable physical meaning. It was found that the linear functional relationship (Using PVG Airport as an example, Supplementary Fig. 1) between the taxi in/out time (\({T}_{{taxi}}\)) and aircraft schedule numbers (Ns) had distinct physical meanings of the slopes (\(\Delta T\)) and intercepts (\({T}_{0}\)). The value of \(\Delta T\) indicates an increase in the taxi time per unit of the Ns, and \({T}_{0}\) represents the initial taxi time without any aircraft ready for take-off. According to the modeling results, \(\Delta T\) was always a positive number, signifying a positive correlation between \({T}_{{taxi}}\) and Ns. This implies that taxiway congestion as a result of an increasing number of aircraft schedules will lead to additional taxi in/out time. In contrast, \(\Delta T\) and \({T}_{0}\) exhibited declining trends with an increasing hourly Ns (Fig. 1a). Additionally, when using the annual average hourly aircraft numbers (N) for the statistical analysis, we found a significant exponentially decreasing relationship between the \(\Delta T\), \({T}_{0}\), and N, as shown in Fig. 1b, with the other airports demonstrating a similar tendency (Supplementary Fig. 2). This indicated that as the number of departing/arriving aircraft increases, the taxi time will not initially increase consistently, but instead the runway utilization rate will gradually improve until the runway capacity becomes a constraint, and then the taxiway congestion status will result in a higher taxi time. This phenomenon would be more pronounced at mega airports that have reasonable ground control and maintain high runway productivity levels.

Fig. 1: Relationship between ALTFEM-estimated taxi time and aircraft numbers at representative airports.
figure 1

a In typical hours, a functional relationship exists between the taxi in-time and Ns for the Shanghai Pudong International Airport (PVG). b The correlation of \(\Delta T\) and \({T}_{0}\) with annual average hourly aircraft numbers (N) during taxi-in. c The ALTFEM-estimated annual average hourly taxi in- and out-time and corresponding weighted average Ns for the Beijing Capital International Airport (PEK) and PVG and a comparison with the ICAO-referenced values.

Full size image

The dynamic Ns-based taxi in/out time for each flight significantly improved the time-in-mode accuracy compared to the ICAO default. We found significant discrepancies between the airport-specific ALTFEM-estimated taxi time and ICAO constants (420 s for taxi-in and 1140 s for taxi-out). Generally, the annual average taxi-in (323 s) and taxi-out (558 s) time for 237 airports in China were overestimated by the ICAO by 23.2 and 51.0% respectively. Furthermore, the annual average hourly taxi-in and taxi-out time ranged from 180 s (2:00, DOY) to 915 s (21:00, PVG) and from 183 s (7:00, ZQZ) to 2730 s (1:00, FUO), respectively. The maximum airport-specific differences in the taxi in- and out-time were as high as 372% and 530%, respectively. For the different airport classes41, the annual average hourly taxi-in time for 4 F, 4E, 4D, and 4 C were 520 s (316 s–923 s), 337 s (219 s–658 s), 295 s (181 s–822 s), and 302 s (213 s–523 s), respectively, and for the taxi-out time were 927 s (537 s–1779 s), 686 s (309 s–1389 s), 553 s (287 s–1525 s), and 468 s (151 s–2036 s), respectively (Fig. 2). In particular, the mean taxi-in time of the 4 F airports was instead underestimated (up to 23.7%) by the ICAO.

Fig. 2
figure 2

The ALTFEM-estimated annual average hourly taxi in- and out-time (s), along with the average Ns values during taxi-in and taxi-out, across different airport classes.

Full size image

The drivers behind taxi time variations are different for different moments and airports. The airfield activity could be divided into low-traffic (primarily at night) and high-traffic (primarily in the daytime) period based on the hourly Ns at the airport. In general, significant variation in Ns is observed across airport classes. Larger airports (e.g., 4 F) exhibit higher Ns due to greater flight volumes and infrastructure capacity, while smaller airports (e.g., 4 C) show lower values. We observed that high values of taxi time occurred not only during high-traffic, but also during low-traffic periods (e.g., taxi-out time at 4:00 at PEK (Fig. 1c)). During the high-traffic period, the taxi in- and out-time was primarily affected by the traffic on the taxiways, as the growing number of flights led to runway congestion and increased taxi time for aircraft in the queue. Controlling the number of aircraft departing/arriving from airport surfaces based on the configuration of a runway can reduce the taxi time20. As for the low-traffic time period, the taxi time was less correlated with the number of aircraft on the runway, while the tower controller operations played a crucial role. Optimizing an aircraft’s taxi path is a potentially effective way to decrease the taxi in- and out-time16.

