KAZR and MWR observations
In Fig. 2 the time–height plots of the first three moments (effective radar
reflectivity factor Ze, subsequently called reflectivity; mean Doppler
velocity Vd, and spectral width (σ)) of the primary peak of
the KAZR Doppler spectrum are shown. The primary peak is defined to be the
noise-separated peak containing the bin with maximum power spectral density
(cf. Fig. 4; Kollias et al., 2007b). This is the peak used in the standard
ARM radar moments data products. The ARM MicroARSCL data product (Kollias et
al., 2007b) extends the reported moments to skewness and kurtosis for both
the primary peak and an additional noise-separated secondary peak, if one
exists. The peak power densities and modal velocities of up to two local
maxima occurring within the primary peak are also reported. Mean and maximum
spectral noise power are determined using the technique described in
Hildebrand and Sekhon (1974). If only one hydrometeor population is present
in the radar volume (cf. Fig. 4a), the radar Doppler spectrum is usually
characterized by a single peak above noise floor, which is controlled by the
width of the PSD and sub-volume turbulence. In cases with more than one
hydrometeor class in the radar volume (e.g, liquid droplets and snow) it is
possible to have sufficient fall velocity separation between the two
hydrometeor classes so that the radar Doppler spectra is bimodal. As
mentioned in Luke and Kollias (2013), strongly multimodal situations can be
considered to be “golden” samples as they make it easy to separate the
contributions of the individual hydrometeor populations to the total radar
return. Here, we classify all peaks which are separated by the mean noise
floor into three categories (liquid droplets, freshly generated ice, and
(rimed) snow). Peaks are grouped into these classes according to their mean
Doppler velocity and spectrum width. However, in many observations, the
terminal fall velocity difference between two different particle size
distributions is not large enough to produce individual peaks separated by
the mean noise floor. Instead, broad merged peaks consisting of the
contribution of two or more PSDs occur (cf. Fig. 4d). This is also the reason
why only a short time period is analyzed in this study; synoptic situations
in which ice particles from a “seeder” cloud above are falling through a
SLW layer where they experience riming occurred at least half a dozen times
during the BAECC-SNEX period but unfortunately, for all other events, only
merged peaks were observed in the KAZR Doppler spectra.
Examples of KAZR Doppler spectra at different times and heights.
Notice the different x and y scales. Maximum and mean noise floor
determined according to Hildebrand and Sekhon (1974) are indicated by grey
and black horizontal lines, respectively. The primary peak is labeled as 1st,
the secondary peak as 2nd. The mean Doppler velocity of primary peak is shown
by the vertical grey line. The plots are created with the Doppler Spectrum
Visualizer, a visualization toolkit which is publicly available at
http://www.gim.bnl.gov/armclouds/specvis_java_toolkit/.
![]()
An objective way to distinguish spherical supercooled liquid droplets from
freshly generated nonspherical ice would be the use of spectral linear
depolarization ratio (LDR). Unfortunately though, no KAZR cross-polarization
channel data were gathered during BAECC-SNEX and thus no LDR could be
determined. However, microphysical modeling sensitivity tests (not shown)
showed that unrealistically high ice particle number concentrations – on the
order of a few hundreds to a few thousands per liter for ice particle sizes
of a few hundred microns – would be required to produce a peak of about
-20 to -15 dBZ at 0.15 m s-1. For this regime of sizes, the
observed ice particle number concentration reported in the literature is
below 10 per liter (e.g., Zhang et al., 2014; Lloyd et al., 2015). We can
thus assume that power spectrum peaks at small velocities are due to liquid
droplets and not ice particles.
In this study, the focus is on the period when the upper level snow band
moves in, roughly from 22.4 UTC when first detected by the KAZR to the end
of the first snow shower at around 22.77 UTC. In the KAZR reflectivity field
(cf. Fig. 2a) the onset of snowfall is clearly visible by the high Ze
(> -5 dBZ) area extending from a fall streak feature at a height of
8 km downwards. The Ze structure is strongly tilted above 3.5 km due to
horizontal wind shear. The decrease in Ze at 22.7–22.77 UTC below 1 km can
likely be attributed to sublimation in an ice-subsaturated layer (cf.
humidity profile in Fig. 1).
