The air traffic data contain waypoint data for all civil flights that traversed the Shanwick Oceanic Area Control Centre (OACC) from 1 January 2016 to 31 March 2021, totalling 2.1 million flights. Flight data include the call sign, operator, origin and destination airports, and the ICAO aircraft type designator. Waypoint data include 4D positions (longitude, latitude, altitude and time) recorded when the flight passes through a series of fixed waypoints along the route and when a climb/descend air traffic command is executed by the pilot in between the fixed waypoints.

There are spatio-temporal heterogeneities in the dataset stemming from two distinct sources: (i) waypoints for westbound flights are only recorded in the Shanwick OACC (10 to 40∘ W), whereas additional waypoints are available prior to entry into Shanwick for eastbound flights (10 to 70∘ W); and (ii) the temporal resolution between the recorded waypoints ranges from 30 to 3600 s. We resample the dataset by performing a great-circle interpolation between waypoints to produce flight segments with a uniform temporal resolution (60 s). Full flight trajectory coverage in the adjacent Gander OACC is approximated by a great-circle interpolation between the first (final) recorded waypoint and the origin (destination) airport, with the subsequent removal of waypoints outside Shanwick and Gander. Therefore, the processed dataset approximates the air traffic activity in the North Atlantic flight corridor between 40–75∘ N and 50–10∘ W. Further information on the NATS dataset is detailed in the Supplement (Sect. S1).

Meteorological and radiation data are downloaded from the ECMWF Copernicus Climate Data Store (ECMWF, 2021). The ERA5 HRES reanalysis is obtained at a 0.25∘ × 0.25∘ horizontal resolution for 37 pressure levels and at a 1 h temporal resolution, whereas the ERA5 10-member ensemble (0.5∘ × 0.5∘ horizontal resolution, 37 pressure levels, 3 h temporal resolution) provides an estimate of uncertainties in the reanalysis (Hersbach et al., 2020) and is used to evaluate uncertainty in contrail forcing due to meteorology (Sect. 2.6). For each waypoint, we use a quadrilinear interpolation to estimate the local meteorological values (Schumann, 2012).

The relative humidity with respect to ice (RHi) determines the contrail persistence and lifetime (Schumann, 1996; Kärcher, 2018). However, existing studies have highlighted that the ERA5 humidity fields are generally weakly supersaturated (RHi ≈100%) and underestimate regions with very high supersaturations (RHi >120%) (Gierens et al., 2020; Schumann et al., 2021; Rädel and Shine, 2010; Reutter et al., 2020; Tompkins et al., 2007). We present a comparison of the ERA5 humidity fields with in situ measurements from the In-service Aircraft for a Global Observing System (IAGOS) database (Petzold et al., 2020; Boulanger et al., 2018) in Sect. S3, and our analysis supports these earlier findings (Fig. S9a). To address these issues, we scale the ERA5 humidity fields throughout the domain by division of coefficient a; data points within ISSRs (RHi >100%) are then scaled up using a power-law function with coefficient b: 1RHiCorrected=RHia,whenRHia≤1minRHiab,1.65,whenRHia>1.

The coefficients (a=0.9779 and b=1.635 for the ERA5 HRES; a=0.9666 and b=1.776 for the ERA5 10-member ensemble) are found by minimising the Cramér–von Mises test statistic (Parr and Schucany, 1980), a measure of the goodness of fit between two empirical distributions, between the probability distribution of RHiCorrected and that of the IAGOS observations in the spatial domain of the air traffic dataset in 2019, comprising 262 flights and 43 919 data points. Further details are included in Sect. S3. We explore the sensitivity of contrail predictions to meteorological data in Sects. 2.6 and 3.5.

The BADA Family 3 and 4 aircraft performance models are used to simulate the physical forces that act upon an aircraft and the resultant fuel consumption (EUROCONTROL, 2016, 2019). BADA 3 covers over 1400 aircraft types, whereas BADA 4, which improves upon BADA 3 by including proprietary manufacturer information and accounting for the compressibility and wave drag (Nuic et al., 2010), covers 105 aircraft–engine combinations. The simulated fuel mass flow rate (m˙f) and net thrust are used to estimate the overall propulsion efficiency (η), which influences contrail formation (Schumann, 1996).

