Co-observation of strongly convective precipitation using VHF atmospheric radar and dual-polarized microwave radiometer during a typhoon passage

The Chung-Li very-high-frequency (VHF at 52 MHz) atmospheric radar and a dual-polarized microwave radiometer were operated collaboratively to investigate strongly convective precipitation while the typhoon Trami just passed through the 10 Taiwan in Aug, 2013. First, respective Doppler velocities of clear-air and precipitation echoes were identified automatically by the VHF radar as clearly as possible. Two approaches were designed for this purpose: contour-based and peak-finding processes. The two approaches initially determined some major spectral centers or peaks, which were usually redundant, and then proper sifting and clustering were performed for the redundant spectral centers or peaks to yield several mean locations of Doppler velocities for profiling. The outcomes of the two approaches were consistent in general. With the estimated Doppler 15 velocities, a tracing process was developed for Doppler profiling, in which Doppler velocity shear was one of the significant criteria in the tracing process. The radar echoes collected by the VHF radar during rainy and strongly convective atmosphere have been investigated to validate the two approaches and the tracing process. About 98% of the tracings could identify the Doppler profiles of clear air and precipitation, even the atmosphere was disturbed severely. The radar spectral parameters, Doppler profiles, and the information from a dual-polarized microwave radiometer as well as the simulation of weather model, 20 were examined jointly. It signified that strong updraft and turbulent atmosphere could bring the liquid water to the height above the melting layer, and then the Bergeron effect and coalescence process on formation of ice crystal and graupel above the height of the melting layer occurred accordingly.

: Flow chart of contour-based and peak-finding approaches to determination of Doppler velocity and spectral width.

3) Determine the contour centers
With the coordinates of contour lines, C, the contour centers were determined by applying the locating process proposed by Chen et al. (2008). The determined contour centers will be nearly aligned in the dimension of the original 1-D Doppler spectrum because the modelled Doppler spectrum in the orthogonal dimension processed in step 1 hardly biases the locations https://doi.org/10.5194/amt-2021-267 Preprint. Discussion started: 5 October 2021 c Author(s) 2021. CC BY 4.0 License. of the contour centers. Therefore, in the following statement the location of contour center denotes the coordinate of the contour 105 center in the original Doppler spectral dimension.
Five contour centers that are most significant were selected at most through the process mentioned above, but the ones at the lowest contour level were discarded. Then the identified contour centers within a velocity interval (vinterval) were grouped to calculate a weighted mean center, where vinterval was assigned to 1.4 m/s and the respective contour levels of the contour centers were used as the weights in calculation of the weighted mean center (mean center hereafter). 110 At most 5 mean centers can be obtained in each range gate if the separation of the adjacent contour centers are larger than vinterval and the ratio of the spectral power, averaged from the spectral lines centered on the mean center, to a referenced noise level is higher than a specified threshold value (Rp), say, 0.75. Giving a higher Rp will reject more minor mean centers. In this study, the referenced noise level was estimated from the mean of the lowest one fourths spectral power lines, and the averaged spectral power of each mean center was calculated from 6, 10, and 14 spectral lines centered on the mean center, respectively, 115 for the Doppler spectra of 32, 64, and 128-point FFTs.

4) Estimate the Doppler velocities and spectral widths of respective targets with Gaussian fitting
Once the mean centers were obtained, for each mean center we fitted a Gaussian function (Gaussian fitting thereafter) to the original 1-D Doppler spectrum with 5, 7, 9, and 13 spectral lines centered on the mean center, respectively. Among the four Gaussian fittings, the Gaussian function with minimum standard deviation and the separation between the Gaussian-fitted peak 120 and the mean center less than a half of the velocity interval (vinterval=1.4 m/s) was adopted. Finally, the Gaussian-fitted peak denotes the Doppler velocity of the mean center, and the standard deviation is the corresponding spectral width. At most 5 pairs of Doppler velocities and spectral widths can be retained for later investigation if 5 mean centers are identified in the third step. In addition, the normalized amplitudes of the Gaussian fitting were retained for future statistical analysis and Doppler profiling. 125

Peak-finding approach
Peak-finding is commonly employed in several of the aforementioned algorithms for estimate of Doppler velocity and profiling.
