Two advanced lidar instruments, the multiwavelength Raman lidar Polly (POrtabLe Lidar sYstem) and a high spectral-resolution lidar (HSRL) , were operated continuously aboard the drifting Polarstern during the MOSAiC expedition. These two lidars are complementary regarding their capabilities. While the multiwavelength Raman lidar delivered detailed spectrally resolved optical properties of the aerosol in the Arctic and retrieval products regarding microphysical and cloud-relevant aerosol properties from October to March, i.e., during nighttime conditions, the single-wavelength HSRL is of advantage during the summer half year (when Raman lidar observations are of limited use) and allows us to measure profiles of the climate-relevant particle extinction coefficient at 532 nm wavelength even during sunlight conditions. Both lidar systems are polarization lidars and thus permit detailed monitoring of cloud phase and thus of liquid-water, mixed-phase, and ice clouds and morphological features (sphericity) of aerosol particles.

The MOSAiC expedition for the first time provided the unique opportunity to perform lidar observations north of 85∘ N over the entire winter half year 2019–2020. This part of the central Arctic is not covered by any other lidar measurement, neither by observations with the spaceborne CALIPSO lidar (only up to 81.8∘ N) nor by measurements of the ground-based Arctic lidar network . Thus, we add a new data set to the global aerosol database. were the first to run lidars (mounted on an ensemble of autonomous drifting buoys) in the central Arctic, even north of 82∘ N, to perform year-round aerosol profiling, including the winter half year of 2015–2016. These measurements together with respective CALIPSO lidar observations are used in our contrasting analysis regarding the aerosol conditions during the MOSAiC year 2019–2020 and the year 2015–2016 characterized by unperturbed aerosol conditions .

The setup and basic technical details of the Polly instrument are given in , , and . Polly belongs to the lidar network PollyNET , which is part of the European Aerosol Research Lidar Network (EARLINET) organized within the Aerosols, Clouds and Trace gases Research InfraStructure (ACTRIS) project . The lidar transmits linearly polarized laser pulses at 355, 532, and 1064 nm with a pulse repetition rate of 20 Hz. All laser beams point to an off-zenith angle of 5∘ to avoid a bias in the observations of the optical properties of mixed-phase and cirrus clouds caused by strong specular reflection by falling and then frequently horizontally aligned ice crystals .

The receiver unit consists of a near-range receiver part, optimized to provide aerosol and cloud optical properties from 120 m to several kilometers above the surface, and a far-range receiver part which permits accurate aerosol and cloud profiling from about 800 m to 30–40 km height. Thirteen receiver channels are available to collect the following lidar return signals: elastically backscattered photons at the three laser wavelengths, the cross-polarized signal components at 355 and 532 nm at two different FOVs, the vibrational Raman signals of nitrogen at 387 and 607 nm, and those of water vapor at 407 nm. All signals are measured with photon-counting photomultipliers and stored with 30 s of resolution. We introduce “cross” and “co” to indicate the plane of polarization orthogonal and parallel to the plane of linear polarization of the transmitted laser pulses.

The recently introduced and implemented dual-FOV polarization lidar option permits the measurement of the cross and total (cross-polarized + co-polarized) signal components at 532 nm at two different FOVs. The dual-FOV lidar technique enables us to determine multiple scattering by droplets in liquid-water-dominated cloud layers and to retrieve from this multiple scattering information cloud microphysical properties (e.g., effective droplet size and cloud extinction coefficient) . The method is based on the measurement of two volume linear depolarization ratios at two different FOVs. The depolarization ratio is defined as the ratio of the cross-polarized to the co-polarized backscatter coefficient.

The Polly instrument is designed for automated continuous profiling of aerosols and clouds and thus was running around the clock. Well-defined breaks were necessary to exchange laser flash lamps, to run different calibration procedures, to check the full setup, and to perform an overall alignment of the Polly instrument. Five TROPOS lidar scientists took care of the OCEANET instrumentation over the 1-year MOSAiC expedition period.

The measurement and retrieval products obtained from the multiwavelength dual-FOV polarization and Raman lidar Polly are listed in Table . In addition, typical relative errors caused by signal noise and required input parameters in the data analysis are given in the table. The basic data analysis procedure to obtain aerosol and cloud optical properties is described by , , , , and . Height profiles of the particle backscatter coefficient at the laser wavelengths, of the particle extinction coefficient at 355 and 532 nm, respective extinction-to-backscatter ratio (lidar ratio) at 355 and 532 nm, the particle linear depolarization ratio at 355 and 532 nm, and the mixing ratio of water vapor to dry air by using the Raman lidar return signals of water vapor and nitrogen can be determined. The aerosol Raman lidar method was applied to determine the particle backscatter and extinction profiles .

The particle linear depolarization ratio (PLDR) is obtained from the measured volume linear depolarization ratio (VDR, depolarization ratio influenced by Rayleigh and particle depolarization of backscattered laser light). The PLDR can be used to discriminate spherical aerosol particles such as haze particles and liquid droplets, producing a PLDR close to zero, from non-spherical particles such as dust particles or ice crystals, causing a PLDR around 0.3 and typically >0.4 at 532 nm, respectively. In the case of Polly, the cross-polarized and total backscatter signals are measured. The specific approach to obtain the VDR from the Polly observations is described by .

Although PollyNET delivers automatically calculated PLDR profiles, the lidar observations presented in this article and in the accompanying article of were manually analyzed. To keep the uncertainties in the derived aerosol quantities at an acceptably low level of <10 % for the backscatter coefficients and depolarization ratios and of the order of 20 %–25 % for the particle extinction coefficients and lidar ratios, large vertical signal smoothing and regression window lengths of 500 to 2500 m and signal averaging times from 30 min to more than 10 h had to be applied. Fortunately during the winter half year, the nighttime conditions allowed us to apply the Raman lidar methods around the clock. More details are given in .

Auxiliary data are required in the lidar data analysis in the form of temperature and pressure profiles in order to calculate and correct for Rayleigh backscatter, extinction, and light-depolarization contributions to the measured lidar return signals. In the MOSAiC data analysis, we used the Polarstern radiosonde observations. As an important contribution to MOSAiC, radiosondes were routinely launched every 6 h (at 05:00, 11:00, 17:00, and 23:00 UTC) throughout the entire MOSAiC period .

