Assessing the Effects of Microphysical Scheme on Convective and Stratiform Characteristics in a Mei-Yu Rainfall Combining WRF Simulation and Field Campaign Observations

Microphysics parameterization becomes increasingly important as the model grid spacing increases toward convection-resolving scales. Using observations from a field campaign for Mei-Yu rainfall in China, four bulk cloud microphysics schemes in the Weather Research and Forecasting (WRF) model were evaluated with respect to their ability to simulate precipitation, structure, and cloudmicrophysical properties over convective and stratiform regimes.*ese are the*ompson (THOM),Morrison graupel/ hail (MOR_G/H), Stony Brook University (SBU_YLIN), and WRF double-moment six-class microphysics graupel/hail (WDM6_G/H). All schemes were able to predict the rain band but underestimated the total precipitation by 23%–35%. *is is mainly attributed to the underestimation of stratiform precipitation and overestimation of convective rain. For the vertical distribution of radar reflectivity, many problems remain, such as lower reflectivity values aloft in both convective and stratiform regions and higher reflectivity values at middle level. Each bulk scheme has its advantages and shortcomings for different cloud regimes. Overall, the discrepancies between model output and observations mostly exist in the midlevel to upper level, which results from the inability of the model to accurately represent the particle size distribution, ice processes, and storm dynamics. Further observations from major field campaigns and more detailed evaluation are still necessary.


Introduction
e representation of microphysics is of substantial importance in numerical weather prediction (NWP) models because it controls the distributions of latent heating and cooling and condensate loading that influences the dynamics [1,2]. With the increases of computing power and NWP technology, more accurate and detailed descriptions of cloud microphysical processes are proposed and the trend has been toward prognostically calculating hydrometeor types and sizes explicitly [3][4][5][6][7][8]. However, these microphysical schemes have their own advantages in different types of cloud and weather system. Extensive and detailed validations for existing schemes are therefore needed to constrain and reduce uncertainties for different cloud types and regions.
Obvious discrepancies between models and observations are found by previous studies, such as smaller stratiform areal coverage, weaker precipitation [9,10], and heavier precipitation from convective regions [11] in simulated storms. A recent trend in evaluating NWP models is to use remote sensing datasets such as satellites [12][13][14], radars [15,16], and cloud-aerosol lidar [17,18]. ese active sensors makes it possible to evaluate the three-dimensional structure of clouds and precipitation [19], and in particular, the simulated and observational reflectivity allows one to more easily see detailed meso-and storm-scale structures forecasted by finer-resolution NWP models [20,21].
Parameterizing microphysics remains challenging because of fundamental uncertainty in the underlying microphysical processes, especially for ice [15]. erefore, more hydrometeor variables need to be evaluated in addition to the macrophysical quantities just like accumulated rainfall and cloud coverage which are evaluated in previous research. Johnson et al. [22] used the observed polarimetric radar variables to evaluate the microphysics schemes in the ensemble forecasts, and they found that a hail-like rimed-ice category may be necessary to reproduce the observed hail signature. Remillard et al. [23] used cloud radar doppler spectra to evaluate stratocumulus drizzle size distributions in large-eddy simulations with size-resolved microphysics. Detailed observations from field campaigns which can reflect macro-and microphysical characteristics are also used. Naeger et al. [24] evaluated several bulk microphysical parameterizations in combination with observations from a cold season precipitation experiment of an intense warm frontal band. Taufour et al. [25] used the airborne in situ measurements and dual-polarization radar variables to evaluate a two-moment scheme. ey found that improving the hydrometeor representation, focusing especially on the description of the particle size distributions for different water species, is a good way to improve microphysical scheme.
Utilizing multisource datasets from radar and field campaign to evaluate the cloud microphysical schemes has only recently begun in China. With the increases in local severe weather activities, the need for more accurate weather information has become increasingly crucial especially for Yangtze River basin where extreme precipitation occurs frequently during Mei-Yu season. Mei-Yu is one of the most significant events for the hydrological cycle in the East Asia monsoon region [26], which produces a large amount of flood disasters and substantial economic losses in the Yangtze-Huai River Valleys [27]. During Mei-Yu season, a quasi-stationary front, known as the Mei-Yu front, is formed and extends from eastern China to southern Japan [28]. At any given time, Mei-Yu front is associated with the presence of an elongated cloud band, which is usually a mixture of deep convective clouds and stratiform clouds [10]. Different microphysical and thermodynamic features are observed in two cloud regimes. In convective regions, the growth of ice is dominated by riming, whereas in stratiform regions, deposition and aggregation are the primary mechanisms [29]. e resultant latent heating profiles are quite different for the two regions [30,31].
To provide a much needed data source concerning the formation-dissipation processes of mesoscale convective system (MCS) during the Mei-Yu season, a field campaign entitled the "Investigative Monsoon Frontal Rainfall Experiment (IMFRE) 2018" was conducted from 10 June 2018 to 10 July 2018 in the middle reaches of the Yangtze River (MRYR). e purpose of this study is to investigate the impacts of bulk cloud microphysics schemes on the simulation of different cloud regimes (stratiform and convective clouds) along the Mei-Yu front using observations from the IMFRE 2018. We will examine the differences in the evolution, horizontal, and vertical structure of MCS and determine the strengths and weaknesses of different microphysics schemes for Mei-Yu rainfall. Furthermore, the cloud microphysical properties from C-band dual-polarization doppler weather radar (C-POL) are also evaluated. e remainder of this paper is organized as follows. Section 2.2 describes the datasets used in this study. e methods and description of the case study are presented in Section 2.
e results and discussions are presented in Section 3. Finally, the main findings of this study are summarized in Section 4.