We have developed our own process as follows:

1) Find the spectral peaks
This step is similar to the third step of the contour-based approach, except that the function "findpeaks" built in MATLAB 130 was used to find 5 most significant spectral peaks that were separated by at least a velocity interval (vinterval). Note that vinterval was given as that in the contour-based approach. In addition, the spectral peaks at the lowest contour level, as given in the contour-based approach, were discarded.
For each spectral peak, a mean location was further calculated by the moment method with several spectral lines centered on the peak. The numbers of spectral lines used in the calculation of mean location were 6, 10, and 14 for the Doppler spectra 135 estimated from 32, 64, and 128-point FFT, respectively. Such estimate of mean center for each spectral peak is to provide a https://doi.org/10.5194/amt-2021-267 Preprint. Discussion started: 5 October 2021 c Author(s) 2021. CC BY 4.0 License. more representative mean location for the Gaussian fitting executed in the following step. Finally, the calculated mean location was discarded when the ratio of the averaged spectral power to the referenced noise level was smaller than the threshold value, Rp. Here the averaged spectral power, referenced noise level, and Rp were calculated as those in the third step of the contourbased approach. 140

2) Estimate the Doppler velocities and spectral widths of respective targets with Gaussian fitting
The Gaussian fitting performed in the fourth step of the contour-based approach was executed for the respective mean locations of spectral peaks to provide several pairs of mean Doppler velocity and spectral width. After this fitting, we calculated another pair of mean and standard deviation additionally via moment method with the spectral lines centered on the Gaussianfitted mean Doppler velocity; the purpose of this extra calculation is to make a further comparison with the contour-based and 145 peak-finding approaches. The spectral lines included in the extra calculation were within 2 times of the Gaussian-fitted standard deviation on both sides of the Gaussian-fitted mean Doppler velocity. As a result, at most 5 pairs of mean and standard deviation can be obtained additionally if the Gaussian fitting produces 5 results initially.
It should be clarified here that redundant Doppler velocities and spectral widths, obtained from the two locating approaches and the additional moment-method calculation, are to acquire the spectral parameters as complete as possible; that is beneficial 150 for Doppler profiling later.

Validation of spectral parameters
To validate the approaches described in Sect. 2, the Chung-Li VHF radar data collected on 21 Aug, 2013, were examined.
Heavy convective precipitation was prevailing during the observational period while the typhoon Trami just passed through the Taiwan area. The radar parameters were set as follows: vertically transmitted radar beam with multiple carrier-frequency 155 mode (i.e., the carrier frequencies of 51.5, 51.75, 52, 52.25, and 52.5 MHz were transmitted alternately), inter-pulse period 200 μs, number of coherent integrations 128, pulse width 1 μs, sampling interval 1 μs (corresponding to a range resolution of 150 m), lowest sampling height 1.5 km, number of range gates 80, and without pulse coding. As a result, the sampling time of radar returns for each carrier frequency was 0.128 s (=200 μs128 integration times5 frequencies). For the goal of this study, however, only the echoes from the carrier frequency of 52 MHz were analyzed. 160 Fig. 3a shows the height-time distribution of echo intensity and Fig. 3b presents time series of the disdrometer-observed rainfall rate. As indicated, the primary radar returns in region 1 were characterized by a stratified and shallow distribution with height, which were associated with weak precipitation. By contrast, those in region 2 accompanied with a very intense rainfall and extended up to 5 km more. Note that the radar data shown in Fig. 3a have been examined by Chen et al. (2016) for evaluation of range imaging technique (a multiple-frequency analysis). These radar data were used again in our study for 165 demonstrating the effectiveness of the Doppler profiling process. Furthermore, an investigation of raindrop shape and the effect https://doi.org/10.5194/amt-2021-267 Preprint. Discussion started: 5 October 2021 c Author(s) 2021. CC BY 4.0 License. of convective atmosphere on the composition of hydrometeor particles was also made with the information of microwave radiometer and the present radar data.  was done by copying the one fourths spectra on the leftmost side to the rightmost side, making the precipitation spectra more complete when the falling velocity of the precipitation was larger than the Nyquist Doppler velocity in observation. This is 180 beneficial for locating the spectral peaks of precipitation at larger Doppler velocities. The copying, nevertheless, is not always necessary if the sampling time is short enough so that the Nyquist Doppler velocity is much larger than the Doppler velocity of precipitation.
