2.1 Disdrometer Data Sets

Surface DSDs are observed by the Particle size and velocity (Parsivel) disdrometer network that is recently established and operated by CMA (Y. Han et al., 2022). The maintenance procedures of the Parsivel network follow the station equipment maintenance protocols of the CMA, ensuring long-term stable observations. Data sets from a total of 293 Parsivel disdrometers in East China during the four Meiyu seasons (2019–2022) (see Table S1 in Supporting Information S1) are collected and analyzed in this study. The Parsivel is equipped with a laser sheet to estimate the size and fall velocity of raindrops based on the attenuation and duration of the laser signal as they fall through (Tokay et al., 2014). One known limitation of the Parsivel is the possible undercounting of small raindrops due to their blockage by larger drops, which may lead to an overestimation of the raindrop mean diameter and an underestimation of the number concentration (Tokay et al., 2013; Wen et al., 2017). Despite this inaccuracy, the Parsivel has been proven to possess good consistency in rain rate with gauge observations and is widely used for DSD studies globally (B. Chen et al., 2020; G. Chen et al., 2022; Friedrich et al., 2013, 2015; Matrosov et al., 2016).

To minimize measurement error in heavy rainfall, the same processing and quality control (QC) procedures of Parsivel as G. Chen et al. (2022) and Wen et al. (2017) are performed. The first two size bins of the Parsivel data are excluded, spurious particles larger than 8 mm are removed. Particles deviating from the theoretical fall velocity–diameter relationship of Brandes et al. (2002) by more than ±60% were also filtered out. After QC, raindrop size distribution data is stored in 32 size bins with 1-min frequency. Related parameters investigated in this study include the rain rate R (mm h⁻¹), mass-weighted mean diameter D_m (mm), and generalized intercept parameter N_w (mm⁻¹ m⁻³). The mean diameter D_m is defined as the fourth-order to the third-order moment of the DSD, and N_w is calculated using D_m and rainwater content (W, $$\mathrm{g}\,{\mathrm{m}}^{-3}$$) following Bringi et al. (2003): $${N}_{w}=\frac{{4}^{4}}{\pi {\rho }_{w}}\left(\frac{{10}^{3}W}{{D}_{m}^{4}}\right)$$, where $${\rho }_{w}$$ denotes the density of water. To obtain reliable DSD and environmental information from these heavy rainfall events, the heavy rainfall periods (HRPs) are identified from the Parsivel 1-min interval rain rates. An HRP is defined as the continuous observational period with a mean rain rate exceeding 20 mm hr⁻¹ for any consecutive period of 60 min.

2.2 GPM DPR Retrievals

The GPM DPR level-2 data set version 7 (Seto et al., 2021) during the nine Meiyu seasons (2014–2022) is also used to examine the microphysical structure of heavy rainfall. The GPM DPR operating at Ku- and Ka-bands provides rainfall and three-dimensional DSD retrievals based on dual-frequency observations at an approximately 5-km spatial and 125-m vertical resolution (see algorithm document of the official product, Iguchi et al., 2021). GPM DPR products in this study are the attenuation-corrected Ku-band radar reflectivity factor ($${Z}_{e}$$) and the retrieved DSD parameters ($${D}_{m}$$ and $${N}_{w}$$) from the inner swath data. The inner swath corresponds to the central 25 normal scan beams obtained simultaneously and co-aligned from both the Ka- and Ku-band radars. Recent studies have validated the accuracy of GPM DPR retrievals in China, and demonstrated the potential applicability of this method to reveal the microphysical structures of convective systems in statistics (F. Chen et al., 2022; Hu et al., 2022; Huang et al., 2021; Sun et al., 2020).

2.3 Reanalysis Data Sets

Environmental conditions and locations of the Meiyu fronts are both derived from the fifth-generation global atmospheric reanalysis data set from the European Center for Medium Range Weather Forecasts (ERA5; Hersbach et al., 2020). The ERA5 reanalysis data has a spatial resolution of 0.25° × 0.25, a temporal resolution of 1 hr, and provides meteorological variables including temperature, moisture, and winds. To analyze the thermodynamic and dynamic conditions prior to the occurrences of the HRPs, environmental information 2 hr prior to the initiation of HRPs is extracted from the ERA5 reanalysis data sets. On the other hand, location of the Meiyu front is identified using the ERA5 850 hPa equivalent potential temperature ($${\theta }_{e}$$) fields. The position of a front is identified using a two-step method: First, at each longitude, the latitude where $${\theta }_{e}$$ equals 345 K is located following (Luo et al., 2014). Second, when multiple 345-K $${\theta }_{e}$$ values are identified along a given longitude, the latitude with the maximum $${\theta }_{e}$$ gradient is confirmed as the location of the Meiyu front (Sun et al., 2020). All the identified Meiyu fronts are visually verified to exclude any unclear results.

