Feasibility of contrail avoidance in a commercial flight planning system: an operational analysis

We use the CoCiP model to estimate contrail climate forcing of individual flight trajectories before and after optimization in FK5D [9, 13]. CoCiP is a parameterized physical model that efficiently estimates formation, persistence, and climate forcing of contrail segments produced by flights. The details of CoCiP are described elsewhere in the literature [6, 10, 39, 40] and available as part of the open-source pycontrails ⁷ library [30]. Figure 2 (3, 4, 6, 7) illustrates trajectory evaluation with pycontrails using the CoCiP model.

Briefly, CoCiP initializes contrail segments when two adjacent flight waypoints satisfy the SAC [4, 41]. Segments are considered persistent if contrail ice particles survive the initial wake vortex phase. Persistent segments evolve in time using a Runge–Kutta integration scheme until the contrail segment reaches an end-of-life criterion. Contrail end of life is defined as when the ice number concentration is lower than the background ice nuclei (${\lt}10^{3}\ \textrm{m}^{-3}$), contrail optical depth (τ_contrail) is less than 10⁻⁶, or the contrail segment reaches a maximum age of 12 h.

FK5D runs CoCiP through the Contrails API ⁸ [29] (figure 2 points 3, 6), a hosted version of pycontrails with built-in meteorology data available through an HTTP interface. This study ran the Contrails API version v0.15.6 with pycontrails version v0.42.1 [30]. This version of the API included ECMWF HRES meteorology forecast data as input to the CoCiP model. At the time of this study, HRES data was available through the Contrails API hourly out to +36 h for pressure levels ⁹ on a $0.1^{\circ} \times 0.1^{\circ}$ latitude-longitude grid. The specific humidity variable of the ECMWF HRES forecast is scaled using the methodology described in [42].

CoCiP uses jet engine emissions, in particular nvPM soot number emissions index (EI_n ), to estimate initial ice crystal activation in the contrail plume. pycontrails implements models for nvPM EI_n emissions based on aircraft, engine type, and aircraft performance as described in [12, 43]. FK5D provides high-fidelity true airspeed and aircraft mass estimates calculated with proprietary coefficients from the airline to override the standard aircraft performance models built into pycontrails (see section 2.1). One exception is overall propulsive efficiency, which is not provided by OEMs datasets. For this metric, FK5D relies on the default aircraft performance relations implemented in pycontrails, described in detail in [43].

To estimate regions that have the potential to form strongly warming contrails, we use a generalized form of the CoCiP model (termed CocipGrid) that calculates contrail climate forcing for arbitrary vector pseudo-waypoints independently [44]. In this formulation, each waypoint represents a nominal flight segment placed at the center of the vector coordinate. Aircraft performance and emissions are estimated assuming a single aircraft-engine combination and nominal cruising conditions at each vector waypoint. Any contrail formed by the nominal segment is evolved as in the original CoCiP model, independent of segments placed at other vector waypoints.

The output of the CocipGrid model is a field of contrail climate forcing per flight distance at each vector waypoint. When the model is evaluated using forecast meteorology, the output can be interpreted as a forecast of expected contrail climate forcing for a specific aircraft flying nominal cruise conditions through a grid cell. While output accuracy is currently limited by known challenges in forecasting upper tropospheric humidity [25, 26, 28], this output effectively emulates the size and distribution of contrail forecast data for the purposes of quantifying cost and avoidance feasibility.

CocipGrid is part of the pycontrails library and used to serve prototypical contrail forecasts through the Contrails API. Forecasts are generated every 6 h out to +36 h using the latest ECMWF HRES meteorology forecast for a set of 10 standard aircraft. To simplify this study, we use forecast data generated for only one aircraft-engine combination (A320). In the future, we expect contrail forecasts to incorporate aircraft-engine dependencies. Forecast data is initially calculated on the native HRES resolution ¹⁰ and then linearly interpolated in altitude to flight levels 28 000 $\textrm{ft}$ (FL280) to 41 000 $\textrm{ft}$ (FL410) in 1000 $\textrm{ft}$ steps and down-sampled to a $0.25^{\circ} \times 0.25^{\circ}$ latitude longitude grid.