The retrieval of aerosol microphysical properties such as particle volume, mass, and surface area concentration and estimates of cloud-relevant properties such as CCNC and INPC is performed by means of the POLIPHON (Polarization Lidar Photometer Networking) approach . Lidar input data sets are the height profiles of the 532 nm backscatter coefficient and the PLDR. , for example, show the full set of height-resolved POLIPHON aerosol products in the cases of an 18-month Polly campaign in Dushanbe, Tajikistan, for central Asian aerosol. Alternatively to the POLIPHON method, we applied the multiwavelength lidar inversion technique to derive the microphysical properties of aerosol particles including the particle size distribution. The method of is applied to MOSAiC smoke and Arctic haze observations to estimate layer mean microphysical properties in pronounced smoke and haze layers with sufficiently accurate backscatter coefficients at 355, 532, and 1064 nm and extinction coefficients at 355 and 532 nm.

Details of the analysis of the dual-FOV polarization lidar measurements can be found in . The method allows us to retrieve the cloud extinction coefficient and the effective radius of the droplets at a height of 50–100 m above cloud base and, in the next step, to calculate the liquid-water content (LWC, function of cloud extinction coefficient and effective radius) and cloud droplet number concentration from LWC by assuming a gamma size distribution.

Relative-humidity fields are obtained from the mixing-ratio measurements (Raman lidar method) and temperature profiles measured with the Polarstern radiosondes. Quality checks of the continuously obtained water vapor fields were based on comparisons with the relative-humidity profiles obtained with radiosondes launched at 05:00, 11:00, 17:00, and 23:00 UTC.

A good knowledge of the tropopause height is important when profiling aerosol and cloud layers in the troposphere and stratosphere. The tropopause was computed from the MOSAiC radiosonde temperature and pressure profiles by using the approach of the Global Modeling and Assimilation Office (GMAO), Goddard Space Flight Center, Greenbelt, Maryland, USA . In this approach, the tropopause height zTP is found from the height profile of the difference αT(z)-log⁡10p(z) with α=0.03, temperature T (in Kelvin), pressure p (hPa), and height z (m). The tropopause pressure p(zTP) is defined as the pressure where the defined difference reaches its first minimum above the surface. If no clear minimum was found up to z=13000 m over Polarstern, a tropopause height zTP was not assigned. The obtained tropopause heights agree well with the ones we obtained by applying the definition of the World Meteorological Organization to the radiosonde temperature profiles and considering refinements in the determination described by . In most cases, the GMAO approach delivered 20-80 m lower tropopause levels and produced fewer outliers.

The large spectrum of retrieved aerosol and cloud microphysical properties forced us to use any favorable opportunity for comparison with airborne in situ observations of these quantities in the framework of validation experiments in the past. These combined lidar and airborne in situ observations to characterize the quality of the retrieval products and to check the respective uncertainty ranges were usually available during large field campaigns. Regarding the aerosol microphysical properties, the comparisons showed that particle number concentrations, surface area, and volume and mass concentrations can be obtained with an uncertainty of 25 %–50 % (see Table ) in cases with a clearly dominating aerosol type, e.g., in dense desert dust plumes or lofted wildfire smoke layers . The size distributions of aerosol particles can be precisely identified and estimated from multiwavelength lidar measurements in cases with a pronounced accumulation mode as it is the case for aged wildfire smoke and aged Arctic haze.

Validation studies are especially valuable to characterize the potential of lidar to estimate cloud-relevant aerosol parameters such as CCNC and INPC. Comparisons with airborne in situ measurements showed that CCNC can be obtained with an uncertainty of about 30 % (inversion of multiwavelength data) to 50 % (conversion of the 532 nm extinction coefficient) when the aerosol type (and thus the typical aerosol size distribution) is known and about a factor of 2 if the aerosol type is not well known or mixtures of different aerosol types prevail .

In a few efforts, the potential of lidar to deliver trustworthy INPC profile information was investigated based on simultaneous airborne in situ and lidar observations . These few attempts indicated that an INPC estimation is possible with an uncertainty within an order of magnitude. The large uncertainty is caused by the ice-nucleating particle (INP) parameterization used (taken from the literature) and not from the aerosol input parameters obtained from the lidar observations. In this field, much more coordinated field activities are needed. One alternative approach is to compare the derived INP concentrations with the estimated ice crystal number concentration from lidar–radar observations in so-called closure experiments. Good consistency in these closures indicates a high reliability of the selected INP parameterization .

All these uncertainties in the aerosol retrievals can be considerably larger if the aerosol fraction of interest contributes to a minor part to the observed aerosol amount, i.e., when, for example, the interesting mineral dust fraction in Arctic haze is of the order of 10 %. For this reason, accompanying airborne in situ observations are always useful to better characterize the uncertainty range of the aerosol and cloud products gained from active remote sensing.

Concerning microphysical properties of liquid-water cloud layers (CDNC) and cirrus clouds (ICNC), validation studies based on aircraft–lidar comparisons are difficult because of the usually strong vertical and horizontal inhomogeneities in the cloud properties. Here, we make use of different, independent methods and techniques to derive estimates of CDNC and ICNC to check the quality of the retrievals regarding CDNC, see, e.g.,. Product evaluation efforts have been reported by in the case of the retrieval of ice crystal microphysical properties from combined lidar–radar observations and were based on comparison with respective satellite retrievals or airborne in situ observations.

Extreme and long-lasting wildfires in central and eastern Siberia, in closest neighborhood to the Arctic region, were most probably responsible for the UTLS smoke layer over the High Arctic in the winter half year of 2019–2020 . The main burning phase lasted from 19 July to 14 August 2019. In the absence of any pyrocumulonimbus (pyroCb) convection and thus of a very efficient and fast process to transport smoke into the UTLS height range, we hypothesize that the smoke was lifted by the so-called self-lifting process . Details are given in . In self-lifting processes, strong absorption of solar radiation by the black-carbon-containing smoke heats the air in lofted smoke plumes. The buoyancy created then forces the smoke plumes to ascend up to the tropopause (at around 10 km) within about 3–5 d according to our simulation studies .

Once in the UTLS height range, smoke particles become quickly distributed over large parts of the Northern Hemisphere within a few weeks, as observed and documented for the first time in 2001 and recently confirmed after the record-breaking Canadian fires in the summer of 2017 . In the case of the strong Siberian fires, the smoke particles probably became quickly distributed over the entire Arctic region (as suggested by the satellite observations presented by ) and then remained in the UTLS height range for months. The decay of smoke-related stratospheric perturbations usually takes more than a half year .

