2.2. FURUNO X-Band Weather Radar and Its Data
The FURUNO weather radar installed at the Milešovka observatory (FURUNO WR2120) in late 2020 is a Doppler polarimetric X-band radar. Its basic parameters are given in
Table 1. The radar is located at the top of the observatory tower (
Figure 1), which provides an unobstructed view of the surroundings with the exception of the few metal rods needed for other measuring devices placed on the tower that cannot be removed due to the need for maintaining measuring continuity. Our experience regarding the operation of the radar since 2021 has shown that these metal rods probably cause, under certain conditions, artefacts in the radar measurements, which are manifested at higher elevations and in the vicinity of the radar, and are currently being solved in cooperation with the radar manufacturer. The artefacts are not standard errors in radar measurements, which are known, thus standard automatic corrections, e.g., [
18], cannot be applied. As we are interested in the microphysical processes of convective clouds and their relation to cloud electrification, we focus on the data from the mid and upper troposphere, so the described artefacts are not crucial for us.
The radar scanning strategy consists of two steps: (i) the radar performs 7 horizontal ppi (plan position indicator) clockwise scans for elevations of 1.1°, 1.7°, 2.5°, 4°, 6°, 10° and 25°; (ii) the radar makes 6 RHI (range height indicator) scans, i.e., 6 vertical cross section scans, for elevation angles from 3° to 90°. The difference between adjacent elevation angles is non-equidistant but not larger than 0.5°. The azimuths of the scans are set in a way that they can be folded into three cross sections, the first of which is oriented from south to north, the second rotated 60° clockwise to the first, and the third rotated 120° to the first.
We set the radar scanning strategy to correspond to our research of summer convective clouds which occur near and above the Milešovka meteorological observatory. Specifically, we combine PPI and RHI scans and use the final scan parameters resulting from a trade-off between two following contradictory requirements: (i) to perform as many spatial measurements as possible, which is important due to the local character of convective phenomena; and (ii) to obtain the highest possible time resolution, which is important due to the rapid development of the convective phenomena.
The lowest PPI scan is set relatively high intentionally since our primary target is not precipitation estimatation in the radar domain but investigation of cloud structure. In addition, the mountain ridge, with heights of about 1000 m at a distance of about 10 km from the radar in the upper left quadrant of the radar domain (north, northwest), does not allow effective use of scans with lower elevations.
The vertical cross sections in the figures presented below are oriented in a way that the negative distance from the Milešovka observatory is in the direction to the south, while the positive distance is in the direction to the north. Other radar scan parameters were set as follows: radar horizontal resolution is 150 m, radar range is 50 km, and Doppler velocity range is 49 m/s. One cycle of all the scans lasts 170 s.
The radar measures and records the following types of data:
R [mm/h]: Rainfall intensity.
Zh [dBZ]: Reflectivity intensity factor of horizontal polarization wave.
Zh_corr [dBZ]: Attenuation corrected Zh of the horizontal polarity data.
V [m/s]: Doppler velocity.
Zdr [dB]: Differential reflectivity.
Zdr_corr [dB]: Corrected differential reflectivity.
Kdp [deg/km]: Specific differential phase.
Φdp [deg]: Differential Phase Shift (cross polarization).
Rhohv: Co-polar correlation coefficient.
W [m/s]: Doppler velocity spectrum width.
Rhohv is primarily influenced by the hydrometeors’ shape and canting angle distributions and is used to identify the type of the target (Rhohv < 0.8 is typical for clutters; values 0.8 ≤ Rhohv < 0.97 indicate non-uniform meteorological targets such as hail or melting snow; and values of Rhohv ≥ 0.97 are typical for uniform meteorological targets like rain and snow). Zdr depends on the hydrometeors’ shape and it is insignificant at particle sizes that are small compared to the radar wavelength, but can be substantial for hail and wet snow. Thus, it can be used to distinguish some types of hydrometeors. Φdp values can be used to identify shapes of hydrometeor, however it is a cumulative characteristic, which is difficult to interpret. From a meteorological viewpoint, it is more suitable to identify places where Φdp is changing. Therefore, the Kdp, which is the range derivative of the Φdp, is used to estimate the hydrometeor type, based on typical Kdp values for the hydrometeor types. Further details on the attributes of the measured quantities can be found, for example, in the book by Ryzhkov and Zrnić [
19] and in the references therein.
