remote sensing

Article
Melting Layer Detection and Characterization based
on Range Height Indicator–Quasi Vertical Profiles
Shaik Allabakash , Sanghun Lim * and Bong-Joo Jang
Korea Institute of Civil Engineering and Building Technology, Ilsan 10223, Korea; shaik@[Link] (S.A.);
roachjbj@[Link] (B.-J.J.)
* Correspondence: slim@[Link];




Received: 16 October 2019; Accepted: 28 November 2019; Published: 29 November 2019


Abstract: The melting layer (ML) is an important region used to describe the transition of hydrometeors
from the solid to the liquid phase. It is a typical feature used to characterize the vertical structure
of the stratiform precipitation. The present study implements a new automatic melting-layer
detection algorithm based on the range-height-indicator–quasi-vertical profile (R-QVP) in the X-band
dual-polarization radars. The algorithm uses the gradients of the polarimetric radar variables
reflectivity factor at horizontal polarization (Zh ), differential reflectivity (Zdr ), and copolar correlation
coefficient (ρhv ), and their combinations to describe the ML characteristics. The melting layer heights
derived from the radar were compared and validated with the heights of the 0 ◦ C wet-bulb temperature
derived from the Modern-Era retrospective analysis for research and applications (MERRA) reanalysis
datasets and obtained high correlation coefficient 0.96. The R-QVP combined with this algorithm led
to spatial and temporal variabilities of the melting layer thickness. The thickness of the melting layer
was independent of the seasonal, spatial, and temporal variabilities of the precipitations. Intriguing
polarimetric signatures have been observed inside, above, and below the ML, based on the phase of
the precipitation particles. The statistics of the polarimetric variables were evaluated for ML, rain,
and snow. Further, the linkage between enhanced specific differential phase shift (Kdp ) and Zdr in the
dendritic growth layer (DGL) and surface precipitation was also described.

Keywords: melting layer; precipitation; R-QVP; X-band dual polarization radar

1. Introduction
The ML is an important region in the precipitation used to describe the transition of the
phase of hydrometeors. The ML or the bright band (BB) occurs at different heights in different
precipitation events (cold and warm) owing to the temperature variability at different seasons.
However, the temperature of the ML is typically 0 ◦ C or slightly higher. During rainfall events,
when the snow/ice particles are exposed to higher temperatures (T > 0 ◦ C), they melt and produce
liquid precipitation on the surface. Extreme precipitation events, such as typhoons and winter storms
frequently occur over the Korean peninsula. These natural disasters significantly impact the lives of
people and animals and damage public and private properties. Further, storms are hazardous for
aviation, transportation, and communication lines. Thus, quantitative precipitation estimation (QPE),
quantitative precipitation forecasts (QPF), and interpretation of the microphysical characteristics of the
storms are particularly important for numerical weather prediction models.
Weather radar is a crucial tool used to study the cloud properties and the microphysical
characteristics of precipitation events. A high-resolution X-band dual polarization radar can play
a significant role in the accurate estimation of the ML, ice growth processes, and QPE. The ML is the
main parameter used to classify the different types of hydrometeors. Several researchers described
various algorithms to detect and characterize the ML using plan position indicator (PPI) scans.

Remote Sens. 2019, 11, 2848; doi:10.3390/rs11232848 [Link]/journal/remotesensing

Remote Sens. 2019, 11, 2848 2 of 23

Giangrande et al. [1] detected the ML based on high Zh , high Zdr , and low ρhv values for volumetric
PPI scans of S-band radars. The range gates which correspond to the ML have Zh values in the
range of 30–47 dBZ, Zdr values in the range of 0.5–2.5 dB, and ρhv values in the range of 0.90–0.97.
Subsequently, Boodoo et al. [2] modified the algorithm which was proposed by Giangrande et al. [1]
to estimate the ML more accurately. The modified thresholds of Zh were found to lie in the range of
20–47 dBZ, Zdr values were in the range of 0.8–2.6 dB, and the ρhv values were in the range of 0.9–0.97,
and were used for the identification of ML in the C-band dual-polarized radar. Matrosov et al. [3] used
a simple ρhv approach for the detection of the top and bottom parts of the ML in the X-band radar.
The ρhv values ranged from 0.9 to 0.95. The end values of this range represented the ML boundaries.
Shusse et al. [4] explored the spatial distribution of the melting layer in a winter stratiform precipitation
over Japan using X-band polarimetric radar using the pseudo-range-height-indicator (RHI) scans.
The minimum value of ρhv (<0.94) was associated with the maximum Zh values and with temperatures
slightly above 0 ◦ C, as suggested by the ML signature. They observed that the ML was consistent with
the boundary between dry snow and rain regions. The ML in convective rain events over a tropical
region was studied by Yuan et al. [5] using the S-band radar PPI measurements. The authors proposed
the ρhv = 0.95 threshold to detect the convective rain melting layer (CRML). Further, the thickness of
the CRML was ~2 times greater than the stratiform melting layer thickness. They reported that the
ML could not be identified at higher altitudes through the high reflectivity and low ρhv (0.95) owing
to the mixing of raindrops with the melting layer. The above-mentioned ML detection algorithms
were based on PPI scans. Wolfensberger et al. [6] described an algorithm based on the RHI scans of
the X-band polarimetric radar. The abrupt vertical gradients of Zh and ρhv were used to characterize
the ML signatures. They reported that the thicker MLs contained deep BB and increased numbers of
particles results in more cooling because of the melting process. Van den Heuvel et al. [7] presented
the spatial and temporal variability characteristics of the ML based on RHI scans in an Alpine valley
using the X-band polarimetric radar. The distribution of the ML heights varied according to the season,
while the thickness was constant with respect to the season and location. The top parts of ML varied
owing to the wind speed, and its directions, and the topography of the Alpine region.
All of the above presented ML detection algorithms were based on PPI and RHI scans. It is not
possible to comprehensively describe the microphysical processes and ML signatures for the entire
precipitation event in time-height form using ML detection algorithms based on RHI and PPI scans.
Hence, Ryzhkov et al. [8] introduced a new quasi-vertical profile (QVP) method for PPI scans to
describe and display the polarimetric radar variables in spatial and temporal form of the storms.
QVP is more beneficial at describing the dendritic growth layer [9], ML [10], and saggy bright band
structures [11], refreezing rain signatures [12], the characteristics of the liquid–ice phase transitions [13],
identification of the dendrites, aggregation, and riming signatures [8,9,14]. QVP is the azimuthal
averaging of the PPI scans at high and fixed elevations. For QVP, Ryzhkov et al. [8] and other researchers
used PPI scans with high elevation angles >10o . Further, QVP needs conical volume PPI scans for
azimuthal averaging. In our study, PPI scans were performed at low elevations of 5◦ and 6◦ . The X-band
dual polarization radar used in this study (see Section 3) contains RHI volume scans rather than PPI.
Therefore, the QVP methodology is not possible with the available PPI scans. The scanning strategy of
the radar is shown in Table 1 (see Section 3). Recently, a novel range-height-indicator–quasi-vertical
profile (R-QVP) methodology was proposed by Allabakash et al. [15] for processing and presenting
the RHI data in spatial and temporal forms. R-QVP is an effective method used to examine the snow
growth processes and describe the different ice particle habits. The R-QVP method is beneficial for
the characterization of the evolution of the precipitation events with data obtained from scanning
weather radars. Allabakash et al. [15] focused on the snow growth processes and their microphysics
using R-QVP. Therefore, we utilized the R-QVP methodology to describe the structure of the ML and
elucidate several microphysical processes of the precipitation events. ML detection algorithm [6] for
RHI scans based on Zh and ρhv was complex and not robust for R-QVPs, especially when the low ρhv
and high Zh values present at multiple height/range levels of R-QVP. Thus, we propose a new and
Remote Sens. 2019, 11, 2848 3 of 23