Besides the strong and long-lasting fires in Siberia, the Raikoke volcano erupted in June 2019 . The amount of the emitted SO2 suggested that a stratospheric aerosol layer with a maximum 532 nm AOT of around 0.025 at latitudes >50∘ would form. It was further expected that the AOT would decrease to values of 0.01 over the High Arctic in autumn 2019 and to 0.005 during the winter months, as discussed in .

From the first days of the MOSAiC observations on, in late September 2019, we observed a pronounced and well-aged stratospheric aerosol layer with a backscatter maximum around or just above the tropopause, smooth internal structures, and clear wildfire smoke signatures. The 532 nm AOT was around 0.1 in autumn 2019 over the Polarstern and thus much higher than the expected sulfate AOT of 0.01. Further lidar observations at Leipzig (51.3∘ N, 12.4∘ E), Germany, and Ny-Ålesund, Svalbard (78.9∘ N, 11.9∘ E), Norway, indicated a strong increase in the stratospheric AOT in August 2019, probably as a result of the record-breaking Siberian fires and the spread of the smoke over large parts of the Northern Hemisphere.

Figure  presents the optical properties of the smoke layer as measured on 11 December 2019 (Fig. g and h). The Polarstern was at 86.6∘ N and 120.7∘ E at 12:00 UTC. The lidar observations were averaged over the entire day; thus 24 h mean height profiles are shown. The UTLS aerosol layer extended from 8 to more than 18 km height. The internal vertical structures were rather smooth. The layer base as indicated in Fig.  was determined by the altitude at which the 1064 nm backscatter coefficient started to increase with height. The smoke layer top was set to the height where the total-to-Rayleigh backscatter ratio at 1064 nm dropped below a value of 1.1.

A clear and unambiguous optical fingerprint for the presence and dominance of aged smoke particles (after long-range transport) is the unusual spectral dependence of the extinction-to-backscatter ratio L (Fig. d) together with a high 532 nm particle lidar ratio L532 typically exceeding 70 sr. L532 is much larger than the 355 nm lidar ratio L355, typically by more than 20 sr . Another smoke signature is the weak wavelength dependence of the particle extinction coefficient σ, expressed in terms of the extinction-related Ångström exponent Åσ,355,532=ln⁡(σ355/σ532)/ln⁡(532/355). For the extinction profiles in Fig. c, Åσ,355,532 is about 0.5. In the case of volcanic sulfate aerosol the Ångström exponent is typically clearly >1.0. Note that such an unambiguous identification of the wildfire smoke type is only possible with multiwavelength Raman lidars and high spectral-resolution lidars , which allow us to determine the particle extinction coefficient at several wavelengths.

The particle depolarization ratios at 532 nm (around 1 %) and at 355 nm (1 %–2 %) in Fig. b were very low. Such low values are indicative of spherical particles. We hypothesize that the comparably slow ascent (over several days) of the smoke particles up to the UTLS regime over Siberia in the summer of 2019 allowed the smoke particles to complete the aging process in the humid troposphere (with high levels of condensable gases). At the end of the particle aging process, the majority of the smoke particles showed a spherical core–shell structure and thus caused the low particle depolarization ratios. In contrast, in the case of pyroCb-related ascents (fast lifting within 30–90 min up to the tropopause) there is no time for particle aging. Then, the particles widely keep their original, non-spherical shapes and thus produce large depolarization ratios of about 20 % at 532 nm.

Figure  presents several volume size distributions of the smoke particles. The volume size distributions were obtained from the Polly observation by applying the lidar inversion method to the layer mean three backscatter and two extinction coefficients . As typical for aged wildfire smoke, a well-defined accumulation mode was found. A distinct coarse mode was absent. The findings agree well with in situ observations of aged smoke transported over long distances .

An overview of the smoke conditions during the MOSAiC winter half year (October to April) is presented in Fig. . Most of the time, the smoke layer was observed between 7 and 17 km height with the backscatter maximum just above the tropopause. A trend of downward movement of the layer is not visible. The maximum extinction coefficients (532 nm) decreased with time from values >10 Mm-1 in October and November to <5 Mm-1 in April 2020 (Fig. a).

The AOT in Fig. b was computed from the particle backscatter height profiles. The directly determined extinction profiles were too noisy, especially in the upper part of the smoke layer. The 532 nm backscatter coefficients were multiplied with the smoke mean lidar ratio of 85 sr (computed from all smoke observations measured during the winter half year) . Subsequently, we integrated the extinction values between the smoke layer base and top heights as given in Fig.  to obtain the AOT. During the winter months, especially in January to February 2020, polar stratospheric clouds (PSCs) were frequently observed at the top and above the smoke layer, from 17 to 25 km height. We removed these PSC-affected parts from the height profiles of backscatter and extinction coefficients before we calculated the vertical column-integrated smoke optical properties. However, several weak PSCs developed within the smoke layer, and in this case, the backscatter and extinction contributions from these optically thin PSCs were not removed. The PSC-related uncertainty in the 532 nm AOT was estimated to be of the order of 5 % .

In terms of the 532 nm AOT in Fig. b, the perturbation decreased from 0.05–0.12 in October and November to values of 0.03–0.06 in December to the middle of March and dropped to 0.01–0.02 in April 2020. Almost constant AOT conditions were observed from 10 December to 10 March. During that time period the strong polar vortex was established, dominated the airflow, even below the vortex, and controlled the horizontal and vertical exchange of gases and particles. As mentioned by , from mid December 2019 to mid April 2020, the weather pattern was rather stable and meridional air-mass exchange was widely suppressed. The vortex started to collapse around 20 April .

The 532 nm AOT of the volcanic sulfate aerosol was estimated to be about 0.01 in October 2019 and 0.005 during the winter months January and February 2020. According to the estimations in , the Raikoke aerosol fraction was always lower than 15 %.

The layer mean 532 nm smoke extinction coefficients in Fig. c (obtained from the ratio of AOT divided by the layer geometrical depth in Fig. a) were on the order of 10 Mm-1 in October, around 4–5 Mm-1 during the main winter months, and mostly ≤3 Mm-1 at the end of the lifetime of the smoke layer. From the measured layer mean extinction coefficients, mass concentrations of the smoke particles were derived and ranged from 0.4–2 µgm-3 during the autumn and winter months. Note that AOT values for a clean stratosphere are around 0.001–0.002 , and minimum extinction coefficients are of the order of 0.1–0.2 Mm-1. Minimum stratospheric mass concentrations (at midlatitudes) are close to 0.01–0.02 µgm-3 .