Precipitation Data and Reanalysis.
A precipitation product [32], based on the rain gauge precipitation from automatic weather stations in China and the Climate Prediction Center morphing technique (CMORPH) precipitation [33], was used in this study.
is product has a 0.1°× 0.1°spatial resolution with hourly temporal resolution. What is more, large-scale air flow and thermodynamic conditions of Mei-Yu front are analyzed with the hourly ERA-5 reanalysis [34] in uniform grid of 0.25°× 0.25°, which is the latest and fifth generation of European reanalysis produced by the European Center for Medium-Range Weather Forecasts (ECMWF). e ERA-5 reanalysis was also used as model initial and lateral boundary conditions.

Datasets from Field
Campaign. Various datasets have been collected to study the mechanism of MCS during Mei-Yu season from the field campaign of IMFRE 2018. reedimensional structures and microphysical properties of MCS measured by IMFRE provide critical observations to examine the impacts of different bulk schemes on cloud microphysics and precipitation. e red dot (Figure 1(a)) represents the Xianning site of the field campaign, where convective systems can easily start and develop because of the impact of mesoscale topography, primarily influenced by the Mufu Mountain to the southeast of Xianning. e primary instruments used in this study include the China New Generation Weather Radar (CINRAD) and C-POL. In the studying area, there are 15 S-band operational Doppler radars (the black circles in Figure 1(a)) operating at 10 cm wavelength and providing 9 unique elevation scans every 6 minutes). e reflectivity datasets underwent critical quality control [35,36] and were gridded to 3 km horizontally and 0.5 km vertically in an attempt to make a comparison with model simulations. C-POL is located in Xianning site (the red dot in Figure 1(a)); the specifications of measurements from C-POL are shown in Table 1. e radar data are preprocessed by a group of algorithms [37] so as to provide hydrometeor categories. e algorithms include corrections of system biases of reflectivity and differential reflectivity, suppression of ground clutter and anomalous echoes, and filtering of differential phase shift.  (Figure 1(b)). ere were 50 layers from surface to 50-hPa. All simulations started at 0600 UTC 29 June with 30-hour run. And the model outputs were saved at a 12-min interval. In this paper, four bulk microphysical schemes were tested in WRF to investigate their performance in simulating the structure and evolution of MCSs and cloud microphysics, with a focus on the convective-stratiform precipitation processes. e schemes are ompson (THOM) scheme [6], the Morrison graupel/hail (MOR_G/H) scheme [7], the Stony Brook University microphysics scheme (SBU-YLIN) [38], and the WRF double-moment six-class microphysics (WDM6) graupel/hail (WDM6_G/H) scheme [8]. Each scheme is described briefly in Table 2. Other physics parameterization schemes include the Yonsei University (YSU) planetary boundary layer (PBL) scheme [39], the Rapid Radiative Transfer Model for General Circulation Models (RRTMG) scheme for longwave radiation calculations [40], the Goddard shortwave scheme for shortwave radiation calculations [41], and the Noah land surface scheme [42].