The ALTFEM-estimated climb and approach time can reduce the uncertainty in time-in-mode estimation by considering changing mixing layer height compared to the ICAO-referenced constant (132 s for climb and 240 s for approach). We found distinct monthly changes at the different airports, with averages ranging from 69 s (January) to 160 s (June) for climb and from 219 s (January) to 471 s (June) for approach. Changes in the MLH primarily drove these variations42. A high correlation (R2 > 0.9) between the MLHs and estimated climb and approach time variations at the different airports was found (Supplementary Fig. 3). Moreover, the monthly disparity of the MLH-derived climb and approach time varied for the different airports. For example, the maximum difference between the summer and winter climb/approach time at PEK (in Beijing) was 2.4 times higher than at CAN (in Guangdong) (Supplementary Fig. 3). This difference was due to the MLH being significantly higher in China’s northern summer than in the winter, but not noticeable in the south.

Fuel consumption, CO2, and pollutants emissions estimated by the ALTFEM

The fuel consumption, CO2, and multiple-pollutant emissions from the LTO for civil aircraft in mainland China were estimated using the ALTFEM. The aviation-related fuel consumption and CO2 emissions were 5148.45 Gg and 16217.61 Gg for 237 domestic airports in 2019. The total emissions of NOx, CO, HC, SO2, and PM were 66.21 Gg, 50.67 Gg, 3.85 Gg, 19.91 Gg, and 1.25 Gg, respectively.

Distinct day-high (7:00–20:00) and night-low (2:00–5:00) diurnal characteristics were found (Supplementary Fig. 4a), which was consistent with the LTO numbers and the increased activities of the airport43. Monthly emissions (Supplementary Table 1), co-influenced by the LTO numbers and the variable MLHs, showed a maximum fuel consumption and emissions in July and a minimum in February.

The Climb, approach, taxi, and take-off consumed 26.4%, 26.5%, 35.1%, and 12.0% of fuel, respectively (Supplementary Fig. 4b). For the taxi mode, nearly 90% of the CO and HC emissions were attributed to it. This was mainly due to the lower engine thrust settings, inadequate combustion states with an unfavorable fuel-air mixture, and longer flight durations17, resulting in higher emissions of CO and HC. When climbing and approaching, the engines operate in an efficient situation, and consequently higher NOx emissions.

The B738, A320, A321, B737, and A319 contributed more fuel consumption and emissions (Supplementary Fig. 4c) because of a significantly higher number of LTOs. Taken together, the five types accounted for 78.0% of the total fuel consumption and 63.5–80.3% of the total emissions. The A380, A340, and B747 burned more fuel and emitted more CO2 and pollutants during each LTO cycle due to their higher emission factors equipped with four engines44. The disparity of the aircraft type-specific fuel flow (kg/s) was as high as 11.5–19.9 times.

Airport-specific difference in results between the ALTFEM-estimated and other methods

The ALTFEM-estimated fuel consumption and emissions showed an uneven distribution across various airports, with 20% of the airports contributing ~89% of the fuel consumption and emissions. High emissions (NOx, Supplementary Fig. 4d) were primarily concentrated in the Beijing, Guangdong, Shanghai, Sichuan, and Yunnan. This attributed to economic development, population density, transportation distance, and traffic inconvenience45. Therefore, it is of great significance to investigate airport-specific fuel consumption and emissions.

The accuracy of airport-specific emissions was significantly improved by the ALTFEM compared to the ICAO and NG methods. The MAE of taxi emissions between the ALTFEM-estimated and actual taxi time-derived emissions was an average of 4% (from −21 to 25%), which significantly reduced the uncertainty of the ICAO-estimated 85% (from −23 to 163%). According to the data, the ICAO and NG methods tended to overestimate the aircraft LTO emissions in Chinese airports. The mean relative error (MRE) between the HC, CO, NOx, PM, SO2, CO2 emissions and fuel consumption according to the ALTFEM and ICAO methods were an average of 40.1% (from −60.4 to 28.4%), 38.7% (from −59.3 to 27.0%), 11.6% (from −49.4 to 42.3%), 13.7% (from −48.5 to 31.7%), 14.1% (from −48.5 to 32.5%), 14.1% (from −48.5 to 32.5%), and 14.1% (from −48.5 to 32.5%), respectively. In addition, the NG method dramatically overestimated the HC, CO, NOx, and PM emissions by 85.1%, 39.7%, 45.4%, and 67.3%, respectively (Fig. 3).

Fig. 3: Hourly fuel consumption and emission relative error (RE, %) found using the ALTFEM and the ICAO and the yearly fuel consumption and emission RE values found using the ALTFEM and NG at the 237 airports.
figure 3

(The size of the circle indicates the range of the LTO numbers at the airport; and the mean relative error (MRE) for different pollutants is labeled as a different color at the top of the graph).