Before the arrival of the snow band in Hyytiälä, a liquid-topped
mixed-phase cloud with cloud top at 3.4 km is observed (see Fig. 3). Its
roughly 500 m thick SLW layer can clearly be distinguished by mean Doppler
velocities of around 0.0 m s-1 in Fig. 2b, as well as very narrow
spectrum width values below 0.08 m s-1 and low reflectivity values
(Ze < -15 dBZ). A typical KAZR Doppler spectrum example within this
liquid layer is shown in Fig. 4a. Within this SLW layer new ice formation
took place at about -12 ∘C; subsequently, the ice sedimented and
grew in size as reflected by a gradual increase of Ze (22.4–22.69 UTC)
between 2.9 and 0.9 km, as well as an increase of Vd and σ. A
Doppler spectrum example of this freshly generated ice mode is shown in
Fig. 4c. The rapid increase of Ze below 0.9 km prior to the snowfall
indicates ice particle riming which is confirmed by the ground-based in situ
observations of rimed ice crystals during that time period. The riming most
likely occurred in another SLW layer at the lower inversion at 0.7–0.9 km,
characterized by mostly non-noise-floor-separated liquid peaks in the Doppler
spectra (not shown). Strong surface turbulence below 0.8 km is obvious in
highly variable Vd and high σ. The strong turbulence in the
surface layer led to a broadening of the Doppler spectra peaks which resulted
in broad merged peaks (not separated by the noise floor) and thus hampered
the application of our microphysical retrievals for that SLW layer.
When the snow band starts falling through the SLW layer, the primary peak
moments are not sufficient to capture the radar view of the microphysics. The
coincidence of liquid and ice particle size distributions within the KAZR
sampling volume leads to multimodal KAZR radar Doppler spectra due to the
terminal fall velocity difference between the liquid droplets and the falling
snow (Fig. 4b). As obvious in Fig. 4b, the snow falling through the SLW layer
has a higher dynamic range than the liquid mode and is thus classified as the
principle peak, leading to a sudden change of Vd values in Fig. 2b
after 22.69 UTC and below 3.4 km. Similarly, the sudden high values of
σ at 22.68–22.7 UTC and below 2 km can be explained by a merged
peak of the freshly generated ice and the snow which are no longer separated
by the mean noise floor, as illustrated in Fig. 4d. The strong increase of
Vd (on the order of 0.5 m s-1, cf. Fig. 2b) within the SLW
layer indicates riming of particles, as also seen in the in situ measurements
(cf. Sect. 3.2).
Observations supporting the presence of riming when the snow fall streak
intercepts the SLW layer are provided in the temporal evolution of the MWR
liquid water path (LWP) shown in Fig. 5b. While the LWP varies between
320 and 400 g m-2 before 22.69 UTC when only the liquid-topped
mixed-phase cloud is present, it rapidly decreases from 400 to
250 g m-2 within 3 min (22.69–22.74 UTC). The observed reduction in
the LWP can partly be attributed to the capturing of SLW droplets by the
falling snowflakes, leading to rimed particles with high density and fast
fall velocities.
Zoomed view (2–4 km, 22.54–22.77 UTC) of the snow front on
21 February 2014. Panel (a) shows KAZR total reflectivity,
(b) microwave radiometer (MWR) liquid water path (LWP),
(c) reflectivity of the liquid peak, (d) mean Doppler
velocity of the liquid peak used as vertical air motion tracer. The dotted
lines depict individual fall streaks starting at different generating levels
zgen(zgen=6 km (black), zgen=2.9 km
(blue)). All subsequent averaged profiles refer to the area between the two
black fall streaks.
