Introduction

Heavy precipitation presents critical concerns as it can trigger serious natural catastrophes and obstruct economic development worldwide1,2,3. As the climate continues to warm, a growing trend in the occurrence frequency of extreme precipitation events is found at many locations despite substantial regional variations4,5,6. Therefore, accurately forecasting the rainfall intensity has become more important than ever for mitigating the impact of extreme precipitation disasters. One of the challenges lies in better understanding the relationship between rainfall intensity and convective microphysical processes7,8,9.

Strong updrafts in intense convection drive vigorous ice-phase and mixed-phase microphysical processes, producing a large number of sizable ice particles that can be converted into heavy surface precipitation. Consequently, intense convection was widely recognized to be closely related to heavy rainfall10,11,12. However, warm-rain microphysical processes can also produce heavy precipitation with relatively weak convection13,14,15,16. Over land, where most of the intense convection events occur10, only about 30% or fewer events are characterized by both extreme rainfall and extreme convection13,14,17. As Xu et al.14 indicated, weak convection contributes to extreme rainfall almost as much as intense convection, and the stark contrasts in vertical structures and atmospheric conditions between these two types of extreme rainfall underscore distinct physical mechanisms. In this way, only the metrics of convective intensity observed from spaceborne or ground-based radars are insufficient for monitoring or nowcasting the rainfall intensity18.

Although previous studies have emphasized the distinct contributions of varying convective intensities to near-surface rain rates, these analyses have primarily focused on the most extreme precipitation events11,13,14,15,16,17,19. By analyzing the vertical structure of convective systems, it is possible to elucidate the primary microphysical processes that determine the surface rain rates of a wide range20,21,22. Notably, earlier regional investigations have documented fundamental differences between oceanic and continental tropical convection in terms of convective strength and the dominance of warm-rain or ice-phase processes23,24,25,26,27,28, but a systematic understanding of how these differences relate to rainfall intensity across the full convective spectrum remains incomplete. Moreover, although many studies on the impact of future climate change on precipitation have found an increase in extreme events under global warming with large regional variations4,5,6, the projections remain less reliable due to significant precipitation biases in climate models29 and those studies rarely accounted for the complexities introduced by precipitation events with different microphysical features. Since large-scale circulation changes tend to be more accurately simulated in climate models30, a discussion about the environmental conditions related to these microphysical differences can provide valuable insights into addressing the complexities of future changes of heavy precipitation events with distinct convective intensities. Therefore, a comprehensive understanding of the microphysical processes and environmental conditions governing the variability in rain rates produced by convective clouds is critical for improving weather forecasting models and mitigating the impacts of climate change on global water cycles and extreme weather patterns. However, such knowledge is still lacking.

Here we reveal the microphysical discriminations and associated environmental mechanisms in global convective precipitation events (CPEs; see “Methods”), using long-term records from spaceborne precipitation radars of Tropical Rainfall Measuring Mission (TRMM)31 and Global Precipitation Measurement (GPM)32, cloud-permitting ensemble simulations, and the European Center reanalysis data (ERA5)33. The CPEs (Table S1 in Supplementary) are classified into the “intense” CPEs (InCPEs), “moderate” CPEs (MoCPEs), and “weak” CPEs (WeCPEs) based on the maximum height of the 30-dBZ echo (MAXHT30), used as a proxy for convective intensity (see “Methods”). The differences between those generating “heavier” rainfall (HR), “medium” rainfall (MR), and “lighter” rainfall (LR) in each convective-intensity category are investigated. This study aims to bring an in-depth understanding of microphysical processes in determining the disparate surface rain rates of different CPE types.

Results

Global distributions and convective characteristics

The CPE distributions observed by GPM (Fig. 1) and TRMM (Fig. S1 in the Supplementary) show similar patterns in the tropics and subtropics, despite the different satellite platforms. The InCPEs exclusively occur over land, consistent with previous studies11,14,34, while a large portion of MoCPEs and WeCPEs are located over the maritime ITCZ and monsoon regions (Fig. 1a–c). The well-recognized convective hotspots over land (black boxes in Fig. 1), including southern North America, the Amazon, Central Africa, southern Asia, southeastern China, and the maritime continent islands, exhibit a high occurrence of all CPE categories and thus are selected for the following analysis. In these regions, the InCPEs contribute nearly equally (about 30–50%) to the HR and LR events (Fig. 1d, f). However, as the convection weakens, the proportion of LR increases, exceeding 70% for the WeCPE (Fig. 1l). As moderate events lie between the intense and weak categories, subsequent analyses focus on four representative combinations to avoid redundancy.

