Two airborne field campaigns were performed during February and March 2010, and January and February 2011, based at Rothera Research Station (67∘34′ S, 68∘08′ W) on the Antarctic Peninsula. During these two periods, flights were made to study a variety of meteorological phenomena, including boundary layer (Fiedler et al., 2010), orographic flow (Elvidge et al., 2014), and cloud studies. A total of 64 flights were completed during these periods. This paper mainly considers 24 of these flights (12 per year) where detailed cloud microphysical measurements were collected. Fig. 1 shows the flight tracks of these 24 cloud flights over the two periods. The sampling took place on both sides of the peninsula, flying between 61 and 73∘ W. The predominant cloud types by far were stratus or altostratus, normally in multiple thin layers. A cross-section, across the peninsula, of the flight tracks is shown in Fig. 2. It should be noticed that the flights in 2010 did not go as far west and that sampling on the western side of the peninsula was largely limited to altitudes below 2000 m west of 69∘ W.

The observations were made with the British Antarctic Survey's instrumented Twin Otter aircraft (see King et al., 2008). This aircraft is fitted with a variety of instruments to measure temperature, humidity, radiation, turbulence, and surface temperature. The aircraft was also fitted with a Droplet Measurement Technology Cloud, Aerosol, and Precipitation Spectrometer (CAPS) (Baumgardner et al., 2001) carried on a wing-mounted pylon.

The CAPS instrument contains three discrete instruments: the Cloud and Aerosol Spectrometer (CAS), the Cloud Imaging Probe (CIP), and the Hotwire LWC Sensor. Data from the LWC sensor were only used in this study to help validate the CAS data. The CAS and CIP are described in turn below.

The CAS measures the diameter of particles between 0.5 and 50 µm at a frequency of 1 Hz. While the CAS used in this campaign did not have a full anti-shatter inlet, some modifications had been made to reduce the effect of shattering on the inlet by removing the shroud that was originally fitted to the inlet. A previous study (Grosvenor et al., 2012) has already looked at a small subset of these data and reported errors with the data from the CAS instrument. In particular, it appeared to be over-counting when integrated water content from the CAS was compared with the LWC sensor. After investigation this was found to be due to the air accelerating in the tube of the CAS instrument. Studies in the Cambridge University Markham wind tunnel using a fine pitot tube to measure the speed-up within the tube showed an increase that would result in the count being increased by 1.47, and this has been accounted for in this latest study. When this correction is applied, the liquid water content calculated by integrating the CAS data for most flights agrees within 15 % with the Hotwire LWC Sensor – part of this discrepancy is attributed to the Hotwire LWC Sensor's tendency to under-read at high values of LWC.

The CIP images particles between a diameter of 25 µm and 1.5 mm, with 25 µm pixel resolution, and had not at the time of this campaign been fitted with anti-shatter tips. However, a study of the particle inter-arrival times indicated very few shattered particle, and these were removed by eliminating particles that arrive within 1 µs. This study only uses data from the CAS and CIP, although the hotwire was used to help validate the CAS data.

The CIP instrument produces shadow images of the larger cloud particles and small precipitation size particles onto a charge-coupled device (CCD) array. Data processing is performed on these images to derive size-segregated ice crystal and large liquid drop number concentrations. Particles that are imaged by the extreme ends of the CCD array are rejected, and this means that the effective collection volume, used to calculate the concentration, gets smaller as the particles get larger. Further details concerning the data processing and quality control of the CIP images can be found in Crosier et al. (2011). Particles are separated into ice and liquid categories based on their circularity, C: C=P2/4πA, where P is the measured particle perimeter and A is the measured particle area (a minimum area of 50 pixels – equivalent to a spherical particle with a diameter of 200 µm – is used as smaller particles are not sufficiently resolved to discriminate between drops and crystals). Following previous studies (Crosier et al., 2011; Taylor et al., 2016), particles with circularities between 0.9 and 1.2 are classified as circular and therefore liquid drops. For particles with values from 1.2 to 1.4 the decision on whether to count the particles as liquid or ice was made on a flight-by-flight case after looking at images. Particles with values over 1.4 were always counted as ice. The efficacy of the phase separation and choice of the circularity thresholds was confirmed by examining the sorted images “by eye”. The ice water content was calculated using the Brown and Francis mass-dimension parameterization (Brown and Francis, 1995).

