During ML-CIRRUS, the German research aircraft HALO was equipped with a comprehensive suite of novel particle measurement sondes and obtained a broad data set of microphysical properties of contrail cirrus and natural cirrus for process studies and climatological analyses. N, Reff, size distributions, ice or liquid water content (IWC and LWC) and extinction are derived from measurements of CAS-DPOL (Cloud and Aerosol Spectrometer with Detector for Polarization) for particles from 3 to 50 µm and CIP (Cloud Imaging Probe) for the size range of 15 and 960 µm (diameter as the maximum dimension). CAS-DPOL measures the forward scattered light of particles when they pass through a laser beam (Baumgardner et al., 2011). The uncertainty of the particle size measurements is ±16 % (Kleine et al., 2018) for the lower size range and can be up to ±50 % for Reff if the shape of the particles is not known. This translates into an error of up to ±100 % for the IWC derived from scattering cloud probes (Baumgardener et al., 2017). Using 64-element linear photodiode arrays, the CIP acquires two-dimensional shadow images of particles (de Reus et al., 2009). The size resolution is 15 µm, with the uncertainty decreasing considerably with diameter, reaching ±15 µm when particles are larger than 50 µm. N from CAS-DPOL, denoted by NCAS in the following, and from CIP (NCIP) are also combined to an overall N between 10-4 and 102 cm-3. Validation of the measurements has been performed, taking atmospheric and cloud chamber measurements into account (Braga et al., 2017a, b).

As for ambient conditions, the AIMS (Atmospheric Ionization Mass Spectrometer; Jurkat et al., 2016; Kaufmann et al., 2016) was applied to measure the actual water vapor concentration from ambient air using a backward heated inlet. The range of detection is between 1 and 500 ppm with an overall accuracy from 7 % to 10 %. Static pressure and temperature, measured by the BAHAMAS (Basic HALO Measurement and Sensor System; Krautstrunk and Giez, 2012) with an accuracy of 0.3 hPa and 0.5 K, were used to convert water vapor concentration to RHi with an uncertainty of 10 % to 20 % (Kaufmann et al., 2018). During the mission, nitrogen oxide (NO) and the sum of all reactive nitrogen species (NOy) were measured using the AENEAS instrument (AtmosphEric Nitrogen oxides mEAsuring System). This dual channel instrument is based on the well-established chemiluminescence technique combined with catalytic conversion of the NOy components. It has been regularly operated on HALO during several missions (e.g., Ziereis et al., 2022). A detector identical in construction has also been operated on a commercial aircraft as part of IAGOS-CARIBIC (Stratmann et al., 2016). The statistical detection limit is about 7 pmol mol-1 for NO. The overall uncertainty depends on the ambient concentration of NO and ranges from about 8 % to 6 % for volume mixing ratios between about 1 and 0.5 nmol mol-1. The time constant of the instrument for measurements in the upper troposphere is of the order of 1 s. The time offsets of NO have been shifted based on ice number.

Backscatter profiles of clouds and aerosol were acquired by the lidar system WALES (Water vapor Lidar Experiment in Space; Wirth et al., 2009) at the wavelengths of 532 and 1064 nm. In this study backscatter is used to extract information about the cirrus cloud structures, such as cloud top height (CTH), geometrical depth and others. WALES also provides 2D measurements of the water vapor mixing ratio and aerosol particle linear depolarization ratio. The backscatter ratio and aerosol depolarization (perpendicular to parallel) (Biele et al., 2000; Esselborn et al., 2008) are used to create a cloud mask, which helps to discriminate ice clouds. For these clouds the RHi is calculated from the measured water vapor mixing ratio and collocated model temperatures from the European Centre for Medium-Range Weather Forecasts (ECMWF). This instrument and method have also been applied by Groß et al. (2014) and Urbanek et al. (2018), who found that the lidar measurements were accurate when compared with in situ data. The statistical error in the retrieval of water vapor by WALES is estimated to be about 5 % (Kiemle et al., 2008), and the ECMWF temperatures induce an error of around 10 %–15 % in the final RHi values (Groß et al., 2014).

SEVIRI is the primary instrument aboard the geostationary MSG satellites, which provides observations of the Earth disk every 15 min from three solar and eight thermal channels with 3 km sampling distance at nadir, as well as its High-Resolution Visible (HRV) channel with 1 km spatial resolution (Schmetz et al., 2002). We use MSG-3/Meteosat-10 observations for the study on 26 March 2014 with a temporal resolution of 15 min.