Also shown in Fig. 4a and b are the mean Doppler velocities and the spectral widths of clear-air and precipitation echoes, which were estimated from the contour-based (marked in red) and peak-finding (marked in black) approaches, respectively. 185 In Fig. 4a, only the clear-air echoes were visible, while the clear-air (distributed around 0 m/s) and precipitation (characterized by salient positive Doppler velocities) echoes were both present in Fig. 4b. As seen, the spectral parameters estimated by the https://doi.org/10.5194/amt-2021-267 Preprint. Discussion started: 5 October 2021 c Author(s) 2021. CC BY 4.0 License. two approaches were generally in agreement, although there were still some minor discrepancies in them. For a clearer inspection, Fig. 4c displays the spectral power in curve for several sampling gates for comparison. As seen in the upper panel of Fig. 4c, where shows the motion of clear air, the peak-finding approach determined four Doppler velocities at the first gate 190 (black dots), but the contour-based approach provided only one (red dot). Apparent discrepancies in the calculated Doppler velocities can also be found at the heights around 8.4 km, 8.7 km, and 9.2 km in the upper panel, and at around 4.8 km in the lower panel of Fig. 4c. In view of this, a further classification and grouping of these Doppler velocities are necessary for tracing the Doppler profiles automatically; this will be addressed in the next section. the signal to noise ratio (SNR; indicated by thick black curve) was much lower than 10 dB (denoted by the dotted vertical line).
Here the SNR was calculated from the 64-point data length and self-normalized for presentation. By contrast, the distribution 210 of VC shown in Fig. 5b highly varied with height and SWC was relatively broad.
A noticeable feature in the distributions of SWC in Fig. 5b was a bias of distribution peaks to smaller values as the FFT point increased. A detailed examination showed that in the environment of intense precipitation, there was a tendency for the mean of spectral width to become lager (smaller) if a FFT with less (more) data points was performed to generate the Doppler spectra, as shown in the left panel of Fig. 6b. As seen, the mean of SWC was the largest for 32-point FFT (black thin curve), and the 215 smallest for 128-point FFT (blue curve) throughout the range. Such feature can be attributed to the smoothing effects on the spectral lines caused by the windowing function in time domain used to perform FFT and the practice of the incoherent integration in spectral domain. By contrast, the mean of SWC shown in Fig. 6a, where the clear-air echoes dominated, were nearly the same. This is expected because in region 1 the atmosphere with little precipitation aloft was relatively stable, resulting in narrow spectral width. On the other hand, irrespective of region 1 or region 2, the standard deviations of SWC were 220 essentially independent of the point of FFT used in the calculation of Doppler spectrum, as shown in the right panels of Fig.   6a and b, suggesting the robustness of the process in the determination of the spectral width.
A further comparison between the mean Doppler velocities and spectral widths obtained from the contour-based and peakfinding approaches are shown in Fig. 7, in which VC and SWC are the Doppler velocity and spectral width of the contour-based approach, Vpg and SWpg are the corresponding results of Gaussian fitting in the peak-finding approach, and Vpm and SWpm 225 denote the additional estimates using the moment method in the peak-finding approach. The histogram in each panel shows the distribution of the difference between the two spectral parameters compared, which is shown between -2 and 2 m/s. As shown in Fig. 7a for clear air echoes, no matter what the methods were used, the estimated spectral parameters were in good agreement, although the slope (~0.96) of the regression line for the values of SWC and SWpm (at the second row and middle column) was slightly lower than the others (close to 1). Note that, for the moment method executed in the peak-finding 230 approach, the spectral lines within 2 times of the standard deviation, obtained from the Gaussian-fitting process and centered on the mean Doppler velocity, were used in the calculation, which included more spectral lines than those of the Gaussian fitting method. As a result, the standard deviations (i.e., spectral widths) estimated from the moment method could be larger than those from the Gaussian-fitting method. Such situation could be worse under the conditions of strongly turbulent atmosphere and precipitation. As shown in Fig. 7b, the slopes of the regression lines for the values of SWC and SWpg, SWC 235 and SWpm, and SWpg and SWpm were smaller than those in Fig. 7a, which reduced to about 0.98, 0.81, and 0.95, respectively.