simple algorithm to detect the top and bottom parts of the ML based on the gradients of Zh , Zdr , ρhv ,
and their combinations. The new method was applied to the R-QVP vertical profiles of the polarimetric
variables to describe the ML and BB. The characteristics of the ML, BB, and ice habits were studied
based on the polarimetric signatures, and these were verified with respect to wet-bulb temperature,
temperature (T), relative humidity (RH), liquid water content (LWC), and ice water content (IWC).

Table 1. Korea Institute of Civil Engineering and Building Technology (KICT) radar scanning strategy
for 10 minutes duration.

Time Scan Strategy CAPPI

1. PPI EL:5◦ , 360◦
AZ: 0 -> (1min)
2. RHI EL: 0 -> 180◦ , AZ: 0: 180◦
3. RHI EL: 180 -> 0◦ , AZ: 162:342◦
1. PPI EL: 6◦ , AZ: 324 -> 323.9◦ (1, 2 min)
2. RHI EL: 0 -> 180◦ , AZ: 324:144◦
3. RHI EL: 180 -> 0◦ , AZ: 126:306◦
1. PPI EL: 5◦ , AZ: 288 -> 287.9◦ (2, 3 min)
5 min 2. RHI EL: 0 -> 180◦ , AZ: 288:108◦
3. RHI EL: 180 -> 0◦ , AZ: 90:270◦
1. PPI EL: 6◦ , AZ: 252 -> 251.9◦ (3, 4 min)
2. RHI EL: 0 -> 180◦ , AZ: 252:72◦
3. RHI EL: 180 -> 0◦ , AZ: 54:234◦
1. PPI EL: 5◦ , AZ: 216 -> 215.9◦ (4, 5 min)
2. RHI EL: 0 -> 180◦ , AZ: 216:36◦
3. RHI EL: 180 -> 0◦ , AZ: 18:198◦
PPI EL: 10◦ , AZ: 198 -> 197.9◦
1. PPI EL: 6◦ , AZ: 180 -> 179.9◦ (5, 6 min)
2. RHI EL: 0 -> 180◦ , AZ: 180:0◦
3. RHI EL: 180 -> 0◦ , AZ: 342:162◦
1. PPI EL: 5◦ , AZ: 144 -> 143.9◦ (6, 7 min)
2. RHI EL: 0 -> 180◦ , AZ: 144:324◦
3. RHI EL: 180 -> 0◦ , AZ: 306:126◦
1. PPI EL: 6◦ , AZ: 108 -> 107.9◦ (7, 8 min)
5 min 2. RHI EL: 0 -> 180◦ , AZ: 108:288◦
3. RHI EL: 180 -> 0◦ , AZ: 270:90◦
1. PPI EL: 5◦ , AZ: 72 -> 71.9◦ (8, 9 min)
2. RHI EL: 0 -> 180◦ , AZ: 72:252◦
3. RHI EL: 180 -> 0◦ , AZ: 234:54◦
1. PPI EL: 6◦ , AZ: 36 -> 35.9◦ (9, 10 min)
2. RHI EL: 0 -> 180◦ , AZ: 36:216◦
3. RHI EL: 180 -> 0◦ , AZ: 198:18◦
PPI EL: 10◦ , AZ: 198 -> 197.9◦

The remainder of this study is organized as follows. Section 2 describes the R-QVP methodology
and the ML detection algorithm used in this study. The observational data is presented in Section 3.
The characteristics of the ML are described in Section 4 based on five case studies. The statistics of the
polarimetric variables in ML, rain, and snow are presented in Section 5. The summary and conclusions
of this study are presented in Section 6.

2. Methods

2.1. R-QVP Methodology

The R-QVP methodology is an efficient way to process and represent the RHI scans in the
time–height form. This method provides the vertical structure of the precipitation event and temporal
evolution of the storm. In this method, initially the range coordinate of the RHI data for a fixed
azimuthal angle is converted/transformed into height (slant range converted into vertical height).
Remote Sens. 2019, 11, 2848 4 of 23

Then, data from the multiple elevation angles within the 100 m grid spacing were averaged together.
In the present study, we utilized the data which corresponded to ρhv > 0.6 and Zh ≥ 0 dBZ for the
averaging process to reduce the non-meteorological contamination. The RHI data corresponding to
the 109◦ azimuthal angle (Goyang region) were utilized for the averaging process. Further, we used
data in the 6–20 km range and at low elevation angles (<45◦ ) to reduce the ground clutter and beam
broadening effects. The 100 m vertical and 5 min temporal resolutions were used in this strategy.
The schematic diagram of the R-QVP strategy is shown in Figure 1. The R-QVP of Zh , Zdr , Kdp , and ρhv
provided significant information related to the precipitation particle type and growth of the ice crystals.
This method mitigated the errors and noise of the polarimetric radar variable. For more details of the
R-QVP strategy the reader can refer to [15].