Convective-Stratiform Partitioning
Algorithm. In order to analyze the convective-stratiform precipitation processes in a Mei-Yu front storm, a partitioning algorithm, based on Steiner et al. [43], was used to separate areas of convection from stratiform regions for each synthesized radar volume. Several input parameters for the partitioning scheme are tuned to fit the midlatitude scenarios because the scheme is originally developed for tropical convection [44]. e partitioning algorithm is based on the 3 km horizontal radar reflectivity gradient, which is a background exceeding technique using a low-level 2D horizontal reflectivity field [45]. e algorithm has three steps. First, any grid points having reflectivities greater than 43 dBZ are classified as convective, according to the Z-R relationship in midlatitudes [44]. Second, Reflectivity at each radar grid is compared to its background intensity, computed using the linearly averaged reflectivity with an 11-km radius centered on the grid. Any grid point that exceeds its background reflectivity by an intensity-dependent value is classified as convective. Last, for each grid point classified as convective by either of the two tests, an intensity-dependent radius of influence is used to assign an area around the point as convective. e rest of the echoes that exceed 10 dBZ are classified as stratiform.
is algorithm is also suitable for the WRF-simulated radar reflectivity so that the model results are evaluated at the same metrics as the observations. WRF-simulated radar reflectivity is calculated assuming Rayleigh scattering following the approach of Smith [45], using the specified or predicted size distribution and particle-density parameters consistent with each scheme.

Results and Discussion
In this section, the simulations of the above four microphysical schemes are evaluated from rainfall among different cloud regimes, horizontal and vertical structure of MCS, and cloud microphysical properties, determining their strengths and weaknesses for Mei-Yu rainfall over the MRYR.  June. e large-scale meteorological conditions derived from ERA-5 reanalysis at 0000 UTC 30 June are presented in Figure 2. e field of air flow at 700-hPa was characterized by a northeast-southwest oriented shear line formed at 29-30°N (the green line in Figure 2(a)). South of this shear line, there was a warm and moist southwesterly low-level jet (LLJ) extending from the South China Sea to the JiangHuai Basin. e large pseudo-equivalent potential temperature gradient at 850-hPa, which is often referred to as the location of Mei-Yu front, is also near 30°N. ese large-scale weather conditions are typical features of Mei-Yu front rainstorm. Figure 3 describes the life cycle of the Mei-Yu front rainstorm via the hourly radar reflectivity at 3 km from 0000 to 0800 UTC 30 June, 2018. During the developing stage (1800 UTC 29 June-0200 UTC 30 June), two small MCSs along the northwest-southeast oriented shear line formed and strengthened when moving eastward. At the mature stage (0300-0600 UTC 30 June), the two MCSs merged into one convective system and remained between 112 and 114°E. After 0600 UTC 30 June, the MCSs weakened and gradually dissipated, while the stratiform region remained quite extensive until 0800 UTC 30 June.
e MCSs occurred along the Mei-Yu front, causing moderate precipitation in Central Hubei Province with northwest-southeast oriented rain bands and the rain center exceeded 100 mm adjacent to Xianning site (Figure 4(a)). As shown in Figure 5, rainfall mainly occurred during 0000-1200 UTC on 30 June precipitation and the peak occurred during 0400-0600 UTC on 30 June.