To calculate broadband solar and thermal irradiances at TOA for ice particles, the sophisticated radiative transfer package libRadtran is used (Mayer and Kylling, 2005; Emde et al., 2016). Water and ice clouds are represented in this model in detail and realistically. Optical properties of water droplets are computed using the Mie theory and tabulated as a function of wavelength and Reff. Ice crystals are not spherical in shape and habits (Letu et al., 2016), and for this simulation the parameterization of Baum et al. (2014) for ice crystal habits has been employed to define the conversion from optical to microphysical properties. In analogy to the MODIS products (Yang et al., 2018), we select rough aggregates for ice crystal shape (see Sect. 4 for a discussion about this choice). The selected one-dimensional radiative transfer solver is DISORT (Discrete Ordinate Radiative Transfer) 2.0 by Stamnes et al. (2000) with 16 streams. LibRadtran recommends the REPTRAN band parameterization with a spectral resolution of 15 cm-1 for spectral calculations (Buehler et al., 2010; Gasteiger et al., 2014).

We exploited positions of ice clouds from ML-CIRRUS measurements and CTHs estimated from WALES observations to collocate the clouds probed by HALO with the SEVIRI observed contrail cirrus and natural cirrus. We notice that satellite observations are functions of both time and space. Since the temporal frequency of in situ observations is 1 Hz, various HALO measurements are located inside each SEVIRI pixel. In general, time series of observations are produced with the original temporal resolution of the given instrument.

In the night and the very early hours of 26 March 2014 a ridge cirrus cloud band built up north of Ireland down to the southern tip of Portugal (Fig. 1). Close to Iceland as well as to the southwest of Ireland, this cirrus cloud thickened (optical thickness larger than 1 at 10:45 UTC, Fig. 1d). While exactly off the coast of Ireland, in coincidence with the eastbound morning and westbound afternoon air traffic to and from the US, many linear structures can be seen in SEVIRI observations (Fig. 1, especially Fig. 1b where brightness temperature differences – BTDs – between the SEVIRI channels at 10.8 and 12.0 µm are shown) and the ice cloud is thinner (optical thickness around 0.3 at 10:45 UTC). Please notice that Fig. 1 is enlarged in Fig. 2. Considering that the peak of eastbound morning air traffic is approx. at 03:00 UTC (Graf et al., 2012), under favorable conditions with low temperature and high humidity contrails induced from these aircraft are expected to form and live for hours such that they can be identified in MSG observations in the morning of the same day. This cloud band evolves with time towards the south, and in correspondence of the Ireland coast the ice clouds dissipate around noon. In addition, this thin bluish/violet ice cloud band partly overlaps with a liquid water cloud field below (yellowish clouds in the false color composite in Fig. 1a).

Due to its approx. 3.5 km × 4.5 km spatial resolution, SEVIRI can only observe contrails that have already grown larger (and thicker), while young contrails cannot be detected (Mannstein et al., 2010). Nevertheless, the satellite observations indicate that because of these favorable meteorological conditions and the relatively high air traffic density in this region, contrails from various aircraft could form over a long time period (∼12 h) that overlaps with “the remnants” of previous contrails. Thus, air traffic in this area has a strong impact on cloudiness on 26 March 2014.

On 26 March 2014 the HALO aircraft started from Oberpfaffenhofen in Germany at approximately 05:30 UTC and probed the cirrus over NAR from around 08:00 to 11:30 UTC with a race track pattern between approx. 51.5 and 54∘ N at a longitude of ca. -14∘E (-13.6 to -14.4∘E), see the flight track in Fig. 1a and also Voigt at al. (2017, Fig. 4). In this area, HALO flew three lidar legs almost perpendicular to the NAR tracks (07:57–08:35 UTC, south to north; 09:17–09:30 UTC, north to south; and 10:21–10:52 UTC, south to north), each followed by in situ legs at different altitudes.