We use BADA 4 where possible, covering 91 % of flights in the NATS dataset, and BADA 3 for the remaining flights. BADA also provides a range of mass values for each aircraft type. As aircraft mass is not known, we set the initial aircraft mass (at the first waypoint) to the nominal (reference) value for flights occurring before April 2020. For flights affected by the COVID-19 pandemic, from April 2020, we assume a “low” aircraft mass by taking the average of the zero-fuel weight and reference values to account for the reduced load factor. We describe and quantify the sensitivity to these aircraft mass assumptions in Sects. 2.6 and 3.5.

An estimate of the nvPM EIn and NOx emissions index (EINOx) requires inputs of m˙f and engine-specific data. Here, the m˙f and aircraft–engine assignments are obtained from BADA, and the ICAO EDB provides the pressure ratio, nvPM EIn and EINOx for available engine types at the four certification test points (EASA, 2021).

Three approaches are used to estimate the nvPM EIn at cruise: (i) a new approach which utilises nvPM EIn measurements from the ICAO EDB to perform a linear interpolation relative to the non-dimensional engine power (66.4 % of flights in the NATS dataset); (ii) the fractal aggregates model that was used in earlier studies (Teoh et al., 2020b, a, 2019) to estimate the nvPM EIn from the mass emissions index (Stettler et al., 2013; Abrahamson et al., 2016), particle size distribution and morphology (33.3 % of flights); and (iii) a constant nvPM EIn of 1015 kg-1 when data on the aircraft–engine pair are not available (0.3 % of flights). Further methodological details are described in Sect. S2. Method (i) is the preferred approach because it captures the distinct emissions profile from different engine types (Sect. S2.1). However, the new ICAO EDB nvPM database only provides data for 47 aircraft–engine pairs, thereby necessitating the use of method (ii) or (iii). Method (ii) was formulated based on the emissions profile of singular annular combustors. In a comparison against cruise measurements from Voigt et al. (2021), we found good agreement for the nvPM EIn estimated using method (i) for one aircraft–engine pair (Sect. S2.3). While limited, this indicates that the emissions indices that are corrected for system line losses are most appropriate.

We estimate the cruise EINOx using the Fuel Flow Method 2, where EINOx from the ICAO EDB is fitted linearly on a log–log scale and interpolated using the equivalent m˙f at sea level (DuBois and Paynter, 2006). The CO2 emissions are estimated with a constant emissions index of 3.159 kg CO2 kg-1 (Wilkerson et al., 2010).

CoCiP is used to evaluate the life cycle of contrails produced from individual flights (Schumann, 2012). If two consecutive flight waypoints satisfy the SAC, a contrail segment is formed. A Runge–Kutta integration simulates the contrail evolution with time steps of 1800 s until its end of life, which is defined as the point when the ice number concentration is lower than the background ice nuclei (<103 m-3), τcontrail is less than 10-6 or the maximum contrail lifetime of 24 h is reached (Schumann, 2012). Further information on CoCiP can be found in the literature (Schumann, 2012; Schumann et al., 2012).

The initial contrail properties are dependent on the nvPM EIn (Kärcher, 2018; Schumann, 2012). For this study, we calculate the nvPM EIn for different aircraft–engine types (Sect. 2.4) and set a lower bound of 1013 kg-1 to account for the activation of organic volatile particles and ambient natural aerosols (Kärcher, 2018). We have also used the latest evidence from in situ measurements of contrails that indicates that the activation of nvPM particles to form contrail ice particles is not complete when ambient temperatures (Tamb) are near the SAC threshold temperature (TSAC) (Bräuer et al., 2021a). Specifically, the proportion of nvPM that activates to form contrail ice crystals is calculated as follows:

2pactivation=-0.661edTSAC+1,ifdTSAC<0wheredTSAC=Tamb-TSAC.

Equation (2) asymptotically approaches unity and attains values > 0.99 for dTSAC<-4.2K.

CoCiP simulates the unique properties and shape of individual contrail segments over time (Schumann, 2012), and it accounts for overlapping of contrails above or below clouds present in the meteorological data (Schumann et al., 2012) (Fig. S17 in the Supplement). The model is set up in its original form without humidity exchange between contrails and the background air and without radiative effects caused by contrail–contrail overlapping. We acknowledge that the lack of atmospheric interaction and feedback is a limitation of CoCiP relative to general circulation models (Burkhardt et al., 2018; Chen and Gettelman, 2013; Bickel et al., 2019) and is a topic for further research. Previous studies that approximated the contrail–atmosphere humidity exchange in CoCiP have since found a 10 %–20 % reduction in τcontrail and contrail net RF (Schumann et al., 2015, 2021), while a parametric analysis suggested that the effects of contrail–contrail overlapping on the net RF is small (∼0.3 mW m-2) in regions such as the North Atlantic and can likely be neglected (Sanz-Morère et al., 2021).