Evaluation of Convective-Stratiform Precipitation.
In spite of minor location errors for the rainfall as compared with the observations, all simulations captured the distribution of northwest-southeast oriented rain bands ( Figure 4). ompson scheme (Figure 4(b)) simulates precipitation centers lager than 75 mm as a little bit      Advances in Meteorology northwest with a larger scope, with the 15-50 mm centers being smaller than the observation (Figure 4(a)), and so do the Morrison (Figure 4(c)) and SBU-YLIN scheme (Figure 4(d)). e rain scopes less than 15 mm simulated by the first three schemes ( ompson, Morrison, and SBU-YLIN scheme) are close to the observation. In contrast, the locations of heavy precipitation (more than 50 mm) simulated by the WDM6_G/H scheme (Figure 4(e)) are scattered with a larger scope than the observation, while the rain scope less than 15 mm is smaller than the observation. To quantitatively evaluate the simulated precipitation, Figure 5 shows the time series of hourly surface rain rate from the observations and WRF simulations with 4 microphysics schemes. As shown, the tendencies of simulated curves are close to observations but all schemes underestimate the total mean precipitation by 23%-35%.
Total precipitation values from WDM6_G/H, THOM, and MOR_G/H scheme are more close to the observation than the SBU-YLIN scheme.
In order to evaluate the WRF-simulated MCS structure in those convective and stratiform regions, the classification algorithm used by Steiner et al. [43] has been applied to both radar observations and WRF simulations. Figure 6 shows the hourly rain rates which are averaged over the convective and stratiform regions from both observations and the WRF simulations. Compared with the mean observed convective rain rate (4.43 mm/h), the first three simulations (Figures 6(b)-6(d)    8 Advances in Meteorology overestimate the convective rain except the WDM6_G/H scheme, which produces more stratiform precipitation.

Horizontal and Vertical Structure of Mei-Yu Front
Rainstorm. To evaluate the horizontal distribution of simulated MCS, 3 km radar reflectivity (Ze, dBZ) at developing (0100 UTC), mature (0400 UTC), and dissipating (0700 UTC) stage are presented in Figure 7. THOM, MOR_G/H, and SBU-YLIN schemes capture the two smaller MCSs at the developing stage, although the MCSs location is a little bit northwest than the observations. At the mature stage, the two simulated MCSs are merged into one just like the radar observations, but there is a time delay in the three schemes. e simulated centers with large Ze (Ze > 40 dBZ) are also scatted by the WDM6_G/H scheme. At the dissipation stage, the WDM6_G/H simulations are much stronger than the observations and the others are close to observations. Figure 8 shows the comparisons of Ze frequency at 3 km altitudes from observations and simulations to further examine the horizontal distribution of Ze. Compared to the observations, all schemes dramatically underestimate the occurrence frequency of lower Ze (0 < Ze < 10 dBZ) but overestimate the occurrence frequency of higher Ze (Ze > 20 dBZ). Except for WDM6_G/H, the other schemes also overestimate the frequency of medium Ze (10 < Ze < 20 dBZ). In general, all schemes underestimate the occurrence frequency of lower Ze but overestimate frequency of higher Ze, especially for the WDM6_G/H scheme.
We further examined the vertical structure of MCS between the observations and simulations. e cross sections of Ze indicated by a black line in Figure 7 are chosen to study the evolution of the vertical structure ( Figure 9). It can be found that the large reflectivity (Ze > 40 dBZ) within the deep convective cores from radar extended to about 6 km at the developing stage. e simulated heights of deep convective cores are close to observations except for MOR_G/H scheme with the large reflectivity extended to about 9 km. At the mature stage, deep convective cores of THOM and WDM6_G/H schemes extended to about 12 km which is almost the same distance that occurred in radar observations. However, all simulations underestimate radar reflectivity in the stratiform region at heights of 9-13 km during the whole lifetime of MCSs. e Contoured Frequency by Altitude Diagrams (CFADs) of reflectivity are shown in Figure 10 from both radar observations and WRF simulations for stratiform and convective cloud, which are produced from the reflectivity volume at 12-min intervals from 03 to 05 UTC on 30 June 2018. is period corresponds exactly to the mature stage of MCS. Discrepancies in the reflectivity distribution characteristics between simulations and observations are obvious. For stratiform cloud, the simulated maximum frequency of Ze (except SBU-YLIN) at lower levels is much smaller than the observations, which are located around 25-30 dBZ and 10 dBZ. Near 6 km, the observed maximum frequency of Ze is located around 20-30 dBZ and decreases with height. e simulations of THOM and MOR_G/H scheme are close to this observation. e simulated Ze (Ze > 10 dBZ) frequency from all schemes at upper levels (10-14 km) is smaller than the observations. For convective cloud, the observed maximum frequency of Ze at middle and lower levels (0-6 km) is located around 35-40 dBZ, and the simulations of THOM and SBU-YLIN are as much as the observations. e simulated frequency of Ze around 30-50 dBZ from MOR_G/H scheme is stronger than other schemes. Simulated frequency of Ze around 20-35 dBZ above 6 km is much weaker than the observations. is kind of vertical distributions of the simulations may cause the simulated MCS to be lower than the observed.