A PSC observation is shown in Fig.  at the end of this section. According to the PSC classification scheme , we observed type Ib PSCs. This type is made up of supercooled liquid ternary solutions that likely consist of H2SO4, HNO3, and H2O. In contrast to type Ia and II PSC particles (crystals) the liquid droplets of PSC type Ib only slightly depolarize incoming laser light. The temperature at PSC base height showed values of -78 ∘C, and at the backscatter maximum the MOSAiC radiosonde measured a temperature of -86 ∘C. We observed a much lower number of PSCs over the North Pole region (86 to 88.6∘ N) than the CALIPSO lidar (60 to 81.8∘ N) during the main PSC period from January to March .

The original and primary goal of the shipborne MOSAiC lidar measurements was to provide, for the first time, a height-resolved characterization of tropospheric aerosols and clouds over the North Pole region during the winter half year. Because of its importance for the climate and environmental conditions, Arctic haze has been intensively studied for more than 50 years . However, knowledge about the vertical layering structures of these aged haze aerosols and about the composition and microphysical properties is still limited and mostly based on sporadic aircraft observations performed during field campaigns, preferably in spring . Furthermore, presented a ground-based multiwavelength polarization Raman lidar study on Arctic haze in terms of backscatter, depolarization, and lidar ratios. The situation has improved since 2006 with the start of the CALIPSO mission .

As summarized by , long-range transport of cold, polluted air masses from northern Eurasian source regions (mainly north of the Arctic front) prevails in winter and leads to the formation of Arctic haze with the highest mass concentrations in late winter and early spring. Such a low-level aerosol transport is missing in summer. pointed out that the winter transport of aerosols towards the High Arctic occurs at low heights. Long-transported aerosols may reach heights up to the middle troposphere (5–7 km height).

In Fig. , we present two MOSAiC cases of Arctic haze observed on 4 February and 4 March 2020. The Polarstern was drifting with the ice at latitudes of 87.5∘ N (4 February) and 88.1∘ N (4 March). The most striking feature in both figures is that aerosol layers occurred almost everywhere within the troposphere and in the lower stratosphere. Remnants of PSCs were visible around 13.5–14 km height on 4 February 2020 (Fig. a). Temperatures were around -76 ∘C at 13.5 km height and thus sufficiently low to allow the formation of type Ia PSC . Type Ia PSCs are thought to consist of nitric acid trihydrate (NAT) crystals and produce significant depolarization of backscattered laser light. Note that the Polarstern was fully below the strong polar vortex from the beginning of January to mid April 2020 .

According to the backward trajectory analysis shown in Fig. , the aerosol pollution in the pronounced aerosol layers around 5 km (4 February, 11:00 UTC, Fig. a) and 4 km height (4 March, 11:00 UTC, Fig. b) originated from central and western European regions (4 February) and from Russia and the Black Sea area (4 March). The height-resolved trajectory analysis indicated that most of the aerosol circled around in the Arctic (at latitudes >70∘ N) for several days before crossing the Polarstern.

In Fig. , the optical properties of Arctic haze for the two cases are illustrated. The figure shows 12 h (4 February) and 18 h (4 March) mean height profiles of the basic lidar products (backscatter, extinction, extinction-to-backscatter ratio). The long signal averaging times were needed to keep the signal-to-noise-related uncertainty below 10 % (backscatter coefficient), 20 % (extinction coefficient), and 25 % (lidar ratio). In both cases, we found a near-surface layer of up to about 2.5 km height and a separate lofted aerosol layer of up to 5 and 7 km height. A clear wavelength dependence of the backscatter and extinction coefficients was found on 4 March, as expected for fine-mode-dominated particles in the Arctic consisting of a mixture of anthropogenic haze and fire smoke . The Ångström exponent for the extinction coefficient was around 1.7 in the lofted layer above 3 km height, and the lidar ratios were high with values of up to 100±15 sr. This is indicative of the presence of small, strongly light-absorbing particles.

On 4 February 2020 (Fig. a), the aerosol optical properties were less well defined, the extinction coefficients were almost equal at both wavelengths (355 and 532 nm), and the noisy lidar ratios at 532 nm were larger than at 355 nm in the lofted layer above 3 km height, which is typical for aged wildfire smoke as discussed in the foregoing section. The lidar ratios were lower than on 4 March; i.e., less absorbing particles were present. The volume depolarization ratios (not shown) were rather low in both cases and indicated the dominance of aerosol pollution. The 532 nm AOT was close to 0.025 (4 February, for the lowest 7 km height) and 0.02 (4 March, for the lowest 5 km) and thus considerably lower than the UTLS smoke AOT of about 0.04–0.05. The uncertainty in the tropospheric AOT values is about 20 %.

The values for the 532 nm extinction coefficients of 2–8 Mm-1 in the lower layer (up to 2.5 km height) and 1–7 Mm-1 in the lofted layer above 3 km height are in good agreement with CALIPSO lidar observations . showed 14 d mean and layer mean values of 2–8 Mm-1 (0–2 km layer), 2–10 Mm-1 (2–5 km), and 1–2 Mm-1 (5–10 km layer) measured in the area from 5–25∘ E (north of Svalbard) and 80–82∘ N in February and March 2015. analyzed 13 years of CALIPSO observations (June 2006 to December 2019) for the Arctic (entire area from 65 to 81.8∘ N) and found winter mean extinction values of 2–4 Mm-1 (2–6 km height) and around 2 Mm-1 (6–10 km height).