Fig. 1: Geographical distributions of the CPEs from the GPM data.
figure 1

a–c the event numbers of InCPE, MoCPE, WeCPE; (d–f) the fractions (%) of HR-, MR-, LR-producing InCPEs in all InCPEs; (g–i) the fractions of HR-, MR-, LR-producing MoCPEs in all MoCPEs; (j–l) the fractions of HR-, MR-, LR-producing WeCPEs in all WeCPEs. In (d–l), only grids with more than 20 events are plotted. Black boxes indicate the regions of southern North America (115°W–76°W, 10°N–40°N), the Amazon (80°W–50°W, 20°S–6°N), Central Africa (5°E–35°E, 10°S–10°N), southern Asia (67°E–108°E, 6°N–25°N), southeastern China (108°E–122°E, 19°N–37°N) and the maritime continent islands (96°E–150°E, 9°S–6°N).

Full size image

The convective characteristics in both TRMM and GPM data show consistent results (details in Supplementary Discussion 1). The average horizontal scale of global CPEs significantly increases with increasing convective intensity or rain rate (p < 0.01 for differences between the CPE types; Fig. S2a). The higher rain rate also corresponds to a larger scale of the CPE-embedded parent precipitation system and a slightly higher length-to-width ratio, across all CPE categories (p < 0.01; Fig. S2b, c). Of notice is that the HR-producing InCPEs are more convectively strong than the LR-producing InCPEs (Fig. S2e–i). Specifically, the maximum height of the 20-dBZ echo in 13% of the HR-producing InCPEs extends beyond the tropopause (>15 km; Fig. S2e), which is consistent with the presence of a larger amount of precipitation-size solid particles (Fig. S2g, h) with vigorous mixed-phase processes and more active lightning activity (Fig. S2i). This proportion is notably higher than in the Darwin monsoon regime25, where echo tops during active periods of monsoon rarely exceeded the tropopause, unlike the break period when most intense cells reached or overshot it. On the other hand, the results of WeCPEs suggest limited precipitation-size solid particles and a dominance of warm-rain processes in producing near-surface rainfall, consistent with previous studies on oceanic and monsoon convection23,24,25,26,27,28.

Vertical structures and microphysical processes

Figure 2 (Fig. S3) shows the GPM (TRMM) composite vertical profiles of the rain rate (details in Supplementary Discussion 1), the maximum reflectivity (MAXREF; details in Supplementary Discussion 1), liquid water content, ice water content, mass-weighted mean diameter (Dm), and the normalized drop size distribution (DSD) scaling parameter for concentration (Nw) for the various CPE types. The particle concentration of HR-producing CPEs is generally higher than that of LR-producing CPEs within the same convective-intensity category, while the corresponding mean size of particles is slightly smaller. These DSD characteristics are probably related to a more humid environment for the HR-producing CPEs, preventing evaporation and promoting the growth of numerous small liquid particles35. Meanwhile, the stronger convective updrafts in HR-producing InCPEs, compared to those producing LR (Fig. S2e–i), transport more moisture and ice nuclei to higher altitudes, facilitating the rapid formation of a large number of small ice crystals above 10 km. The abundance of these ice crystals leads to increased competition for available water vapor, thereby limiting crystal growth and resulting in a smaller mean diameter36,37.

Fig. 2: Composite vertical profiles for InCPEs and WeCPEs from the GPM data.
figure 2

a, g rain rate, (b, h) maximum radar reflectivity, (c, i) radar-estimated ice water content, (d, j) radar-estimated liquid water content, (e, k) average Dm, and (f, l) average Nw. The dashed black line denotes the approximate melting level (5 km). The 50th percentiles and 90th percentiles are represented by solid and dotted lines, respectively. Due to the contamination of surface clutter, only profiles above 1 km from the ground are applied.

Full size image

Above the melting level (about 5 km), both the rain rate and MAXREF show minimal differences between HR and LR in WeCPEs (Fig. 2g, h). In contrast, for InCPEs, the HR-producing cases exhibit a more rapid increase of MAXREF downward from 7 to 5 km compared to the LR-producing cases (Fig. 2b), i.e., the 50th percentile profiles increase by 8 dBZ and 5 dBZ for HR-InCPEs and LR-InCPEs, respectively, consistent with the increase of ice water content (Fig. 2c). Meanwhile, the Dm value in HR-InCPEs increases downward faster than LR-InCPEs in this height range, accompanied by a rising Nw (Fig. 2e, f). These suggest that the solid particles in HR-InCPEs are augmented by stronger riming processes than LR-InCPEs from 7 to 5 km altitudes38,39. Consequently, significantly higher liquid water content of HR-InCPEs compared to LR-InCPEs is observed just below the melting level due to the larger ice water content (Fig. 2d), and the MAXREF differences reach 5 dBZ between HR-InCPE and LR-InCPE near the melting level.