Full size image

There were differences in the estimation errors found using the ICAO method for the different airports and modes (Fig. 4a). For the taxi in and out modes, mega airports with higher LTO numbers were underestimated (e.g., PVG, annual LTOs: 252635) and small airports (e.g., JHG, annual LTO: 28920.5) were overestimated. In particular, the underestimated and overestimated emissions of CO were more significant. However, for the climb and approach modes, the PVG airport were instead overestimated, and the JHG airport were underestimated, with a higher estimation error for the NOx emissions. This might have been related to changes in the MLHs at the airports. The MRE for taxi-in, taxi-out, climb, and approach of the 237 airports were an average of 26% (from −44 to 100%), 51% (from −72 to 9%), 17.2% (from −76 to 58%), and 52% (from −48 to 171%), respectively.

Fig. 4: Comparison and uncertainty analysis of aircraft emission estimates at major airports.
figure 4

a The NOx, CO, and CO2 emissions found using the ALTFEM, ICAO, and NG for the Shanghai Pudong International Airport (PVG) and the Xishuangbanna Gasa International Airport (JHG). b The daily and hourly NOx emission changes and the range of uncertainty in the 95% prediction intervals for the Beijing Capital International Airport (PEK).

Full size image

The ALTFEM captured the hourly varying characteristics of emissions better than the ICAO method (Fig. 4b). The study also showed the hourly NOx emission variations for typical periods, with January, April, July, and October representing the different seasons. For instance, when MLHs were lower in winter46, the overall LTO emissions were also lower.

Analysis of the fuel conservation and emission reduction during taxi

Shortening the taxi time by optimizing the taxi routes is a more feasible way to reduce fuel consumption, carbon emissions, and exhaust emissions during the LTO47, as it is difficult to achieve aviation-related fuel replacement within a short period of time48,49, and the taxi mode is the major contributor to LTO fuel consumption and emissions50. We analyzed the hourly emission reduction potentials of the different airports during taxi based on a synergistic consideration of the total fuel consumption (\({kg}\)) and the unit LTO-based fuel consumption (\({FC}\), \(kg\cdot {{LTO}}^{-1}\)) using the ALTFEM estimation results. The Mitigation Potential Index, which can be calculated using the methodological framework shown in Supplementary Fig. 5, has been proposed to assess mitigation potential.

The airport-specific total annual and unit LTO-based fuel consumption during taxi of the 237 airports were standardized and analyzed using the quadrant diagram (Fig. 5a). It was found that although mega airports with high LTOs generally showed higher total fuel consumptions, the unit LTO-based fuel consumptions of the different airports showed significant differences. For example, the PVG, PEK, and CAN airports had comparable total fuel consumptions, the unit LTO-based fuel consumption of CAN was significantly lower than that of the other two airports, with annual unit LTO-based fuel consumption 34.3–34.6% lower than the others. The difference in the \({FC}\) of taxi mode was driven by the aircraft types of the LTO and the ground taxi operating time. Actually, the fleet composition at the CAN airport had a relatively lower aircraft type-specific fuel flow, and the average hourly taxi duration was also shorter, which might have been influenced by the airport’s runway configuration, traffic congestion, and ground taxi management. This indicated that there is still potential for improvements in the aircraft taxi emissions at airports with higher \({FC}\) values (e.g., PVG and PEK).

Fig. 5: Synergistic relationships and reduction potential of taxi-related fuel consumption and emissions across airports.
figure 5

a The airport-specific synergistic relationship between the total yearly and unit LTO-based fuel consumption during taxi of the 237 airports and the corresponding changes with a 1-min taxi time reduction (The horizontal and vertical axes of the quadrant diagram are the standardized index). b The correlation between the changes in the unit LTO-based fuel consumption and emissions and the LTO under a 1-min taxi time for the 237 airports. c The airport-specific reduction potential on an hourly-scale using the examples of the Shanghai Pudong International Airport (PVG), the Beijing Capital International Airport (PEK), and the Guangzhou Baiyun International Airport (CAN).