![]()
It should be noted that the vertically integrated value of LWP cannot be
entirely attributed to the layer of SLW at 2.9–3.4 km. Unfortunately, the
entire vertical distribution of SLW cannot be reconstructed because the lidar
signal was already extinguished by a thin SLW layer at 0.2–0.4 km. However,
two more thin SLW layers were detected by KAZR Doppler spectrum analysis at
0.8–0.9 and 1.5–1.7 km (see Fig. 3). These layers however were
intermittent and coincide in time with the periods of the highest values of
LWP around 22.6–22.65 and 22.7 UTC. When these intermittent layers were
observed, the LWP increased by 60–80 g m-2; we use this values as the
estimate of their combined contribution to the total LWP. The SLW layer that
had formed at the lower temperature inversion at 0.8 km is a remnant
signature of the long-lived thin mixed-phase cloud with cloud top at just
below 0.9 km that was present since before 14 UTC.
Panels (a, d) show KAZR reflectivity (Ze) field. There, the
location of the SLW layer is indicated by thin horizontal black lines. The upper
middle and upper right panels show range and time spectrograms of vertical
profile which is indicated by a black vertical line in (a); the lower
middle and lower right panels along slanted fall streak are shown in (d)
as a dashed line. Panel (b) shows vertical range spectrogram along vertical black
line in upper left panel, (c) time spectrogram at 3.16 km,
(e) slanted range spectrogram along dashed line in (d).
Panel (f) shows the air-motion-corrected time spectrogram at 3.16 km.
![]()
Based on the liquid peak reflectivity (Fig. 5c) of the SLW layer at
2.9–3.4 km the mean liquid water path of this layer (averaged between the
two slanted paths in Fig. 5) was estimated to be 80–90 g m-2.
Fall streak tracking
When vertical wind shear is observed, the analysis of vertical profiles is
not sufficient to correctly trace the paths of evolution of hydrometeor
populations, as already stated in Marshall (1953). Instead, it is necessary
to follow the falling hydrometeor populations along slanted fall streaks if
we want to estimate their microphysical evolution from the particle-generating level (zgen) to height z. As illustrated in Hogan and
Kew (2005), the slanted fall streak patterns can be simulated in the radar
time–height observation space by taking the mean Doppler velocity
Vd as average fall velocity of the particle population and using
the horizontal wind profile u(z), e.g., from radio soundings closest in time
to account for advection:
dz/dtrad=(Vd(z)u(zgen))/(u(z)-u(zgen)).
Here, u(zgen) is the horizontal wind velocity at the particle-generating level zgen which is assumed to be close to cloud top
(zgen=6 km) and trad is the radar time (x axis in
time–height plots in vertically pointing mode). It is important to note that
trad is usually different from the sedimentation time of the
particle population and that fall streaks are not identical to particle
trajectories but are a result of overlapping trajectories as they are
advected over the radar (Bohren and Fraser, 1992). As in Hogan and
Kew (2005), we use the finite difference equivalent of Eq. (1) and work our
way downward level-by-level considering the displacement at level z in the
determination of displacement at level z-1. We use this technique to
determine a fall streak for each radar time step. Subsequently, all averaged
profiles (means and standard deviations) are determined for the fall streaks
spanned between the two black slanted fall streaks in Fig. 5. Our focus of
interest will be the uppermost SLW layer at the height range 2.9–3.4 km.
Slanted profile moments of SLW (blue), ice generated in the SLW
layer (red), and frontal snow (black). Means and standard deviations of
(a) reflectivity, (b) mean Doppler velocity, and
(c) mode width are shown for the slanted paths encompassed by the
two dashed lines in Fig. 5. Within the SLW layer, the velocities are
corrected for vertical air motion.
![]()
The closest radio sounding in time was launched at 23.2 UTC, approximately
45 min after the snow front was first observed with the KAZR. Ideally,
horizontal wind profiles at the time of interest should be used; however,
overall agreement of the slope of the black simulated fall streaks in Fig. 5
with the slope of the reflectivity features (above 2.9 km) seems to confirm
that the horizontal wind field did not change considerably within these
45 min. Below 3 km the slope of the black simulated fall streaks does not
fit well with the Ze feature, which is tilted in the opposite direction
compared to the fall streak above. Additional fall streak simulations
(depicted in blue in Fig. 5a) show that this backward tilted slope is well
matched by simulations if we assume zgen=2.9–3 km and the mean
Doppler velocity of the rimed snow mode in Eq. (1). The lower-level Ze
feature thus corresponds to the fall streaks of those snow particles that
experience riming in the SLW layer and indicates – as expected – the SLW
layer as the correct generating level for the rimed particle population.