For all convective-intensity categories of CPE, the MAXREF differences between HR and LR become larger with decreasing height below the melting level, highlighting the importance of warm-rain microphysical processes in determining the near-surface rain rate. The liquid-phase processes primarily include collision-coalescence, breakup, evaporation, and size sorting. To investigate the different roles of these processes, we adopt the method of Huang and Chen40, using the differences of Dm and MAXREF between 1 km and 3 km (∆Dm and ∆Z; Fig. 3). In Fig. 3a, b, compared to the LR-producing InCPEs, the HR-producing InCPEs have a higher combined fraction of coalescence and coalescence-breakup balance processes (from 23% to 55%) with a lower percentage in the 3rd quadrant (breakup processes). In WeCPEs, coalescence combined with breakup-coalescence balance accounts for 85% of the HR-producing events (Fig. 3c), larger than the 60% for the LR-producing events (Fig. 3d). And WeCPEs are accompanied by much weaker breakup processes (10–28%) compared to InCPEs (40–69%), likely due to smaller raindrop sizes (Fig. 2) and the absence of strong convective updrafts and downdrafts. These results highlight the importance of coalescence in producing heavier rainfall in both InCPEs and WeCPEs. However, in HR-producing InCPEs, the combination of strong riming and a near balance between coalescence and breakup contributes to more extreme MAXRR. Specifically, the 50th and 90th percentiles of MAXRR reach 115 mm/h and 299 mm/h, respectively, which are substantially higher than those in HR-producing WeCPEs (74 mm/h and 164 mm/h; Fig. 2a, g).

Fig. 3: The occurrence frequency in 2-dimensional space of ∆Dm and ∆Z for CPEs from the GPM data and CM1 simulations.
figure 3

a HR-InCPEs, (b) LR-InCPEs, (c) HR-WeCPEs and (d) LR-WeCPEs are derived from the GPM data, while (e–h) are the results from the CM1 simulations of four cases. The text in each quadrant denotes the primary microphysical process and the total occurrence frequency in the quadrant (%).

Full size image

To help examine the results of microphysical processes inferred from the space-borne radar observations, four sets of idealized ensemble simulations using the Cloud Model 1 (CM1) with different environments are conducted (CASE1-4; details in Methods and Supplementary Discussion 2). CASE1 corresponds to an HR-InCPE, CASE3 also produces HR but through a We-to-Mo CPE, while CASE2 represents an InCPE producing LR-to-MR, and CASE4 corresponds to a LR-WeCPE (Figs. S4–6). Among the ensemble members of the same case, the evolutions of CPE show consistency, with some ensemble spreads beyond the convective core caused by the minor differences in initial temperature perturbations (Figs. S4, 5).

Figure S7a shows the updraft intensity in CASE1-4 defined as the 90th percentile of vertical velocity within a 2.5 km radius centered on the location of maximum near-surface rain rate. CASE1 (red) shows the strongest updrafts, consistent with the highest convective intensity in HR-InCPEs (Fig. S2e–i). Elevated snow content at 6–8 km and peak graupel near 5 km (Fig. S7c, d) suggest active riming processes. Nevertheless, CASE2 (black) also exhibits strong updrafts, with the overwhelmingly highest cloud ice and graupel mixing ratios compared to the other three cases (Fig. S7b, d). In contrast, CASE3 (blue) and CASE4 (gray) exhibit weak updrafts and limited ice-phase processes. Furthermore, as shown in Fig. 3e–h, the simulation results generally reproduce the quadrant distributions of various liquid-phase microphysical processes seen in the satellite observations (Fig. 3a–d). Compared to CASE2, CASE1, with a similar intense convection but producing higher surface rain rates, is characterized by the dominance of coalescence and coalescence-breakup balance (51% in Fig. 3e), instead of breakup processes (57% in Fig. 3f). Moreover, the result of CASE3 shifts toward an overwhelming dominance of coalescence processes (72% in Fig. 3g), whereas CASE4 shows only a 24% contribution from these processes (Fig. 3h), consistent with the satellite-observed HR- and LR-producing WeCPEs.