Contrail forecasts are served through the Contrails API in two formats: (1) as a NetCDF file containing a single continuous variable of climate forcing per flight distance or (2) as a GeoJSON format containing polygon isolines of climate forcing at each flight level. Polygons are calculated using a threshold of $5\times10^8$ J m⁻¹ chosen based on the results of a global contrail simulation for 2019–2021 [42, 45]. For this simulation, the contrail polygons are calculated and then ingested into FK5D navigation service database as hard restrictions (figure 1 point 1). The threshold chosen for this study reflects the 80th percentile of contrail climate forcing per flight distance in the global simulation [42]. This study does not explore the impact of the polygon threshold on route modification and overall cost, but this is an important area of future work.

This study includes the analysis of 84 839 AAL flights conducted during two seasons: one in summer, encompassing flights departing between 3 and 18 June 2023; and another in winter, covering flights taking place between 3 and 18 January 2024. Of the 49 411 summer flights, 39 691 are domestic flights within the USA, while 9723 are international. Of the 35 429 winter flights, 27 454 are domestic, while 7920 are international.

Flight data includes regular flights to various countries in South America, Europe, and East Asia. The most frequent international destinations include Mexico, the United Kingdom, and the Dominican Republic. More details on the flight dataset are included in appendix. An overview of the origin-destination pairs for domestic and international flights can be seen in figure 3.

Zoom In Zoom Out Reset image size

Contrail avoidance regions are generated every 6 h by the CocipGrid model using ECMWF HRES forecast data (section 2.2) out to +36 h. Every hour, FK5D imports new contrail avoidance polygons as restrictions by calling the Contrails API (figure 2, point 1).

Each evening, FK5D imports flights scheduled for the following day and creates optimal trajectories according to the default customer settings. Flights are first calculated without contrail avoidance as a basis for comparison (figure 2, point 2). After the first calculation, FK5D automatically runs each trajectory through the /trajectory/cocip endpoint of the Contrails API to estimate the EF$_{\mathrm{contrail}}$ for each segment of the route (figure 2, point 3).

The total net EF$_{\mathrm{contrail}}$ of the trajectory is calculated by summing the EF$_{\mathrm{contrail}}$ of each segment (figure 2, point 4). Flights that generate net warming contrails ($\sum$EF$_{\mathrm{contrail}} \gt 0$) are automatically re-routed around the contrail avoidance regions implemented as polygon restrictions (figure 2, point 5). While the optimizer does consider ATM restrictions, the post-calculation filing of flights and subsequent validation of trajectory feasibility is not simulated. Flights were selected for rerouting based on their contrail warming potential [50, 51].

The optimization is performed vertically, laterally, or a combination of both, depending on the most cost-effective solution (figure 4). Once re-optimization is complete, each new trajectory is rerun through the Contrails API to obtain an updated EF$_{\mathrm{contrail}}$ for the modified trajectory (figure 2, points 6–7). The final step involves comparing the cost-optimal and contrail-optimal flights (figure 2, point 8).

Zoom In Zoom Out Reset image size

The simulation employs flight data from Flightkeys' customer, AAL. This data includes origin-destination pairs, flight schedules, payload, cost index, and aircraft types for all simulated flights. To approximate real-world scenarios as closely as possible, aircraft performance metrics used by the airline were imported into our testing environment from the customer's production environment. While the initial flight data import is the same in both environments, the final trajectory flown might differ due to last-minute changes in flight data or routing decisions made by the flight dispatchers after the initial data import.

Upon trajectory calculation, the system evaluates several key performance indicators (KPIs), including trip fuel, measured in kilograms (kg); total cost of operation (including overflight costs), in U.S. Dollars (USD); and flight duration, in minutes. For each trajectory, positive and negative contrail EF values in megajoules (MJ) are retrieved and stored from the Contrails API.

The principal aim of this study is to characterize the trade-offs between these KPIs, EF$_{\mathrm{contrail}}$, and GWP. Ultimately, the analysis seeks to pinpoint scenarios wherein reductions in EF$_{\mathrm{contrail}}$ (and combined EF$_{\mathrm{contrail}}$ + EF$_{\mathrm{CO_2}}$) can be achieved with the least impact on other KPIs.