Figure 1. Schematic of the R-QVP strategy similar to Allabakash et al. [15], except 100 m vertical resolution.

2.2. ML Detection Algorithm

The polarimetric variables, especially Zh , Zdr , and ρhv , are sensitive to the precipitation type,
particle habits, and represent the ML characteristics. Therefore, we used a methodology based on
Zh , Zdr , and ρhv , to detect the ML or BB. The Zh values increased in the ML owing to an increase in
the dielectric factor of the particles, and possibly enhanced aggregation rates, before they melt into
smaller raindrops and decrease in concentration because of flux divergence. The Zdr can easily exceed
that of typical rain for pristine ice crystals, while snow aggregates tend to have Zdr below that of the
equivalent rain distribution. In the ML, the ice particles transition to rain drops within the melting
layer; hence, their axis ratio evolves during the particles change. Additionally, the raindrops are mostly
aligned in the horizontal direction, and Zdr increases for smaller canting angles of raindrops compared
to snow. The ML contains heterogeneous hydrometeors with various axis ratios of melting ice particles,
raindrops, and wetted particles. Therefore, ρhv decreases in the ML. Wolfensberger et al. [6] proposed a
gradient-based method for ML detection based on Zh and ρhv . However, for the ML detection, the Zdr
and ρhv were important compared to Zh [2]. Furthermore, Zdr , ρhv , and Kdp , are more sensitive than Zh
for precise ML detection [16] because of large and wetted particles presented in the ML. Thus, the Zh -
and ρhv -based methods were unable to reliably detect the ML in some cases (when low ρhv and high Zh
values present over the multiple height/range levels). Therefore, Zh , Zdr , and ρhv are robust descriptors
of the shape, size, and type of the precipitation particles. Thus, in the present study, we refine the
Remote Sens. 2019, 11, 2848 5 of 23

algorithm proposed by Wolfensberger et al. [6] and present a method based on Zh , Zdr , and ρhv and
their combinations to detect reliably the ML. In this method, the values of Zh , Zdr , and ρhv were initially
normalized by the maximum value in the respective profiles, and their gradients were then estimated.
These gradients could not detect the true ML thickness in low ρhv and high Zh cases as described above.
Therefore, the Zh and Zdr values were individually multiplied/combined with the complementary ρhv
values, and the product of the Zh and Zdr values was multiplied with the complementary ρhv value
that were used to describe the top and bottom ML parts. The combined parameters are shown below.

Zh ·(1-ρhv ) (1)

Zdr ·(1-ρhv ) (2)

Zh ·Zdr ·(1-ρhv ) (3)

Finally, the gradients were estimated for the absolute values of the normalized components of (1),
(2), and (3). The corresponding heights of the maximum and minimum values of these gradients were
then computed.
The ML boundaries were detected as follows:
Step 1. The coinciding height of the maximum values of any two gradients among (1), (2), and (3),
represent the bottom part of the ML.
Step 2. The coinciding height corresponds to the minimum of any two gradients among (1), (2),
and (3), and signifies the top ML part. However, in some cases, the Zdr peak is below the Zh peak and
the ρhv minimum because the value of Zdr is low in ice and high in liquid. For this reason, the top part
of the ML is associated with considerably high Zh and low ρhv values, while the bottom ML part is
associated with increased Zdr values. Thus, the top part of the ML should be corrected.
Step 3. If the coinciding height presented in “Step 2” is consistent with the height of the minimum
Zh and the maximum ρhv gradients, the corresponding height is then considered as the top part of
the ML.
Step 4. If “Step 3” is not satisfied, the maximum height of either the height of the minimum
Zh or maximum ρhv gradients should be greater than the coincident height (presented in “Step 2”).
Correspondingly, this height represents the top part of the ML.
Step 5: If “Step 3” and “Step 4”are both unsuccessful, then the height above the coinciding height
presented in “Step 2” is considered as the top part of the ML.
This method estimates the top and bottom parts of the ML efficiently. The flow chart of the
proposed method is shown in Figure 2.
This algorithm was applied to the polarimetric variables, and the relevant example is shown
in Figure 3. Figure 3a illustrates the profile obtained at 14:24 [coordinated universal time (UTC)]
on 24 July 2014. The left panel represents the absolute normalized values (hereafter absolute normalized
values are referred to as AN) of Zh (for simplicity referred to as AN(Zh )), Zdr (for simplicity referred
to as AN(Zdr )), ρhv (for simplicity referred to as AN(ρhv )), and their gradients are illustrated in
the middle panel. The gradients of AN of Zh (1-ρhv ), AN of Zdr (1- ρhv ), and AN of Zh Zdr (1- ρhv ),
are shown in the right panel. The maxima of the gradient of AN(Zh (1-ρhv )) and the gradient of
AN(Zh Zdr (1-ρhv )) are coincident at 4 km. Thus, the 4-km height signifies the bottom of the ML.
Similarly, the minima of the gradients of AN(Zh (1-ρhv )) and AN(Zh Zdr (1-ρhv )) are coincident at a
height of 4.5 km. However, the minimum and maximum gradients of Zh and ρhv are respectively
presented at the different heights of 4.5 and 4.85 km. Therefore, the ML top represents the maximum
height level of 4.85 km, which is presented above the coinciding height (4.5 km). The magenta color
“x” marks represent the ML boundaries detected by the method based on Zh and ρhv [6], which shows
that the bottom of the ML is similar to the proposed method. However, the top of the ML, at 4.5 km,
differs from the current method. Figure 4c shows the height of the 0 ◦ C wet-bulb temperature at 4.9 km,
at 15:00 UTC (data extracted from MERRA; 14:30 UTC data not available). This example signifies that
the proposed method estimates reliable ML boundaries.
Remote Sens. 2019, 11, 2848 6 of 23

Figure 2. Flow chart of proposed ML detection method.

