The FM100 observations may be liquid fog, ice fog (note that we do not distinguish between ice fog and clear-sky precipitation, also known as “diamond dust”), blowing snow, or snow. Thus, it is desirable to classify the probe observations in order to identify the scenes containing liquid fogs. To do this, information about particle phase, precipitation occurrence, the presence of elevated cloud layers, and the likelihood of blowing snow is needed.

The classification procedure is as follows: (1) scenes containing near-surface particles are separated from clear boundary layer scenes using a number concentration (Nc, cm-3) threshold in the 10 m FM100, (2) elevated tropospheric cloud layers are identified, followed by (3) occurrences of blowing snow and snow, and then (4) the phase of the particles is determined for all cases where particles were observed near the surface (step 1) but the sky was otherwise clear (step 2) and there was no blowing snow (step 3). Next, we will describe the individual classifications.

Figure 5 summarizes the fractional occurrence of the classified cloud types. Particles were observed in the lowest 10 m for a majority of the time in all months, from 65 % of the time in July to >85 % of the time in winter. Liquid fogs under otherwise clear skies were classified between 1 % of the time (April) and 12 % of the time (September); if liquid fogs could be reliably detected in the presence of tropospheric clouds, these percentages would likely be higher. While most common in late summer, liquid fogs were also identifiable 3 %–7 % of the time during January–March. The fact that there were fewer ice identifications than liquid identifications does not necessarily indicate that in situ formation of ice near the surface at Summit is less common than liquid formation for three reasons: (1) the number density of liquid fog is expected to be higher, so liquid is more likely to be identifiable using the AERI, (2) some ice fog events may be incorporated within the blowing snow or snow categories, and (3) the phase partitioning for events from which phase could not be retrieved using the AERI is unknown.

The classification scheme reveals several key features of hydrometeor occurrence over the Greenland Ice Sheet:

Precipitation occurred predominantly from July through to October (similar to the study by Pettersen et al., 2018). This seasonal cycle is shifted later in the year compared to total precipitation amount, which peaks in July (Castellani et al., 2015).

Clear skies persisted from 15 to 24 June 2013, and liquid fogs were observed in the early morning on most days during this period. While this case represents an ideal illustration of fog formation at Summit, several similar cases can be found in the dataset. Figure 6a and b show the time–particle-size cross sections of Nc from the FM100s beginning at 20:00 UTC on 15 June extending through 14:00 UTC on the 16 June. During this period, the 10 m air temperature was -15 to -22 ∘C with light southerly winds (Fig. 6e). At 10 m, condensate first occurred ∼00:10 UTC as the solar elevation angle (SEA) dipped to 10∘ (<20 W m-2 SWnet, Fig. 6d) and 7 h into development of the surface-based inversion (Fig. 6d). Initiation was followed by a period of uniform growth lasting 4–5 h that closely aligns with a theoretical growth curve (Houghton, 1985), calculated assuming a constant supersaturation of 0.1 % (Fig. 6a). During the first hour of growth, the droplets reached approximately 25 µm in diameter after which the growth slowed and deviated from the theoretical curve, with particles eventually reaching ∼40 µm. The curve neglects the gradual decrease in supersaturation that accompanies condensation without replacement of moisture, implying a decrease in supersaturation during development and indicating that the moisture source did not introduce new vapor as quickly as it was condensed. While most of the droplets closely followed the main growth curve, nucleation of new droplets continued until about 03:00 UTC. The fog disappeared completely by ∼10:30 UTC when the sun reached an elevation angle of ∼30∘.

The sodar detects thermal turbulence where mixing occurs within a vertical temperature gradient. Note that structure in the sodar record may only reflect structure in temperature and not necessarily the boundaries of the fog; although in some cases radiative cooling at the top of the fog may account for the thermal contrast that is observable with the sodar. The sodar record shows a typical pattern of a “nighttime” stable surface layer 10–25 m deep before 07:00 and after 20:00 UTC (Fig. 7a). This surface layer represents a shallow, but strong, temperature inversion embedded within a deeper inversion that extended about 100 m above the surface at initiation, further deepening to several hundred meters 12 h later. The surface layer increased in depth from about 15 to about 65 m during the course of the duration of the fog layer, transitioning from statically stable to a shallow convective layer in association with the diurnal cycle (Fig. 7a). The convective plumes are visible in Fig. 7a as vertically oriented echoes below the red dashed line with intervals on the order of minutes. Dissipation occurred shortly after the onset of convection ∼08:45 UTC (SEA ∼23∘, SWnet=61 W m-2, SWD =426 W m-2), and the fog disappeared when the convection reached a developed stage around 10:00 UTC (SEA ∼29∘, SWnet=87 W m-2, SWD =562 W m-2).