For this study it has been assumed that the clouds are mostly mixed-phase clouds and that the particles observed by the CAS (less than 50 µm) are all liquid, while the CIP particles are characterized as either liquid drops or ice depending on the value of their circularity (see above). The number of large liquid drops (greater than 50 pixels – 200 µm equivalent diameter) seen by the CIP is very small on all flights. The total number of particles seen by the CIP is much lower than that observed by the CAS, and the number of particles seen by the CIP (less than 50 pixels – equivalent to a circular drop of 200 µm diameter) is for all flights less than 2–3 % of the number seen by the CAS; thus we have ignored these particles whose phase we do not know. The version of the CAS probe used for this study was not able to measure polarization, and so it was not possible to attempt to discriminate between solid and liquid with the CAS. However, it seems likely that the assumption that all CAS particles are liquid is valid since we find that the liquid water calculated assuming the CAS particles are all liquid agrees reasonably with that calculated from the hotwire probe on the CAPS (as stated above). Moreover we see a distinct peak (not shown) in the CAS size spectra when in mixed-phase clouds, indicative of drop formation.

In this study the average cloud properties over all the flights shown in Fig. 1 are considered. Each flight did not go to every longitude bin, and even when a flight visited a longitude bin it did not necessarily enter a cloud. For example for the bin nearest Rothera the temperature and humidity show a large number of points as most flights will have data in this bin. For the cloud parameter graphs there are fewer points as it was normal to avoid clouds during take-off and landing and flying close to the mountains.

The CAS instrument has also been used to examine the aerosol concentrations outside clouds. To improve statistics, all flights made in the study area during 2010 and 2011 when the CAPS probe was operational have been used. This includes an extra 31 flights that were conducted primarily to investigate the boundary layer or the large-scale flow but still had clouds present in the sky. To ensure that these measurements only include cloud-free conditions, the data were filtered by removing periods when there were particles larger than 1 µm. The CAS instrument only measures particle larger than 0.5 µm and so only gives us a measure of the larger aerosols, but it is assumed that this will bear some relation to the number of CCN and INPs available. In the case of INPs the relationship between aerosols greater than 0.5 µm and INPs is represented in the parameterizations developed by DeMott et al. (2010).

Figure 3 shows the mean sea level pressure from the ERA-Interim reanalysis for the periods of the two campaigns. The flow during 2010 is generally slack, while in 2011 the Amundsen Sea low to the west of the Antarctic Peninsula has intensified and moved east. This brought a more northerly flow across western side of the peninsula, and this could be expected to bring warmer air. However, looking at the temperatures as a function of longitude from the aircraft flights in Fig. 4, we see that 2011 is actually colder in the west, and this is also seen in the temperature fields from the ERA-Interim reanalysis as well as in the radiosonde ascents performed daily at Rothera Research Station (not shown). The cold in the west is a result of air being pulled around the tip of the peninsula from the Weddell Sea, and this is confirmed by the back-trajectory analysis reported later in this paper. The relative humidity (RH) plotted as a function of longitude (Fig. 5) shows more variability than the temperature record, and there is no clear difference between the years except an increase of RH in the west in 2010 in spite of the larger temperatures (Fig. 4), indicating an increased amount of water vapour at that time.

The number concentration of liquid drops (cm-3) averaged first over 1∘ longitude bands for each flight as well as each band averaged for the 2 years is shown in Fig. 6. This means that each flight has the same weight, giving a more representative final value. This averaging method also means that a long flight in a particular longitude bin does not dominate the overall average. Each year is plotted separately with 2010 in black and 2011 in red. At each longitude bin the average from each individual flight is shown as a point – the small number of points at each longitude means that it is not reasonable to calculate the standard deviation, but the spread of the points gives some idea of the data variability. To improve statistics the data were binned further into two large bins, one on each side of the peninsula: one from 67 to 74∘ W and one from 60 to 65∘ W. These values, along with their standard deviations, are reported in Table 1. The CAS size spectrum showed peaked distributions around 8–12 µm, illustrating the condensational growth of supercooled droplets (not shown). The average number of droplets varies from around 60 to over 200 cm-3 (Fig. 6), which is typical of concentrations found over the open ocean away from possible sources of CCN from continental land masses (Pruppacher and Klett, 1997; Chubb et al., 2016). Table 2 gives the statistical significance of differences between both sides of the peninsula as well as differences between both years, using a t test. The most significant difference (at the 99 % level) for liquid drops is in the east of the peninsula between 2010 and 2011. A less significant but still noticeable difference exists between either side of the peninsula in 2011 (90 %). These relationships were not found in 2010 (< 80 %).

The LWC in clouds averaged by longitude for the 2 years is shown in Fig. 7. Again to get better statistics, the values of LWC have been averaged on both sides of the peninsula using the same size longitude bins that were used for liquid drop numbers. A significant difference between the 2 years is found at the 90 % level in the east and 95 % level in the west (see Table 2), although the east–west difference in both years is not significant. The single point of high LWC in 2011 at 73∘W is the result of a small number of flights into active frontal systems with large amounts of liquid water.