Our findings are also in good agreement with the results previously published and obtained during major field activities such as POLARCAT-IPY (Polar Study using Aircraft, Remote Sensing, Surface Measurements, and Models, of Climate, Chemistry, Aerosols, and Transport, International Polar Year) as well as during ARCTAS (Arctic Research of the Composition of the Troposphere from Aircraft and Satellites) . Both campaigns took place in the spring of 2008. Arctic haze was mostly found up to 7 km height. The aerosol composition was analyzed in great detail based on aircraft observations.

summarized and reviewed earlier Arctic observations and compared observations with model calculations and found that the ratio of black carbon (BC) to organic aerosol (OA) is high with values of 0.1–0.15 for aerosols advected from Russia. This aerosol is a mixture of anthropogenic haze (sulfate aerosol) and domestic, forest, and agricultural fire smoke (organic aerosol). Sulfate aerosol prevails in the near-surface air, whereas the OA fraction becomes comparably large in the free troposphere. Agricultural fires during spring (Europe, Asia) and flaring of natural gas (Russian oil industry) were found to be important sources for BC and OA particles. The fire smoke mixes with anthropogenic haze mainly from Asia during the long-range transport towards the central Arctic, beginning in late winter with peak occurrence in the spring season. Agricultural and forest fires produce BC/OA ratios of typically <0.1, whereas anthropogenic haze may cause ratios >0.15.

Two MOSAiC case studies of aerosol–cloud interaction are presented next. In this section, the evolution of a long-lasting mixed-phase altocumulus layer in Arctic haze within the lower free troposphere is discussed, and, in the next section, we present a cirrus development in the lower part of the UTLS wildfire smoke layer.

The mixed-phase cloud system was shown in Fig. e and f. The cloud layer was observed for more than 7 h over the Polarstern on 10 December 2019. The dark band in the depolarization ratio panel between 2–3 km height in Fig. f indicates the liquid-water-dominated cloud top layer. The increase in the depolarization ratio above the dark zone at the liquid cloud base is caused by multiple scattering by cloud droplets. Favorable conditions with cloud top temperatures around -28.5 ∘C at 2.6 km height (at 03:00 UTC) were given for heterogeneous ice formation via immersion freezing, i.e., ice nucleation on INPs immersed in the water droplets . After nucleation in the cloud top layer, ice crystals grow fast to sizes of 50–100 µm within minutes and immediately start to fall out. As visible in Fig. e and f, the crystals formed long virga below the shallow altocumulus layer. The ice crystals partly evaporated on the way down but partly reached the ground as precipitation. The liquid-water-dominated cloud top layer was not depleted at any time during the 7 h period.

About 40 long-lasting mixed-phase cloud events (ice-precipitating shallow altocumulus decks) with durations from 4–30 h were observed from October 2019 to March 2020. Because of their sensitive influence on radiative transfer and the water cycle, they have been the focus of research for more than 15 years . However, because of the complexity of influencing meteorological and aerosol aspects, there are still many open questions concerning their long lifetime, especially of the longevity of liquid-water layers and thus of water droplets in the presence of ice crystals. MOSAiC contributes to this research field by means of combined lidar and radar observation.

We applied our recently developed dual-FOV polarization lidar method to derive the microphysical properties of liquid-water droplets in the cloud top layer. The results are shown in Fig. . The dual-FOV lidar technique was originally designed for pure liquid-water cloud observations but can be applied to mixed-phase clouds as long as backscattering by ice crystals is negligible compared to droplet backscattering in the cloud top layer. This condition holds here with ice crystal backscatter coefficients of 5–10 Mm-1sr-1 in the virga (not shown) and thus probably also in the cloud top layer with droplet backscatter coefficients of the order of 700 Mm-1sr-1 (not shown). In the case of crystal-to-droplet backscattering of 0.01, the contribution of ice crystals to the observed multiple scattering features, from which the microphysical properties of the droplets are retrieved, can be ignored. Note that the method delivers the time series of droplet effective radius, CDNC, and cloud extinction coefficients in Fig.  for the height of 75 m above cloud base. Thus, the properties of freshly formed droplets are mainly observed.

As can be seen, the retrieved CDNC values were around 20 cm-3 in the beginning and around 100 cm-3 in the cloud base region later on. With increasing CDNC the effective radius (a characteristic droplet size) decreased and vice versa as expected when assuming a constant water vapor reservoir for droplet nucleation. The cloud extinction coefficient showed typical values from 10–20 km-1 in the first half of the cloud lifetime. Uncertainties in the lidar products are indicated by error bars (1 standard deviation) and are of the order of 20 %–25 % (cloud extinction coefficient, droplet effective radius) and 50 % (CDNC).

In the next step, we wanted to know how many cloud condensation nuclei (CCN) were available. According to the HYSPLIT backward trajectories in Fig. , the altocumulus layer developed in aged Arctic haze, originating from northern parts of Europe. In order to estimate the CCN concentration (nCCN in Fig. ) just below the altocumulus layer, the lidar observations in clear skies (before 03:00 UTC) were analyzed. The result is shown in Fig. . In this specific measurement, we used the multiwavelength Raman lidar technique and applied the lidar inversion method (see Sect. ) to a set of three backscatter and two extinction coefficients. Here, the data analysis was based on the mean backscatter and extinction values (for the height range from 1500–2000 m height). The goal was to obtain (a) a value for the mean aerosol particle number concentration n50 (particles with radius >50 nm) and (b) a ratio of n50 to the 532 nm backscatter coefficient. We assumed that the obtained n50 value is a good estimate for nCCN (for a supersaturation level of Sw=0.2 %) . By using the obtained ratio of n50 to the 532 nm backscatter coefficient, the entire profile of the 532 nm particle backscatter coefficient was converted into a nCCN profile as shown in Fig. . The profile of the 532 nm particle extinction coefficient shown in Fig.  was obtained by multiplying the backscatter coefficients with a lidar ratio of 30 sr as obtained from the backscatter and extinction Raman lidar observations. The dashed nCCN curves indicate an uncertainty of 30 % in the n50 and nCCN estimation. The uncertainty of 30 % is much lower than indicated in Table . The uncertainty in the Table  holds in the case of the simpler conversion method .

Figure  shows that the estimated nCCN values of 25–55 cm-3 around 2500 m height were in the same range as the numbers of the retrieved CDNC (10–40 cm-3 in Fig. b) in the beginning of the cloud evolution. The actual updraft characteristics at cloud base determine the actual supersaturation levels. Strong updrafts may produce water supersaturation levels exceeding 0.5 %. Then the CCNC values are 50 %–100 % larger than our estimates in Fig.  as discussed in . All in all, we can conclude that the lidar observations provide a clear link between low aerosol particle concentration and therefore low CDNC in the mixed-phase cloud top layer. The aerosol conditions controlled the microphysical properties in the liquid-water-dominated cloud top layer.