Particles were observed at 2 m 10–20 min after initiation at 10 m. Both the Nc and the width of the size distribution were larger at 2 m than 10 m, consistent with particle formation near 10 m followed by settling and evaporation (∼0.01 m s-1). At Summit, condensation in radiation-induced fog frequently occurs near 10 m because nonlinearity between temperature and saturation within the inversion leads to supersaturation first between 3 and 18 m, with lower vapor pressures both above and below this layer (Berkelhammer et al., 2016). A lag in Nc of 3 to 4 min between 10 and 2 m near the peak particle size is evident in the time series (Fig. 6c), implying a maximum settling rate of 0.03 to 0.04 m s-1. Theoretical calculations of settling rates assuming laminar conditions following Pruppacher and Klett (2010) agree with these estimates. Specifically, the calculations indicate settling rates up to 0.01 m s-1 during the first hour of the case and 0.03–0.035 m s-1 during the mature phase of the fog between 06:00 and 08:00 UTC.

Between 02:00 and 05:00 UTC, the 2 m air temperature deviated by ∼+5 ∘C from the smooth diurnal cycle that is evident both before and after the fog, and the LWP increased to ∼15 g m-2 with no clouds observed above the fog layer by the radar or lidar. This corresponded to an increase in CRF from <10 to ∼65 W m-2, caused primarily by increased downwelling longwave radiation, as evidenced by the LWCRE (Fig. 6f). These values are within the range of the thickest clouds at Summit (Cox et al., 2014; Miller et al., 2015) and large enough to drive latent and sensible heat fluxes to near-neutral conditions (Fig. 6g), indicating a well-mixed surface layer. As we will see later, fogs generally produce closer to ∼10–20 W m-2, but larger values such as the 16 June case occur occasionally. Because of these factors, the temperature inversion within the surface layer completely eroded by around 03:30 UTC. Surprisingly, this did not dissipate the fog, and, in fact, an increase in Nc at 2 m occurred in conjunction with a brief interruption in the otherwise smooth growth rate. Following the brief interruption, the particle size and Nc were larger than before with Nc at 2 m exceeding 200 cm-3; as we will see later, such high Nc as observed from 04:00 to 05:00 UTC are atypical.

Thus, while the fog was likely induced by radiation initially, it was maintained, and ultimately continued to grow, without additional infrared loss at the surface driving saturation in the air column. Indeed, with warming of the surface layer, introduction of large quantities of vapor would have been necessary to maintain the relative humidity supporting the fog. The large LWP (Fig. 6f) implies significant cloud-top radiative cooling may have occurred; therefore, buoyancy-driven mixing is a plausible mechanism to have supplied the moisture that maintained the fog. However, the moisture was more likely introduced to the surface layer from above by mixing generated by wind shear. This is supported by the sodar data, which shows signatures of turbulence in the form of Kelvin–Helmholtz instabilities (Fig. 7b) driven by shear at the top if the surface layer (30–50 m) between 03:00 and 05:00 UTC, corresponding to the period of time with enhanced LWP, CRF, and Nc, and generally more variable particle size at both measurement heights. Indeed, in June, the firn is generally colder than the surface (Miller et al., 2017) so vapor transfer tends to be downward, toward the surface, from the atmosphere because the mixing ratio is higher in the (saturated) air immediately above the surface than in the firn (Berkelhammer et al., 2016). The latent heat flux (LH) was positive (defined positive into the surface) for the duration of the case study, consistent with the hypothesis that the fog was driven by moisture from aloft (Fig. 6g). Therefore, the moisture source for this case was likely the atmosphere and not the local surface. Interestingly, the Nc at 10 m was consistently lower than at 2 m. The reasons for this are unclear but since there were also more small particles at 2 m, it is possible that the concentration of particles at 2 m was associated with partial evaporation and subsequent slowing of the particles descent. The difference may also be associated with the heights of the instruments relative to the height of maximum supersaturation (see Fig. 2 by Berkelhammer et al., 2016).