The contrail outputs are saved at waypoint, hourly, flight and gridded levels (Schumann, 2012; Teoh et al., 2020b). Contrail cirrus and natural cirrus cover in a grid cell are assumed when their respective optical depth (τ) values are above a threshold of 0.1, which corresponds to the satellite detectability threshold (Mannstein et al., 2010). Contrail cirrus coverage (as a percentage of sky area) in a region is defined as the total cirrus cover minus the natural cirrus cover (Schumann, 2012). The local contrail RF (RF', instantaneous change in energy flux per contrail area) (Schumann et al., 2012) of each waypoint is aggregated to obtain the contrail RF in the domain, and the annual contrail ERF is approximated using an ERF / RF ratio of 0.42 (Lee et al., 2021). The contrail energy forcing (EFcontrail) from individual contrail segments and flights is calculated as follows:

3EFcontrail=∫0TRFnet′t×L(t)×W(t)dt, where L, W and T are the contrail length, width and lifetime respectively. Essentially, the RF', RF and ERF (in W m-2) quantify the instantaneous changes in the radiative energy balance at one point in time over a defined spatial domain, whereas EF (in J) provides the cumulative radiative effect of individual contrail segments.

The simulated contrail properties and climate forcing are sensitive to various input parameters, assumptions and model processes. To perform an uncertainty and sensitivity analysis on our results, we re-run the simulation with five distinct set-ups: (i) the ERA5 10-member ensemble is used to evaluate the uncertainty in contrail properties and forcing due to uncertainty in meteorology and radiation (Hersbach et al., 2020); (ii) a simulation where no corrections are applied to the ERA5 HRES humidity fields; (iii) a simulation to evaluate the sensitivity to EIn, in which we assume a constant nvPM EIn of 1015 kg-1 for all waypoints; (iv) two simulations where we assume a respective “low” aircraft mass (average of the reference and zero-fuel weight) and “high” aircraft mass (average of the reference and maximum take-off weight) for all flights at the initial waypoint; and (v) a simulation to test the sensitivity to the soot activation near threshold conditions by assuming that all nvPM particles activate to form contrail ice crystals (i.e. pactivation=1). Due to computational resource constraints, the uncertainty and sensitivity analysis is conducted for 2019 only.

Table 1 summarises the annual air traffic, emissions and contrail statistics in the North Atlantic region. Between 2016 and 2019, we found the following: (i) the growth in total fuel consumption and CO2 emissions (+3.13% yr-1) outpaced the total flight distance (+3.05% yr-1) due to the use of larger aircraft types (mean aircraft mass, +1.3 % yr-1) (Fig. S6); (ii) total NOx emissions grew by +4.5 % yr-1, which is likely attributable to the higher combustion pressure and temperature in more fuel-efficient engines (Kyprianidis and Dahlquist, 2017; Freeman et al., 2018); (iii) the mean nvPM EIn increased by +0.5 % yr-1; and (iv) the contrail cirrus net RF showed significant inter-annual variability (up to ±19% relative to the mean, ranging between 204 and 280 mW m-2), consistent with Wilhelm et al. (2021), indicating a stronger dependence on meteorology than the total flight distance (Fig. S12).