Moreover, the scatter dots in Fig. 7b spread out more than in Fig. 7a.    To estimate respective spectral widths of clear air and precipitation, we categorized the spectral widths into two different 255 groups in terms of mean Doppler velocity: between -3 and 3 m/s, and larger than 3 m/s. This is because that most of the Doppler velocities larger than 3 m/s were found to result from precipitation echoes. Fig. 8 shows the results calculated from the spectral widths (SWC) of the contour-based approach for the radar echoes in the region 2 of Fig. 3a. In Fig. 8a, all the outcomes of SWC were used to estimate the mean (μSWc) and standard deviation (σSWc) of SWC. It is apparent that the mean of spectral widths of precipitation echoes (red solid curve) were slightly larger than those of clear-air echoes (blue solid curve) below ~6 km; notice 260 that strong precipitation occurred mainly below ~6 km in the region 2 of Fig. 3a. Even if the outcomes having the Gaussian-https://doi.org/10.5194/amt-2021-267 Preprint. Discussion started: 5 October 2021 c Author(s) 2021. CC BY 4.0 License. fitted amplitudes larger than 0.5 or 0.8 (a normalized value) were adopted only, the mean of spectral widths of precipitation echoes were still larger than those of clear air echoes below ~6 km, as demonstrated in Fig. 8b-c. Moreover, the deviations (σSWc) of spectral widths were small for clear-air and precipitation echoes both. In view of this, the precipitation echoes indeed distributed over a broader velocity interval than the clear-air echoes. This result is expected because the precipitation was 265 accelerated from rest to a terminal velocity of about 10 m/s in the vertical direction, and the falling velocities of precipitation spread in a wider velocity interval due to different raindrop sizes.

Doppler profile of clear air 280
Starting from the lowest sampling gate, the two fitted Doppler velocities of greatest significance within ±2 m/s, that were determined respectively by the two proposed approaches, were averaged as an initial mean Doppler velocity (v1) of clear-air motion. The respective amplitudes of the two fitted Doppler velocities were employed as the weights in the calculation of v1.
In case of no fitted Doppler velocities available in the first sampling gate, the second sampling gate was examined. This process could continue through several sampling gates until a value of v1 was determined. In case no v1 was obtained from the lowest 285 15 sampling gates (the number is user defined), the value of 0 m/s was assigned to v1. Mostly, the first or several preceding sampling gates can provide the value of v1.
Determination of mean Doppler velocity for the second sampling gate was based on v1, i.e., by taking the fitted Doppler velocities within a velocity window of -vdev and vdev that centered on v1, the weighted mean Doppler velocity was calculated as v2. In the present calculation, vdev=3 m/s. v2 became the reference velocity of the next trace in the third sampling gate. In a 290 similar fashion, a Doppler velocity profile for clear air can be obtained. However, it happened occasionally that there was no reference velocity for the present sampling gate when the calculation failed in the preceding sampling gate due to the diagnosis of Doppler velocity shear (described below). In case of this, the tracing process continued searching the existed mean Doppler velocity in the lower sampling gates. If no value was available in the 10 sampling gates lower than the present sampling gate, the searching process stopped. 295 In the tracing, the Doppler velocity shear is a crucial criterion. Assigning vn, vn-1, vn-2, and vn-3 to the mean Doppler velocities of the present and the lower three sampling gates, and vshear to the threshold of Doppler velocity shear, vn will be discarded if |vn-vn-1|>vshear or |vn-vn-2|>1.5vshear or |vn-vn-3|>1.5vshear. For clear air, the value of 2 m/s was given for vshear in this study. Use of these Doppler shear criteria in tracing can increase the reliability of the traced Doppler profile although intermittent breaks in the profile may happen occasionally. It should be stated here that all values assigned for tracing are changeable, considering 300 the atmospheric condition is variable. Once the tracing of clear-air motion finished, the fitted Doppler velocities within -vdev/2 and vdev/2, centering on the respective mean Doppler velocities of the sampling gates, were discarded in the tracing of precipitation velocity described below.