In order to better understand the entire role of aerosol particles in the evolution of mixed-phase clouds, we also estimated the ICNC in the virga from combined lidar–radar observations and the INPC for the cloud top temperature of -28.5 ∘C from the 532 nm extinction coefficient in Fig.  .

Without going into too much detail, weakly enhanced PLDR values above 2 km height indicated a minor contribution of mineral dust particles (about 5 %) to the overall Arctic haze backscatter coefficient. This finding is in good agreement with studies of and , who analyzed spaceborne and ground-based lidar observations in combination with backward trajectory analysis and concluded that mineral dust typically contributes to the continental aerosol mixtures in the lower and middle free troposphere in the Arctic that were transported over long distances. Dust is the most favorable although not the only relevant INP type in the case of immersion freezing at temperatures >-30 ∘C during the winter half year when biogenic and biological components are absent . We used the INP parameterization scheme of to estimate the dust INP concentration for immersion freezing. Here, the particle number concentration n250 of dust particles with diameters >500 nm is an input parameter and obtained from the respective lidar observation of the dust-related 532 nm backscatter coefficient . The retrieval finally yielded INPC estimates in the range from 0.1–0.5 L-1 for the altocumulus top temperature of -28.5 ∘C. Uncertainties in the INPC estimates are of the order of a factor of 3 (i.e., within an order of magnitude). The main uncertainty source is the INP parameterization and not the retrieval of n250 values.

To obtain also an estimate for ICNC, we used the lidar observations of the 532 nm extinction coefficient in the first strong ice virga zone (04:00–05:00 UTC) below the liquid-water cloud layer together with radar reflectivity (8 mm wavelength) and Doppler information on the ice crystal fall speed spectrum in the virga from the AMF-1 KAZR (Ka-band zenith-pointing radar) . These observations were compared with comprehensive model simulations of the lidar and radar observations as a function of ice crystal number concentration, size distribution, and shape . The match between simulations and observations provide the range of retrieved ICNC values. In the case of the mixed-phase cloud we obtain 0.1–1.0 ice crystals per liter within the virga. Again, the retrieval uncertainty allows us to provide the order of magnitude of ICNC only.

Figure  summarizes our data analysis efforts and highlights the overall potential of our advanced aerosol/cloud lidar to contribute to cloud research. Since radiosondes were launched every 6 h (Fig. a and c) over the entire MOSAiC year, excellent conditions for a detailed lidar-based aerosol and cloud monitoring (Fig. b), including a coherent monitoring of the relative-humidity field (Fig. b), were given during the MOSAiC winter half year.

The temperature and relative-humidity (RHw) profiles measured with four radiosondes before, during, and after the cloud event indicate that the liquid-water-dominated cloud top layer was never thicker than a few 100 m. The cloud system developed in a warm and moist air mass and vanished when the moist air mass was replaced by very dry air (see 17:00 UTC radiosonde RHw profile in Fig. c). A strong drop in the relative humidity at all heights occurred around 15:00 UTC according to the lidar observations in Fig. d. The relative humidity (from lidar) is obtained from the Raman lidar observation of the specific humidity (water-vapor-to-dry-air mixing ratio) and radiosonde temperature profiles permitting the computation of the respective water-saturation-related specific humidity levels .

As mentioned above, favorable conditions for ice nucleation (via immersion freezing on dust particles) were given at cloud top temperatures of -28.5 ∘C in the beginning of the cloud lifetime and at around -25 ∘C later on. Strong ice crystal evaporation in the virga zone caused the strong moistening of the air mass below the main altocumulus deck after 03:00 UTC and also caused reduced crystal evaporation later on so that ice crystals could partly reach the ground as precipitation.

All estimated values for CCNC, CDNC, INPC, and ICNC are included in the lidar panel in Fig. b. We can summarize our observations as follows: during the early altocumulus development, the estimated CCNC values outside of the cloud were in a similar range to the estimated CDNC levels within the cloud as was also the case for the estimated INPC and ICNC levels. Thus, our data suggest that the estimated cloud active particle levels could be high enough to control the case-study cloud at given favorable moisture conditions and in the absence of other processes that might influence CDNC and ICNC levels (e.g., secondary ice formation, crystal-crystal collision and aggregation processes, or droplet collision and coagulation events). This hypothesis would be in line with numerous other Arctic studies that have previously observed this phenomenon e.g.,. However, higher-resolution in-cloud microphysical data are required to verify this lidar-based hypothesis.

It is worth mentioning that the estimated INP concentrations of 0.1–0.5 L-1 in Fig. b are also in line with previous Arctic in situ observations at similar temperatures. The measured INP values ranged from 0.001 to around 2.5 L-1 for temperatures from -25 to about -28 ∘C . As part of the MOSAiC data analysis, we plan to analyze many winter as well as summer altocumulus events and thus cloud formation under contrasting aerosol conditions to obtain an improved view of the role of aged aerosol pollution that has been transported over long distances in the evolution of mixed-phase clouds in the High Arctic.

According to a study of , Arctic ice clouds tend to form almost exclusively by heterogeneous ice nucleation with a contribution of only 10 % by homogeneous freezing. MOSAiC now offers an excellent opportunity to investigate the potential impact of aged wildfire smoke particles on cirrus formation. More than 50 cirrus systems developed in the upper troposphere in a smoke-influenced environment during the winter half year. The importance of such a smoke impact study arises from the fact that OA, besides dust and marine particles, is ubiquitous in the atmosphere . However, in contrast to dust and marine particles the INP potential of OA is not well understood. Scarce field data are the main reason for the lack of clarity and knowledge.

Because of the complex chemical, microphysical, and morphological properties of aged fire smoke particles, which can occur as glassy, semi-liquid, and liquid aerosol particles, the development of smoke INP parameterization schemes is a crucial task. Smoke particles from forest fires are largely composed of organic material (organic carbon, OC) and, to a minor part, of BC. The BC mass fraction is typically <5 % . Biomass burning aerosol also consists of humic like substances (HULISs) which represent large macromolecules. The particles and released vapors within biomass burning plumes undergo chemical and physical aging processes during long-range transport such as coagulation, condensation, and heterogeneous reactions, resulting in changes in their morphological characteristics (size, shape, and internal structure) and mixing state.