Comparing statistics during the COVID-19 period (1 April 2020 to 31 March 2021) with the prior 1-year period, we find that the reductions in total fuel consumption and CO2 emissions (-71%) are greater than the reduction in total flight distance (-66%), partly because we assumed a lower aircraft mass to account for lower passenger load factors resulting from COVID-19 travel restrictions (Sect. 2.3). The proportion of the flight distance forming persistent contrails (pcontrail) during the COVID period was higher relative to pre-COVID (16.3 % vs. 14.7 %), partly due to the use of more efficient aircraft types with a higher mean η (+3.1%), which facilitates contrail formation. There is a 66 % reduction in the annual contrail cirrus net RF, but the reduction in SW RF (-74 %) is larger than that in LW RF (-70 %) for reasons that are consistent with recent COVID studies (Gettelman et al., 2021; Schumann et al., 2021): reductions in total flight distance and contrail cirrus cover were largest during the spring months (-80% and -83% relative to pre-COVID respectively) at a time when the contrail SW RF tends to be at its maximum (Fig. 1f). We also simulated contrails for the COVID-19 period with pre-COVID traffic to approximate the likely contrail climate forcing under normal traffic conditions; in this scenario, the annual contrail cirrus net RF increased from 69.6 mW m-2 (actual COVID scenario) to 235 mW m-2, which is 15 % higher relative to the pre-COVID period (204 mW m-2).

On average (from 2016 to 2020), around 12 % of all flights in this region accounted for 80 % of the annual EFcontrail (Table 1, Fig. 2); this is approximately 5 times larger than an earlier study which found that 2.2 % of flights over Japan caused 80 % of the total EFcontrail (Teoh et al., 2020b). Several factors are likely to have contributed to this difference: (i) the small sample size of the study over Japan, which only utilised 6 weeks of air traffic data; (ii) the fact that 58 % of all flights over Japan are domestic short-haul flights, meaning that more time is spent in climb and descent, leading to a smaller mean pcontrail (7.2 %) relative to the North Atlantic (16.6 %; Table 1); and (iii) the use of radar surveillance in Japanese airspace which results in more randomised traffic patterns, relative to the organised track structure (OTS) in the North Atlantic where flight paths often traverse the same air parcel and have a smaller variation in flight distances and headings.

The air traffic activity, meteorology and contrail characteristics in the North Atlantic exhibit seasonal patterns (Fig. 1). Air traffic activity peaks in the summer (Fig. 1a), but there is a higher occurrence of persistent contrails during the winter despite traffic levels being 30 % below peak values: pcontrail is higher in wintertime (23 %) relative to summertime (13 %) (Fig. 1b), consistent with the pcontrail derived from in situ measurements (Gierens et al., 1999), and these contrails persist for longer (3.2 in winter vs. 2.6 h in summer) (Fig. 1c). These phenomena can be attributed to seasonal variations in the ISSR coverage (Fig. S13). In winter, the larger horizontal extent of ISSRs increases pcontrail, and the thicker vertical extent of ISSRs reduces the probability of contrails encountering warm/dry air after forming and sedimentation, thereby increasing their lifetime (Agarwal et al., 2022; Kärcher, 2018; Schumann and Heymsfield, 2017). The contrail ice crystal volume mean radius (rice) and τcontrail averaged over all contrail-forming flights in wintertime are around 25 % smaller relative to the summer (Fig. S14g, h), and these are likely caused by seasonal variations in the tropopause and temperature (Hoinka et al., 1993; Lewellen, 2014): a higher proportion of flights cruise above or close to the tropopause in wintertime, due to the lower tropopause height (Fig. S13), and contrails are formed with less condensable water (Fig. S14d), whereas a higher tropopause height in the summer and the moist upper troposphere facilitate the formation of contrails with larger ice water content, rice and τcontrail.

An earlier study highlighted that contrails that form over the North Atlantic can exhibit a net cooling effect under cloud-free conditions (Sanz-Morère et al., 2021). However, our analysis of the hourly mean contrail cirrus net RF suggests that these cooling periods occur infrequently (∼13 % in 2019; Fig. 3) because around 70 % (summer) to 90 % (winter) of the contrail area overlaps with natural cirrus (Fig. 1d). The mean ERA5 natural cirrus coverage in this region varies between 40 % (summer) and 59 % (winter), and the contrail cirrus coverage generally peaks at around 0.7 % in the summer (Fig. 1e) because of the minimum natural cirrus cover and cloud–contrail overlap during this period.