Doppler spectrum evolution
Figure 6 illustrates the necessity of tracking particle populations along
slanted paths in conditions of wind shear by contrasting the KAZR Doppler
spectrum evolution along the vertical line at 22.69 UTC in Fig. 6a versus
along the estimated slanted fall streak in Fig. 6d. Clearly, the vertical
range spectrogram in Fig. 6b shows some non-microphysical features – such as
the discontinuity of spectral reflectivity at around 4 km – caused by
taking a vertical profile when a slanted one is more representative. The
range spectrogram along the slanted path in Fig. 6e tells a much more
consistent microphysical evolution story of the particles when the frontal
system moves in: above the 3.4 km height, snow with a mean Doppler velocity
of around 1 m s-1 is observed. A few shallow layers of increased
turbulence as well as thicker layers of up-/downdrafts result in a shift of
the entire Doppler spectrum to more positive/negative values, respectively.
At 2.9–3.4 km, the layer of SLW with Vd fluctuating around
0 m s-1 is obvious in Fig. 6e as well as in the time spectrogram in
Fig. 6c. Due to their small size, the terminal fall velocity of SLW
droplets is negligible, so their Vd (also illustrated in Fig. 5d)
can be used as an air motion tracer as done in previous studies (e.g., Shupe
et al., 2004; Rambukkange et al., 2011). The Vd offset of the
liquid mode from 0 m s-1 gives us an estimate of radar volume mean
vertical air motion. Thus, in the SLW layer, the fall velocity of ice
particles can be corrected for vertical air motion as illustrated in Fig. 6f.
In Fig. 6e, the same air-motion correction is applied in regions with SLW.
Again, the increase of Vd of the snow falling from higher layers
into the SLW layers indicates riming. Also, the new ice mode generated in the
SLW layer is revealed, as its Vd quickly increases from a few
tens of cm s-1 to 0.8 m s-1 between 2.8
and 2 km.
Profiles and probability density functions (PDFs) of radar
moments
Although the given observations cannot fully disentangle microphysical and
dynamical effects, a fairly consistent picture of the evolution of the
present hydrometeor populations can be formed (cf. Fig. 6). The means and
standard deviations of Ze, Vd, and σ of slanted profiles of
the three hydrometeor populations (SLW, ice generated in the SLW layer, and
frontal snow) are shown in Fig. 7. The average profile and standard
deviations are based on all 110 simulated fall streaks encompassed by the two
dashed lines in Fig. 5. Since in the Rayleigh scattering regime radar
reflectivity is proportional to the number of particles N and to D6
(with D being the diameter of the droplets), the reflectivity of the SLW
droplets is very low (-18 to -22 dBZ on average, cf. Fig. 7a). The
reflectivity of the ice generated in the SLW layer increases between 3.2 and
2 km from -16 to -10 dBZ because the ice particles grow due to water
vapor deposition and aggregation. At the same time, the Vd of the
ice mode increases from 0.2 to 0.7 m s-1 (cf. Fig. 7b). The absence of
an air motion tracer outside of the SLW layer does not allow for vertical air
motion contributions to the observed Vd to be accounted for. As a
result, the observed Vd is not equal to the terminal fall velocity
of the ice particles but could be higher (in updrafts) or lower (in
downdrafts). Due to its large particles, the greatest contribution to the
total radar return is of course given by the frontal snow mode. The total
increase of mean Ze of this mode from 6 to 2 km is 12 dBZ; however, from
6 km to the top of the SLW layer, the mean increase of Ze and Vd
of the snow mode is only moderate (5 dBZ and 0.15 m s-1), suggesting
only moderate growth of the snow particles, which is likely due to water
vapor deposition in higher parts of the cloud. Vd of the frontal
snow mode increases from 1 to 1.65 m s-1 between 3.2 and 2 km.
Spectrum width profiles in Fig. 7c show several thin turbulent layers in
which the standard deviation of σ is high. σ of the SLW mode is
smallest (0.05–0.07 m s-1) while σ of the new ice and frontal
snow mode are on the order of 0.1–0.2 m s-1. As previously mentioned,
larger σ can be caused by a superposition of broad PSD and sub-volume
turbulence.
