For the in situ measurements we used a Cessna Grand Caravan 208B (denoted Caravan), which is equipped with a meteorological sensor package . This small research aircraft combines several advantages for the investigation of small-scale phenomena in the ABL such as the strong single-engine power, high manoeuvrability and robust design. It is equipped with a high-accuracy inertial reference system (IRS) for position and attitude determination and a meteorological sensor package mounted under the left wing. describe the details of the measurement instrumentation and the corresponding uncertainties for the high-frequency, 100 Hz, measurements of pressure, temperature, humidity and wind vector. The main results of the measurement accuracy are summarized in Table .

We conducted six measurement flights during two campaigns in June 2012 and July 2013 as listed in Table . Flights 1, 2 and 6 were conducted over relatively flat terrain north of the Alps and west of Munich (Germany), with smooth hills covered by fields and woodland. On the first two flight days a high-pressure influence was dominating. The wind and wind shear were moderate from the western direction during flight 1 and very weak during flight 2. Examples for the clouds during the first two flights are shown in Fig. . During flight 6 the wind was moderate from the north-west and in the rather humid surrounding (rh>70 %) the cloud cover was higher and the cumulus clouds were situated in lower levels compared to the other flight days. Flights 3 to 5 were devoted to the investigation of convective clouds over alpine orography. The clouds developed above the mountain peaks during strong high-pressure influence with weak southerly wind. The convection tended to start above distinct points above the mountain ridges drifting north during its life cycle. The flight tracks are shown in Fig.  and information about the flight conditions can be found in Table .

We chose a similar flight strategy for all flights in order to achieve comparable data sets. Each flight started and ended with a vertical profile to obtain information about the undisturbed atmosphere outside the cloud. During ascent the cloud base and top were defined visually and a mean wind direction was estimated from the on-board quick-look data. With this information the operator defined the flight directions “along” and “across” the mean wind and up to three height levels within the cloud. In some cases, also transects below cloud level were flown. Figure  shows the definitions of flight levels and directions as well as the main flight pattern, which is shaped like an 8. We also performed a simple reverse-heading pattern, which allows for a high transect rate and facilitates the relocation of the target cloud. Besides the single-cloud sampling we also performed longer straight flight legs in different directions and levels in order to gain broader statistics of the cloud properties.

The target clouds were selected visually during the flight. The identification of the cloud boundaries is realized in two steps. First, a digital time mark set by the operator during the flight gives a rough estimate of the location. As a second step, we take the signal of relative humidity to determine the exact cloud boundaries. Thus, the cloud starts and ends with humidity saturation as measured by a Ly-α absorption hygrometer , which has a response time faster than the acquisition frequency of 100 Hz.

We require a cloud diameter of at least 200 m to avoid very small cloud filaments. Such a cloud transect typically includes about 300 data points. This limit left us with 191 cloud transects including 17 different individual clouds which were repeatedly penetrated. Other authors required different minimum cloud lengths; the scarce resolution of models or earlier measurements required higher thresholds of ≈500 m e.g.. More recent measurements, for example , required a minimum length of 200 m, and one of 50 m.

Several factors complete the identification of a cloud. A single cloud often consists of more than one updraught. It can contain large gaps above its base, which makes it difficult to distinguish it from other clouds in the vicinity. Figure a shows an example. The cloud consists of an active updraught near the upwind side of the cloud separated by a gap at higher levels from an older, already decaying updraught further downwind, but joined through a common cloud base. For the data evaluation, we have used the flight protocol and video recording to confirm the common cloud base. We also use a subset of 94 transects for which gaps in the transects above common cloud base were at most 150 m and less than 30 % of the cloud diameter. The cloud definition is summarized in Table . The existence of cloud gaps is in line with recent measurements e.g.. The detection of the common cloud base it hardly possible with a fully automatic cloud analysis, but inevitable with our observations during the measurement campaign.

We classified the cloud transects in terms of cloud region (bottom, middle, top), along- or crosswind transects and terrain (lowland, mountains). A further criterion considers the activity status of the cloud, where we require a positive median buoyancy inside the cloud for active clouds. The numbers of selected cloud transects representing the different criteria are listed in Table .

No agreement exists what for constitutes a subsiding shell. originally defined a 50–100 m range of negative vertical wind directly outside the cloud. use a range of 50 m within and 200 m outside the cloud. split the subsiding shell in an inner and outer shell, where the inner shell has negative vertical velocities and negative buoyancy. It is driven by the negative buoyancy after mixing and evaporation at the cloud boundary , and thus can partially also appear inside the cloud. The outer shell has still negative vertical velocity but positive buoyancy. Generally, the existence of the subsiding shell is identified by a negative peak of the mean (or median in a non-Gaussian process) vertical velocity right outside the cloud boundaries. Due to the turbulent character of the cloudy environment, a single representation of a cloud transect will usually not exhibit the characteristics of a subsiding shell. The mean distribution of the vertical velocity gives insight into the strength and depth of the subsiding shell. We investigate the width of the shell relative to the cloud diameter which accounts for the strong variability in cloud size. A circular subsiding shell with a width of 20 % of cloud diameter has an area approximately equal to the embedded cloud.