Our knowledge on the potential of smoke particles to serve as INP is mainly based on laboratory and field studies of the ice nucleation potential of fresh smoke (with a focus on black carbon or soot particles) . However, aged smoke in the upper troposphere (at cirrus level, several days, weeks, or months after emission) has fundamentally different properties to the ones of wildfire particles emitted just a few minutes to hours ago . As pointed out, freshly emitted soot or BC particles are typically hydrophobic, lacy fractal-like aggregates. During transport, lacy soot undergoes compaction upon humidification. mention that, during the aging process, the particles can also accumulate secondary sulfate mass from condensation of gaseous sulfuric acid that is present in the background atmosphere.

In this section, we present a MOSAiC cirrus case study to illuminate, for the first time, the potential of aged wildfire smoke to influence ice formation in the upper troposphere. An overview of organic particles, their complex properties, and ice-nucleating efficacy is given in the review article of . It is assumed that aged smoke particles show an almost perfect spherical core–shell structure and that the ability of smoke particles to serve as INP mainly depends on the organic material (OM) in the shell of the coated smoke (or soot) particles. At low temperatures, e.g., in the UTLS region, where the atmospheric temperature can be as low as 180 K, it is conceivable that the particles are in a glassy state. Aerosol particles serving as INPs usually provide an insoluble, solid surface that can facilitate the freezing of water . Deposition ice nucleation is defined as ice formation occurring on the INP surface by water vapor deposition from the supersaturated gas phase. When the supercooled smoke particle takes up water or its shell deliquesces, immersion freezing can proceed, where the INP immersed in an aqueous solution can initiate freezing . If the smoke particle becomes completely liquid (and no insoluble part within the particle is left), homogeneous freezing will take place at temperatures below 235 K .

Our MOSAiC observation was performed on 6 December 2019. As shown in Fig. c and d, a long-lasting cirrus event was monitored. The full cirrus lifetime was about 36 h. Large virga of falling ice crystals permanently removed ice crystals from the cloud top region where they were nucleated. The cirrus system developed in the lower part of the UTLS smoke layer. The bow-like feature of the lower boundary of the extended virga zone was caused by a high-pressure ridge (with dome-like temporal evolution of the influenced height range). The ridge crossed the Polarstern during this day and was characterized by a warmer and very dry air mass in the lowest few kilometers of the troposphere. Ice crystals falling into this dry air mass immediately evaporated.

Figure  shows the mean cirrus layer structures measured from 06:00–12:00 UTC and in the early afternoon (12:30–13:30 UTC) together with the meteorological conditions in terms of air temperature, relative humidity, and ice supersaturation measured with three MOSAiC radiosondes at about 05:30, 12:30, and 17:30 UTC (30 min after launch). Very constant meteorological conditions were found from about 4 km up to cirrus top at 8–9.5 km height. Temperatures were below -40 ∘C at heights >5.5 km so that the development of high-altitude liquid-water layers was impossible. Clear ice supersaturation conditions (Si>1.1) were present in the upper part of the cirrus system.

It is worth mentioning that equilibrium in ice supersaturation conditions as observed in the extended virga zone over the whole day (Fig. c) is a sign of a low crystal number concentration (<35 L-1) . Such a low amount of ice crystals is not able to quench the supersaturation, which is in turn indicative of heterogeneous ice nucleation. Homogeneous freezing would produce crystal concentrations of >500 L-1 so that equilibrium at ice saturation level would occur within a short time period. Figure  is further discussed below.

The HYSPLIT backward trajectories in Fig.  indicate that the humid air mass, in which the cirrus started to form, originated from the remote northern Pacific. The air mass ascended to upper-tropospheric heights 8–10 d before reaching the Polarstern at 86∘ N and 122∘ E. During the last 5 d of travel the humid and probably unpolluted Pacific air mass had the chance to mix with wildfire smoke (entrained from above). We hypothesize that these smoke particles then served as ice nuclei in the cirrus formation process. Remaining marine particles may have served as INPs as well but are usually regarded as inefficient INPs .

The data analysis with respect to ice nucleation on smoke particles is explained in Figs.  and . We applied the water-activity-based immersion freezing model (ABIFM) that allows predicting of immersion freezing under cirrus conditions . The following equation is used to compute the number concentration of smoke INP for the immersion freezing mode : 1nINP,I=sJhet,IΔt, with the surface area concentration s of the smoke particles in cm2 m-3, the ice crystal nucleation rate coefficient Jhet,I (in cm-2 s-1), and the time period Δt (in seconds) for which constant or almost constant ice supersaturation conditions are assumed. This can be the time period of a short updraft event (of a few minutes, 120–300 s) or the duration of the lifting period of a gravity wave of typically 600±200 s . The ice nucleation rate coefficient Jhet,I is a function of the organic material in the (liquid) shell of the smoke particle and the water-activity criterion Δaw . Δaw, shown in Fig. c, describes the difference between the ice melting conditions and the observed freezing conditions as a function of temperature and humidity as measured with the radiosonde. In the computation, Δaw is obtained from the difference between the actually occurring RHw (over water) and the ice-saturation-related RHw,sat value for the observed (actually occurring) temperature (with RHw and RHw,sat in decimal numbers). Δaw must reach values of 0.2–0.27 or even 0.29 to initiate significant ice nucleation (see vertical lines in Fig. c). Homogeneous ice nucleation is characterized by Δaw≈0.313 . As can be seen, Δaw takes values of 0.1–0.2 according to the radiosonde observations. Δaw values of 0.22–0.29 are only reached here during vertical motion of air parcels.

For a demonstration of our method, we chose to apply the ABIFM for leonardite (a standard humic-acid surrogate material) to represent the amorphous organic coating of smoke particles. An alternative natural organic substance would be Pahokee peat . Leonardite, an oxidation product of lignite, is a humic-acid-containing soft waxy particle (mineraloid), black or brown in color, and soluble in alkaline solutions. Both substances (leonardite or Pahokee peat) served as surrogates for HULISs in extended immersion freezing laboratory studies.

Similarly, we calculate the INP number concentration for the deposition ice nucleation (DIN) mode : 2nINP,D=sJhet,DΔt. The ice nucleation rate coefficient Jhet,D requires information on temperature T and ice supersaturation Si, also shown in Fig. c, and parameters describing the ice nucleation potential of the organic material. Also in the case of DIN, laboratory results for leonardite were used . A more detailed description of smoke INP computation may be found in .