Full size image

Given that decreasing taxi time will reduce fuel consumption and carbon and pollution emissions to some extent, we used a 1-min reduction in taxi time at the different airports to better understand the role of taxi time on the LTO51,52,53,54,55. The re-estimated results showed that the total and unit LTO-based fuel consumption and emissions had differential reduction driven factors as taxi time decreased. The reduction in the total fuel consumption and emissions was primarily driven by the LTO numbers, with mega airports showing a more pronounced reduction. However, the unit LTO-based fuel consumption and emissions under the 1-min taxi time were driven by the fleet composition. Mega airports with higher LTOs (e.g., PEK) had a higher impact than small airports with relatively shorter taxi time on the reduction of unit LTO-based fuel and emissions, of which mega airports had a positive correlation with the LTO and small airports fluctuated (Fig. 5b). At airports with the most significant mitigation effects, such as PEK, if the taxi time was reduced by 1 min, the unit LTO-based fuel consumption declined approximately 41.1 kg/LTO.

Furthermore, the airport-specific reduction potential was identified and analyzed on an hourly-scale, using the PVG, PEK, and CAN airports as examples (Fig. 5c). We found that although the total fuel consumption during the high-traffic period was 2.4–2.9 times higher than that of the low-traffic period, A higher mitigation potential was found during the low-traffic period. Fewer aircraft queues during the low-traffic period indicated that longer taxi distances may result in increased taxi time. For PVG, the total fuel consumption during the low-traffic period was only one-third that of the high-traffic period. However, the unit LTO-based fuel consumption (\(kg\cdot {{LTO}}^{-1}\)) and unit LTO-based fuel consumption per minute (\(kg\cdot {{LTO}}^{-1}{\cdot }{\min }^{-1}\)) were 1.1 and 1.2 times higher, respectively. This indicated that aircrafts operating during low-traffic periods may have higher fuel flows, and airport ground taxi routes management should be further optimized during low-traffic periods (nighttime) to improve the efficiency of the taxi. In addition, mitigation strategies should be tailored to each airport’s unique taxiway and management circumstances.

ALTFEM promotion schemes for the multiple available data scenarios

The ALTFEM model is versatile and can be promoted to multiple available data scenarios. We have provided the usage guideline of four promotion schemes (Fig. 6) for calculating aircraft LTO emissions using the ALTFEM in Supplementary Note 1.

Fig. 6
figure 6

Flow chart for establishing and applying the ALTFEM and the emission factors (kg/LTO) for multiple available data scenarios.

Full size image

Considering different data foundations, it can be customized for regional-specific needs by inputting original modeling data and estimating emissions. We also unfolded readily available databases based on different forms to provide multi-option support for users (Supplementary Tables 2 and 3). Additionally, the most practical method to estimate aircraft LTO emissions is using the unit LTO-based emission factor of the LTO cycles (\({EF}\), kg/LTO) method. However, significant differences in the \({EF}\) for different airports and temporal resolutions (Supplementary Fig. 6) and significant disparities ( − 60–42%) between the NG and ALTFEM-recommended emission factors indicated the need to revise the NG, which only provides \({EF}\) values for different pollutants. Therefore, we proposed an airport-aircraft type-specific and airport-specific \({EF}\) for the multiple available data scenarios. Note that the yearly and monthly emission factors may not fully reflect the effects of diurnal variations in airport operations, and detailed quantitative error analysis of these schemes is presented in Supplementary Note 1.

The ALTFEM model provides not only a flexible tool for emission estimation across diverse data environments but also a valuable decision-support framework for climate-related aviation policy. By producing airport-specific, time-resolved LTO emission inventories, ALTFEM enables policymakers to identify high-emission airports, time periods, and aircraft types, which can guide the prioritization of mitigation efforts (e.g., implementing low-emission taxi technologies, optimizing flight scheduling, or targeting specific aircraft categories). Furthermore, ALTFEM outputs can be directly integrated into air quality models and emission reporting systems, supporting compliance with international frameworks and national carbon inventory protocols. This positions ALTFEM as a scientifically grounded and policy-relevant tool to support the aviation sector’s transition toward low-carbon development.

Model sensitivity testing and uncertainty analysis

The frequently-used Monte Carlo method was applied to quantify the uncertainty of the time, fuel consumption, and emissions calculated using the ALTFEM. The uncertainty ranges were obtained from 20,000 Monte Carlo simulations with a 95% prediction interval.

The uncertainty of taxi time was primarily attributed to calculation processes from Nq to Ns, and the uncertainty ranges were confirmed by comparing the difference between the estimated and actual taxi time (Supplementary Fig. 7a). The airport-specific (PEK) uncertainty ranges at given hour (12:00) in 95% prediction intervals were [241, 2959 s] for taxi-in and [66, 1471 s] for taxi-out. In addition, the climb and approach time uncertainty was measured using the actual flight heights and time from the AMDAR (Supplementary Fig. 7b). The airport-specific (PEK) uncertainty ranges in the given month (January) were [4, 398%] for the climb and [4, 273%] for the approach.

