Processes contributing to Arctic cloud dissipation and formation events that bookend clear sky periods

The Arctic is predominantly cloudy with intermittent clear sky periods. These clear periods have profound impacts on the surface energy budget and lower atmospheric stratification, connected to a lack of downwelling longwave radiation in the absence of cloud. Despite the importance of clear sky conditions, an understanding of the atmospheric processes leading 10 to low-level cloud dissipation and formation events is relatively limited. A strict definition to identify clear periods at Utqiagvik (formerly Barrow), Alaska, during a five-year period (2014-2018) is developed. A suite of remote sensing and in situ instrumentation from the high-latitude observatory are analysed; we focus on comparing and contrasting atmospheric properties during low-level cloud dissipation and formation events to understand the processes controlling clear sky periods. Vertical profiles of lidar backscatter suggest that aerosol presence across the lower atmosphere is relatively invariant around 15 the clear period bookends, which suggests that a sparsity of aerosol is not frequently a cause for cloud dissipation. Further meteorological analysis indicates two active processes ongoing that appear to support the formation of low clouds after a clear sky period and have a link to surface aerosol concentrations; namely, horizontal advection which was dominant in winter and early spring and quiescent air mass modification which was dominant in the summer. During summer, the dominant mode of cloud formation is a low cloud/fog layer developing near the surface. This low cloud formation is driven 20 largely by air mass modification and pooling of aerosol particles near the surface under lower-atmosphere stratification.

The approach taken here is to utilise 5 years of measurements from a long-term measurement site at Utqiagvik, on the north coast of Alaska. The measurements include lidar backscatter (a proxy for aerosol concentration profiles), cloud radar, radiosondes, surface meteorology, and surface measurements of total aerosol concentration. Most of the analysis focuses on the ~1 hour period following cloud dissipation or preceding cloud formation that 'bookend' periods that are entirely cloud-free.
The analysis first considers the relationships between cloud dissipation/formation and aerosol profiles, comparing the profiles immediately after/before the transition with those for the clear periods as a whole (broken down by month), going on to consider the surface aerosol concentration either side of cloud transitions, and relationships between aerosol and lower tropospheric stability under both clear and cloudy conditions. This analysis provides no significant evidence for a causal link between aerosol properties and cloud dissipation/formation at the measurement site.
The analysis then considers thermodynamic and dynamic processes. This analysis leads the authors to conclude that "the onset of clear sky periods, and subsequently the end of clear periods, are primarily responsive to transient atmospheric forcing". For the onset of cloud they essentially conclude that under clear skies radiative cooling causes a fall in temperature and associated increase in relative humidity; ultimately saturation point is reached and provided there are sufficient aerosol present low cloud or fog will form. No firm conclusions are drawn about the processes resulting in the dissipation of cloud, other than the association with 'transient atmospheric forcing'.
These conclusions are rather generic and unlikely to help improve modelling of Arctic cloud.
The results remain of interest in providing a picture of typical conditions and some seasonal variations thereof, for periods of clear air bookended by low level clouds. There is considerable scope to improve this picture, however, and I recommend major revision before publication is considered.

General/major comments
While the aim of the paper is very worthwhile, I feel it ultimately fails to deliver robust conclusions. In part this is a, perhaps inevitable, result of the limitations of the data set. The aim is to understand what the processes are that lead to cloud dissipation/formation -transient events that are inherently linked to changes in local air mass properties over time. Measurements from a fixed site are, however, unable to distinguish between temporal evolution of the air mass properties resulting from in situ processes and the simple advection of a pre-existing spatial gradient in properties past the measurement site. This is a perennial problem for intensive, and/or long-term measurements. The authors attempt, but I think ultimately fail, to work around this problem by studying the statistics of an ensemble of cases. This provides correlations between measured properties associated with cloud transitions, and the hope is that probable processes can be inferred from these correlations. It is quite possible that observed behaviours might only be explicable by specific processes, and a fairly robust conclusion may be drawn. Sadly I don't think that is the case here.

Aerosol Analysis
The analysis of links with aerosol properties is quite extensive, but ultimately finds no causal links with cloud dissipation/formation. The extensive initial focus on aerosol is (I assume) prompted by results from the central Arctic Ocean where very low aerosol concentrations (< 10 cm -3 ) have been found to result in clear sky conditions even when the boundary layer is saturated, and several modelling studies have found that it is essential to accurately represent the aerosol in order to effectively represent the cloud and boundary layer structure. (as a side note, I find it odd that while the authors cite 3 modelling studies, all of which utilise the same observed case from the ASCOS project, they don't cite the original observational paper that first documented such CCN limited conditions and on which Sedlar is a co-author).
The CCN limited conditions in the central Arctic, are from a very different environment from the coastal site used here. The surface aerosol measurements in figure 6 and 7 show that concentrations rarely fall much below ~100 cm -3 , and are often much higher -far too high for aerosol to be the limiting factor on cloud formation. I think this possibility could have been ruled out much more easily by simply evaluating the surface concentrations (and perhaps relating them to the lidar profiles) for clear sky cases, without the need for the extensive analysis presented here.
The aerosol backscatter profiles show a consistent decrease with altitude through the boundary layer and across the top of the boundary layer and (former) cloud top. This is consistent with a surface source of aerosol. A surface source such as wind-blown dust would include some quite large particles with a significant sedimentation velocity, this would result in the sort of decrease with altitude observed here. No modification of aerosol concentrations by cloud is required.