Doppler profile of precipitation
It is assumed that the Doppler speed of precipitation near the ground is apparently larger than that of clear-air motion in 305 observation of using a vertical radar beam. Therefore, the initial mean Doppler velocity (v1) of precipitation was calculated from the fitted Doppler velocities larger than 3 m/s. In case of no v1 obtained in the lowest 15 sampling gates, 5 m/s was assigned to v1. The process of tracing was the same as that for clear air. However, considering a wider range of Doppler velocities during the falling of precipitation, 3 m/s was given as the threshold of Doppler velocity shear (vshear), which was larger than 2 m/s used for clear air. Notice that the fitted Doppler velocities used in tracing the Doppler profile of clear air had 310 been discarded here. Totally, 109 Doppler spectra were produced from the radar data shown in Fig. 3. Fig. 9 exhibits several typical tracing results, where the black and red curves are the traced Doppler profiles of clear air and precipitation, respectively. Original Doppler spectra are displayed in the background. Fig. 9a and b are two examples of pure air echoes, showing the tracing was good, although the airplane echoes might interfere the tracing slightly, e.g., around the height of 7 km in Fig. 9b. In Fig. 9c, 315 one more profile (red curve) was traced automatically for precipitation, showing the precipitation echoes occurred aloft and was much weaker than clear-air echoes. More successful tracings for both clear air and precipitation are exhibited in the panels (d)-(k) of Fig. 9, in which the strong interferences just above 2 km in the panel (j) have been treated properly in the tracing so that they were discarded in the profiling.
Two cases that were not traced completely are shown in the panels (l) and (m) of Fig. 9. The two Doppler spectra were 320 similar to that in the previous time period (the panel (k)), but the Doppler profile of clear air broke around the height of 5 km and continued again since ~6 km. In the height interval of 5 and 6 km, the clear-air echoes were traced as precipitation echoes, and the weak precipitation echoes between 5 and 7 km, which were traced successfully in panel (k), were not identified. This precipitation around the altitude of 5 km. In view of this, the success rate of tracing was really high. The 109 tracing results can be found via the supplement link. It is expected that the spectral parameters and tracing process validated in this study can be helpful for more scientific studies in the future. We provide a case study in the following section.

Observation campaign of convective precipitation aloft
As shown in Fig. 3, dynamic echoes were present in a height range between about 7 and 10 km in region 2. From the Doppler 335 spectral tracing results shown in Fig. 9d-o, these echoes were characterized by very broad spectral width combined with intense upward air motion (sometimes larger than 5 m/s), and were identified as clear air echoes rather than precipitation echoes.
Nevertheless, we observed that these echoes were followed by the presence of precipitation echoes from the supercooled droplets between about 5.5 and 7 km (the temperature of 0C occurred at the height of about 5.7 km according to the radiosonde data nearby; shown later). This feature seems to suggest that there existed hydrometeors in the height range of 7 and 10 km 340 that were responsible for the formations of the supercooled droplets below 7 km. To verify this, the measurement of a collaborative dual-polarization microwave radiometer was examined to study the properties of the targets responsible for the echoes observed by the VHF radar. study. It is noteworthy that according to Hou and Tsai (2019), the accuracy of measurement decreased gradually when the rain rate was higher than 20 mm/h, and became doubtful when the rain rate reached the value of about 60 mm/h. More descriptions of the radiometer and measurement characteristics can be found in the study of Hou and Tsai (2019). Several key parameters 360 of the radiometer are listed below. The receiving bandwidth is 400 MHz; range of the brightness temperature measurement is 0-350 K with a precision of 0.05 K; integration time (or data resolution) is 1 s; half power beam width (HPBW) of the antenna beam is 10.25 with a sidelobe level less than -30dBc; precision of PD measurement is 0.25 K. Therefore, the observation zone of the radiometer at the distance of 13.52 km lay about 5 km to the east of the VHF radar site, and approximately in the height range between 6.33 and 9.57 km above the sea level of the VHF radar station. As a result, there was no overlapped field of view between the obliquely pointed radiometer beam and the vertically pointed VHF radar beam. Nevertheless, by taking account of finite beam widths, the separation of observational volumes between the radiometer and the VHF radar was less than 5 km in the height range of 6.33 and 9.57 km. A separation of 5 km was not a considerable 370 distance for the convective structure examined later that covered a horizontal region of over ten kilometers. Therefore, it was justified to compare the concurrent measurements of convective precipitation aloft made by the Chung-Li VHF radar and the radiometer.   Fig. 11a provides five consecutive maps of meteorological radar reflectivity, which were trimmed from those released by the Central Meteorological Bureau of Taiwan. As shown, a strong front-like convective precipitation structure (or rainband), tilted in northwest-southeast direction, was moving from north to south at a phase speed of about 20 m/s. The rainband approached the VHF radar (black circle) and the radiometer (white circle) around 05:15 and 05:22 UTC, respectively, and left the VHF radar after 05:45 UTC approximately. The rainband was a strong convective structure, according to the large and prevailing upward Doppler velocities found in the VHF radar spectra shown in Fig. 9d-n. In the meantime, the PD shown in Fig. 11b increased from about -1.5 to around zero between 05:18 and 05:42 UTC and decreased quickly after 05:42 UTC, although the PD dropped intermittently around 05:31 UTC. Such variation in PD implies that the shape of liquid water particles varied mostly from horizontally oblate shape to almost round shape during the passage of the convective structure, which can thus be speculated to be caused by an uplift of liquid water particles in the convective structure, i.e., the liquid water particles 390 got more round during uplift.