To process measured radar data, we used the default settings provided by the manufacturer. We applied attenuation correction of both Zh and Zdr and calculation of R using Zh or Kdp using relationships recommended by the manufacturer [
20]. Corrected Zh
c and Zdr
c are calculated along radar rays as:
where Δr is the radar horizontal resolution in km and i is the number of the radar ray bin.
Rain rate is calculated dependent on Kdp values. If Kdp > 0.3 [deg/km], and Zh > 30 [dBZ], then:
else:
For our purposes, we have developed the XCLASS algorithm to identify 7 types of hydrometeors: light rain, rain, wet snow, dry snow, ice, graupel, and hail. The XCLASS algorithm is based on the procedure proposed by [
21], which is modified to achieve more reliable results for data from our region. Our experience has shown that the original algorithm almost never identifies wet snow; instead, it overestimates the existence of graupel near the ground, where temperatures above 0 °C prevail.
Our XCLASS uses Zh, Zdr, Kdp, Rhohv, and estimates of air temperature T [°C] as input data. XCLASS is based on the fuzzy technique, which is a frequently used approach assigning weights to each hydrometeor class based on individual input information. The weight is proportional to the probability that the value would be measured in that class. In the following text, we use the word “weight” instead of “probability”, as it is frequently applied.
The total weight WT for hydrometeor class iclass is given by the relation:
where iclass denotes the type of hydrometeor and qs
iclass, FT
iclass, FZdr
iclass, FKdp
iclass, and FRho
iclass are weights for iclass determined dependent on input T, Zdr, Kdp, and Rhohv, respectively. In the following steps, light rain is not separated from rain and graupel from hail. The term qs
iclass is equal to 1, except for wet snow, where qs
iclass = 1.2. The reason for the increased weight for wet snow is explained later.
The function FT
iclass is a trapezoidal function of T and expresses that individual types of hydrometeors occur at certain temperatures T only. The trapezoidal function is described by points T1 < T2 < T3 < T4 (
Table A1 (
Appendix A)). FT
iclass (T) = 0 for both T ≤ T1 and T ≥ T4, and FT
iclass (T2) = FT
iclass (T3) = 1. FT
iclass (T) is obtained by linear interpolation of T between T1 and T4.
FZdr
iclass, FKdp
iclass and FRho
iclass are crucial to XCLASS. They give the weights of the occurrence of a given hydrometeor for combinations of Zh-Zdr, Zh-Kdp and Zh-Rhohv values.
Table A2,
Table A3 and
Table A4 (
Appendix A), which were taken from [
21], give values of
Z1 and
Z2 for measured Zh for a given hydrometeor. In contrast to [
21], there is a larger overlap of non-zero weights for individual hydrometeors in our tables (
Table A2,
Table A3 and
Table A4 (
Appendix A)). This is the case, for example, for ice and graupel. This overlap gives more natural transitions between hydrometeors.
Similar to [
21], we applied two half-Gaussian functions
f(
x) to define weights. Each of the two half-Gaussian functions is characterized by three parameters: position of the maximum m, which is equal to 1, and two half-widths,
Z1 (left) and
Z2 (right):
Selection of the hydrometeor depends on the WTiclass value and on the position of the classified point relative to the melting layer (ML), which we define as a layer with T between −1 and 1 °C. The hydrometeor with the highest WTiclass value is the result of the classification, and the following conditions must be met:
- (i).
Dry snow or ice cannot be identified within the ML;
- (ii).
Below the ML, only rain, graupel, and hail can be detected;
- (iii).
Wet snow cannot occur above the ML;
- (iv).
If there is dry snow above the ML, then there should be wet snow in the ML;
- (v).
If hail occurs below the ML with no connection to hail above the ML, then the classification is changed to rain. Specifically, this rule tests whether graupel/hail occurs at (i, j), where i is the horizontal coordinate and j the vertical coordinate (oriented upward). If it does, then graupel/hail must occur at at least one point (I − 1, j + 1), (i, j + 1), and (i + 1, j + 1) as well.
The first three conditions were also applied in [
21]. The impact of the fourth and the fifth conditions is illustrated in
Figure 2 together with the application of qs
iclass. The difference between the original classification [
21] and our XCLASS is mainly that the original one gives very little wet snow. In contrast, XCLASS reduces graupel and hail below ML and replaces them with rain. We cannot objectively validate these results; nevertheless, subjectively, we consider them more likely. In particular, the original method seems to generally overestimate the occurrence of near-ground graupel and hail below the ML, which is not confirmed by ground-based observations.