Figure 2 presents the temporal variation of contrails and surrounding clouds with one HALO in situ/lidar leg at close time and air traffic data 2 to 3 h before, from 08:30 (the first in situ leg) to 10:30 UTC (the third lidar leg). In addition, the area around the flight path is investigated in Sect. 4.2 in more detail, but that serves here as orientation to easily capture the temporal evolution of the ice clouds. The wind speed in the east–west direction fluctuates around 0 while winds in the north–south orientation drive the ridge cloud to move southwards (blue arrows in Fig. 2 indicate wind direction at 225 hPa from ERA5). The aircraft fly to and from Europe. In the BTD images – where SEVIRI brightness temperatures at 10.8 and 12.0 µm are subtracted from each other – the black areas are caused by the low-level clouds, while the bright pixels correspond to thin cirrus. Small ice crystals, for instance in contrails, correspond to the largest BTDs. From 08:30 UTC, many thin lines (contrails) are seen in the false color RGBs in the first column in Fig. 2, where the HRV channel was also used. They run parallel to most air traffic routes (blue diamonds in the second column, the description of the air traffic data set in Sect. S1 in the Supplement) and are intersected perpendicularly by the HALO route. Some of the contrails are also visible in the BTD pictures (the second column in Fig. 2), that have a lower spatial resolution than the HRV channel, thus indicating the presence of small ice crystals. These contrails form a thin cirrus layer, potentially of anthropogenic origin but also with natural ice clouds (see discussion of Fig. 3 below). In the south, the ridge cirrus is thicker and no contrails can be observed. At 09:30 UTC, the flight area is dominated by contrails. With time, from 08:30 to 10:30 UTC the typical contrail lines become always fainter and fewer in number. Thus, contrails begin either to dissipate or to lose completely their linear shape due to wind shear such that they turn out to be undistinguishable. However, contrails can still be observed in the northeastern part of the satellite images at later times.

Figure 3 presents the backscatter ratio at 532 and 1064 nm from the WALES lidar measurements in that sequence. The lidar data show that the geometrical thickness of high-level clouds reduces from ∼2.0 km at 08:30 UTC to ∼1.5 km at around 11:00 UTC. The temporal evolution of CTH from WALES shows that the cloud first reached up to approximately 12 km and slowly descended to slightly above 11.5 km, with backscatter values becoming smaller with time, in line with the passive observations that indicate dissipation of the cloud during the day. For the investigation of contrails from in situ measurements we concentrate on the time period between 08:30 and 11:30 UTC, but we extend this time to cover the period from 06:00 to 14:00 UTC in Sect. 4.2.

Considering the three WALES legs in more detail (Fig. 3), in leg 1 between 08:10 and 08:25 UTC there is a series of connected fall streaks with high backscatter at a height of 11 km below a thin ice cloud layer top at 12 km. These structures resemble those in the large eddy simulations by Unterstrasser et al. (2017a, b) where they considered contrails within natural cirrus. Thus, these bright spots with their fall streaks can probably be identified as contrails, with many small ice particles causing high backscatter and with larger ice crystals sedimenting down. For leg 1, these contrails make up the majority of the cloud in this temporal interval. The presence of developed fall streaks is a hint that these contrails are aged (>30 min). Unfortunately, the high WALES backscatter inhibits the determination of RHi (Fig. S1) for these bright spots, but the upper level of the clouds, just below the potential contrails, shows occasional ice supersaturation, with RHi above 100 %.

Lidar leg 2, much shorter than the other two, also shows the presence of some spots with elevated backscatter ratios, although less numerous, at an altitude of 11 km at 09:26 UTC. Unlike leg 1, many high backscatter spots do not show extended fall streaks, suggesting the presence of younger contrails in addition to older ones. Furthermore, the top ice layer above 11.5 km also contains some very high backscatter areas. Finally, also above 9 km some clouds are visible that resemble contrails (or a dissipating cirrus). Thus, in this leg the contrail-like structures do not fill the entire cloud as in leg 1, but there are indications of young and older contrails, in addition to natural cirrus. The last lidar leg 3 took place 1 h later and also shows various bright spots at different levels, from 10 to almost 12 km, and elongated vertical structures that remind of those in leg 1. The cloud as a whole is slightly lower than that in leg 1, with ragged edges especially at its lower border suggesting that the cloud is sublimating and thinning out. This is confirmed by the RHi observations of WALES for leg 3 in Fig. S1 in the Supplement, where subsaturation is indicated especially at the lower and upper edges, while around 10.5 km at selected locations RHi reaches saturation. These WALES measurements are indicative for the cloud probed between 08:00 and 11:30 UTC but cannot directly be compared to in situ observations taken at different times in between.