In this section, we explore two heuristic approaches to identify flights to prioritize for rerouting. The first heuristic aims to minimize EF$_{\mathrm{contrail}}$ with minimal flight modifications (target big hit flights); the second prioritizes maximizing the ratio of EF$_{\mathrm{contrail}}$ change to added cost (target low cost reroutes).

3.2.1.1. Targeting big hit flights

The studies reported in [6, 10, 42] present the concept that a small subset of flights, roughly 2%–10%, cause approximately 80% of the total contrail forcing. The results of this study show a similar lever for targeting flights with outsize climate forcing (section 3.1, figure 5(b)). This approach aims to address the majority of climate benefits while minimizing the number of modified flights and, therefore, operational disruptions. Reducing the total number of modifications reduces the total fuel and time costs as well as the workload for ATC.

One way to implement this concept in practice is to only re-optimize flights with the original $\sum$EF$_{\mathrm{contrail}}$ above a certain threshold. To determine the appropriate value for this threshold, we can sort the original flight data by descending $\sum$EF$_{\mathrm{contrail}}$ and find the flight with the $\sum$EF$_{\mathrm{contrail}}$ where the curve crosses 80% of the total possible reduction. Figure 6 displays the change in total contrail EF, fuel consumption, cost, and time as a function of the original $\sum$EF$_{\mathrm{contrail}}$ sorted from the highest to the lowest.

Zoom In Zoom Out Reset image size

The $\sum$EF$_{\mathrm{contrail}}$ threshold for the flights responsible for 80% of the total warming EF$_{\mathrm{contrail}}$ is $1.27\times10^{14}$ J. There are 2438 flights exceeding this threshold, representing 2.87% of the total flight count. The optimizer was able to identify alternative routes for 2260 of these flights. The results of rerouting this subset of flights are detailed in table 3. Compared to the original set of flights, the total contrail EF is reduced by 65.68% with only 0.05% added fuel and 0.03% added cost.

Table 3. Flight statistics for targeting big hit interventions and comparison with full optimization scenario.

Statistics	Units	Cost optimal	Big hits	Δ (%)
Number of flights	—	84 839	2260	—
Time	103 hr	209.6	209.6	0.00%
Fuel	109 kg (Mt)	0.660	0.661	0.05%
Cost	106 $	1285	1286	0.03%
Total $\mathrm{CO_2}$ emissions	109 kg (Mt)	2.086	2.087	0.05%
Total EF contrail	1018 J	0.874	0.300	−65.68%
Contrail warming in $\mathrm{CO_2}_{\textrm{e},20}$	109 kg (Mt)	0.906	0.311	−65.68%
Total warming, in $\mathrm{CO_2}_{\textrm{e},20}$	109 kg (Mt)	2.986	2.398	−19.85%

Even though in the example, we focus on the top 2.87%, other possible selection strategies could be considered. Rerouting can be performed based on a contrail EF threshold value or on pre-determined geographical or seasonal/hourly criteria (see section 4.1.2), as discussed in [10, 53]. This criterion could be further strengthened by incorporating a probability ratio; the higher the probability of a flight generating contrails, the greater the likelihood of rerouting. The probability could be assessed by comparing the agreement of different models over a specific contrail forecasted area.

3.2.1.2. Targeting low-cost re-routes

While the previous method aims to limit the number of modified flights, an alternate strategy is to target the most cost-effective re-routes. Airlines can set a specific cost threshold and all flights with rerouting costs below this threshold will be selected for optimization.

The first step involves determining the specific cost, measured in USD, per unit of total warming (accounting for both contrails and $\mathrm{CO_2}$) in tonnes of $\mathrm{CO_2}_{\textrm{e},20}$ (T$\mathrm{CO_2}_{\textrm{e},20}$). This specific cost for each flight is quantified as:

$$\begin{equation} \textrm{Specific cost} = \frac{\Delta\textrm{Cost}}{\Delta\mathrm{TCO_{2{\textrm{e},20}}}} \end{equation} \tag{ 4 }$$

Figure 7 shows the reduction in contrail EF as a function of specific cost. Notably, roughly 10% reduction in EF$_{\mathrm{contrail}}$ can be achieved at minimal added cost. Above these low-cost interventions, EF$_{\mathrm{contrail}}$ reductions show a similar trend to targeting bit hit flights, dropping steeply and then leveling off beyond a specific investment. This shape indicates an optimal cost range, where most reduction in EF$_{\mathrm{contrail}}$ is achieved for a moderate investment.