Microphysical Properties of Hydrometeors.
In this paper, a fuzzy-logic, C-POL hydrometeor classification algorithm following the method of Dolan et al. [46] is employed to carry out the hydrometeor classification of the Mei-Yu rainfall. Five hydrometeor types are identified as IC � ice crystals, GA � graupel, DS � dry snow, WS � wet snow, and RN � rain. e vertical cross sections of hydrometeor categories derived from C-POL along 284.0°azimuth at 0925 BJT on 30 June 2018 and WRF simulations at the same position are shown in Figure 11. Due to limited C-POL datasets in this field campaign, the distributions of different hydrometers frequency are not shown in this study. e middle and upper regions (above 6 km) of the cross section from C-POL are characterized by dry snow and regions of ice crystals. Regions of graupel are concentrated at heights from 4 to 8 km, which then transition to rain drops. For the vertical distribution of rain particles, all simulations are close to the observation, although SBU-YLIN scheme simulates greater top height of rain than the observation. e   MOR_G/H and WDM6_G/H schemes simulate more graupel and ice than those in the observations. Because the SBU-YLIN scheme does not distinguish between graupel and snow, only the snow distribution is shown in Figure 11. Overall, the discrepancies between model output and observations exist in the midlevel to upper level (above 6 km), which results from the inability of the model to accurately represent the particle size distribution, ice processes, and storm dynamics. Similar to the method of analyzing clouds and precipitation, hydrometeor mixing ratio of stratiform and convective clouds is evaluated over the same study area. Figure 12 shows the time-and domain-averaged hydrometeor mixing ratio vertical profiles over the stratiform and convective regions for different microphysical schemes. We can clearly find a common feature in different schemes: less cloud and rain are distributed in the stratiform region than in the convective region. Meanwhile, there are more snow and cloud ice particles distributed in the stratiform region than in the convective region, and less graupel is seen in the stratiform cloud regime than in the convective region. ese features indicate that the convective-stratiform partitioning algorithm used in this paper is reasonable. Compared to the other three schemes, WDM6_G/H scheme produced less cloud in both cloud regions and more rain in stratiform region. In convective regions, more rain particles are also found above 2 km for WDM6_G/H scheme. is indicates a large autoconversion of cloud water to rain in the WDM6_G/H scheme than the other schemes. Table 3 shows integrated vertically mixing ratio of cloud, rain, and icephase hydrometeors which is the sum of ice, snow, and graupel over stratiform and convective regions. e icephased particles in the WDM6_G/H scheme are less than those in the other schemes especially for the snow particles in both cloud regimes. is is becuase the smallest maximum vertical air velocity of WDM6_G/H scheme in the convective regime, cloud, and rain particles cannot be brought to upper level air (Figure 13). e THOM and SBU-YLIN scheme simulate less ice but more snow than the other schemes, resulting in moderate mixing ratio of ice-phase particles. e MOR_G/H scheme simulates the most icephase particles in convective region, corresponding to the highest radar reflectivity within 0-6 km and the strongest maximum vertical velocity in convective region, while the mixing ratio of ice-phase particles in stratiform region shows a moderate level, and this is consistent with the underestimated reflectivity above 6 km.