3.1 Surface DSDs

A total of 1,757 HRPs are captured by the disdrometer network during the four Meiyu seasons (2019–2022). To quantitatively examine the DSD difference associated with the Meiyu front, all HRPs are classified into three types according to their latitudes with respect to the identified fronts (Figures 1b and 1c). A total of 539 HRPs (blue) located in the frontal area (within 1° to the front) are defined as the Front type. Following previous studies, the warm sector is defined as the area more than 2° south of the Meiyu front (B. Han et al., 2021; M. Zhang & Meng, 2019). To ensure a comparable spatial extent between the warm sector and the frontal area, 673 HRPs (red) located 2–4° south of the front are classified as the Warm type. The remaining 545 HRPs (yellow) in the transitional zone between the warm sector and front (i.e., 1–2° south of the front) are classified as the Trans type.

Consistent with previous studies (G. Chen et al., 2012; R. Yu et al., 2007; F. Zhang et al., 2023), an obvious double-peak pattern in the diurnal occurrence frequency of all 1,757 HRPs is observed, with one peak in the early morning and another in the late afternoon (dashed gray line, Figure 1d). Interestingly, when the HRPs are divided into individual types, the double peaks are separated accordingly. Specifically, the Front (Warm) type exhibits an obvious single morning (afternoon) peak at around 07:00 (19:00) LST. The diurnal cycle of total rainfall across the three HRP types closely resembles that of occurrence frequency (Figure 1e). Over 60% of the total rainfall during the afternoon peak is contributed by the Warm type, while nearly 50% the morning peak rainfall is attributed to the Front type. This result supports previous findings that convection in different areas relative to the Meiyu front exhibits distinct diurnal cycles (G. Chen, 2019; G. Chen et al., 2017, 2019). The potential differences in cloud microphysical properties associated with the double-peak pattern formed by convection at various frontal positions warrant further study. Two typical HRP cases belonging to the Warm and Front types are shown in Figures 1f–1k, which seem to display notable differences in DSD characteristics. Compared to the Front HRP case, the Warm case exhibits a greater amount of medium (D within 2–4 mm) and large raindrops (D > 4 mm), and it is characterized by higher $${D}_{m}$$ but lower $${{\log }_{10}\,N}_{w}$$ values from both disdrometer and GPM DPR observations. This result is consistent with previous case studies (B. Han et al., 2021; Oue et al., 2010). However, robust conclusions need to be derived from a more comprehensive analysis.

The Warm, Trans, and Front HRPs encompass a total of 22,749, 18,267, and 18,219 heavy rainfall (R > 20 mm hr⁻¹) minutes, respectively. The related DSD parameters are compared in Figure 2. As for the composite raindrop spectrum, the Warm type is identified with the highest number concentrations of medium and large raindrops. At D = 4 mm, the ratio between Front and Warm in N(D) is approximately 0.7. Conversely, the Front type shows higher N(D) for small raindrops within the 1–2 mm range as compared to the other two types. The results indicate that Warm (Front) heavy rainfall is composed of larger (smaller) raindrops. Correspondingly, about 20% rainfall comes from raindrops with D > 3 mm for the Warm type, compared to 14% for Front (Figure 2b). Small raindrops with D within 1–2 mm contribute ∼38% (33%) of the total rainfall for the Front (Warm) type, respectively. Relatively higher fractions of rainfall are produced by larger (smaller) raindrops in the warm sector (frontal) area.

Histograms of $${D}_{m}$$ and $${{\log }_{10}\,N}_{w}$$ for the three types of heavy rainfall are shown in Figures 2c and 2d. The mean $${D}_{m}$$ is 2.10, 2.03, and 1.91 mm for the Warm, Trans, and Front types, respectively. It is evident that raindrop mean size is larger for the warm-sector heavy rainfall compared to the other two types. In contrast, the mean values of $${{\log }_{10}\,N}_{w}$$ are 3.61, 3.68, and 3.76 for the heavy rainfall types, the lowest mean number concentration is observed in the warm-sector heavy rainfall. The two-tailed Welch's t tests were employed to examine the differences in $${D}_{m}$$ and $${{\log }_{10}\,N}_{w}$$ between any two heavy rainfall types. The p-values, effect sizes, and t-statistics are provided in Table S2 of Supporting Information S1. It indicates that the differences in the two parameters are statistically significant at the 0.01 level. Although the differences in $${D}_{m}$$ and $${{\log }_{10}\,N}_{w}$$ between the frontal and warm-sector heavy rainfall are not as large as the comparison between typical continental and maritime convection (Bringi et al., 2003), the findings still indicate obvious difference of surface DSDs in heavy rainfall during the Meiyu season.