In order to assess the structure of the subsiding shell we analyse the properties of the up- and downdraughts in the vicinity of the cloud. In accordance with , we define a downdraught (updraught) as the region where the vertical velocity is below -0.2 m s-1 (above +0.2 m s-1). The small deviation from zero accounts for small-scale turbulence inside the up- and downdraughts and also corresponds to the measurement uncertainty in the system. Thus, regions with small vertical velocity are disregarded. We omit up- and downdraughts narrower than 10 m. Furthermore, where the gap between two neighbouring downdraughts (updraughts) is smaller than 10 m and the vertical wind does not exceed +0.2 m s-1 (fall below -0.2 m s-1) the up- or downdraught is considered a single one. They are estimated for the cloud region and up to 0.5 cloud diameters away from the cloud boundary. Three different categories of up- and downdraughts are distinguished: inside the cloud, at the cloud boundary and in the environment. The up- or downdraught at the cloud boundary is situated partly inside and outside the cloud.

During the day of flight 2 shallow convection formed around midday in a low-wind situation with weak high-pressure influence. Compared to the other flight situations the horizontal wind and wind shear of ≈ 1 m s-1 km-1 were very weak – at least up to the highest flight level. Some meteorological parameters of the environmental air are listed in Table .

Figure a shows an example cloud of this flight including a narrow cloud turret, which grew fast above the broader and longer persisting cloud base. After 5–10 min the turret (the upper part of the cloud) dissolved in the relatively dry surrounding air and gave way to a new updraught, while the cloud base persisted.

Figure  shows measurements along a crosswind transect flown in the upper part of another cloud during flight 2. The relative humidity in Fig. 5a shows a compact cloud with small cloud gaps in the western part indicated by subsaturation. Here, also the vertical wind velocities (Fig. 5b) are small compared to the eastern half where updraughts of up to 5 m s-1 are present. Also, the buoyancy shown in Fig. 5c is increased in the updraught region, while the pressure perturbation in Fig. 5d is significantly negative in the dissolving (or decaying) part of the cloud.

Outside the cloud boundaries, a clear signal of sinking air with magnitudes up to 3 m s-1 is present. On the left boundary an ≈ 200 m wide region of downdraughts starts already within the cloud. On the right side the downdraught region is ≈ 300 m wide with a distinct minimum about 150 m away from the cloud boundary followed by a weak subsidence region. It is important to note that due to the turbulent character of the cloudy environment the representation of a single cloud transect cannot give distinct information about the existence of the subsiding shell.

Not many of the investigated individual transects possess such distinct downdraughts directly outside of both cloud boundaries. For example, Fig.  shows humidity and vertical wind for four different transects for the same cloud in north–south direction (along the main wind direction). From the video recording and operator's notes there is strong evidence that all cloud parts have a common base, even though rather large subsaturated regions occur (e.g. Fig. c). Such gaps occur very frequently when weaker decaying cloud parts and regions with stronger updraughts tend to line up along the mean wind direction. It is almost impossible to recognize the vertical wind structure from one transect to the other, which is due to the turbulent nature, the high spatial and temporal variability, and transient behaviour of the flow in the cloud. Apparently, not even the updraught (downdraught) regions can be identified as quasi-steady “coherent structures” as is sometimes the case in small-scale turbulent flows. However, in Fig. c and d the main updraught might be the same, but for the rest of the transects the vertical velocity structures are different. This is similar for many transects in other clouds (not shown).

Figures  and  exemplify the large variations in strength and diameter or distance of the downdraughts in the vicinity of the cloud boundaries. We also find updraughts or vast regions of downdraughts near the cloud. Downdraughts are also frequent within the cloud itself, especially in the vicinity of cloud gaps (see Fig. c near position 0.25).

We find a similar distribution of the vertical wind also for the transects of the other flights. The turbulent character of the cloud environment is obvious. The up- and downdraughts seem to be rather randomly distributed with strong up- and downdraughts within the cloud as well as in the environment.

Figure shows the median vertical velocity distribution for all the cloud transects. Note that the spatial coherence of the individual transects is lost with the representation of the percentiles. The median vertical velocity has a distinct maximum within the cloud, which is slightly shifted towards the upwind side. The vertical wind minimum outside of the cloud boundary is the subsiding shell. The vertical velocity becomes already negative well within the cloud. Thus, the average cloud boundary experiences downward motion. The minimum slightly outside of the cloud boundaries is stronger on the downwind side. Further away from the cloud the downdraughts become weaker. The 75th and 90th percentiles have no downdraughts at all while the 10th and 25th percentiles show continuous negative vertical velocity. The minimum near the cloud boundary is visible for all percentiles, but is weaker for the 75th and 90th percentiles.