Figure 8 is similar to Fig. 6, but for a wintertime case. The fog was detected at ∼05:30 UTC at 10 m, and the particles grew quickly to ∼35 µm within approximately 30 min (Fig. 8a, b). The Nc at both heights was about an order of magnitude smaller than the June case (Fig. 8c). The troposphere was clear with intermittent ice clouds above 2 km (Fig. 8d). Similar to the June case, the wind was southerly and light with lower wind speeds during the period of the fog (Fig. 8f). The net atmospheric heat flux was generally negative (Fig. 8h) due to radiative cooling at the surface that maintained a temperature inversion with a gradient of ∼10 ∘C between 2 and 10 m (Fig. 8e). There was not enough LWP for reliable detection by the MWR (uncertainty ∼5.5 g m-2), but the CRF was between 10 and 40 W m-2; note that the upper-level clouds contributed some to this forcing, in particular after 14:30 UTC. The CRF quickly spiked to 40 W m-2 during the initial fog development around 06:00 UTC, which was also the time period when the Nc was the highest. Later increases in CRF also corresponded in time to increased Nc. Consistent with variable wind direction, the apparent intermittence of Nc may be explained by a spatially heterogeneous fog periodically passing the tower instruments.

This case was likely a mixed-phase fog with a liquid formation layer precipitating into an underlying settling layer composed of homogeneously frozen ice. The air temperature at 10 m (near the formation height) was between -35 and -38 ∘C. While the homogeneous freezing point for liquid is imprecise and dependent on conditions (generally about -40 ∘C), the 2 m air temperature and skin temperatures were ∼-47 and -50 ∘C, respectively, both too cold to support liquid particles. Thus, it is likely that the particles observed at 2 m were frozen droplets. While we can only infer the phase of the underlying layer from the measured temperature, we do have confidence that the droplets that formed aloft were liquid: the AERI classifications were not ambiguous as 176 of 177 samples for this event were identified as liquid; the AERI classification should not have been confounded by the presence of the ice because the viewport for the AERI is a few meters above the surface, and therefore near the top or above the ice layer; the AERI results are also supported by the MPL data, which indicate mostly liquid at levels where the MPL has sensitivity, between 100 and 150 m (Fig. 8d).

The ceilometer (not shown) and MPL (Fig. 8d) data indicate a deeper fog layer than the previous case, which may have been enabled by generally deeper and more persistent inversions in winter (Miller et al., 2013); the height of the maximum temperature within the troposphere for this case was 300–400 m compared to ∼100 m at initiation for the June case. However, the sodar record (Fig. 7b) shows that the depth of the surface layer embedded within the deeper inversion for the winter case was actually much shallower than in the summer case, <10 m during the duration of the fog. The particles within the surface layer were sampled by the measurements made at the tower (e.g., Fig. 8a, b) and may be distinct from the deeper fog layer visible in the MPL data: the sodar record also shows a significant amount of structure throughout the upper fog layer indicating additional stable layers (Fig. 7c). This structure is physically decoupled from the surface layer, separated by a thin layer of low reflectivity that may be indicative of a low-level jet, which could suppress mixing between the layers. However, the layers may be dynamically coupled in other ways. Buoyancy waves with periods of a few minutes to 15 min in the upper layer have an observable remote influence on the surface layer through fluctuations in the horizontal pressure field that produce a moving pattern of convergence and divergence (not shown). It is unknown if the observed microphysics below 10 m is representative of the fog above the surface layer. However, radiatively, the combined physically thick layer of fog compensates somewhat for the low Nc observed near the surface, enhancing the CRF relative to the summer case with a larger optical depth. The strength of the inversion also contributed to enhancing the CRF because the particles were warmer than the surface.

Interestingly, the fog development coincides closely with the weak early-season diurnal cycle. In mid-January, the sun does not rise above the horizon at Summit but is about 3∘ below the horizon at solar noon (Fig. 8e). Atmospheric scattering in these twilight conditions produced 1–3 W m-2 of diffuse incident solar radiation (Fig. 8e) that corresponded in time with the fog initiation. Though the mechanism is unknown, there is a possibility that the fog was diurnally forced, similar to the June case; yet, unlike summer, the timing was coincident with a peak in solar radiation rather than the minimum.

The stably stratified surface layer was largely isolated from the free troposphere in this case. The temperature gradient within the firn (which is warmer at depth than at the surface) produces a constant supply of vapor towards the surface from below, providing moisture for the fog, which is then returned via settling (Berkelhammer et al., 2016). As before, we observed higher concentrations and more small particles at 2 than 10 m.