The functions FTiclass and the height of the ML depend on T, which is difficult to determine accurately, especially in convective storms. The height of the ML can be estimated by measurements of Doppler polarimetric radars. Despite that, we chose a simpler method, making use of the current temperature measurements available at the Milešovka observatory, and applied a standard vertical temperature gradient, i.e., the decrease of 6.5 °C per 1 km of altitude, on the measurements. The reason for using this simple method is that our research is focused on severe summer convective storms and we are mainly interested in cloud areas at very high altitudes; therefore, determining the exact height of the ML is not crucial for us. We are aware that the chosen procedure has evident shortcomings; however, the estimation of temperature is difficult and we have left more thorough estimation of T for future study.
In the final step, we distinguished rain based on Zh value (Zh < 10 dBZ), which we call light rain in this paper, from rain (Zh ≥ 10 dBZ) in places where the rain was identified, and we identify graupel (Zh < 50 dBZ) and hail (Zh ≥ 50 dBZ). The threshold values of Zh were selected subjectively. For instance, in the case of hail, we deliberately chose a lower value than the usual 55 dBZ, e.g., [
22], because our experience has shown that the FURUNO radar rarely gives values above 55 dBZ in vertical scanning, probably due to partial attenuation.
We calculated Zh in a constant altitude PPI level of 2 km (CAPPI 2km). The value for a given point of the CAPPI 2km was set to the value measured in the nearest PPI level and the resulting field was smoothed using a median filter applied to 3 by 3 radar pixels.
2.3. Satellite Data Meteosat Second Generation
In this study, we used satellite data measured by geostationary satellites Meteosat Second Generation (MSG-9, -10 and -11), which are operated by EUMETSAT (European Organisation for the Exploitation of Meteorological Satellites) and located at geostationary orbit (approximately 36,000 km above the Earth’s surface). Specifically, we used data from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) sensor on the board MSG, recognized as a useful tool for monitoring dynamical and microphysical properties of developing storms, cloud-top temperature, cloud-top cooling rate, etc.
In this paper, we focused on: (i) the brightness temperature (BT) in the infrared channel with a wavelength of 10.8 μm (IR10.8), (ii) the water vapor channel with a wavelength of 6.2 μm (WV6.2), and (iii) their difference (BTD = WV6.2 − IR10.8). The BT observed in IR10.8 measures thermal radiation emitted by the surface in the case of clear-sky conditions, or by clouds. This means that the BT in IR10.8 is directly related to cloud top temperatures if a cloud is present at a given place. WV6.2 is used for interpretation of water vapor content in high tropospheric layers.
Based on BTD values, it is possible to estimate whether there is an overshooting top (OT) in the monitored storm or not. OT is a domelike protrusion above a cumulonimbus anvil, which is caused by a strong updraft through its equilibrium level near tropopause. OT indicates the existence of a deep convective storm with an updraft of sufficient strength to penetrate through the tropopause into the lower stratosphere. Existence of OT is often associated with severe weather, such as heavy rainfall, damaging winds, large hail, lightning, and serious turbulence which endangers air safety. When an updraft is strong enough to penetrate through the tropopause, the penetrating warmer and moister air adds additional radiance in the WV6.2 to the thermal emission originating from the cold storm top, but remains transparent to the IR10.8. Therefore, BTD gains positive values, which indicate the presence of a deep convective storm, e.g., [
23,
24,
25]. There are more effective but also more demanding methods recognizing the existence of OTs [
24]. We used BTD in this paper as an indicator of severe convective storms.
The reason for using the above-described satellite data is that they supplement the radar information well for the case of convective events and they also provide data which can be used for validation of our radar data. Other satellite data do not provide fundamentally new information about the development of the studied storms, which is why we did not use them in this study. In addition, we have experience with using these data [
5,
26].
It should be mentioned that the horizontal resolution of the satellite data for the considered region is approximately 5 km in both south-north and west-east directions. Their relatively low resolution is their main disadvantage. We corrected the measured satellite data using parallax correction, which also can contribute to a decrease in the horizontal accuracy of the data.
However, despite the described shortcomings, the use of MSG data is important in our analysis because it enables validation of vertical radar scans. Specifically, it provides the opportunity to test whether the radar data correctly identify cloud tops.