Figure  shows the median vertical wind distribution for different cloud categories stratified by cloud activity, level within cloud, underlying terrain, and along or crosswind transects. The 95 % confidence interval was computed at each point along the scaled transect by bootstrapping (1000 repetitions with replacement) and is shown in grey. Even taking the uncertainty resulting from the limited sample size into account, the median vertical velocities for all active transects in Fig. a, except the bottom one, are clearly distinguishable between the interior and exterior of the cloud. Such a distinction is not possible for the inactive transects shown in Fig. b, especially since even bootstrapping will underestimate the uncertainty due the smaller sample sizes for this class .

Active clouds (except at the bottom level) have pronounced updraught regions and a subsiding shell at the boundaries. The strongest updraughts are found at cloud top level. The most distinct downdraught regions at the cloud boundaries are present on the downwind side of the transects of the centre level and the clouds above mountains. They have a broad region of sinking air, which already starts well within the cloud. Looking at the active crosswind transects we find this wind minimum as well, but here the vertical wind almost vanishes within half a cloud diameter. The upwind side of the active along-wind transects have almost no downdraughts with a very narrow minimum right outside the cloud boundary. The inactive transects show high variability of the wind signals inside and outside of the cloud. At cloud mid and top level they do not show any strong updraughts.

Figure  shows the histograms of the vertical velocity inside the cloud and within 20 % outside of the cloud diameter. The distributions obviously differ in size and shape, with some statistical values summarized in Table . In the cloud the mean vertical velocity is ≈ 0.5 m s-1 and the skewness of the distribution is directly visible in the figure with increased frequencies of fast-rising parcels. However, only one of the selected transects has no negative vertical velocity at all within the cloud. Except eight cases, all the transects have downdraughts stronger than -1 m s-1 inside the cloud. In the shell the mean vertical velocity is significantly below zero for all four cloud boundaries. Especially on the upwind side the distribution is narrow compared to the other investigated parts. In the downwind and crosswind shells we find stronger downdraughts and higher variability compared to the upwind side. The highest variability in the vertical velocity is present within the clouds, which is also visible in Fig. . In Fig. a the stronger downdraughts in the downwind shell compared to the upwind shell become visible. The frequencies and magnitude of the updraughts are similar for the shell region on both sides. A separated analysis of the left and right crosswind shells does not lead to any significant differences neither for the median distributions nor for the histograms.

Figure  presents the median distribution of the relative humidity and horizontal along-wind perturbation as well as the buoyancy and the horizontal pressure perturbation for the 191 selected cloud transects. The median relative humidity within the cloud is saturated, but the 10th percentile is significantly below. Due to the definition in Table , all values are rh=100 % at the cloud boundaries. Outside of the cloud boundaries the relative humidity decreases rapidly. The gradients are stronger on the upwind side and the median value is significantly enhanced on the downwind side for at least half a cloud diameter, which can be explained by the humidity halo on the downwind side of the cloud . The mean horizontal wind component along the flight track (uac) in Fig. b is significantly reduced within the cloud, where also the strongest updraughts are found. For the 10th and 25th percentiles this signal is most pronounced. It is enhanced on the upwind side and matches the mean values on the downwind side. This feature is only present in the along-wind transects, while it is not visible in the crosswind transects. It is strongest in the bottom level transects and vanishes in the top level (not shown). The distribution of uac is also characterized by a high variability which is similar to the vertical wind variability.

Figure c shows that within the cloud a mean upward motion coincides with enhanced buoyancy, while on both sides outside of the cloud the buoyancy is almost zero on average. On the upwind side a weak negative peak is indicated with a strong and clear gradient through the cloud boundary. This gradient is much weaker on the downwind side of the cloud where values near zero are present well within the cloud. The median pressure perturbation (Fig. d) is small and with magnitudes of a few pascals similar to the sensor resolution (2 Pa). A weak negative anomaly is visible within the cloud which is counteracted by a positive contribution especially on the upwind side of the cloud. However, the percentiles show that significant deviations of the hydrostatic equilibrium are frequent both inside and outside of the clouds.

Even though the clouds were actively chosen during the flight with a focus on vital clouds, many of them contain big cloud gaps. Different rising plumes, decaying cloud parts with strong downdraughts, and also subsaturated air parcels entrained into the cloud coexist and build the entity of a single cloud. From the chosen cloud transects nine cases have no cloud gaps at all. For 25 cases the fraction of cloud gaps relative to the cloud diameter exceeds 50 %. For the 25th percentile, the median and the 75th percentile we estimate a cloud fraction of ≈10 %, ≈20 % and ≈40 %, respectively.