Figure 9 shows distributions of the particle sizes for liquid fogs (blue) compared to those for the ice categories; the figure displays sensor height as rows (10 m – top row, 2 m – bottom row) and season as columns (left column – low-light winter season, NDJF, and right column – sunlit summer season, JJAS). Ice particles are nonspherical and, thus, the distributions represent an effective size with reference to the scattering properties of spherical liquid; asphericity and orientation are important factors in the sizing of ice using a scattering spectrometer that imposes significant uncertainties (Borrmann et al., 2000). Thus, the distributions of ice classes should be treated cautiously and are shown here for context. The liquid classification stands out distinctly from the ice both in the shape of the distribution and the overall Nc within each size bin. For smaller particles, liquid particles are present in higher concentrations than ice particles in summer and are smaller in winter, while the opposite is generally true for larger particles. Despite the uncertainties in sizing ice particles, the relative Nc within each size bin separates logically between the different types. For example, there are more particles in blowing snow at 2 m than at 10 m, while the two heights show similar concentrations of ice fog. Also, the occurrences of snow have consistently low Nc.

As implied by the case studies, fogs in winter have lower Nc overall compared to summer. This relationship between temperature and fog Nc is evident when analyzed directly: Nc is smaller in colder fogs, with temperature being correlated with the log of Nc (r=0.54). Gultepe et al. (2002) and Gultepe and Isaac (2004) reported similar findings in tropospheric Arctic clouds containing supercooled liquid. Note that while aerosol concentration is likely a factor, dynamical and thermodynamical processes can also play a role (Gultepe et al., 2002; Gultepe and Isaac, 2004).

The distribution of liquid particles in summer peaks between 20 and 25 µm in diameter at 2 m (Fig. 9b) while the distribution at 10 m (Fig. 9a) is broader with more small particle sizes. This is consistent with particles preferentially forming near 10 m before settling to 2 m. Specifically, multiple growth stages are likely represented in the distribution at 10 m, while the distribution at 2 m is more idealized because it is composed primarily of mature particles. Though the winter liquid distributions are more difficult to interpret owing to a limited number of data, this finding is supported by a calculation of the effective diameter (the ratio of the third and second moments of the size distribution) for all liquid fog scenes in all months, which shows a correlative relationship between effective diameter and Nc at 10 m (Fig. 10a) (r=-0.55), but not at 2 m (Fig. 10b). This indicates that at 10 m, when the Nc is high, the particles are small (forming), while at the same time low Nc for large particles is consistent with loss via settling out of the layer. While a similar relationship might be expected from a range of aerosol concentrations, if aerosols were the explanation for the observations at 10 m, a similar result should be found at 2 m, but it is not.

The overall distributions of Nc (Fig. 11) at both heights show concentrations in winter that are smaller than in summer. In winter, there are typically more particles at 2 m than 10 m, whereas during summer the differences between the heights are less discernible. The median value of Nc was typically <7 cm-3 in winter and between 5 and 20 cm-3 in summer (Fig. 11).

Figure 12 shows the temperature distributions for times classified as liquid fog and ice fog (the two types associated with in situ formation). In Fig. 12, Nc was only used for identification of events. Thus, precise estimates of Nc are not important, and the threshold for wind direction is less restrictive than the microphysical results shown in Figs. 9–11. Consequently, larger sample sizes are incorporated into this analysis. Three different temperatures are plotted beginning with brightness temperatures derived from near-saturated CO2 emission between 675–680 cm-1 measured by the AERI in Fig. 12a. These frequencies are sensitive to the lowest few tens of meters above the instrument and may be most similar to the fog thermodynamic temperature. Since liquid freezes homogeneously near -40 ∘C, observations at lower temperatures are not possible; less than 1 % of observations classified as liquid occur at temperatures <-40 ∘C in Fig. 12a. The air temperatures at 10 m (Fig. 12b) are similar to the AERI-derived temperatures. Air temperatures closer to the surface at 2 m (Fig. 12c) are generally colder and include many more instances with temperatures below -40 ∘C. Note that this is not an indication of water existing in a liquid phase below -40 ∘C but rather indicates that the liquid layer lies above this cold air where the temperature is warm enough to permit the existence of liquid droplets. Between November and March, two-thirds of the liquid fog identifications occurred when the 2 m air temperature was below -40 ∘C. This result indicates that the 16 January case is typical. A stretch of such occurrences was observed when the 2 m air temperature was <-45 ∘C during an extended clear-sky period during the last 2 weeks of March 2013. Unlike the 16 January case, for cases during the spring and autumn, there was sufficient daylight for images acquired by cameras mounted to the tower to show evidence of optical phenomena caused by liquid droplets. Images of fogbows appeared at times that coincided with the identification of liquid fogs in March 2013. Fogbows are formed by scattering processes dominated by diffraction when the size of spherical particles is within the Mie regime (e.g., Lynch and Schwartz, 1991). Note that the presence of a fogbow is suggestive of the presence of liquid, but it does not rule out the presence of ice. Additionally, spherical or quasi-spherical ice formed by freezing of supercooled liquid has been reported by other studies (e.g., Thuman and Robinson, 1954), though the preferred habits of ice fog particles remain controversial (see Gultepe et al., 2015) and we are unaware of any studies linking ice fogs to optical phenomena normally associated with liquid droplets.