Between 2016 and 2019, the mean contrail cirrus SW RF in springtime is around 16 % larger than summertime (-323 vs. -280 mW m-2 respectively; Fig. 1f) despite air traffic levels being 11 % below the summer peak and both seasons having a comparable mean solar direct radiation (SDR) (∼394 W m-2; Fig. S14i). This can be attributed to summertime meteorological conditions that are generally less favourable for contrail formation (pcontrail of 13 % in summer vs. 17 % in spring) and persistence (mean age of 2.7 vs. 3.0 h respectively). The contrail LW RF is dependent on the surface temperature and outgoing long-wave radiation (OLR), contributing to smaller seasonal fluctuations in the LW RF relative to the SW RF (Fig. 1g). When taken together, the mean contrail net RF peaks during the winter (302 mW m-2) and reaches a minimum in spring (196 mW m-2) (Fig. 1h). Figure 1 also shows outliers in pcontrail, contrail cirrus cover and RF values in April 2017, and this was primarily caused by large intersections between the ISSR and OTS (Fig. S15).

Figure 3a shows the contrail net RF and EFcontrail per contrail distance for each hour of 2019. There is a clear diurnal effect, where the mean net RF is ∼199 mW m-2 during daylight hours, increasing to ∼385 mW m-2 (and up to 2.6 W m-2) around ±1 h of the sunrise and sunset time, and then falls to 293 mW m-2 at night-time (periods where SDR =0; Fig. S16). The peak net RF around sunrise and sunset is partly attributable to the flight scheduling in this region: eastbound and westbound traffic activity is highest at around 03:00–06:00 and 12:00–16:00 UTC respectively (Fig. S5a). At night, the mean net RF is 24 % smaller than the peak values around sunrise and sunset because air traffic is at a minimum between 18:00 and 02:00 UTC, but contrails that persist during these times have the largest EFcontrail per contrail length, which is up to 1 order of magnitude larger than daytime values (Fig. 3b).

The daytime contrail cirrus net RF also exhibits a large variability (Fig. 3a) and can be strongly warming (1086 mW m-2 at 14:00 UTC on 18 September 2019) or cooling (-594 mW m-2 at 10:00 UTC on 28 August 2019). Taking these two 1 h periods as a demonstrative example, they are similar in that the mean RHi and OLR at the contrail waypoints agree within 1.7 % and 6.7 % respectively (Fig. S18a, b). The SDR at 10:00 UTC on 28 August 2019 (net cooling) was 31 % smaller than at 14:00 UTC on 18 September 2019 (net warming) (Fig. S18c), which would work in favour of greater (negative) SW RF for the warming period. However, there was a lower percentage of cloud–contrail overlap on 28 August 2019 (68 % for August vs. 76 % for September; Fig. S17) and, thus, lower mean albedo along contrail waypoints (0.27 for August vs. 0.48 for September; Fig. S18d). Both of these differences were likely caused by a lower occurrence of low-level optically thick water clouds, which ultimately led to the different contrail net RF in the respective periods. A larger cloud–contrail overlap and albedo reduces the SW RF attributable to contrail cirrus (Schumann et al., 2012; Meerkötter et al., 1999; Sanz-Morère et al., 2021) and means that the LW RF dominated at 14:00 UTC on 18 September 2019.

The full dataset contains 1.27 million contrail-forming flights, and contrails formed by 79.7 % of these flights are warming (EFcontrail>0 J). Figure 4 compares the distribution of aircraft, traffic, meteorology and radiation variables for all contrail-forming flights as well as separately for flights with strongly warming contrails (EFcontrail>99th percentile for each year) and strongly cooling contrails (EFcontrail<1st percentile). Here, we show that the differences in EFcontrail can be grouped into four categories: (i) seasonal changes in meteorology and radiation, (ii) time of day, (iii) background cloud fields, and (iv) the nvPM number emissions from different aircraft types.

The ERA5 ensemble spread represents the observation uncertainties (provided to the data assimilation system) and model state uncertainties in the reanalysis (Hersbach et al., 2020). We propagate these uncertainties to estimates of the annual emissions and contrail properties for 2019. Figure 6 shows that uncertainties derived from the ensembles are small for the total fuel consumption, CO2, NOx and nvPM number emissions (within ±0.01% relative to the ensemble mean), whereas the fleet-aggregated contrail properties have larger uncertainty bounds of up to ±8%. Uncertainty in the 2019 contrail cirrus net RF (216–238 mW m-2; Table S6) is smaller than the inter-annual variability between 2016 and 2019 (204–280 mW m-2; Table 1).