In this section we illustrate the new method used to derive instantaneous cirrus RF and show that it compares well to observations of RSR and OLR by RRUMS. In particular, we explain a 2-step method to compute both cirrus-free and cirrus influenced TOA irradiances. First, an atmosphere is set up as input to the RTM that contains a realistic representation of the situation observed, including in particular all clouds. Then in a second step the ice clouds are removed, for which the RF is to be computed. This provides the cirrus-free reference TOA irradiance for the calculation of cirrus RF. The basic setup of the RTM is described in Sect. 2.3. Every RTM calculation needs an atmospheric state as input. This is in part obtained from ECMWF ERA5 reanalysis data (Hersbach et al., 2020). All 137 model levels are used. The horizontal and temporal resolution are 0.25∘×0.25∘ and 1 h. We derived temperature profiles, logarithm of surface pressure, specific humidity, ozone mass mixing ratio and land or sea mask. Densities of gaseous water (H2O) and ozone (O3) are derived from specific humidity and O3mass mixing ratio. CO2 is set to a volume mixing ratio of 400 ppm. Vertical profiles of liquid clouds are also extracted from ERA5 data. We choose vertical profiles close to the HALO flight time, i.e., with a time difference of 30 min at most between model and in situ measurements. We rely on the fact that reanalysis data should provide a realistic description of cloud properties and cloud positions, but we accept that this procedure might cause small temporal shifts such that observations close to cloud edges might not be represented in an optimal way through the model data. However, we refrain from interpolating cloud properties in time since also this procedure would create artificial clouds that do not exist in reality, especially in locations where no cloud is present at a given time but it is there at the next time. For liquid clouds, the parameterization by Bugliaro et al. (2011, 2022) is applied for creating Reff profiles using IWC and temperature from ERA5. The solar zenith angle corresponds to synchronous SEVIRI observations. Besides, the albedo of the ocean is parameterized following Cox and Munk (1954a, b) and Nakajima and Tanaka (1983), especially involving the wind speed from ERA5. For ice clouds, another procedure is applied. Since SEVIRI observations with CiPS are able to account for the entire cirrus cloud layers but are only dependent on thermal channels and not affected by low lying clouds (Strandgren et al., 2017b), SEVIRI provides accurate ice cloud properties (IOT) that can be used in the RTM.

The representation of cirrus at the flight level is complemented by adding their Reff from in situ measurements, CTH and cirrus bottom height (CBH) from lidar legs, as well as CiPS IOT into libRadtran in the way described in the next lines. This way we simulated a vertically homogeneous ice cloud with the correct IOT obtained by CiPS. We assumed IOT to be constant over an entire SEVIRI pixel while Reff changes according to in situ information (at the altitude levels probed by HALO). Since the RTM needs IWC as input, we determine it from IOT. First, Extinction “Ext” for IWC = 1 g m-3 (i.e., extinction in km-1 g-1 m3) for each measurement is interpolated to the given in situ Reff with the parameterization of optical properties of Baum et al. (2014) for rough aggregates. The selection of this shape for all cloud types is motivated by the fact that each cloud column, even those containing contrails, encompass ice crystals with various temporal evolution, e.g., young, small – probably round – ice particles in contrail cores together with larger sedimenting ice crystals in the fall streaks with different shapes or with evaporating ice crystals, in part still large in size, in subsaturated air that are starting to lose asphericity or maybe even natural ice crystals with unknown shape. To avoid an additional arbitrary choice with respect to ice particle shape we decided to keep the method as simple as possible and selected this shape (rough aggregates) as, e.g., for the MODIS optical property products Collection 6. A sensitivity study with respect to ice particle shape is conducted in Sect. 4.2. Note that the parametrization of this shape only covers Reff from 5 to 60 µm, which results in the inexecutable RTM calculations for larger or smaller ice crystals. Twenty cases in total are removed but have a negligible effect on the estimation of radiative effects as Reff of contrail cirrus and natural cirrus always fall in the range where RTM could simulate as indicated in Fig. 5. The IWC for each measurement of Reff corresponding to a vertically homogeneous ice cloud with given IOT is derived using the following equation: 6IWC=IOTExt⋅(CTH-CBH). CTH and CBH are obtained from the lidar legs since they seem to vary only slowly with time and space (Fig. 3). Thus, this IWC is used to simulate a homogeneous ice cloud layer between CBH and CTH and corresponds to the IOT observed by CiPS.