Dynamics/thermodynamics analysis
The analysis in figure 8 reveals an interesting difference in thermodynamic behaviour in the hours prior to cloud formation between summer months (May-August) and the rest of the year. In the summer a decreasing trend in temperature (cooling) prior to cloud formation is accompanied by a decrease in dew point suppression -an increase in relative humidity. No such association is found for the rest of the year, where dewpoint suppression is more or less constant regardless of trends in temperature. The potential link to cloud formation in the summer is clear -increasing relative humidity will eventually result in saturation and condensation. The lack of change in dew point suppression in winter is ascribed to the cooling temperature trend resulting from advection (of increasingly dry air) rather than local cooling. No additional evidence is provided to support this supposition, and it is not clear why there should be a seasonal separation between local cooling and advection of cooler airmasses. Another possibility is that during the winter months the temperature is below freezing and the humidity of air is controlled by the saturation vapour pressure with respect to ice not water. Cooling will enhance this, resulting in growth of ice/frost by vapour deposition and keep the relative humidity with respect to water suppressed.
It is not clear that radiative cooling at the surface will necessarily explain cloud formation -cooling at the surface will tend to lead to increasing stable stratification, suppressing turbulent mixing and keeping the cooling localised to a shallow layer close to the surface. Air aloft might remain unaffected and at constant temperature. Eventually we might expect cooling to result in fog formation, but the formation of an elevated low level cloud depends on more than just surface cooling -mixing sufficient to maintain a more or less well mixed layer that cools as a whole, and an adiabatic profile so that the upper part of the layer saturates first. No attempt is made to distinguish fog and elevated cloud layers in the analysis, although this would seem to be an important distinction from the perspective of the process for cloud/fog formation.
The analysis of geopotential layer thickness trends I find wholly unconvincing. The data points in Figure 10 are mostly very scattered, and in most cases it would be hard to make out a convincing trend by eye. A line can always be fit to the points, but does not imply a robust relationship.
Further, I have serious doubts about whether the calculated tendencies are meaningful, even on a case by case basis. The trends are calculated from 2 consecutive radiosonde profiles prior to the cloud transition. This means, usually, over a 12-hour interval. The example clear sky case shown in figure 1 is barely 9 hours long. The 2 closest sondes preceding the onset of cloud at the end of the clear event actually span the dissipation of the preceding cloud. The later of the two sondes is 1.5 hours after the dissipation, and about 7 hours prior to cloud formation. I would suggest that the geopotential height trend calculated here is more relevant to the dissipation event than to the formation event to which it is actually applied.
Given that we have both clearing and cloud formation both occurring within an interval less than that over which a single geopotential height trend estimate is calculated, that rather suggests that any correlation between the two is suspect at best, and potentially entirely spurious. To make a really meaningful evaluation a much higher time resolution is required for the geopotential height trends. Maybe the output from an operational forecast model would provide a better measure here. Line 173: the authors note how low cloud and fog can be distinguished here, but never use this to separate out the cases, which I think is relevant for some of the process identification.