Figure 3. Melting layer detection on 24 July 2014. Profiles at (a) 14:24 [coordinated universal time
(UTC)] and (b) at 14:34 UTC. The left and middle panels are the normalized values of Zh , Zdr , and ρhv,
and their gradients, respectively. The right panels are the gradients of the combined parameters.
The melting layer (ML) boundaries estimated using the Zh and ρhv gradient based method are indicated
by the “x” marks.
Remote Sens. 2019, 11, 2848 7 of 23

Figure 4. Cont.
Remote Sens. 2019, 11, 2848 8 of 23

Figure 4. (a) R-QVP profiles of Zh , Zdr , Kdp , ρhv , and ML boundaries obtained on 24 July 2014. The black
solid lines and diamond markers represent the top and bottom parts of the ML by the proposed method
and Zh and ρhv gradients-based method; (b) variation of surface parameters, including temperature,
relative humidity, precipitation, and horizontal wind speed extracted from Modern-Era retrospective
analysis for research and applications 2 (MERRA2) on 24 July 2014; (c) vertical parameters of wet-bulb
temperature, relative humidity, liquid water content, and ice water content extracted from MERRA
on 24 July 2014.

Figure 3b corresponds to 14:34 UTC on 24 July 2014. The right panel shows that the gradients of
AN(Zdr (1−ρhv )) and AN(Zh Zdr (1−ρhv )) maxima are coincident at a height of 4 km, which signifies the
bottom part of the ML. The minima of the gradients of AN(Zdr (1−ρhv )), AN(Zh (1−ρhv )), and AN(Zh Zdr
Remote Sens. 2019, 11, 2848 9 of 23

(1−ρhv )) are presented at a height of 4.85 km, and are also consistent with the minimum gradient of
Zh and the maximum gradient of ρhv . Therefore, 4.85 km was designated as the top part of the ML.
While, the Zh and ρhv gradient based method [6] illustrates the 4.25 km and 4.85 km boundaries
(magenta “x”) of the ML. The top of the ML, at 4.85 km, is similar to the proposed method, whereas the
bottom is different.

3. Observational Data and Processing

In this study, we used the data obtained from the X-band dual polarization radar operated by the
Korea Institute of Civil Engineering and Building Technology (hereafter referred to as KICT radar),
located approximately 20.6 km Northwest of Seoul, South Korea. The radar operates in the simultaneous
transmission and reception mode at a frequency of 9.41 GHz. It provides one PPI and two RHI scans
every minute, and a total of six PPI and 10 RHI scans in a 5 min period. The elevation angle of PPI
alternates between 5◦ and 6◦ , while the RHI scans are produced every 18◦ in the azimuthal direction.
The collected data incorporated the attenuation correction (Φdp- based), data quality check [17],
bias/error correction, and Kdp estimation [18] algorithms. For more details of the KICT radar location,
the scanning strategy, and data quality control algorithms, the reader can refer to Chen et al. [19] and
Allabakash et al. [15].
The atmospheric parameter data used in this study were collected from the MERRA and MERRA2
reanalysis datasets. The MERRA data was generated by the Global Modeling and Assimilation Office
(GMAO) of the National Aeronautics and Space Administration (NASA) for the Goyang region using
the Goddard Earth Observing System, version 5 (GEOS-5), data assimilation system [20]. The MERRA2
reanalysis data were also produced by the NASA GMAO with the use of the GEOS-5.12.4 system [21].
We used MERRA and MERRA-2 for vertical/spatial and horizontal/temporal data, respectively.
The vertical profiles of wet-bulb temperature, RH, cloud liquid water mixing ratio (QL), and cloud ice
mixing ratio (QI), and the surface/horizontal profiles of T, RH, precipitation (P), and horizontal wind
speed (WS) were derived for the Goyang Province.

4. Results and Discussion

4.1. Case 1: 24 July 2014

On 24 July 2014, a precipitation event occurred in the Goyang region. The polarimetric radar
signatures defined the structure of this event. The R-QVP profiles provided the spatial and temporal
error mitigated polarimetric variables. Figure 4a shows the R-QVP vertical profiles of the Zh , Zdr , Kdp ,
and ρhv corresponding to this event. The ML was estimated using the method described above and
illustrated as black solid lines. On the other hand, the diamond markers indicate the ML estimated
using the gradients of Zh and ρhv values [6]. The upper and lower solid/marker lines indicate the top
and bottom parts of the ML, respectively. It can be noted that at 14:15 and 15:15 UTC, ρhv was greater
than 0.98. Accordingly, it was difficult to identify the top and bottom parts of the ML using Zh and
ρhv , and the markers (Zh and ρhv based method) showing unstable high/low values are not reliable.
However, the combination of the Zh and Zdr values in conjunction with the values of ρhv wa3s more
useful in these cases. The solid lines (proposed method) show the reliable and stable ML boundaries.
It is clearly observed that low ρhv values (< 0.98), high Zh values (≥ 28 dBZ), and high Zdr values
(≥ 1.5 dB) are obtained in the ML. Furthermore, the ML also contains high Kdp values (> 0.5◦ km−1 ).
To describe and verify the ML structure, the surface parameters of T, RH, P, and WS, and the vertical
parameters of wet-bulb temperature, RH, QL, and QI were extracted from MERRA2/MERRA and are
illustrated in Figure 4b,c, respectively. T decreases and RH increases from 13:30 UTC onward, as shown
in Figure 4b. Furthermore, RH increases above 90%, which was conducive to rainfall. Correspondingly,
the precipitation exhibits increasing trend. The vertical profiles show the corresponding height of
the ML. The 0 ◦ C wet-bulb temperature at a height of 4.9 km, corresponds to the top part of the ML.
Furthermore, RH values increased for heights in the range, 3.5 to 5 km. The LWC (represented by QL)
Remote Sens. 2019, 11, 2848 10 of 23

and the IWC (represented by QI) yield maximum values for heights in the range from 4 to 5 km, thus
signifying the mixing of the liquid and solid (ice) particles in the ML. Thus, based on this analysis,
we infer that the proposed algorithm was reliable in detecting the true ML. Figures 3a and 4a show
that the Zdr peak is always present below the Zh peak. The Zdr peak was present at lower heights
and at T > 0 ◦ C. Therefore, the melting of the hydrometeors led to a Zdr maximum occurs at lower
height levels than the Zh peak. Thus, this offset signifies the maximum eccentricity of the melting
particles [16]. Further, at lower level of the ML, the smallest particles had already collapsed into
nearly spherical raindrops that were small and contributed less to Zh , leaving the largest aggregates
dominating Zh and thus occur Zdr magnitude.