In order to judge the robustness of the results in terms of cloud definition, we have repeated the analyses for the stricter criteria including restriction 4 and 5 as defined in Table . Thus, we omit the transects with a fraction of cloud gaps of more than 30 % or a cloud gap exceeding 150 m. For the new analysis we select the more homogeneous clouds and neglect the less active or more complex ones, so that just 94 out of 191 cloud transects remain. In Table  the numbers of total occurrences are listed by the numbers in brackets. In Fig.  the respective median distribution for the reduced sample of 94 ideal clouds is represented by the green line. It is obvious that neglecting the less active clouds leads to stronger updraughts. However, the distribution at the cloud boundary and in the cloud-free region remains almost unchanged. Also the histograms of the vertical velocity (not shown) remain qualitatively unchanged. The frequencies of the vertical velocities within the cloud are shifted towards higher values. The vertical velocity distribution in the shell regions are narrower compared to the results in Fig. . After all, the selection of the clouds does not substantially change the results.

In Fig.  we find a pattern similar to a subsiding shell on both sides of the cloud boundaries. However, due to the turbulent character of the cloudy environment the subsiding shell is not visible in most of the individual transects (e.g. as shown in Figs.  and ). Instead, we find a large variety of up- and downdraughts of different strength and size inside the cloud, in the environment and also around the cloud boundaries. According to the definition in Sect.  we analysed the 191 cloud transects and found 1735 downdraughts and 1495 updraughts. This corresponds to an average of nine different downdraughts and eight updraughts for each cloud transect including half a cloud diameter around the clouds. Within the clouds, the average number of up- and downdraughts is approximately equal (≈4 of each), but the updraughts are about 25 % larger. In the environment, on the other hand, downdraughts are slightly more frequent and larger than the updraughts. Directly at the cloud boundaries, finally, the downdraughts are twice as frequent and larger as compared to the updraughts. There are only 328 up- and downdraughts in the cloud boundary region and in 56 cases no significant up- or downdraught was identified. The increased frequency of the downdraughts at the cloud boundary leads to the subsiding shell in the mean distribution of the vertical wind. A summary of the exact numbers and the main properties is given in Table . Most remarkably, the median sizes of the draughts are much larger directly at the cloud boundary compared to the other regions. There, also the median and variance in the vertical wind are strongest. This indicates that we find a large number of relatively small draughts within the cloud and also in the environment. These smaller draughts are less frequent at the cloud boundary which leads to the characteristic distribution shown in Fig. . The distributions of the draughts within the cloud and in the environment are very similar, only the frequency of large diameters is reduced in the environment. The distribution within the cloud corresponds well to the results of . They find some larger draughts (i.e. D>2 km) than in our sample, because our investigation is limited to shallow convection. The distribution at the cloud boundary is different. While at the cloud boundary the small diameters are rare, sizes of several hundred metres are most frequent for the downdraughts and somewhat larger for the updraughts. The largest of these updraughts often cover significant portions of the cloud and can form the main (i.e. the largest) updraught of the cloud. In Table  also the statistics for the main updraughts as a subset of the cloud updraughts are listed. These main updraughts have a median length similar to the draughts at the cloud boundary and a strength and variability in the vertical wind that is higher compared to the other categories.

The numbers of downdraughts at the cloud boundary are almost equally distributed around the cloud but they have smaller diameters at the upwind side compared to the crosswind and downwind sides as listed in Table . The updraughts are slightly more frequent on the upwind side. They are smallest at the crosswind side and twice as large on the downwind side. While the major parts of the downdraughts lie outside of the cloud, the updraughts are situated more inside. Table  in the Appendix provides detailed information about the mean properties of the up- and downdraughts at the cloud border with respect to the different transect categories. A comparison of the draught diameter and median vertical velocity in the scatterplot of Fig. a shows that for all three categories larger draughts often have stronger vertical winds. For the larger diameters above D>200 m the smaller magnitudes of the vertical velocity are most often in the environment. The distribution and the high variability of the larger draughts inside the cloud and at the boundary are very similar, but the occurrences of smaller draughts at the cloud boundary is clearly reduced. In the comparison of the median vertical velocity and buoyancy in Fig. b it becomes clear that updraughts and downdraughts have both positive and negative buoyancy and thus the median values are small. The draughts in the cloud have more cases with positive buoyancy and the draughts in the environment more cases with negative buoyancy. However, it is clearly visible that the negative buoyancy is more frequent with downdraughts and positive buoyancy is more frequent with updraughts. The exact values are given in Table . There, also the correlation coefficients of the vertical velocity and buoyancy are listed. The correlation for the downdraughts is small and at the cloud boundary even slightly negative. Thus, stronger downdraughts do not necessarily have more negative buoyancy. The updraughts have a higher correlation except for the cloud-free environment.