The flight-specific emissions uncertainties were calculated by examining 100 flights at the PEK airport (Supplementary Fig. 7c). Within the 95% prediction intervals, the uncertainties for CO2, NOx, and CO for each flight had ranges of −59–59%, −79–87%, and −61–62%, respectively, during departure and −61–62%, −69–71%, and −69–72%, respectively, during arrival. As flights increase, the uncertainty range will shrink considerably. We further estimated the hourly uncertainties for CO2, NOx, CO, HC, PM, and SO2 emissions during the entire year, and they were found to be within an average of −6–6%, −10–10%, −10–10%, −10–10%, −6–6%, and −6–6%, respectively.

Discussion

This study primarily aimed to improve the accuracy of aircraft emissions estimates during LTO by calculating the dynamic time-in-mode for each flight, which is crucial to investigate the impact of aircraft emissions on local air quality and human health. Fuel consumption was considered a secondary objective, as it directly underpins emissions and provides additional insight into aviation-related energy use and carbon intensity.

An integrated model, the ALTFEM, was developed to accurately estimate the aircraft fuel consumption and emissions during the LTO. The novelty of the method was its ability to dynamically calculate the time-in-mode for each flight compared to the ICAO-based method with constant parameters. The Ns-based taxi in/out time calculation method and the MLH-derived climb/approach time estimation method were proposed. Significant discrepancies between the airport-specific ALTFEM-estimated and ICAO-referenced time-in-mode were revealed. Compared with reality, the Ns-based taxi in- and out-time improved by an average of 30.2% and 118%, respectively, compared to the results of the ICAO method. In addition, the driving factors of the airport-specific differences in the hourly taxi in/out and monthly climb/approach time were elucidated.

The ALTFEM is an easy-to-use and promoted model because of the broad applicability of its methodological framework and the unit LTO-based emission factors proposed in this study for multiple available data scenarios. In particular, the ALTFEM was applied for 237 airports in mainland China in 2019 to estimate the refined aircraft LTO fuel consumption and emissions inventory. This can help to identify the aircraft LTO-related fuel consumption, CO2 emissions, and pollutant emissions in China, as well as their spatial and temporal characteristics. A major benefit of using the ALTFEM is that it considers the effect of runway congestion on increasing taxi time at each airport, which can provide new insights for fuel conservation and emission reductions during taxi time. The emission reduction potential of taxi time was quantified using the hourly unit LTO-based fuel consumption and the 1-min emission reduction scenarios testing. The fuel and emission reductions during taxi time at PVG, with evidence that the low-traffic period should be prioritized, had a 1.2 times higher mitigation potential. This could inspire airport emissions regulation policies.

This study combined actual MLHs and taxi time data to quantify the uncertainty, which helped improve the accuracy of emission estimations. However, some limitations remained due to data and technical restrictions. The certificated engine emission indices derived from the engine manufacturers and reported in the ICAO failed to consider the life expectancy of an aircraft and meteorological conditions56. Future research should be conducted on the dynamic emission factors based on the machine age and flight conditions57. Moreover, a few flights operate with a single engine during the taxi-in phase for fuel consumption-saving purposes16, which was not considered in this study, but it has become widespread.

Methods

Methodological framework and data

The methodological framework of the aircraft landing-takeoff time, fuel, and emission model (ALTFEM) is shown in Fig. 7. The aim of this model is to provide accurate and reliable estimates of mode-specific operating time, fuel consumption, and emissions for each flight. The ensemble ALTFEM is an integration of the emission index and fuel flow localizing model, the time-in-mode calculating model, the fuel and emission estimating model, and the database.

Fig. 7
figure 7

Methodological framework of the Aircraft Landing-takeoff Time, Fuel, and Emission Model (ALTFEM).

Full size image

The ALTFEM was developed using a massive database of more than 11,000,000 records of landing-takeoff information for domestic and international flights at 237 airports in mainland China in 2019. Hourly information was obtained from the Variflight real-time flight performance database (http://www.variflight.com/) and included approximately 90 aircraft types. We also accessed the actual taxi in/out time in collaboration with Variflight based on the on/off block time, as tire manufacturers are required to assess the extent of tire wear and count the taxi time of tires in contact with the ground. On average, 45% (from 3 to 85%) of the actual taxi time was available from different airports due to an incomplete collection of statistics. Additionally, the system used 770,000 pieces of Aircraft Meteorological Data Relay (AMDAR) and continuously observed MLHs from over 110 monitoring stations during the ALTFEM modeling process.