The ice particle numbers and ice water content, averaged in the same way as the drops in Fig. 6, are shown in Figs. 8 and 9 for areas in the cloud that were at least partially glaciated – that is, non-circular particles were observed in the CIP. The number of ice particles observed is roughly 5 to 6 orders of magnitude lower than the number of cloud droplets, and the relative amplitude of the variability higher (the standard deviation of liquid droplet number concentration is about 50 % of the average values, while it is as high as or higher than averages for ice crystal number concentrations). Also, not all the clouds investigated were glaciated to any extent, and so there are slightly fewer measurements (Table 1) for ice crystals than for the drops – except for the east of the peninsula in 2010 when there was an observation of a completely glaciated cloud. First looking at the number concentration of crystals, Fig. 8 and Table 1 show that in 2011 there were more crystals on both sides of the peninsula than in 2010. However, Table 1 shows the standard deviation of the crystal numbers is large. Table 2 shows that differences are not significant between the 2 years (< 80 %) in the west, but significant in the east (90 %). Differences are also significant between either side of the peninsula in 2011 (at the 95 % level), however not in 2010 (77 %).

The ice water content (Fig. 9) shows a similar trend to crystal numbers in Fig. 8. However, in this case using the averaged values on each side of the peninsula (Table 1), the difference between both sides in 2010 is significant at the 90 %level.

The distribution of crystals with atmospheric temperature for the 2 years is shown in Fig. 10. Median ice crystal number concentrations have been derived over 0.5 ∘C bins along with the associated median absolute deviations. The distributions are very different in the 2 years and between the east and west of the peninsula. In the west, both years have a peak ice number concentrations around -5 ∘C. As reported by Grosvenor et al. (2012), for some of the 2010 flights only, this peak is most probably related to a secondary ice production process known as the Hallett–Mossop process (Hallett and Mossop, 1974). This consists of droplets shattering upon freezing when impinged by existing crystals (riming process), thus causing a cascade of crystals. This process only operates efficiently in a narrow temperature range (approximately -3 to -8 ∘C) with an optimum around -5 ∘C and requires a large enough ratio of number concentrations of small drops (nS; < about 13 µm) to large drops (nL; > about 24 µm) (Mossop, 1985). Typical observed concentrations by Mossop (1985) of large (small) drops were 10–80 cm-3 (5–60 cm-3), with the ratio nS / nL ranging between 0.1 and 5. In the present work, median values per flight for nS, nL, and nS / nL (where peaked concentrations are observed at -5 ∘C) are in the following respective ranges: 16–90 cm-3, 9–33 cm-3, and 0.5–13 (with flight minima always above 0.1). Note that no constrain is available in the present dataset to properly estimate the velocity of the riming particles. However, Mossop (1985) suggests that splintering can occur at velocities as low as 0.2 m s-1 and observes no abrupt drop of splinter production from 1 to 0.55 m s-1, meaning that the rimer velocity does not show a lower cut-off preventing the Hallett–Mossop process from happening. In addition, other mechanisms known to trigger ice multiplication do not operate in the -8 to -3 ∘C temperature range but at colder temperatures (< -10 ∘C), for example ice–ice collision (Takahashi et al., 1995 and Yano and Phillips, 2011); freezing of large drops with ice spicules formation (Lawson et al., 2015); or breakup of crystals, which preferably requires irregular shapes like dendrites that are favoured at low temperatures (Bacon et al., 1998). The main limitation to an absolute identification of the process as Hallett–Mossop is the resolution (both temporal and spatial) of the CAPS probe to observe the process in detail from its start as only the resulting larger crystals are observed and not the first ice.

In the east there is no clear peak at the high temperatures (> -10 ∘C) in 2011; there is only a small peak in 2010. It suggests that relatively less secondary ice production is observed for the eastern side of the peninsula. It can be seen that during 2011 the number of ice crystals increases from -10 ∘C to around -20 ∘C, and this probably corresponds to primary ice production (where an INP active at a given temperature interacts with a droplet or vapour to form one crystal). Ice nuclei parameterizations relying either only on the temperature (Cooper, 1986) or on both the temperature and the aerosol (> 0.5 µm) concentration (DeMott et al., 2010) are plotted in Fig. 10. They are meant to account for primary ice nucleation and predict increasing INP concentrations from -10 to -20 ∘C. For the DeMott et al. (2010) parameterization, aerosol concentrations bracketing the present average observations (Sect. 3.3, Fig. 11) were used as input. Comparing the trends of either of the parameterizations to the ice crystal distributions also strengthens the idea that the ice crystal distributions peaking around -5 ∘C are related to secondary ice production processes.