4.2. Case 2: 11 May 2015

The R-QVP profiles related to this event are shown in Figure 5a. The solid lines (proposed method)
show the stable ML boundaries, while the markers (based on Zh and ρhv ) show unstable ML thickness.
The top of the ML estimated using the current method is consistent with the 0 ◦ C wet-bulb temperature
(shown below). The ML boundaries can be observed for heights in the range of 3–4 km. The thickness
of the ML was approximately 0.7 km. If there was an abrupt change in the ML thickness, then the
top and bottom parts of the ML were interpolated with their adjacent values. Alternatively, it was
considered that BB was absent. It can be observed that from 13:00 to 14:00 UTC, the BB was absent.
The ML contains increased Zh , Zdr , and Kdp values with decreased ρhv values that illustrate the different
hydrometeors types. The increased Kdp values in the ML indicate the large melting aggregates of ice
particles. Intriguing signatures can be observed in Figure 5a. The Zh is higher on the bottom side of the
BB (>15 dBZ) than the top (≤15 dBZ), and represent rain and ice, respectively [22]. Conversely, the Kdp
trend was opposite in the cases of Zh and Zdr (above and below of ML). Correspondingly, the Kdp
values increased for solid particles and decreased for liquid particles. Furthermore, slightly depressed
ρhv values were present above ML owing to depositional growth of the pristine ice crystals [23]. It is
interesting to note that the Kdp value increased from 09:00 to 13:00 UTC at the cloud top temperature
(−20 ◦ C) associated with the moderate Zdr values, thus signifying the large concentration of isometric
particles [9,15].
The corresponding surface parameters are shown in Figure 5b. The respective decreasing and
increasing T and RH trends from 06:30 UTC signify the formation of precipitation. The enhancement of
precipitation and WS can also be observed in Figure 5b. It is interesting to note the high precipitation
at/followed by the high Zdr and Kdp values present in the dendritic growth layer (DGL) between the
temperature range, −10 to −15 ◦ C [15,23,24]. The Zdr and Kdp values between −10 and −15 ◦ C were
averaged for each height and presented in the bottom panel of Figure 5b. The high Zdr and high Kdp
presented at 08:00–09:00 UTC, and precipitation also started during this period (see middle panel
of Figure 5b). Again, the enhanced Zdr and Kdp values occurred at 11:00–13:00 UTC, where high
precipitation was observed on the ground. The vertical profiles of the meteorological parameters are
shown in Figure 5c. The plot of wet-bulb temperature shows the 0 ◦ C isotherm in the range of 3.5–3.8
km at different times that correlates with the ML heights. It is interesting to note that the LWC and
IWC show enhanced values at the ML boundary in the range of 3–4 km and illustrate the mixed phase
hydrometeors. At 06:00 UTC, P and ML had small amplitudes, but had increased amplitudes at 09:00,
12:00, and 15:00 UTC. Correspondingly both QL and RH also showed low values at 06:00 UTC and
high values (100%) at other times.
Remote Sens. 2019, 11, 2848 11 of 23

Figure 5. Cont.
Remote Sens. 2019, 11, 2848 12 of 23

Figure 5. (a) R-QVP profiles of Zh , Zdr , Kdp , ρhv , and ML boundaries recorded on 11 May 2015.
The black solid lines and diamond markers represent the top and bottom parts of the ML by the
proposed method and Zh and ρhv gradients based method. (b) Variation of surface parameters,
including the temperature and relative humidity (top panel), precipitation and horizontal wind speed
(middle panel) extracted from MERRA2 on 11 May 2015. Bottom panel corresponds to averaged Zdr
and Kdp within the dendritic growth layer (DGL) (−10 to −15 ◦ C). (c) Vertical parameters of wet-bulb
temperature, relative humidity, liquid water content, and ice water content extracted from MERRA
on 11 May 2015.
Remote Sens. 2019, 11, 2848 13 of 23

4.3. Case 3: 2 December 2015

This case corresponds to a winter precipitation event. The R-QVP profiles are illustrated in
Figure 6a and the corresponding surface and vertical parameters are shown in Figure 6b,c, respectively.
Figure 6a clearly shows the proposed method (solid lines) evaluated a reliable ML, in contrast to the Zh
and ρhv based method (diamond markers). The ML shows significant polarimetric signatures. In the
winter, the top and bottom parts of the ML are present at lower heights in the range of 1–2 km. It is
clearly shown that the solid phase particles exhibit low Zh , low Zdr , high Kdp , and low ρhv values,
while the opposite is true for rain hydrometeors.

Figure 6. Cont.
Remote Sens. 2019, 11, 2848 14 of 23

Figure 6. (a) R-QVP profiles of Zh , Zdr , Kdp , ρhv , and ML boundaries obtained on 2 December 2015.
The black solid lines and diamond markers represent the top and bottom parts of the ML by the proposed
method and Zh and ρhv gradients based method. (b) Variations of surface parameters, including
temperature and relative humidity (top panel), precipitation and horizontal wind speed (bottom panel)
extracted from MERRA2 on 2 December 2015. Bottom panel corresponds to averaged Zdr and Kdp
within the DGL (−10 to −15o C temperature). (c) Vertical parameters of wet-bulb temperature, relative
humidity, liquid water content, and ice water content extracted from MERRA on 2 December 2015.

The surface parameters (Figure 6b) exhibit increased RH values from 00:00 to 04:00 UTC, while the
surface precipitation is also shown to be greater than 1 mm on the ground. The averaged Zdr and
Kdp values of DGL are shown in Figure 6b bottom panel. In DGL, at some instances high Kdp was
present without high Zdr [25]. The averaged Kdp values of DGL are shown at 00:00–01:00 UTC and at
02:00–2:30 UTC, and high precipitation was also observed during this time, especially during the later
time period. The vertical wet-bulb temperature (Figure 6c) illustrates the 0 ◦ C isotherm at a height
of 2 km, which corroborates the ML findings. Furthermore, the RH was enhanced (100%), while the
QL and QI were also high in the 1–4 km height range and represent the liquid and ice hydrometeors.
Remote Sens. 2019, 11, 2848 15 of 23

From 12:00 UTC onward, rainfall changed over to snowfall owing to the negative temperatures,
as observed by the surface T values.