Additional comparisons of DSDs from different rain rate intervals are depicted in Figure S1 (see Supporting Information S1), the contrasts in DSDs generally decrease with rainfall intensity, but higher (lower) fractions of small (large) raindrops are always seen from Front compared to Warm. This implies that distinct DSD characteristics are observed among the Warm, Trans, and Front types across various intensities of heavy rainfall. On the other hand, observations suggest that $${D}_{m}$$ and $${{\log }_{10}\,N}_{w}$$ show clear diurnal variation across three HRP types (Figure S2 in Supporting Information S1). For all HRP types, the $${D}_{m}$$ ($${{\log }_{10}\,N}_{w}$$) values in the afternoon is higher (lower) than that in the morning. This may be attributed to variations in convective evolution, with convection more likely to be in the initiation and mature stages during the afternoon, favoring the falling of large raindrops to the ground. For the same time of day, the $${D}_{m}$$ of the Warm type is still higher than that of Front, likely due to its more favorable thermodynamic environment for developing convection with larger raindrop mean sizes.

3.2 Microphysical Structures

Microphysical structures of the three heavy rainfall types are further elucidated using GPM DPR observations spanning nine Meiyu seasons, which yielded 3,141, 2,792, and 3,718 DPR observed heavy rainfall (R > 20 mm hr⁻¹) pixels for the Warm, Trans, and Front types, respectively. The contoured frequency by altitude diagrams (CFADs; Yuter & Houze, 1995) and mean profiles of $${Z}_{e}$$, retrieved $${D}_{m}$$ and $${{\log }_{10}\,N}_{w}$$ are portrayed in Figure 3. Generally, deeper convection is observed in the Warm type compared to the Front type (Figures 3a1 and 3c1). The mean 20-dBZ height is approximately 12-km for the former and 8-km for the latter. Additionally, mean $${Z}_{e}$$ values of the Warm type are basically 2-dBZ higher than the Front type below the freezing level, and the corresponding $${Z}_{e}$$ values are 46 and 44 dBZ at 1-km altitude (Figure 3d1). Nevertheless, the mean reflectivity values of all three heavy rainfall types rapidly increase from the freezing level to the surface, indicating the importance of warm-rain growth processes for heavy rainfall during the Meiyu season.

Regarding the retrieved DSD parameters, it is worth noting that the majority of heavy rainfall samples (with a normalized frequency exceeding 50%) exhibit $${D}_{m}$$ values below 2 mm across all three heavy rainfall types (Figures 3a2, 3b2, and 3c2). However, there are obviously greater proportions of raindrops with $${D}_{m}$$ > 2 mm for the Warm type as compared to the Front type, aligning with disdrometer observations (Figure 2c). A higher (lower) fraction of samples with $${D}_{m}$$ > 2 mm can also be recognized above the freezing level (∼5 km) for the Warm (Front) type, indicating more (less) large-size ice particles (e.g., graupel and hail) produced by ice-phase processes. The melting of these ice particles is the primary pathway for generating large raindrops in deep convection (G. Chen et al., 2023; Ryu et al., 2021). Additionally, all three types have the majority of samples with logarithmic values of the generalized intercept parameter $${N}_{w}$$ between 4 and 5 at lower altitudes, whereas an obviously higher proportion of raindrops with $${{\log }_{10}\,N}_{w}< 4$$ is observed for the Warm type. For ice particles above the freezing level, the Front type appears to have a higher number concentration, indicating the generation of small size aggregates or ice crystals.