Detailed comments
Line 189: the authors note a peak in the variability in backscatter between a few 100 metres and ~1km. This is presumably a result of variability in BL top, and the associated gradient in aerosol & backscatter across it. This is not mentioned here, and throughout the discussion of figure 3 the profiles are discussed in isolation from any consideration of BL depth. I found this frustrating -there are several places where a feature of these profiles is discussed and some inference made, where my first reaction was that this was a result of variation in BL depth and this point was apparently being missed (see notes below). Same with figure 4. Only much later, at figure 5 is this point acknowledged, and profiles normalised to cloud top height. Given the importance of cloud/BL top in relation to aerosol profiles I think too much is made of the results from figures 3 and 4, when it could be stated up front that to properly interpret the profiles they need to be plotted against altitude normalised to BL top -maybe both true and normalised heights are needed to fully interpret them, but the issue needs acknowledging up front.
Line 201: 'most obvious is a reduction in backscatter in November just before cloud formation (Fig.  4d)' -this doesn't apply at all altitudes, only 200-600m. This might result from, say, subsidence causing BL depth to decrease -change is then not in situ, but movement of layers. It is also not clear that this reduction is relevant to the subsequent cloud formation since we are given no information as to what altitude that cloud/fog formed at. It is perhaps also worth noting that there are only 6 cases for analysis in November, so a single strong case may dominate the statistics.
Line 204: 'It is interesting that the level where backscatter transitions to its quasi-constant value is at or above where low cloud formation (base < 400 m or surface fog) occurred' a) this is exactly what we would expect for any scalar quantity with a surface source (eg water vapour in marine environment)…so reassuring rather than interesting? b) to properly assess this you need to plot against a normalised altitude -you know where cloud top was/will-be so don't need to approximate to 'at or above where low cloud occurred'.
Lines 206-209. "Consistency in aerosol backscatter structure from start to end of these clear periods seems to mimic the behaviour of a residual layer of relatively well-mixed aerosol trapped across the lowest few hundred meters of the atmosphere. This mixed layer may have been an artifact of the previous sub-cloud mixed layer prior to dissipation." a) it is not clear what altitude the authors refer to here -assuming they refer to the 'quasi constant' value from 2 lines up, then they refer to the layer above the BL/cloud, i.e. in the free troposphere. Here aerosol profiles depend mostly on advection and conditions upwind, perhaps far upwind. The reference to a previous subcloud layer then seems rather spurious. And again, you know where the cloud layer was (and will be) so you can pin point this, you don't need to speculate. Normalised altitudes would help again. If the reference is really to within the BL, then this needs making clear.
Line 208: "since the transition to a quasi-constant value is occurring at or above cloud base"physically we expect the transition to quasi-constant free-troposphere values at cloud top, the rather vague, and physically misleading, phrasing 'at or above cloud base' would be unnecessary if the profiles were assessed against a normalised altitude.
The following statement "the data suggest that suface aerosol properties such as number concentration are likely often unrepresentative of aerosol properties at cloud level" I agree with, but not because the 'transition to a quasi-constant value is occurring at or above cloud base' but because there is a general decrease in backscatter with altitude in the lowest levels.
Line 224 & figure 5: only Feb-May are shown in figure 5 'because these months had the most frequent clear sky periods'. This is irritating, since it omits November, the one month in figure 4 which showed a behaviour distinct from the other months shown, and which might be explained by the normalised altitude used here. In general, given the very sparse data set, the limiting of data shown to specific months seems counter productive -better to use all of it all the time -combine months to reduce issues with poor stats in single months. Define season boundaries rather than using whole months to better group consistent seasonal behaviour. If you insist on using only a subset, then at least be consistent and use the same subset throughout.
While the full 2D RFD in figure 5 is useful -it really highlights the variability and that this is clustered (on individual cases?) rather than uniform, it isn't easy to directly compare these plots with figures 3 and 4. The addition of median profiles would help.
Line 233: the words 'and above ( fig. 5a-d)' don't fit grammatically with any of the rest of this sentence.
Line 237: '…cutoff between aerosol and clear sky (Shupe, 2007)' -here 'clear sky' appears to be being used to mean something different than every other occurrence…a complete (?) lack of aerosol? I would rephrase or risk this being interpreted as just 'cloud free'.
Line 241: "Being that aerosol backscatter near and above cloud top (zn=1) was at a minimum suggests that low aerosol concentrations near cloud top could have played a role in its dissipation" -only aerosol below cloud top are directly relevant to its properties, those above can't affect its microphysics. They can only play a role if entrained into cloud, but since the measurements are obtained after dissipation, aerosol above the former cloud top clearly were not entrained. This contradicts the statement on line 239 and is again contradicted (or at least…amended) on line 245.
Line 266-274: The discussion of aerosol concentration at the surface needs more nuance.
In the case of low cloud -formation should not impact aerosol concentration at the surface -CCN lifted above LCL will nucleate a droplet, but if the drop is moved down again it will evaporate leaving the aerosol particle -number of particles is conserved. Loss of particles requires: i) coalescence of droplets -evap would then tend to consolidate all the original aerosol into a single large particle. ii) scavenging of aerosol by droplets -evap as in (i) iii) precip -loss of CCN & scavenged aerosol to surface. All these are possible, but not discussed.
In fog the CPC might undercount total particles, even when conserved, if droplets don't make it through the inlet into counter (quite probable). Again, it would be useful here to distinguish between low (but elevated) cloud and fog.
Line 290: "LWN is primarily proportional to cloud liquid…" -only for liquid water paths below the black body limit of ~50 g m -2 , above that there is little impact on LW radiation.
Line 318: "These seasonal and sky condition differences in particle concentrations suggest different mechanisms are responsible for aerosol numbers near the surface" -this is interesting. Is this simply a result of having an exposed local surface during summer, which may be a strong source or aerosol, and a snow covered or frozen surface for the rest of the year?
Line 391: "least squares linear regression of the tendencies between the layers reveal a moderate agreement to the monthly cases" -'with the monthly cases' or 'for' the monthly cases depending on your intended meaning.