Estimation of fuel, carbon and pollutants emissions

The fuel and emission estimating model in the ALTFEM calculates the fuel consumptions and emissions of CO2 and multiple pollutants from each mode for each flight using the Eqs. (1) and (2).

$${Fuel}=\sum _{l}({{FF}}_{m,l}\times {T}_{m,l})$$
(1)
$${E}_{j}=\sum _{l}\sum _{m}{({EI}}_{m,l,j}\times {{FF}}_{m,l}\times {T}_{m,l})$$
(2)

where \({Fuel}\) is the fuel consumption during the LTO (kg); \({E}_{j}\) is the emissions of CO2 and pollutant \(j\) (NOx, HC, SO2, CO, and PM) from all domestic and international flights in mainland China (g); \({{EI}}_{m,l,j}\) is the aircraft type-specific (~90 aircraft types) emission index in mode \(m\) (climb, approach, taxi-in, taxi-out, and takeoff) of the LTO \(l\) of CO2 and pollutant \(j\) (g/kg); \({{FF}}_{m,l}\) is the aircraft type-specific fuel flow in mode \(m\) of the LTO \(l\) (kg/s); \({T}_{m,l}\) is the time in mode \(m\) of the LTO \(l\) (s), and \({T}_{{takeoff}}\) = 42 s, the ICAO-suggested value was used due to the very short takeoff time.

The emissions can also be estimated using the Chinese National Guideline (NG) method24 based on the emission factors (kg/LTO) (See Supplementary Table 4).

Localized emission index and fuel flow estimation

The aircraft type-specific emission index (\({EI}\)) and fuel flow (\({FF}\)) datasets were constructed for the localized Chinese fleet. Nearly 90 aircraft types were included, consisting of regional aircraft, such as the CRJ, ERJ, and ATR, and approximately 13 other types. To our understanding, this is the most comprehensive dataset of \({EI}\) and \({FF}\). Compared with current studies20,25,58 of Chinese airports that consider emission factors for 10–25 main aircraft types (Supplementary Table 5), the comprehensive dataset reduced the uncertainty in emission estimates.

Moreover, we obtained the engine-specific \({EI}\) and \({FF}\) test results of the engine from the ICAO Engine Emissions Databank (EEDB, https://www.easa.europa.eu/), which are derived from standardized certification tests performed under controlled conditions simulating real-world engine operations. Therefore, \({EI}\) reflects combustion efficiency, while \({FF}\) represents engine thermodynamic performance59. The matching information23 between aircraft and engines was investigated and collated based on civil aviation fleet information in China (http://www.xmyzl.com/).

The \({EI}\) of an aircraft type in different modes were calculated by Eq. (3).

$$E{I}_{i,m,j}={n}_{i}\times \sum _{k}E{I}_{k,m,j}\times {P}_{i,k}$$
(3)

where \({{EI}}_{i,m,j}\) is the emission index of aircraft type \(i\) in mode \(m\) (g/kg) of pollutant \(j\) (NOx, HC, and CO); \({n}_{i}\) is the number of engines fitted to aircraft type \(i\); \({{EI}}_{k,m,j}\) is the emission index of engine \(k\) in mode \(m\) of pollutant \(j\) (g/kg); and \({P}_{i,k}\) is the proportion of aircraft type \(i\) equipped with engine \(k\).

The engine-specific \({EI}\) of PM was excluded in the EEDB and recalculated using the update method, first-order approximation 3.060. The SO2 emissions were determined by the fuel sulfur content (FSC) instead of the engine performance61,62. This study used the localized FSC according to the national standard in China (GB6537, 2006)63. The FSC of aviation jet fuel No. 3 is 0.2%. Assuming 96.7% sulfur is converted to SO2 after a complete fuel burn, the engine \({EI}\) of SO2 was calculated using Eq. (4).

$${{EI}}_{{{SO}}_{2}}=2\times {FSC}\times \eta \times {10}^{3}$$
(4)

where \({{EI}}_{{{SO}}_{2}}\) is the engine \({EI}\) of SO2 (3.868 g/kg fuel); FSC is 0.2%; and η is the conversion ratio to sulfur dioxide (96.7%).

For CO2, its emissions are dependent on fuel only, and the emission index was considered to be fixed based on previous studies17,64, which was 3.150 kg CO2/kg fuel. Therefore, the carbon emissions can be estimated based on the fuel consumption derived from Eq. (1) and the corresponding emission factor.

The \({FF}\) of an aircraft type in different modes were estimated with Eq. (5).

$$F{F}_{i,m}={n}_{i}\times \sum _{k}{{FF}}_{k,m}\times {P}_{i,k}$$
(5)

where \({{FF}}_{i,m}\) is the fuel flow of aircraft type \(i\) in mode \(m\) (kg/s); \({{FF}}_{k,m}\) is the fuel flow of engine \(k\) in mode \(m\) (kg/s); and the definitions of other parameters are similar to those used in Eq. (3).