Probability density functions (PDFs) of the moments of the three hydrometeor
populations are shown in Fig. 8. Only data points from below the SLW layer
top (3.4 km) down to the surface and between the two black dashed fall
streaks in Fig. 5 are considered. For Ze and Vd there is very
little overlap in the PDF of each hydrometeor population. In uniform
beam-filling conditions and sub-volume turbulence, the Doppler spectrum of a
cloud droplet PSD is symmetrical and near-Gaussian, resulting in zero
skewness (Kollias et al., 2011). In contrast, the PDF of the observed SLW
mode is skewed towards negative values, indicating the presence of
supercooled drizzle at around -12 ∘C. The presence of drizzle is
also indicated in Fig. 9 where the joint PDF of skewness and Ze is shown.
While liquid-mode skewness fluctuates around zero for Ze below -20 dBZ, it
becomes increasingly negative at higher Ze values of -20 to -16 dBZ. The
existence of drizzle in SLW layers at the temperature range of -5 to
-12 ∘C was also found by Verlinde et al. (2013) for Arctic
multilayered mixed-phase clouds; the skewness-reflectivity signature as
shown in Fig. 9 is qualitatively similar to the signatures observed in warm
drizzle clouds (Luke and Kollias, 2013).
Probability density functions (PDFs) of noise-separated peaks of SLW
(blue), ice generated in the SLW layer (red), and frontal snow (black) for
the slanted paths encompassed by the two dashed lines in Fig. 4. Only data
from surface to below top of the SLW layer (3.4 km) are considered. Please
note that the PDF of SLW Vd is cut at 20 %.
![]()
1-D microphysical bin modeling
The comprehensive observations were used as input to a one-dimensional (1-D)
bin microphysical model. The model is based on the warm rain model first
presented in Szyrmer et al. (2005) and modified to include the ice processes
of deposition, aggregation, and riming. The leading question is whether it is
possible to reproduce the evolution of the observed radar moments of the
rimed snow mode in the SLW layer using radar forward modeling of the
microphysical model output. Forcing the model output to agree with the
observations could help us to evaluate different proposed riming efficiency
schemes that have been implemented as options in the model (see Appendix A).
The microphysical modeling was limited to the SLW layer between 2.9 and 3.4 km
and focused on the evolution of the moments (Ze, Vd, σ) of
the frontal snow mode.
Model description
The 1-D steady-state model simulates the height evolution
of the bin-resolved snow PSD introduced at the model's uppermost level (which
in this study is the top of the SLW layer at 3.4 km). The model input
includes the vertical profiles of temperature, pressure, relative humidity
(taken from the radio sounding, cf. Fig. 1) and vertical air motion (derived
from the SLW-mode Vd; cf. Fig. 10). The cloud droplet PSD at each
level within the SLW layer is calculated from the liquid peak reflectivity
profile, assuming that the cloud droplets follow a log-normal size distribution
with a height-independent prescribed droplet number concentration N and
dispersion parameter σPSD (N=30 cm-3,
σPSD=0.4) which are also taken for calculations of liquid
water content (LWC) in Fig. 10.
A joint PDF of frequency of occurrence (number of pixels) of skewness
and reflectivity of supercooled liquid mode encompassed by the two black fall
streaks in Fig. 5.
![]()
Mean vertical air motion profile (Vair, left) derived from
liquid-mode mean Doppler velocity and liquid water content (LWC, right)
derived from liquid peak reflectivity between the two black fall streaks in
Fig. 5, respectively.
![]()
The evolution of the snow mode along the fall streak is simulated by
explicitly calculating the contribution to particle growth by the
microphysical processes of water vapor diffusion, aggregation, and riming. In
the model setup used in the presented simulations, the water vapor deposition
and aggregation processes conserve the area ratio, aspect ratio, and
mass–size relationships. The deposition calculations use the results of Field
et al. (2008) to describe the capacitance, except for the smaller particles
for which the electrostatic capacitance approximation for thin plates is
assumed, and the ventilation factor proposed by Hall and Pruppacher (1976) is
adapted. The value of aggregation efficiency is set to 0.2. Details of the
riming parameterizations are presented in Appendix A.