Finally, in situ measurements of Reff are assigned to a collocated IOT from CiPS as well as the ERA5 properties listed above for temperature, gas and liquid water clouds. With this atmospheric setup, TOA irradiances are computed and represent the cirrus contaminated RSR and OLR. Then, the ice clouds only are removed from the input of the RTM and other calculations are performed to compute cirrus-free irradiances. Both together are then inserted in Eqs. (3)–(5) to compute the instantaneous net RF of the cirrus cloud under consideration.

The RSR and OLR values influenced by cirrus as obtained from libRadtran simulations (RSR_L and OLR_L) and compared with RRUMS algorithm outputs (RSR_R and OLR_R) along the flight path of HALO are presented in Fig. 6. The two methods agree quite well, but the RTM calculations tend to underestimate large RSR values, probably due to too thin, missing or even mismatched liquid water clouds that are taken from ERA5. Uncertainties in ice cloud properties cannot have such a large effect on RSR since their optical thickness is very low around 0.2. Furthermore, a smaller overestimation of RSR by the RTM compared to RRUMS is also observed for the smallest RSR values below 150 W m-2, related to the bias of estimated ocean albedo but improved by the application of wind speed. For OLR, the agreement between the two data sets is good, with both slight (<10 W m-2) overestimations and underestimations.

The RTM-based TOA radiation estimation method provides a good representation of the real atmosphere and will be used to determine RF of the ice clouds in the following section.

In Sect. 4.1 we presented TOA RF for each waypoint of the probed cirrus. In order to examine microphysical properties and radiative effects of the contrail cirrus outbreak detected in this area and to analyze the corresponding temporal variation, we present the regional cirrus cover (CC), average cirrus IOT and CTH (all three quantities from CiPS), as well as mean TOA SW, LW and net RF in an area of 36×51 pixels (∼16500 km2, the red area in Fig. 2) near the HALO flight track, characteristic for contrail cirrus and cirrus as shown in Sect. 3. This area extends mostly east of the HALO flight and thus contains the contrails or contrail cirrus formed directly east of the ridge cloud. We consider here all ice clouds in the region with high aviation impact.

RF in this section is at the SEVIRI resolution, and calculated using CTH, IOT from CiPS, Reff from in situ mean values of all cirrus (called Reff,mean in the following) and cloud thickness from WALES, assuming the latter does not change in the area with the same procedure presented in Sect. 4.1 based on the RTM libRadtran.

In Fig. 7a, we observe that CC gradually decreases from 0.77 at 06:00 UTC to 0.25 at 14:00 UTC. The positive vertical velocity from ERA5 around that region implies the local downward motion of air mass to warmer temperature layers, and the CTH also decreases. IOT in Fig. 7b decreases between 07:00 UTC (0.41) and 10:00 UTC (0.15), then slowly increases until 12:00 UTC (0.17), then falls to 0.12. CTH decreases during the day and is thus consistent with both the observations of HALO (Fig. 3) and the downward motion. Since an underestimation of CTH by CiPS with respect to WALES (Fig. 3) is observed there, we assume that CTH is also underestimated by CiPS in this area.

Mean net RF over this area in this synoptic situation is positive in the early morning until 09:00 UTC, with the maximum of net RF is at 07:00 UTC when the sun has risen. Hence the contrail cirrus outbreak is warming during night and early morning hours. After 09:00 UTC, the forcing becomes negative. More explicitly, from around 09:00 to 14:00 UTC the net RF is negative, and thus this contrail cirrus outbreak tends to cool during daytime. The strongest cooling is observed at 12:00 UTC. Notice however that the possible underestimation of CTH by CiPS in this area would result in the general underestimation of the LW RF results since a lower CTH reduces the contrast to the cirrus-free OLR. In turn, this would further shift cirrus net RF towards cooling.

To quantify how the estimated uncertainties in Reff (and resulting IWC) as well as assumed crystal shapes influence the radiative forcing, we compute radiative forcing for aggregates (agg) with Reff=Reff,mean±50 % and perform a sensitivity study about ice crystal shape using the general habit mixture (ghm), also available from Baum et al. (2014). The simulated values are recorded in Table 2. The uncertainty of RF due to Reff has an average of 0.2 W m-2 in SW and 0.1 W m-2 in LW. In total the effect on the net RF is approx. 0.1 W m-2. Compared with aggregates, the ghm model has induced a larger net RF of 1.7 W m-2, with the shift in SW and LW of 1.2 and 0.5 W m-2, respectively.