Dynamic time-in-mode calculation

The ICAO’s constant time-in-mode resulted in significant uncertainty in the emission estimates37. We calculated the dynamic time-in-mode for each flight using the ALTFEM, which considered actual traffic on taxiways and MLH variations65.

Taxi in- and out-time calculations

We developed an hour-specific and airport-specific model for estimating the dynamic taxi time. The taxi-in time is estimated based on the arrival schedule and Ns of the arriving aircraft, while taxi-out time is based on the departure schedule and Ns of the departing aircraft. The calculation of the Ns-based taxi time was formulated using a linear equation, as shown in Eq. (6).

$${T}_{{taxi}}=\Delta T\times {N}_{s}+{T}_{0}$$
(6)

where \({T}_{{taxi}}\) is the simulated taxi in/out time (s); \({N}_{s}\) is the hourly number of arrival/departure aircraft on the schedule; \(\Delta T\) is the slope of the linear model that represents the taxi in/out time increased per Ns (s); and \({T}_{0}\) is the intercept of the linear model that represents the initialization of the taxi in/out time without any aircraft in the queue (s), defined as the actual aircraft movement between wheel chock removal and takeoff, or from landing to wheel chock application.

A significant correlation (P < 0.001) between the hourly actual taxi in/out time and the number of aircraft in the queues (Nq)66,67,68,69,70 was found (Supplementary Fig. 8) using a correlation analysis of the different temporal resolutions (Supplementary Fig. 9). The Nq represents the flight-specific number of aircraft prepared for departure, calculated based on the recorded arrival/departure time and taxi-in/taxi-out durations. However, the Nq calculation has some limitations as it heavily relies on taxi time. As a result, we defined the number of aircraft in the schedules (Ns) calculated from the hourly Nq, as shown in Supplementary Fig. 10. Ns is estimated from the hourly maximum Nq, and the corresponding weighted average taxi time is calculated using the number of aircraft associated with each Nq value as the weighting factor. Ns is the number of aircraft taking off/landing at a given airport in a given hour, indicating the cumulative flow of the hourly departures/arrivals of aircraft.

To ensure the availability and reliability of the input data, the actual taxi time collected was pre-processed. There are two main steps, one is processing the outliers, and the other is supplementing the unavailable actual taxi time data.

  1. (1)

    Outliers Removal. Since the existing outliers (see Supplementary Note 2 for details) would have a negative impact on the establishment of the taxi out/in time functional relationships, the acquired actual taxi time underwent a standardization process using a standardized residual method to remove the outliers ( ~ 1% of the total data), as shown in the following Eq. (7).

    $${z}_{{ei}}=\frac{{e}_{i}}{{s}_{e}}$$
    (7)

    where \({z}_{{ei}}\) is the standardized residual of the \(i\) sample; \({e}_{i}\) is the residual of the \(i\) sample, representing the difference between the real and the estimated taxi out/in time; and \({s}_{e}\) is the standard error of the residual of all samples. By retaining samples with a standardized residual of [-3,3], we could retain approximately 99% of the samples for the functional relationship building.

  2. (2)

    Supplementation of unavailable actual taxi time. To address the unavailable actual taxi time, we preliminary supplemented them for different airports according to the known exact taxi time. The airport-specific probability density functions were calculated based on the distribution of available taxi time, employing non-parametric kernel density estimation. The unavailable taxi time was then supplemented using the Monte Carlo random sampling method. This approach preserves airport-specific distributional characteristics and avoids introducing systematic bias, thereby ensuring the robustness of the model even under conditions of low data availability.

After pre-processing the actual taxi time data, the Ns-based taxi in/out time was further estimated using the Nq-based taxi time. We quantified the uncertainty in the derivation process using Eq. (8).

$${T}_{{taxi},{Ns},i}=\Delta T\times {N}_{s,i}+{T}_{0}+{\varepsilon }_{4},{\varepsilon }_{4} \sim N\left(0,{\sigma }_{4}^{2}\right)$$
(8)

where \({T}_{{taxi},{Ns},i}\) represents the Ns-based taxi in/out time of aircraft \(i\) (s); \({\varepsilon }_{4}\) is the random error; and \({\sigma }_{4}\) is the standard deviation of the random error, \({\sigma }_{4}=\sqrt{{{\sigma }_{1}^{2}+{(\Delta T}_{{Nq}}\times {k}_{2}\times {N}_{s,i})}^{2}}=\sqrt{{\sigma }_{1}^{2}+{\sigma }_{3}^{2}}\). \({\sigma }_{1}\) is the standard deviation of the random error in estimating \({T}_{{taxi},{Nq},i}\); \({\sigma }_{3}\) is the standard deviation of the random error in estimating \({T}_{{taxi},Ns,i}\); the detailed derivation is shown in Supplementary Note 3. The definitions of the other parameters were the same as in Eq. (6).