The average $${D}_{m}$$ and $${{\log }_{10}\,N}_{w}$$ profiles are also compared in Figures 3d2 and 3d3. Their values in descending (and ascending) order are the Warm, Trans, and Front type, which indicate generally larger (smaller) hydrometeors with lower (higher) number concentrations in the relatively deeper (shallower) convection over the warm sector (frontal) area. Mean $${Z}_{e}$$, $${D}_{m}$$, and $${{\log }_{10}\,N}_{w}$$ values from less extreme rain rates are compared in Figure S3 (see Supporting Information S1) and confirm a similar conclusion. The changes in R and $${D}_{m}$$ at low levels are provided in Figure S4 (see Supporting Information S1). For all three heavy rainfall types, both rainfall intensity and raindrop mean size increase from 4 km down to the 1-km level. As precipitation becomes more extreme, the percentage increase in rainfall intensity at low levels can exceed 50%, indicating a dominant role for accretion processes (raindrops collecting cloud droplets, G. Chen et al., 2023).

3.3 Environmental Conditions

In essence, convective properties—including the diurnal cycle, cloud depth, precipitation intensity, and associated microphysical characteristics—are primarily governed by the prevailing thermodynamic and dynamic environmental conditions (Doswell et al., 1996; Zhou et al., 2021). To clarify the microphysical contrasts between Meiyu warm-sector and frontal heavy rainfall, a detailed comparison of their thermodynamic and dynamic environments is performed. For each heavy rainfall type (Warm, Trans, and Front), the corresponding ERA5 fields associated with all HRPs were extracted and composited for further analysis. On the other hand, radiosonde observations closely collocated with HRP events in both time and space were utilized, enabling a more comprehensive analysis of the thermodynamic and dynamic environment.

The horizontal wind at 850 hPa (Figures 4a–4c) indicates that the Front type exhibits the strongest wind speed, followed by Trans, with the weakest winds of the Warm type. Following the definition of low-level jet (LLJ) proposed by Du et al. (2012), LLJs were identified and mapped as regions with high LLJ occurrence frequency (exceeding 60%). A large area of frequent LLJs located south of the Meiyu front for Front, whereas no such obvious LLJ area is found for Warm. This difference can be explained in conjunction with the diurnal cycle in Figures 1d and 1e: The Front type primarily occurs at night, when the boundary layer inertial oscillation strengthens the prevailing southwesterly monsoon flow, horizontal wind shear and moisture convergence (G. Chen et al., 2017; Fu et al., 2019; Xue et al., 2018). Due to stronger low-level winds and enhanced nocturnal moisture transport, the column-averaged (surface to 700 hPa) moisture transport flux (shading) associated with Front is generally greater than that of Warm at similar latitudes.

A comparison of convective available potential energy (CAPE) reveals that the Warm type exhibits the highest CAPE, followed by the Trans type, with the lowest values observed in the Front type (Figures 4d–4f). Conversely, the Front type shows a wide spatial extent of strong moisture transport convergence (black dots, less than –10⁻⁵ kg m⁻² s⁻¹), mainly concentrated near the southern side of the Meiyu front. This contrast further highlights the differences in the environmental thermodynamic and dynamic conditions between the Warm and Front types. Convection of Warm tends to develop during the afternoon under the influence of solar heating, whereas convection for Front primarily occurs at night, driven by strong low-level moisture convergence and dynamic lifting near the frontal zone (Xue et al., 2018; R. Yu et al., 2007).

Composite skew-T log-P diagrams constructed from radiosonde data are shown in Figures 4g–4i. For all HRP types, the precipitable water (PW) exceeds 50 mm, indicating favorable moisture conditions for heavy rainfall. However, the average CAPE value in the Warm type reaches 1,469 J kg⁻¹, nearly three times that of the Front type (585 J kg⁻¹). Conversely, the Front type is characterized by an obviously lower lifting condensation level (LCL) and stronger low-level horizontal winds. The findings align well with those derived from ERA5, not only explaining the environmental differences in different HRP types, but also contributing to a better understanding of the mechanisms behind microphysical variability. Specifically, a higher CAPE of Warm favors stronger convective updrafts compared to Front (Doswell et al., 1996), leading to more intense convective development, higher $${Z}_{e}$$ values, and larger raindrop mean sizes. Moreover, the lower LCL of Front provides favorable environment for more efficient accretion processes between raindrops and cloud droplets, as well as for suppressing evaporation of small raindrops within the warm cloud layer (Chang et al., 2015; M. Wang et al., 2016). Differences in environmental thermodynamic and dynamic fields—such as moisture transport flux, CAPE, and LLJ conditions—together account for the variations in diurnal cycle and microphysical properties of convective systems between the Meiyu frontal and warm-sector areas. In other words, the double-peak diurnal pattern of Meiyu heavy rainfall is caused by two types of convective clouds with distinct microphysical characteristics.