4.4. Case 4: 5 March 2016

Figure 7a shows the R-QVP profiles of polarimetric variables corresponding to this event. The solid
lines (proposed method) show smooth/reliable ML boundaries, in contrast to the diamond markers
(Zh and ρhv based method). The ML can be observed within the 1.5–2.8 km height range. The surface
and vertical parameters are illustrated in Figure 7b,c, respectively. From 02:00 UTC onward, the RH
increases that indicates the rain falling into a dry layer near the surface. The surface precipitation also
shows an increasing trend from 02:00 UTC onward. The enhanced Zdr and Kdp of DGL (bottom panel)
shown at 03:30–07:00 UTC illustrates the start of the precipitation (see middle panel). Again, high
values are presented at 08:30–10:00 UTC, signifying heavy precipitation on the ground. The vertical
wet-bulb temperature illustrates the 0 ◦ C isotherm at an approximate height of 2 km, whereby the QL
and QI also show enhanced values for heights in the range from 2–4 km, while RH values greater than
80% refer to mixed precipitation particles in the ML.

Figure 7. Cont.
Remote Sens. 2019, 11, 2848 16 of 23

Figure 7. (a) R-QVP profiles of Zh , Zdr , Kdp , ρhv , and ML boundaries on 5 March 2016. The black solid
lines and diamond markers represent the top and bottom parts of the ML by the proposed method
and Zh and ρhv gradients based method. (b) Variation of surface parameters of temperature and
relative humidity (top panel), precipitation and horizontal wind speed (middle panel) extracted from
MERRA2 on 5 March 2016. Bottom panel corresponds to averaged Zdr and Kdp within the DGL −10 to
−15 ◦ C temperature). (c) Vertical parameters of wet-bulb temperature, relative humidity, liquid water
content, and ice water content extracted from MERRA on 5 March 2016.

4.5. Case 5: 1 October 2016

In this case, the proposed method (solid lines) signifies the true ML boundaries, while the diamond
markers (Zh and ρhv based method) show uncertain ML values, as shown in Figure 8a. The ML was
Remote Sens. 2019, 11, 2848 17 of 23

observed at heights in the range of 2.5–3.5 km. The RH and surface precipitation also exhibit higher
values from 19:30 to 23:30 UTC (Figure 8b), and also high Kdp of DGL also observed during this
interval, especially at 21:00–23:30 UTC time period (where high precipitation presented). The wet-bulb
temperature shows 0 ◦ C isotherm at 3.2 km height. The enhanced signatures of the vertical parameters
of the RH, QL, and QI, are presented for heights in the range of 3–4 km, which correspond to the
ML height.

Figure 8. Cont.
Remote Sens. 2019, 11, 2848 18 of 23

Figure 8. (a) R-QVP profiles of Zh , Zdr , Kdp , ρhv , and ML boundaries obtained on 1 October 2016.
The black solid lines and diamond markers represent the top and bottom parts of the ML by the
proposed method and Zh and ρhv gradients based method. (b) Variation of surface parameters,
including temperature and relative humidity (top panel), precipitation and horizontal wind speed
(middle panel) extracted from MERRA2 on 1 October 2016. Bottom panel corresponds to averaged Zdr
and Kdp within the DGL (−10 to −15 ◦ C temperature). (c) Vertical parameters of wet-bulb temperature,
relative humidity, liquid water content, and ice water content extracted from MERRA on 1 October 2016.

The ML heights obtained from the Zh and ρhv -based methods and the proposed method for all the
cases were compared with the wet-bulb 0 ◦ C isotherm and presented as a scatter plot, shown in Figure 9.
Remote Sens. 2019, 11, 2848 19 of 23

The figure shows that the proposed method produced high correlation (0.96) with lower standard
deviation than that of the Zh and ρhv -based method (correlation = 0.84, standard deviation = 0.18).
Thus, we advocate that the current method is efficient at ML estimation.

Figure 9. Scatter plot between the ML tops derived from the radar (Zh and ρhv based method,
and proposed method) and the heights correspond to 0 ◦ C wet-bulb temperature.

5. Polarimetric Variables Statistics in ML, Rain, and Snow

Based on the above case studies, we evaluated the statistics of Zh , Zdr , Kdp , and ρhv in the ML,
rain, and snow. The distributions of these variables are shown in Figure 10 and their statistics in
different quantiles are presented in Table 2. Zh in the ML shows symmetrical distributions with a mean
value of 21 dBZ. The quantile 10 (Q10), Q50, and Q90 values are 13.23, 21.15, and 28.38, respectively.
Zh in rain is more symmetrical than that of Zh in snow and the ML. Zh in snow contains lower values
and is left-skewed. Zdr in the ML shows a right-skewed distribution with a mean value of 0.69. Zdr in
rain and snow shows similar distributions to Zh . In the ML, ρhv shows symmetrical distributions
and the values are contained in a small area with a mean of 0.978. The ρhv values within the ML is
lower than in the liquid and solid phase particles. The ρhv values in rain is right skewed with higher
values than that of ρhv in the ML and snow. The distributions of Zh , Zdr , and ρhv are similar to the
results presented by [1] and [6]. Kdp in the ML exhibits left skewed distributions with a mean of 0.29.
Kdp in snow and the ML exhibit similar distributions, but the Kdp spread in snow is slightly larger
than in the ML. The phase of the particles were easily described by these distributions. From all the
distributions and statistics shown in Table 2, we infer the following:

1. The enhanced values of Zh , Zdr , and Kdp , and low ρhv values present in the ML are due to the
mixed phase of hydrometeors.
2. Zh is lower in the snow region than in rain due to lower dielectric effects of ice particle [26].
3. The snow aggregates are larger with low density and randomly oriented, producing smaller Zdr
(< 0.5 dB) values in the snow than in rain [15].
4. Kdp is higher in snow due to the increase in particle concentration [27,28].
5. ρhv is lower in pristine ice crystals, especially mixed with aggregates [26], thus the snow region
contains a lower ρhv than that of the rain region.
Remote Sens. 2019, 11, 2848 20 of 23

Figure 10. Distribution of polarimetric variables (a) Reflectivity (Zh ), (b) Differential reflectivity (Zdr ),
(c) Specific differential phase shift (Kdp ), and (d) Copolar correlation coefficient (ρhv) in ML, rain,
and snow.