In addition to PD, the dual-polarization microwave can also measure the total liquid water path (LWP), and distinguish liquid water of cloud (LWC) and liquid water of rain (LWR), as shown in Fig. 11c. LWP is a measure of the total mass of liquid water (i.e., liquid water content) in an atmospheric column above a unit surface area, and LWC and LWR denote respective contributions from cloud droplets and liquid raindrops, in which solid water (snow, ice) is not included. It should 395 be reminded that water vapour also contributes to the value of LWP. However, it is shown in Fig. 11c that LWP, LWC, and LWR varied in accordance with each other and LWP was approximately the sum of LWC and LWR. In view of this, the contribution of water vapour to LWP was minor in the present event. Since LWP, LWC, and LWR decreased apparently between 5.4 and 5.6 h UTC, it implies a decrease of liquid water content in the observation volume of the radiometer in this time period. The mechanism of causing a decrease of liquid water content is worthy of discussion on the basis of radar and 400 radiometer observations during the passage of convective structure. We supposed that an interaction of ice crystals, water vapour, and supercooled water particles could be the candidate of the cause, as discussed below.
According to Pan-Chiao radiosonde measurement at 06:00 UTC (Fig. 12), where is located at about 20 km northeast of VHF radar, the temperatures of 0C, -10C and -20C occurred at heights of around 5.7 km, 7.6 km and 9.2 km, respectively.
Thus, it signifies that in the absence of the vertical air motion, the compositions of the hydrometeors in height ranges 5.7-7.6 405 km and 7.6-9.2 km were essentially dominated by supercooled water particles and the mixture of ice crystals and supercooled water particles in the cloud, respectively. However, in the presence of intense updraft, the updraft-carried water vapour and liquid water particles can significantly determine the growth of ice crystals and/or supercooled water particles above the 0C isotherm level. The so-called Bergeron effect will proceed to transfer water vapour from supercooled water particles to ice crystals due to the difference in relative humidity around supercooled water particles and ice crystals (Wang, 1997), leading 410 to decrease of LWP. Because of the expense of supercooled water particles and/or water vapour during the Bergeron process, LWC and LWR decrease as well, as those seen around 05:30 UTC in Fig. 11c. As the ice crystals become larger, their weight increases and finally cause them to fall. During the fall of ice crystals, a second process, coalescence or accretion, continue increasing the weight of the falling ice crystals. Coalescence during the fall of ice crystals is a process that the supercooled water particles collide and merge together, and then freeze on the surface of ice crystals (Khain et al., 2001;Korolev, 2007), 415 resulting in the formation of graupel. The process of coalescence can also decrease LWP, LWC, or LWR. We especially note that, in association with the updrafts, considerably intense turbulences that were characterized by the presence of very broad Doppler spectral widths (as large as 10 m/s or more) were observed in a height interval between about 7 and 10 km. Such intense turbulent condition can intensify the Bergeron effect and coalescence process.

430
Apparent falling of ice crystals and graupels started from the height of about 7 km, as shown in Fig. 9d-k. When the ice crystals and graupels felt below 0C isotherm level (at around 5.7 km on the basis of the measurement of Pan-Chiao radiosonde at 06:00 UTC), they started to melt and became wet. It is expected that their backscatter will be about 5-15 dB larger than https://doi.org/10.5194/amt-2021-267 Preprint. Discussion started: 5 October 2021 c Author(s) 2021. CC BY 4.0 License. those above the 0C isotherm level (Liu, et al., 2000;Straka et al., 2000). As shown in Fig. 9e-j, a drastic increase in the radar backscatter of the falling ice crystals and graupels was about 5-10 dB in the height interval of 4-5 km, where was approximately 435 below the 0C isotherm level. This was coincident with the theoretical and previous experimental results (Liu, et al., 2000;Straka et al., 2000).