The lower spatio-temporal resolution of the ensembles relative to the HRES leads to differences in the simulated contrail properties. Figure 6 shows that the ensemble mean τcontrail is 2.6 % larger than the nominal HRES simulation, which influenced the SW RF (+31 %) more strongly than the LW RF (+12%) (Schumann et al., 2012). Although the range of net RF from the ensembles (216–238 mW m-2) encompasses the nominal HRES value (235 mW m-2), the net RF from 9 of 10 ensemble member is below the HRES value.

To evaluate the consistency in contrail prediction for specific flights, we compared the set of flights with EFcontrail>95th percentile in the nominal HRES simulation with each ensemble simulation (5 % of all contrail-forming flights) and found that 36.9 % of these flights have an EFcontrail>95th percentile in all 10 ensemble members. The characteristics of flights with strongly warming/cooling contrails in each ensemble member (Fig. S21) are generally consistent with the HRES (Fig. 4). However, unlike the HRES, the ensembles do not predict the occurrence of strongly warming or cooling contrails before dawn or dusk respectively (Fig. 4i vs. Fig. S21i). This is likely due to the lower spatio-temporal resolution of the ensembles relative to the HRES, making the former less capable of capturing sub-grid areas where RHi <1, thereby causing an overprediction of the mean contrail age (+7.7 % relative to the HRES; Fig. 6d) and a change in the sign of EFcontrail as contrails persist through dawn/dusk. This also caused the percentage of flights accounting for 80 % of the annual EFcontrail in the ensembles (∼8.6%; Table S6) to be lower than the HRES (12.0 %; Table 1).

To evaluate the sensitivity of contrail results to the ERA5 humidity correction, we simulated contrails for 2019 without any correction (Figs. 6, S22). With no corrections applied, pcontrail decreased from 16.2 % in the nominal simulation to 14.7 %, and there was a 34 % and 8.9 % reduction in τcontrail (0.08 without vs. 0.12 with correction) and the mean contrail age (3.2 h without vs. 3.5 h with correction) respectively. On the other hand, corrections applied to the humidity fields increase saturation above 120 % (Fig. S11), and contrails formed in these regions could have a shorter lifetime than in the simulation without humidity correction (Fig. S25a) as the ice particles would grow to larger rice and, thus, experience a higher sedimentation rate. When taken together, these effects caused the annual mean contrail cirrus net RF (121 mW m-2) to be 49 % smaller than in our nominal simulation (235 mW m-2). Both results are within the range of earlier studies that estimated the North Atlantic contrail cirrus net RF to be between 100 and 360 mW m-2 (Burkhardt and Kärcher, 2011; Chen and Gettelman, 2013; Bock and Burkhardt, 2019; Schumann and Graf, 2013) and confirm that the contrail climate forcing is highly sensitive to the humidity fields.

The assumption of a constant nvPM EIn (1015 kg-1) for all waypoints leads to higher fleet-aggregated contrail properties and climate forcing than in the nominal case. pcontrail remains unchanged because the SAC does not depend on the nvPM EIn (Schumann, 1996); however, the higher initial contrail ice particle number (+25 % relative to the nominal simulation) led to a 3.4 %, 7.4 % and 14 % increase in the mean contrail lifetime, τcontrail and annual mean net RF (267 mW m-2) respectively. When the comparison is made for individual flights, however, a constant nvPM EIn leads to more significant differences in the initial ice crystal number (up to ±3.5×1013 m-1), rice (±14 µm), τcontrail (±0.78), lifetime (±10 h) and net RF' (±30 W m-2) (Fig. S23) compared with the nominal case. These results suggest that the new ICAO EDB nvPM database, which captures the emissions profile for specific aircraft–engine types, is critical in identifying flights with the largest EFcontrail. A separate simulation assessing the aircraft mass assumptions (high/low mass at the initial waypoint), which influence the nvPM EIn and wake vortex dynamics, leads to a ±7% sensitivity in the annual mean contrail cirrus net RF relative to the nominal simulation.

The simulated contrail outputs are less sensitive to pactivation (see Eq. 2), and an assumption of pactivation=1 changes the annual mean contrail cirrus net RF by +0.81% (237 mW m-2) relative to the nominal simulation because only 24.8 % of all flights form contrails near the SAC threshold temperature (dTSAC>-5K). For these flights, the sensitivity to pactivation was also relatively small for other contrail properties, including initial ice crystal number (+14% on average), rice (-3.7%), contrail age (+2.8%), τcontrail (+5.0%) and net RF' (+0.55%) (Fig. S24).