Zoom In Zoom Out Reset image size

The decision on investment allocation rests with airlines and policymakers. In contrast to the previous section, where the focus was on modifying a minimum number of flights, the specific-cost approach focuses on rerouting the most economically efficient flights, thereby minimizing added fuel. As an example, we could apply a specific cost threshold of 0.5 USD per tonne of $\mathrm{CO_2}_{\textrm{e},20}$ to this simulation data. In this scenario, 2438 flights, constituting 2.87% of the total, would be re-routed. The operational and environmental implications stemming from these reroutings are detailed in table 4.

Table 4. Flight statistics for targeting low specific cost interventions for a cost threshold of 0.5 USD/tonne of $\mathrm{CO_2}_{\textrm{e},20}$ and comparison with full optimization scenario.

Statistics	Units	Cost optimal	Low-cost	Δ (%)
Number of flights	—	84 839	2438	—
Time	103 hr	209.6	209.6	0.00%
Fuel	109 kg (Mt)	0.660	0.661	0.02%
Cost	106 $	1285	1286	0.01%
Total $\mathrm{CO_2}$ emissions	109 kg (Mt)	2.086	2.087	0.02%
Total EF contrail	1018 J	0.874	0.293	−66.38%
Contrail warming in $\mathrm{CO_2}_{\textrm{e},20}$	109 kg (Mt)	0.906	0.304	−66.38%
Total warming, in $\mathrm{CO_2}_{\textrm{e},20}$	109 kg (Mt)	2.986	2.391	−20.08%

Both intervention approaches can be configured to accommodate varying levels of risk (climate and economic), enabling airlines to adopt contrail avoidance measures incrementally. When comparing the 2438 flights selected with the specified cost threshold to those identified using the EF$_{\mathrm{contrail}}$ threshold in section 3.2.1.1, we observe an overlap of 1449 flights, showing that the most warming flights do not necessarily need to be the most expensive to re-route. This coincidence could serve as an additional criterion to minimize the risk of escalating operational costs while concurrently targeting flights with the most substantial warming potential. This paradigm motivates more research and trials into the efficacy of low-cost interventions and the level of certainty necessary to justify climate and economic investments into contrail avoidance interventions.

Throughout section 3, we computed the contrail warming potential and the combined contrail and $\mathrm{CO_2}$ warming potential using $\textrm{GWP}_{20}$ of the simulated flights. As discussed in section 2.3, the choice of time horizon depends on the environmental strategy and policy goals. While this study chooses a 20 yr time horizon, other studies often choose a 100 yr time horizon [54].

Table 5 compares the original and re-optimized contrail and total warming potential results with 20 and 100 yr time horizons. Changing the time horizon from 20 years to 100 years reduces the contribution of non-$\mathrm{CO_2}$ emissions to the combined GWP.

Table 5. Comparison between estimated total $\mathrm{CO_2}_{\textrm{e},20}$ and $\mathrm{CO_2}_{\textrm{e},100}$ for original and re-optimized flights.

Statistics	Units	Cost optimal	Contrail optimal	Δ (%)
$\mathrm{CO_2}$	109 kg (Mt)	2.086	2.089	0.11%
Contrail $\mathrm{CO_2}_{\textrm{e},20}$	109 kg(Mt)	0.906	0.245	−72.95%
Contrail $\mathrm{CO_2}_{\textrm{e},100}$	109 kg (Mt)	0.247	0.066	−72.95%
Total $\mathrm{CO_2}_{\textrm{e},20}$	109 kg (Mt)	2.993	2.334	−22.00%
Total $\mathrm{CO_2}_{\textrm{e},100}$	109 kg (Mt)	2.334	2.156	−7.61%

Using $\mathrm{CO_2}_{\textrm{e},20}$, contrails account for 30.3% of the total CO$_{2e}$, while, in 100 years, the contrail contribution drops to 10.6%. The total $\mathrm{CO_2}_{\textrm{e},20}$ was reduced by 22% following the optimization, whereas $\mathrm{CO_2}_{\textrm{e},100}$ saw only an 7.6% reduction. Nevertheless, it is evident that, regardless of the selected time horizon, the warming potential diminishes in both cases despite the increase in $\mathrm{CO_2}$.