As ice crystals and graupels fall, their downward speeds are expected to increase. This feature was indeed observed clearly in Fig. 9e-j, where the downward speeds increased to 10 m/s at around the height of 4 km. It was also observed in Fig. 9 that variations in the profiles of atmosphere and precipitation were coherent when the convective condition prevailed in the time 440 interval of Fig. 9d-m. Therefore, it is evident that a convective atmosphere played a key role in the precipitation motion.
To verify further the depletion of liquid water content above the 0C isotherm level, we employed the Weather Research and Forecasting model (WRF version 3) to simulate the distribution of liquid water for the precipitation environment investigated in this study, as shown in Fig. 13. Fig. 13a and b present, respectively, the simulated height-time densities of cloud water and rain water at the grid point nearest to the VHF radar location, in which the time and vertical-range resolutions for the simulation 445 were 0.5 h and 1 km, respectively. The initial field for the simulation was obtained from the National Center for Environmental Prediction (NCEP) reanalysis data at 00:00 UTC on 21 August 2013, and the horizontal resolutions were set as 6 and 2 km sequentially in the nested domain of calculation. As indicated by the red frames, both densities of cloud water (left panel) and rain water (right panel) drastically reduced above 6 km around 05:30 UTC, which was consistent with the decrease of LWC and LWR shown in Fig.11c. This simulated result demonstrated further that the co-observational results of VHF radar and 450 microwave radiometer were reasonable, that is, the composition of hydrometeor in the upper atmosphere (above the melting layer) changed due to strong updraft, Bergeron effect, and coalescence of supercooled water particles and ice crystals. One of the motivations of this study is to identify the respective Doppler velocities of clear air and precipitation automatically when both types of signals coexist in the radar echoes. Two approaches have been developed for the purpose of the study, i.e., contour-based and peak-finding processes. To determine the major spectral centers or peaks for calculation of mean Doppler 460 velocity, several parameters such as SNR, contour level, peak amplitude, separation of peaks, and so on, have been considered.
The two approaches are complementary because not any of the two approaches can determine the major spectral peaks/centers completely, at least for the present data set. Comparisons between the spectral parameters (mean Doppler velocity and spectral width) of clear-air and precipitation echoes that appear simultaneously have been made. From the view of statistical distribution, the results of the two approaches were in good agreement. Moreover, it showed that different numbers of data points in FFT 465 produced slightly different spectra widths, i.e., the smaller the FFT number was, the broader the spectral width could be. This feature was more evident in rainy condition and could be attributed to the smoothing effect of using fewer FFT number and/or more incoherent times of spectra.
With the reliable and complementary outcomes of the two approaches, tracing processes have been developed to determine the respective Doppler profiles of clear-air and precipitation, in which the Doppler shear is one of the significant criteria in 470 tracing. In spite of occasional failure in tracing when the Doppler shear was too large, about 98% of the tracings for the present radar data were acceptable. Because the parameters and criteria employed for determination of the spectral peaks/centers and Doppler profiles are adjustable, it is possible to improve the tracings by adjusting these parameters and criteria properly for various atmospheric conditions. An application study using the traced Doppler profiles, radar echo powers, and spectral parameters above the melting layer 475 was also made for rainy and strongly uplifted condition during the period of a typhoon passage. This was achieved via the collaborative observation using a dual-polarization microwave radiometer. Several information such as brightness temperatures of vertical and horizontal polarizations of water, total liquid water path, liquid water of cloud, and liquid water of rain yielded by the radiometer have signified the Bergeron effect, and coalescence/accretion process on formation of ice crystal and graupel above the height of the melting layer under a strongly convective and turbulent atmosphere. The 480 circumstance of strongly convective and turbulent atmosphere was verified from the Doppler profile and spectral width of VHF radar echoes. In addition, simulated calculations of range-time distributions of cloud water and rain water, obtained from the Weather Research and Forecasting (WRF) model, also supported the scenario deduced from the radar and radiometer observations. More efforts in the future are to test the Doppler profiling process for the radar data under different atmospheric conditions and improve the criteria employed in the profiling. Moreover, collaborative observations of VHF atmospheric radar 485 and radiometer will be beneficial for further studies of dynamic atmosphere and precipitation.