Figure  provides an overview of all retrieval products necessary to evaluate the potential of smoke particles to serve as INP. From the aerosol lidar observations (backscattering in cloud-free conditions) we obtain estimates for the particle surface area concentration s . We show the aerosol conditions at different days in Fig. b (before, during, and after the cirrus event) and used the respective lidar backscatter profiles to estimate the surface area concentration of the smoke particles. It can be concluded that the smoke surface area concentration was in the range of 0.05–0.15 cm2 m-3 during the formation of ice crystals at 7–7.5 km and at 9.0–9.5 km in the case of the cirrus layer in Fig. a. In Fig. b, the estimated number concentration of large smoke particles n250 (with radius ≥250 nm, lower axis) is also shown. n250 may be interpreted as the overall reservoir of potential smoke INPs. It is assumed that larger particles provide better ice nucleation conditions than smaller particles. The uncertainty in the aerosol estimates is about 20 %–25 % (Table ).

Figure c shows the results of the INPC calculations with Eqs. () and (). The range of immersion freezing INPC (i.e., the length of the respective two horizontal bars in Fig. c) is obtained by computing nINP,I with Eq. () for Δaw of 0.225 and 0.25. The particle surface area concentration of 0.05 cm2 m-3 is used in these computations and Δt is set to 600 s. In the case of the deposition nucleation INPC, the shown INPC range (length of each bar) is obtained by computing nINP,D with Eq. () for the corresponding ice supersaturation values Si from 1.37–1.41 (7–7.5 km height, -55.4 ∘C) and from 1.41–1.44 (9–9.5 km height, -64 ∘C). As can be seen, nINP,I ranged from 1–63 L-1 at both heights (for the same Δaw range), and for nINP,D, we obtained values from 0.4–3 L-1 at 7–7.5 km height and <0.01 L-1 at 9–9.5 km height. The uncertainty in these estimates is unspecified. This is the first attempt to estimate aged smoke INP under realistic atmospheric conditions with parameterizations developed under laboratory conditions.

In the final step, we compare the immersion freezing INPC values in Fig. c, which can also be interpreted as predicted ice crystal numbers, with estimated ice crystal numbers derived from combined lidar–radar observations. The lidar–radar ICNC retrieval was already explained in Sect. . In this MOSAiC cirrus closure study, we used the strong and most accurate observations of radar reflectivity in the lower part of the virga zones from 4–7 km height. Since Eqs. () and () primarily deliver ice crystal nucleation rates (for Δt=1 s), we compare nucleation rate values in Fig. b. As pointed out by , the ice crystal downward flux (in m-2 s-1) is the most direct result of lidar–radar retrievals because besides radar reflectivity, the falling velocity is also measured with the Doppler radar. And this crystal flux rate can be interpreted as the ice nucleation rate as long as crystal–crystal collision and aggregation processes do not change the ice crystal number concentration too much, on the way from the cloud top region (ice nucleation region) to the lower part of the virga zones. However, this assumption may be strongly violated as concluded from CALIPSO lidar observations. They found that the ice crystal number concentration in the virga may be lower by a factor of 3–5 compared to the ICNC values at cloud top. On the other hand, the crystal nucleation rates we discuss below were quite low so that the aggregation effect may have been small (of the order of a factor of 2 or less for this Arctic ice cloud system). Other ICNC-influencing effects, e.g., secondary ice formation can be ignored in the case of cold cirrus clouds . The low lidar–radar-estimated ICNC values of 40–70 m-3 or 0.04–0.07 L-1 (mean values for the virga between 4–6 km height for the period from 06:00–12:00 UTC, not shown here) and about 150 m-3 or 0.15 L-1 higher up (based on less accurate radar reflectivity measured at 6–7 km height) corroborate the idea that the aggregation-related uncertainty in our lidar–radar estimate of the ice crystal nucleation rate at cloud top may have been low. For the time period from 12:00–14:00 UTC, the lidar–radar retrieval yielded ICNC values of 80–500 m-3 or 0.08–0.5 L-1in the virga layer from 4–6 km height and about 800 m-3 at 7 km height.

In our cirrus closure experiment, the goal is now to check how well the two numbers of the predicted and the estimated ice crystal nucleation rates are in agreement. The question behind this is as follows: can the smoke INP parameterizations reproduce these values given by the lidar–radar observations so that we can conclude that smoke particles most likely controlled the evolution of the cirrus deck and the virga and respective ice crystal microphysical properties? In the case of a successful closure experiment, we may even conclude that the applied INP parameterization is appropriate to be used in models to predict the smoke impact on cirrus formation.

As shown in Fig. b, the lidar–radar retrieval yielded ice crystal nucleation rates of 10–200 m-3 s-1 and 50–1000 m-3 s-1 (derived from the crystal flux observations between 4 and 6 km height) for the cirrus period from 06:00–12:00 UTC and 12:00–14:00 UTC, respectively. The prediction (based on INP retrieval with Eq. () for Δt=1 s) reveals 1–100 m-3 s-1 at cirrus top layer level. This can be regarded as a reasonable agreement keeping in mind that the particle surface area concentrations may have been underestimated, stronger updrafts may have occurred causing significantly higher Δaw values than 0.25 (at a value of Δaw=0.27 we obtain an order of magnitude higher nucleation rates), or leonardite is not representing well the organic material in the liquid shell of the smoke particles. With Pahokee peat parameters we obtain a factor of 3 higher ice crystal nucleation rates (or INPC values).

To sum up, heterogeneous ice nucleation is a complex process, especially in the case of organic aerosol particles. However, the successful closure, indicated by a reasonable match between the predicted ice crystal nucleation rate (from the nINP,I computation) and the estimated ice crystal nucleation rate (from lidar–radar observations) shows that the wildfire smoke was probably able to trigger cirrus formation (before the ice supersaturation onset for homogeneous freezing was reached) and to control the further evolution of the ice cloud system. The facts that (a) the radiosondes showed permanent ice supersaturation conditions in the virga over the day (a sign of low ICNC values in the virga) and (b) that the lidar–radar retrievals pointed to rather low ice crystal nucleation rates are independent indications that the heterogeneous ice formation was most probably responsible for the observed cirrus evolution or at least controlled the microphysical properties on 6 December 2019. It is clear that many more closure studies are needed to obtain a statistically trustworthy view on smoke and the role in cirrus ice nucleation. The respective extended data analysis will be part of our MOSAiC data analysis.