It is to be expected that the number of drops and ice crystals will be controlled by the number of aerosols acting as CCN and INPs, excluding for the moment the role of secondary ice production. The instruments that were fitted to the Twin Otter in 2010 and 2011 did not allow the full range of aerosols to be measured. However the CAS, which is part of the CAPS probe, measures aerosol size distribution down to 0.5 µm. The zonal variation of aerosol particles greater than 0.5 µm and less than 1 µm is shown in Fig. 11, and their average numbers in both years and on both sides of the peninsula are given in Table 1. Note that to improve statistics and have a better picture of the aerosol population in the region, 55 flights were used, i.e. including flights not primarily intended to perform cloud measurements but still equipped with the CAPS probe. Table 2 gives the significance of aerosol differences between years and regions. If the single point in the far west in 2011 is excluded (which is from one flight and is considered an anomaly), the 2 years to the west of the peninsula are very similar. To the east there is a significant difference between the 2 years (at the 95 % level), with 2011 having almost twice the concentration of aerosol as 2010. In 2011 there is also a significant difference between the east and west sides of the peninsula (at the 99 % level), with twice the concentration of aerosol in the east as in the west of the peninsula. Importantly, the same conclusions are drawn when using the 24 cloud flights only.

The average number of cloud droplets in the clouds (see Fig. 6) varies over a large range from just a few to almost 300 cm-3, and this probably reflects the large range in CCN found in the coastal areas of Antarctica. The mean droplet number concentration on each side of the peninsula is not untypical of values found in the mid-ocean (Pruppacher and Klett, 1997; Chubb et al., 2016) and reflects the position of Rothera exposed to air masses with origins in the South Pacific. Table 2 shows that it is only in the east that there is a significant difference (at the 99 % level) between the 2 years in the aerosol concentration, and this might be a result of the different source regions for particles between the 2 years.

To test this hypothesis, back-trajectory analysis using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model has been performed (Stein et al., 2015). Seven-day back trajectories were calculated using the National Centers for Environmental Prediction (NCEP) reanalysis meteorological field with starting points located at 60 s intervals along the track of each flight. Figure 12 shows the position of the low-altitude (≤ 300 m a.s.l.; see below) air masses 48 h backwards, along with their altitude (colour-coded), for both years and sides of the peninsula. The sea ice fractional coverage (at 25 km resolution), obtained from Nimbus-7 Multichannel Microwave Radiometer (SMMR) and Defense Meteorological Satellite Program (DMSP) SSM/I-SSMIS passive microwave data (Cavalieri et al., 1996), has been over plotted in blue. As in Sect. 3.3 (and in Fig. 11) 55 flights were used to investigate the average origin of the air masses in both years. A striking difference appears between 2010 and 2011, with 2011 showing more sources above the sea ice. Table 3 summarizes the relative proportion of air masses passing above sea ice compared to above open water/ice shelf (referred to as “other”) among all low-altitude air masses (i.e. below 300 m, approximate boundary layer height; Fiedler et al., 2010) and along all the back trajectories. The relative proportions are also indicated for the case when only the 24 cloud flights are used (number in parentheses in Table 3). We first comment on the analysis using the 55 flights. A sea-ice-covered region was defined as a region where fractional coverage is larger than 1 % (however the sea ice regions of interest in Table 3 normally have much larger values; see Fig. 12). These low-altitude air masses will have greater sensitivity to surface aerosol emissions. The statistics integrated over different durations are shown to illustrate the consistently dominating or increasing trends of the relative contributions of sea-ice low-altitude air masses among all low-altitude air masses. Whatever the time interval, low-altitude air masses sampled to the east of the peninsula always overpass sea ice regions relatively more often than they overpass other regions – with the influence of the sea ice being greater in 2011. Looking at Table 3, we can see that the sea ice has most influence, over most time periods, in the east in 2011, then less in the east in 2010, less again in the west in 2011, and least in the west in 2010. Thus, Table 3, along with Fig. 12, shows that (a) air sampled in the east is more sea-ice-influenced than air sampled in the west in both years, (b) air sampled in the east in 2011 has had a relatively longer sea ice track at low altitudes than in 2010, and (c) air sampled in the west in 2011 is more sea-ice-influenced than in 2010. Looking at the analysis restrained over the 24 cloud flights only, the above (b) statement becomes less clear as only the analysis over the period covering the previous 48 h shows a clear increase of sea ice influence in the east in 2011 (Table 3, second line), compared to 2010, but statement (a) and (c) still prevail and are even strengthened when considering these smaller statistics. Overall clear differences appear in the origin of the air masses when splitting the analysis into west–east and 2010–2011 comparisons. The larger concentration of aerosols in 2011, especially on the eastern side (Fig. 11), compared to 2010 (Table 2) could be explained by more air masses having longer tracks at low level over sea-ice-covered regions of the Weddell Sea. It is possible that sea salt on the snow-covered surface of the sea ice could easily be lofted into the air as blowing snow which sublimes to form sea salt aerosol. This has been suggested as an efficient mechanism for getting sea salt aerosol, which could act as CCN, into the boundary layer (Yang et al., 2008). This suggests that the increase in droplet number concentrations seen in 2011 on the eastern side of the peninsula (Fig. 6) could result from the increase in aerosol numbers. Figure 13 shows average vertical profiles of aerosol number concentration on both sides of the peninsula for 2010 (blue circle) and 2011 (red triangles), for the 55 flights used in Sect. 3.3 (note that the same vertical profiles plotted with the 24 cloud flights show similar differences between the years). The corresponding shaded areas indicate absolute minimum and maximum (non-null) values at each altitude level over the respective campaigns. Like the latitudinal averages of aerosols shown in Fig. 11, Fig. 13 shows that the interannual differences in aerosol concentration on the eastern side are much more pronounced (see the 99 % significance level in Table 2) than on the western side (93 % significance level). Interestingly, these differences occur almost exclusively below 2000–2500 m. Above that altitude, the concentration of aerosols is similar between both years and both sides (around 0.1–0.2 cm-3). The central part of the peninsula between 67∘W and 65∘W is also characterized by this concentration of about 0.1 cm-3 (not shown). This suggests that a homogeneous mixture of aerosols (at least in terms of number concentration, if not nature) prevails at altitudes above the height of the peninsula mountain barrier. Conversely, at lower levels, the Antarctic Peninsula barrier would help sustain different pools of aerosols on either of its sides. These aerosol pools could give rise to different cloud microphysics on either side of the peninsula, and a possible interpretation of the observations is that the increased number of droplets in the east in 2011 compared to 2010 could be the result of an environment more influenced by particle originating from sea ice regions.