Figure 7 shows EF$_{\mathrm{contrail}}$ reduction compared to specific cost for 20 yr and 100 yr time horizons. We see a steep drop in EF$_{\mathrm{contrail}}$ for a 20 yr TH, achieving the majority of reductions for less than 1 USD per tonne of $\mathrm{CO_2}_{\textrm{e},20}$. The 100 yr time horizon puts a larger emphasis on added $\mathrm{CO_2}$ emissions and attenuates the rate of EF$_{\mathrm{contrail}}$ reduction. The trade-off between short-term and long-term emissions is a topic to be explored in future research and policy and beyond the scope of this work.

The results presented in section 3 are a combination of two distinct simulations conducted during the first two weeks of June 2023 and January 2024. The 84 839 total flight include 49 411 summer flights and 35 428 winter flights. The variation in the number of flights reflects the usual seasonal differences in flight operations, with AAL having 39% more flights in June compared to January.

Despite the higher number of flights operated during the summer, the proportion of winter flights forming persistent contrails (28.9%) was notably higher than in the summer (19.9%). Furthermore, 16.0% of winter flights resulted in a net warming effect (12.9% net cooling effect) compared to 12.0% net warming (7.9% net cooling) summer flights. These statistics align with findings from [10, 42], which reported that, despite higher levels of activity during the summer, the occurrence and impact of contrails peak during the winter months. Higher climate forcing in the winter is attributed to larger ISSR coverage in the northern mid-latitudes (30–60^∘N), specific atmospheric conditions that lead to smaller particles with longer lifetimes, a higher percentage of cloud-contrail, and shorter daylight hours.

In spite of these differences, table 6 shows the operational impact of contrail avoidance strategies remains relatively consistent across both seasons. In both cases, the increase in fuel consumption (+0.11% summer, +0.12% winter) and operational costs (+0.06% summer, +0.11% winter) are similar. The most significant difference across seasons is the number of net-warming flights that were unable to find a contrail-optimal solution. For the summer season, 5945 flights generated warming contrails and were selected for rerouting. From those, 5840 flights were successfully rerouted, 1.8% less than what was originally planned. For the winter season, 5675 flights generated warming contrails, but only 5071 found a viable solution to be re-routed. Although the optimizer accounts for all ATM restrictions, the optimized trajectories were not validated by ATM.

Table 6. Flight statistics and climate impact assessment of flights before and after contrail re-optimization separated by season.

		Summer			Winter
Statistics	Units	Cost Optimal	Contrail Optimal	Δ (%)	Cost Optimal	Contrail Optimal	Δ (%)
Total number of flights	—	49 411	49 411	—	35 428	35 428	—
Number of flights re-routed	—	—	5840	—	—	5071	—
Flights forming persistent contrails	—	9835 (19.9%)	8260	−16.0%	10 230 (28.9%)	9214	−9.9%
Flights forming net warming contrails	—	5945 (12.0%)	4159	−30.0%	5675 (16.0%)	4403	−22.4%
Flights forming net neutral or cooling contrails	—	3890 (7.9%)	4101	5.4%	4555 (12.9%)	4811	5.3%
Total flight time	103 hr	120.6	120.6	−0.02%	89.0	89.0	0.02%
Total fuel burn	109 kg (Mt)	0.3795	0.3799	0.11%	0.2809	0.2812	0.12%
Total cost	106 $	741.6	742.1	0.06%	543.8	544.4	0.11%
Total $\mathrm{CO_2}$ emissions	109 kg (Mt)	1.20	1.20	0.11%	0.89	0.89	0.12%
Total EF$_{\mathrm{contrail}}$	1018 J	0.562	0.175	−68.9%	0.312	0.061	−80.3%
Flights responsible for 80% EF$_{\mathrm{contrail}}$	%	1.96	—	—	1.20	—	—
Contrail warming, in $\mathrm{CO_2}_{\textrm{e},20}$	109 kg (Mt)	0.582	0.181	−68.9%	0.323	0.064	−80.3%
Total warming, in $\mathrm{CO_2}_{\textrm{e},20}$	109 kg (Mt)	1.781	1.382	−22.5%	1.211	0.952	−21.4%