Table 2. Statistics of the polarimetric variables in ML, rain, and snow.

Variable Statistics ML Rain Snow

Mean 21.07 18.12 11.02
STD 5.95 6.21 5.37
Zh (dBZ) Q10 13.23 10.22 3.75
Q50 21.15 18.42 11.32
Q90 28.38 26.19 17.41
Mean 0.69 0.41 0.29
STD 0.52 0.62 0.35
Zdr (dB) Q10 0.16 −0.23 −0.01
Q50 0.60 0.34 0.24
Q90 1.36 1.22 0.67
Mean 0.29 0.13 0.39
STD 0.28 0.20 0.39
Kdp (o km−1 ) Q10 −0.01 −0.01 0
Q50 0.25 0.07 0.30
Q90 0.62 0.33 0.90
Mean 0.978 0.983 0.980
STD 0.04 0.04 0.05
ρhv Q10 0.976 0.978 0.975
Q50 0.978 0.984 0.981
Q90 0.983 0.986 0.984

It is known that the ML varies according to the seasons owing to the variability of the 0 ◦ C isotherm.
We also observed this feature. The top and bottom parts of the ML were high in the summer, followed
by autumn, spring. Additionally, it is shown that ML is low in winter. We also observed that the
thickness is independent of the season and the spatial location within this region. Table 3 shows the
boundaries of the ML during different seasons.
Remote Sens. 2019, 11, 2848 21 of 23

Table 3. Top and bottom parts of melting layer at different seasons.

Melting Layer (average) Top (km) Bottom (km)

Winter 1.73 1.06
Spring 3.24 2.72
Summer 5.05 4.43
Autumn 3.71 3.11

6. Summary and Conclusions

In this study, we proposed a new and refined simple automatic method to detect the ML thickness
based on R-QVPs of polarimetric variables. The algorithm utilized X-band dual polarization radar
data obtained in South Korea. The gradients of Zh , Zdr , ρhv , Zh (1-ρhv ), Zdr (1-ρhv ), and Zh Zdr (1-ρhv ),
were used to describe the top and bottom parts of the ML. The enhanced values of Zh , Zdr , Kdp ,
and depressed ρhv in the ML were indicative of the heterogeneous hydrometeors. The low values of Zh ,
Zdr , ρhv , and high values of Kdp present on top of the ML indicated the solid phase of the precipitation
particles. Conversely, the Zh , Zdr , ρhv values were high and the Kdp values were low below the ML,
thus signifying the liquid phase of the hydrometeors. The Zdr peak present below the Zh peak in the
ML was attributed to the eccentricity of the hydrometeors. The R-QVP profiles presented the error
mitigated polarimetric variables that exhibited spatial and temporal variabilities of the ML signatures.
The ML was high in the summer followed by the autumn, spring, and winter, owing to the variability
of the height of the 0 ◦ C isotherm. The ML thickness was independent of the season and the spatial
location within this region. The detected ML was also verified by the wet-bulb temperature profile
derived from the MERRA reanalysis data. The ML was mostly associated with the enhanced values
of RH, QL, and QI. The increased values of QL and QI in the ML signified a mixed hydrometeor
phase (liquid and solid). The distributions and statistics of the polarimetric variables in the ML, rain,
and snow regions signified the phase of the hydrometeors in these regions. We also observed high Zdr
and Kdp values in DGL, indicative of high precipitation on the ground, similar to the observations of
Trömel et al. [29].
The ML detection in the R-QVP profiles is useful for the QPE estimation and hydrometeor
classification algorithms than the conventional ML detection methods based on PPI/ RHI scans.
These are particularly important to identify the particle types in the precipitation events for aviation
forecasting. The combined methodologies of Allabakash et al. [15] and this method are useful for
forecasting the heavy rain/QPE and snowfall, and provide early warnings to the public.

Author Contributions: Conceptualization, S.A. and S.L.; formal analysis, S.A. and S.L.; writing—original draft,
S.A.; writing—review and editing, S.L. and B.-J.J.
Funding: This work was supported by the Korea Institute of Civil Engineering and Building Technology Strategic
Research Project (Development of Driving Environment Observation, Prediction and Safety Technology Based on
Automotive Sensors).
Acknowledgments: The authors would also like to thank MERRA and MERRA2 for providing the different
datasets used in this paper.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Giangrande, S.E.; Krause, J.M.; Ryzhkov, A.V. Automatic designation of the melting layer with a polarimetric
prototype of the WSR-88D radar. J. Appl. Meteorol. Climatol. 2008, 47, 1354–1364. [CrossRef]
2. Boodoo, S.; Hudak, D.; Donaldson, N.; Leduc, M. Application of dual-polarization radar melting-layer
detection algorithm. J. Appl. Meteorol. Climatol. 2010, 49, 1779–1793. [CrossRef]
3. Matrosov, S.Y.; Clark, K.A.; Kingsmill, D.E. A polarimetric radar approach to identify rain, melting-layer,
and snow regions for applying corrections to vertical profiles of reflectivity. J. Appl. Meteorol. Climatol. 2007,
46, 154–166. [CrossRef]
Remote Sens. 2019, 11, 2848 22 of 23