The estimation scenarios were divided into the following three categories to eliminate the effects of the unavailable actual taxi time data according to the data availability. The airport-specific model parameters (ΔT and T0) can be derived using the following steps.

  1. (1)

    For the case of a high completeness of data within the hour during an entire year, the simulated Ns-based taxi time (\({T}_{{taxi}}\)) was estimated using the Nq-based taxi time. The hourly relationships between the available actual taxi time and Nq were constructed for each airport. Afterward, the \({T}_{{taxi}}\) values were calculated using the weighted mathematical expectation of the Nq-based taxi time corresponding to Ns in each hour.

  2. (2)

    In terms of moments (<20%) without an actual taxi time during a 24 h period, the model-ready parameters of \(\Delta T\) and \({T}_{0}\) were indirectly estimated based on other hours. The equations were airport specific based on the average annual aircraft departures/arrivals number per hour (see Eqs. (9) and (10)).

    $$\Delta T=u\times \exp ({vN})$$
    (9)

    where \(\Delta T\) represents the taxi in/out time increased per Ns (s); \(N\) is the average annual aircraft departures/arrivals number per hour; \(u\) and \(v\) are the airport-specific constants.

    $${T}_{0}=o\times \exp ({dN})$$
    (10)

    where \({T}_{0}\) represents the initialization of the taxi in/out time without aircraft in a queue (s); \(N\) is the average annual aircraft departures/arrivals number per hour; \(o\) and \(d\) are the airport-specific constants.

  3. (3)

    As for airports with no available taxi time data, such as special small airports (e.g., Sansha Yongxing Airport (XYI), Shangrao Sanqingshan Airport (SQD), Zhalantun Genghis Khan Airport (NZL)), the model-ready parameters of constants in Eqs. (9) and (10) can be indirectly calculated. The constants \(u\), \(v\), \(o\), and \(d\) showed a good clustering relationship with the airport class (Supplementary Fig. 11) that was classified by the Civil Aviation Administration of China (CAAC) based on the airfield area class41.

Climb and approach time calculation

The climb and approach time calculation model was constructed in our previous study23 and integrated into the ALTFEM based on the significant correlation (P < 0.01) found between the flight heights and time of aircraft during climb and approach according to the real-time aircraft height data collected by the aircraft navigation system in the AMDAR data. The MLH-derived climb and approach time for each flight was estimated using the actual MLH, which was the daily maximum MLH calculated based on the dry adiabatic method71. The actual MLH was obtained from observed meteorological data or simulations using the meteorological model.

Model evaluation of ALTFEM

The performance and results of the model were considered to be reliable according to validation using actual observations and measurement data.

The taxi in- and out-time calculation model consisted of over 11,000 formulas on an hourly scale, and the statistical significance testing results were acceptable with P < 0.001. The linear regression performed well with an average R2 of 0.914 (0.265–0.999) (Supplementary Fig. 12).

We compared the simulated taxi time according to the ALTFEM with the actual taxi time. The hourly results for the SHA, PEK and PVG airports in December were selected as a case study (Supplementary Fig. 13). From the actual scenario, the ground taxi time showed continuous fluctuations72, and the ALTFEM provided a better characterization of the hourly changes in the taxi time than the ICAO default. Additionally, we calculated the mean absolute error (MAE) and mean absolute percentage error (MAPE) between the ALTFEM and actual taxi-in (56 s and 15.4% on average) and taxi-out (127 s and 23.0% on average) time for the 237 airports (Fig. 8). The MAE between the ALTFEM and actual taxi in- and out-time were 75 s and 462 s, respectively, lower than that of between the ICAO and actual values, and the MAPE values were of 30.2% and 118%, respectively.

Fig. 8: Accuracy evaluation of estimated taxi time compared with actual values.
figure 8

a The MAE (s) and MAPE (%) between the ALTFEM-estimated and actual taxi in/out time and b between the ICAO-referenced and actual taxi in/out time.

Full size image

We simulated the MLH in 2019 using the Weather Research and Forecasting Model (WRF). The detailed WRF setting and evaluation can be found in Supplementary Note 4 and Supplementary Fig. 14a. The observed MLH values obtained from the vertical meteorological-sounding data were used to validate the simulated MLH values23. The observed and simulated MLH disparities were considered acceptable, as shown in Supplementary Fig. 14b.