Interestingly, the ice crystal number concentration (Fig. 8) does not follow the same pattern as the liquid droplets. A significant difference (at the 95 % level) is seen between the 2 years in the west, and a similar significant difference is also seen between both sides of the peninsula in 2011. However, this is not unexpected as, while aerosols are the source of CCN and INPs so that a correlation between the numbers of liquid drops and ice crystals might be expected, there is another process going on to create ice crystals. This is the secondary ice production due to the Hallett–Mossop process for temperatures warmer than -10 ∘C (introduced in Sect. 3.2). The peak of ice crystal numbers at temperatures above -10 ∘C can be clearly seen (Fig. 10), especially in the west, where as might be expected the temperatures are warmer. This is illustrated by Fig. 4, which shows each year's temperatures averaged by latitude along the aircraft tracks. In the east there is much less evidence of secondary ice formation for both years (Fig. 10), and this correlates with the similarly lower temperatures in both years (Fig. 4). At temperatures below -10 ∘C (Fig. 10) it can be seen that in 2011 there is a distinct peak in ice concentration, which suggests more primary ice production (see Sect. 3.2) in both the west and the east at temperatures approximately ranging between -10 and -20 ∘C, while there is almost virtually no primary ice production peak in 2010. Figure 14 shows the entire dataset of crystal distribution with temperatures for the 2 years (both sides at the same time) colour-coded with the altitude of the observation. The primary ice production mainly occurs at altitudes above roughly 2500 m, while the secondary ice production occurs almost exclusively below 2000 m – this of course is related to atmospheric lapse rate. This means that the primary ice production peak presence cannot be solely related to the increase of aerosols, which is observed mostly below 2500 m (see Sect. 4.1). As such, the lower primary ice production in 2010, and its presence in 2011 on both sides of the peninsula, can be linked to the colder 2011 temperatures compared to the 2010 temperature above 2500 m (not shown). Another possible factor in the lower ice production in 2010 is the different nature of INPs due to different air mass histories. It can be seen that in 2011 more air masses were coming from above sea ice regions (see Sect. 4.1). It has been suggested that biogenic particles found in sea ice could act as INPs (Burrows et al., 2013), and the same method for transferring from sea ice to the atmosphere (as the sea salt) could explain the different nature of INPs found in 2011 above 2500 m, leading to more primary ice production in 2011.