4. Shusse, Y.; Maesaka, T.; Kieda, K.; Iwanami, K. Polarimetric radar observation of the melting layer in a
winter precipitation system associated with a South-coast cyclone in Japan. J. Meteorol. Soc. Jpn. Ser. II 2019,
97, 375–385. [CrossRef]
5. Yuan, F.; Lee, Y.; Meng, Y.; Ong, J. Characterization of S-band dual-polarized radar data for the convective
rain melting layer detection in a tropical region. Remote Sens. 2018, 10, 1740. [CrossRef]
6. Wolfensberger, D.; Scipion, D.; Berne, A. Detection and characterization of the melting layer based on
polarimetric radar scans. Q. J. R. Meteorol. Soc. 2016, 142, 108–124. [CrossRef]
7. Van den Heuvel, F.; Gabella, M.; Germann, U.; Berne, A. Characterisation of the melting layer variability in
an Alpine valley based on polarimetric X-band radar scans. Atmos. Meas. Tech. 2018, 11, 5181. [CrossRef]
8. Ryzhkov, A.; Zhang, P.; Reeves, H.; Kumjian, M.; Tschallener, T.; Trömel, S.; Simmer, C. Quasi-vertical
profiles—A new way to look at polarimetric radar data. J. Atmos. Ocean. Technol. 2016, 33, 551–562.
[CrossRef]
9. Griffin, E.M.; Schuur, T.J.; Ryzhkov, A.V. A polarimetric analysis of ice microphysical processes in snow,
using quasi-vertical profiles. J. Appl. Meteorol. Climatol. 2018, 57, 31–50. [CrossRef]
10. Tobin, D.M.; Kumjian, M.R. Polarimetric radar and surface-based precipitation-type observations of ice pellet
to freezing rain transitions. Weather. Forecast. 2017, 32, 2065–2082. [CrossRef]
11. Kumjian, M.R.; Mishra, S.; Giangrande, S.E.; Toto, T.; Ryzhkov, A.V.; Bansemer, A. Polarimetric radar and
aircraft observations of saggy bright bands during MC3E. J. Geophys. Res. Atmos. 2016, 121, 3584–3607.
[CrossRef]
12. Kaltenboeck, R.; Ryzhkov, A. A freezing rain storm explored with a C-band polarimetric weather radar using
the QVP methodology. Meteorol. Z. 2017, 26, 207–222. [CrossRef]
13. Bukovčić, P.; Zrnić, D.; Zhang, G. Winter precipitation liquid–ice phase transitions revealed with polarimetric
radar and 2DVD observations in central Oklahoma. J. Appl. Meteorol. Climatol. 2017, 56, 1345–1363.
[CrossRef]
14. Kumjian, M.R.; Lombardo, K.A. Insights into the evolving microphysical and kinematic structure of
northeastern US winter storms from dual-polarization Doppler radar. Mon. Weather. Rev. 2017, 145, 1033–1061.
[CrossRef]
15. Allabakash, S.; Lim, S.; Chandrasekar, V.; Min, K.H.; Choi, J.; Jang, B. X-Band dual polarization radar
observations of snow growth processes of a severe winter storm: Case of 12 December 2013 in South Korea.
J. Atmos. Ocean. Technol. 2019, 36, 1217–1235. [CrossRef]
16. Brandes, E.A.; Ikeda, K. Freezing-level estimation with polarimetric radar. J. Appl. Meteorol. 2004,
43, 1541–1553. [CrossRef]
17. Scarchilli, G.; Gorgucci, V.; Chandrasekar, V.; Dobaie, A. Self-consistency of polarization diversity
measurement of rainfall. IEEE Trans. Geosci. Remote. Sens. 1996, 34, 22–26. [CrossRef]
18. Wang, Y.; Chandrasekar, V. Algorithm for estimation of the specific differential phase. J. Atmos. Ocean. Technol.
2009, 26, 2565–2578. [CrossRef]
19. Chen, H.; Lim, S.; Chandrasekar, V.; Jang, B.J. Urban hydrological applications of dual-polarization X-band
radar: Case study in Korea. J. Hydrol. Eng. 2016, 22, E5016001. [CrossRef]
20. Rienecker, M.M.; Suarez, M.J.; Gelaro, R.; Todling, R.; Bacmeister, J.; Liu, E.; Bosilovich, M.G.; Schubert, S.D.;
Takacs, L.; Kim, G.K.; et al. MERRA: NASA’s modern-era retrospective analysis for research and applications.
J. Clim. 2011, 24, 3624–3648. [CrossRef]
21. Bosilovich, M.; Akella, S.; Coy, L.; Cullather, R.; Draper, C.; Gelaro, R.; Kovach, R.; Liu, Q.; Molod, A.;
Norris, P.; et al. MERRA-2: Initial evaluation of the climate. NASA Technical Report. Series on Global Modelling
and Data Assimilation, NASA/TM–2015-104606; NASA: Washington, DC, USA, 2015.
22. Doviak, R.J.; Zrnić, D.S. Doppler Radar and Weather Observations; Academic Press: Cambridge, UK, 1993; p. 562.
23. Kennedy, P.C.; Rutledge, S.A. S-band dual-polarization radar observations of winter storms. J. Appl.
Meteorol. Climatol. 2011, 50, 844–858. [CrossRef]
24. Bechini, R.; Baldini, L.; Chandrasekar, V. Polarimetric radar observations in the ice region of precipitating
clouds at C-band and X-band radar frequencies. J. Appl. Meteorol. Climatol. 2013, 52, 1147–1169. [CrossRef]
25. Moisseev, D.N.; Lautaportti, S.; Tyynela, J.; Lim, S. Dual-polarization radar signatures in snowstorms: Role
of snowflake aggregation. J. Geophys. Res. Atmos. 2015, 120, 12644–12655. [CrossRef]
26. Straka, J.M.; Zrnić, D.S.; Ryzhkov, A.V. Bulk hydrometeor classification and quantification using polarimetric
radar data: Synthesis of relations. J. Appl. Meteorol. 2000, 39, 1341–1372. [CrossRef]
Remote Sens. 2019, 11, 2848 23 of 23

27. Bringi, V.N.; Chandrasekar, V. Polarimetric Doppler Weather Radar: Principles and Applications; Cambridge
University Press: Cambridge, UK, 2001; p. 634.
28. Pruppacher, H.R.; Klett, J.D. Microphysics of Clouds and Precipitation, 2nd ed.; Kluwer Academic:
Dordrecht, The Netherlands, 1997; p. 954.
29. Trömel, S.; Ryzhkov, A.; Hickman, B.; Mühlbauer, K.; Simmer, C. Climatology of the vertical profiles of
polarimetric radar variables at X band in stratiform clouds. In Proceedings of the 38th Conference on Radar
Meteorology, Bonn, Germany, 27–31 August 2017.

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license ([Link]