For both ice and liquid identifications in Fig. 12a–c, the temperatures at which they occur are lower in winter and higher in summer. Liquid generally occurs during cooler temperatures than ice in summer and thus the seasonal cycle for liquid is muted compared to that of ice. This is likely a result of diurnally forced radiation fogs occurring during the colder part of the day in summer. There is also a notable difference in the timing of the seasonal cycle with the liquid tending to be warmer than ice in the spring transition and cooler than ice in fall. It is unclear whether this is more closely tied to the seasonal cycle in aerosols (e.g., Schmeisser et al., 2018) or to meteorology. The wind regimes are similar between ice and liquid in winter but liquid fogs in summer occur during particularly calm conditions in association with the diurnal development of stable stratification (Fig. 12d).

The impact on the surface radiation budget from the various classifications is evaluated in Fig. 13. The sample sizes in Fig. 13 are analogous to those in Fig. 12. First, distributions of LWCRE for the classes appear alongside those for all observations in Fig. 13a. The peaks for both liquid fog and ice fog are close to 10 W m-2. Both ice and liquid fogs show long tails to larger values and thus the mean is 19.8 (median =14.8) W m-2 for ice and 26.1 (median =18.1) W m-2 for liquid fogs. It is notable that unlike distributions of temperature and microphysics shown earlier, the distributions of LWCRE for ice and liquid are similar. It is therefore possible that a similar amount of downward longwave radiation is needed, when both liquid and ice fogs form, to achieve radiation balance between the surface and the lower atmosphere. However, note that the sample selection was radiative in origin in the first place, and the proportions of missed classifications, in particular at the low end of LWCRE, may be different for the two types. The other ice classes, snow, blowing snow, and a mixture of both, are also plotted in the panel for context. Blowing snow is distributed across the range of LWCRE with the larger values more likely to be associated with higher wind speeds (r2=0.31). Snow also is distributed over a wide range of values, but generally higher than blowing snow alone, which is expected because LWCRE during snowy conditions is associated with a precipitating cloud, whereas blowing snow may occur under otherwise clear skies. When snow and blowing snow occur together, the distribution is clustered near the largest values, a condition mostly associated with storms.

When the total CRF is considered (Fig. 13b), the results are similar. The forcing is smaller because of the addition of shortwave cloud cooling but only slightly smaller because the high year-round albedo at Summit limits the ability of clouds to cool the surface there (Miller et al., 2015). Notably, there is increased separation in CRF compared to LWCRE between the ice fog class (μ=12.1 W m-2, median =7.9 W m-2) and liquid fog (μ=22.1 W m-2, median =17.1 W m-2). This is because a larger proportion of ice fog occurs at higher sun angles when the shortwave cloud-cooling effect is larger. Therefore, the time of day when liquid fogs typically occur maximizes their net radiative forcing.

Figure 13c and d give the statistics for each month and annually for liquid (Fig. 13c) and ice (Fig. 13d). The annual mean CRF for liquid fogs under otherwise clear skies was 1.5 W m-2 when normalized by the frequency of occurrence. The maximum occurred in July (2.6 W m-2) and the minimum was in April (∼0 W m-2). Due to subsampling, this estimate is probably slightly lower than the total annual CRF from the fogs under all sky types because fogs that were present under optically thin cloud cover may have contributed additional CRF. However, events that were missed for being too thin contributed little radiatively (and consequently, no identification was possible); only a small number of “too thick” (0.8 %) fogs were identified, limiting the influence of this class on the mean CRF; and when ambiguous identifications are included in the analysis (not shown), the annual mean CRF decreases slightly to 1.2 W m-2. For ice, the normalized annual mean is 0.7 W m-2. While the magnitude of the values is similar between ice and liquid, the seasonal cycles are different with ice peaking in July and September (2.2 W m-2). For reference, the mean annual CRF for all sky conditions at Summit is 33 W m-2 and is positive in all months (Miller et al., 2015).