Conclusions
Based on multisource datasets from IMFRE 2018, four bulk cloud microphysics schemes in WRF model were evaluated with respect to their ability to simulate convective and stratiform precipitation, structure, and cloud microphysical properties for a Mei-Yu front rainstorm. e findings from this study can be summarized as follows.
In spite of minor location errors for the accumulated rainfall as compared with the observations, all simulations captured the distribution of northwest-southeast oriented rain bands. However, four schemes underestimate the total precipitation by 23%-35%, and this is mainly attributed to the underestimation of stratiform precipitation. We also found that THOM, MOR_G/H, and SBU-YLIN scheme underestimate the stratiform precipitation and overestimate   For the vertical distribution of radar reflectivity, the large reflectivity (>40 dBZ) within the deep convective cores from radar extended to about 6 km at the developing stage. e simulated heights of deep convective cores are close to observations except for MOR_G/H scheme with the large reflectivity extending to about 9 km. At the mature stage, deep convective cores of THOM and WDM6_G/H schemes extended to about 12 km which is almost the same distance that occurred in radar observations. However, all simulations underestimate radar reflectivity in the stratiform region at heights of 9-13 km during the whole lifetime of MCSs.
Discrepancies in the vertical distribution characteristics between simulations and observations are obvious form the CFADs of reflectivity. For stratiform cloud, all simulated radar reflectivity values at upper levels (8-14 km) are smaller than the observations. For convective cloud, the observed maximum probability at middle and lower levels (0-6 km) is located around 35-40 dBZ, and the simulations of THOM and SBU-YLIN are as much as the observations. e simulated radar reflectivity of MOR_G/H is stronger than other schemes. Simulations of reflectivity above 6 km are much weaker than the observations. All schemes are able to simulate well the rain distribution at low level and snow distribution above 6 km. However, there are many discrepancies between simulations and observation of graupel and ice vertical distribution. THOM and SBU-YLIN scheme simulate less ice but more snow than other schemes, resulting in moderate mixing ratio of ice-phase particles.
e MOR_G/H scheme simulates the most icephase particles in convective region, corresponding to the strongest maximum vertical velocity in convective region, while the ice-phase particles in stratiform region show a moderate level, and this is consistent with the underestimated reflectivity above 6 km. e ice-phased particles in WDM6_G/H scheme are less than those in other schemes especially for the snow particles in both cloud regimes.
In summary, each bulk scheme evaluated in this study has its advantages and shortcomings for different cloud regimes. Overall, the discrepancies between model output and observations mostly exist in the midlevel to upper level, which probably results from the inability of the model to accurately represent the particle size distribution, ice processes, and storm dynamics. erefore, more detailed microphysics schemes, such as higher moment schemes or complicated schemes that can describe more ice-phase particle interactions, may be needed to properly represent these evaluated features.
It is worth noting that, due to the limited Mei-Yu front rainstorm cases observed by IMFRE, the microphysical characteristics of Mei-Yu rainfall will continue to be observed and studied in the later stage. Further observations from major field campaigns and simulations are still necessary. Meanwhile, the errors caused by the nonlinear method of the numerical sensitivity test, the uncertainties of model initials, and the limited ability of the model simulations (e.g., cumulus, land surface, or radiation schemes) all have certain influence on the simulation results.

Conflicts of Interest
e authors declare that there are no conflicts of interest regarding the publication of this paper.