The frontal snow PSD is introduced at the uppermost model level in a
functional form of a generalized gamma function with melted diameter
representing particle size. The values of the two shape parameters are taken
from Delanoë et al. (2005) for the form that is most consistent with the
observations. The initial mass–size relation of a power law form with
exponent 2 and prefactor 0.0012 g cm-2 is used. The rimed fraction of
the snow particles at the model's uppermost level is set to 0. The initial
area ratio is calculated from the empirical relation between area ratio and
particle density in Heymsfield et al. (2002) for side plane aggregates
(particle density = 0.18 area ratio1.5).
A comparison of observed and modeled moments of the frontal snow
mode in the SLW layer. Black solid and dashed lines refer to the mean and
standard deviation of the slanted paths encompassed by the two black fall
streaks in Fig. 5. Model results based on riming efficiencies by Hall &
Cober as well as Lohmann are shown in red and blue, respectively (see text
for details). Model results assuming area ratio increase with mass and rimed
fraction (solid red and blue line) as well as with mass only (dotted red and
blue) are shown.
![]()
Comparison of model output and observations
In Fig. 11, KAZR-observed profiles of snow-mode radar moments and the modeled
moments are compared. Backscattering calculations are performed using Mie
spheres with nonuniform mass distribution (Fabry and Szyrmer, 1999). The
terminal velocity of snow particles – unrimed at the top and partially rimed
below – is calculated based on the method proposed by Heymsfield and
Westbrook (2010). However, two different possibilities for the evolution of
the area ratio with riming are implemented. In the first approach (denoted by
1 in Fig. 11) for a given particle maximum diameter, the increase of the
area ratio from the initial value is parameterized as a function of the
aspect ratio modified by riming. In the second approach (denoted by 2 in
Fig. 11) the increase of the area ratio is smaller and is obtained assuming
that the initial relation between area ratio and particle density is
maintained, i.e., the increase of the area ratio results from the increase of
mass only.
In addition to the sensitivity of the terminal velocity of the rimed snow
particles to the evolution of the area ratio relationship, the sensitivity of
the modeled radar moments of the rimed snow mode to different riming
efficiencies is evaluated. Specifically, in one model realization, the riming
efficiency parameterizations from Hall (1980) and Cober and List (1993) for
small and large ice particles are used, respectively. Model results using
this particular parameterization are denoted as Hall Cober in Fig. 11. In
another model realization, the riming efficiency parameterization proposed by
Lohmann (2004), assuming plates for small D and aggregates for large D is
adapted. Model results using this approach are denoted as Lohmann in
Fig. 11.
Two effects can explain some of the discrepancies between the observed and
modeled profiles. First, dynamical effects such as turbulence are not
included in the radar forward model that estimates the radar Doppler spectrum
width profile. Thus, the forward simulation cannot reproduce the spectrum
broadening of the rimed mode below 3 km (cf. Fig. 11c). Second, there are
uncertainties in the exact slope of the fall streaks due to the fact that the
horizontal wind profile from the sounding is not taken at the time of
interest but 45 min later. Overall, the evolution of the rimed snow-mode
moments in the SLW layer can be reproduced by the model (within the standard
deviations). In Fig. 11a, the observed and modeled Ze profile shows an
increase from the top of the SLW layer to about 3.15 km which is attributed to an increase of
mass of the snow when the liquid droplets attach to the snowflakes. Below
3.15 km, the dZe/dz is slightly reduced because even
with the increase of mass, the increase of backscatter is very low and at the
same time there is a reduction of particle number concentration. Except for
the aforementioned influence of turbulence on spectrum width, the model
reproduces the mean Doppler velocity and spectrum width profiles well. With
the increase in mass and density during the riming process, particle fall
velocities of the rimed snow steadily increase from top to bottom of the SLW
layer from about 0.95 to 1.2 m s-1. The model is also capable of
reproducing the small decrease in the spectrum width due to differences in
the riming efficiency of small and large particles, leading to a small
reduction in the spread of the rimed particles' fall velocities.