At 1 km altitude, the major difference in the GPM DSD between LR and HR, regardless of convective intensity, lies in the much higher droplet concentration in the HR-producing CPEs, facilitating coalescence to produce increased rain rates (Fig. S8). While the LR-producing InCPEs are mostly concentrated in the region with relatively large size and low concentration of precipitating particles (red box), which corresponds to the typical characteristics of the continental regime41, the HR-producing InCPEs shift toward higher concentration with a significant portion above the continental regime (Fig. S8a, b). The former can include graupel or hail-dominant storms as shown in Fig. S7d (black), while the latter is featured with numerous large raindrops produced under the combined effects of mixed-phase processes and coalescence-breakup balance42. Compared to the LR-producing WeCPEs, the HR-producing WeCPEs also show much higher concentration and thus exhibit a more maritime-like characteristic on average (Fig. S8c, d). Examples of the HR-producing WeCPEs include the oceanic-like convective rainfall in humid monsoon areas43, such as the Meiyu rainband in East Asia.

Environmental conditions

To investigate the key differences in environmental conditions of CPEs producing heavier and lighter precipitation, the ERA5 profiles closest to each CPE center in the inflow region are used (Fig. S9 in the Supplementary). The lifting condensation level, the level of free convection, and the differences between temperature and dewpoint temperature at the same height consistently indicate a more humid environment and thicker warm-cloud layer for HR-producing CPEs than LR-producing CPEs (p < 0.01; Fig. S9c–f). Moreover, the HR-producing WeCPEs show a lower cloud base and moister middle troposphere than the HR-producing InCPEs (p < 0.01; Fig. S9c, e), which facilitates more active and efficient coalescence, compensating for the lack of strong mixed-phase processes in WeCPEs for the HR production. The environmental vertical wind shears (VWS) for CPEs are predominantly concentrated within the ranges of 3–8 m/s at the lower levels and 4–10 m/s at the middle levels, respectively (Fig. S9g, h), and they are only slightly stronger for HR-producing CPEs than LR-producing CPEs (p < 0.01). Note that the thermodynamic and dynamic differences in the initial atmospheric sounding states between CASE1-4 show consistent results (Fig. S10), e.g., the much lower lifting condensation level and slightly stronger VWSs in CASE1 and CASE3 compared with CASE2, with the lowest cloud base in CASE3.

The vertically integrated horizontal water vapor flux reveals significant anomalies in HR-producing InCPEs, LR-producing InCPEs, and HR-producing WeCPEs compared to the mean state of all CPEs (Fig. 4). For the Amazon, southern Asia and the maritime continent islands (Fig. 4a–l), the monsoon currents in HR-producing and LR-producing CPEs from upstream oceans are significantly enhanced and decreased (p < 0.05), respectively, contributing to increased (decreased) moisture convergence, and HR-WeCPEs exhibit stronger enhancement of the monsoon airflows compared to HR-InCPEs. Furthermore, as the summer monsoon circulations influencing southern Asia and the maritime continent islands are projected to intensify at the end of the 21st century under both intermediate- and high-emission scenarios (Fig. S11 in Supplementary)44,45,46, the HR-producing CPEs in these regions are very likely to increase, a projection inferred from the aforementioned characteristics.

Fig. 4: Composite fields and anomalies of horizontal water vapor flux and its divergence.
figure 4

The horizontal water vapor flux (kg∕(m s); arrows) and its divergence (10−5 kg∕(m2∙s); shadings) are vertically integrated from the surface to 300 hPa for all CPEs in the six regions (a, e, i, m, q, u), and the anomalies of HR-InCPEs (b, f, j, n, r, v), LR-InCPEs (c, g, k, o, s, w) and HR-WeCPEs (d, h, l, p, t, x) compared with the composite over all CPEs are also shown. The arrows and shading areas indicate that the anomalies are statistically significant at a 95% confidence level. From top to bottom are the regions of the Amazon, southern Asia, the maritime continent islands, southeastern China, southern North America, and Central Africa.

Full size image

In regions of southeastern China, southern North America and Central Africa, the anomalies of water vapor transport are more variable. In southeastern China (Fig. 4m–p), the HR-InCPEs and LR-InCPEs are observed with an anticyclonic and cyclonic anomaly over the Philippine Sea, respectively, whereas HR-WeCPEs are accompanied by a cyclonic anomaly over southeastern China coasts, which results in moisture flux from the South China Sea to converge over southeastern China. In southern North America (Fig. 4q–t), the anomaly of HR-InCPEs is featured by an anticyclonic anomaly off the western coast of Mexico and enhanced southerlies over the Gulf of Mexico, delivering moisture northward and converging with southward airflow from the Northwest Atlantic. For HR-WeCPEs, this anomalous water vapor channel is replaced by a westerly airflow from the eastern tropical Pacific, forming a cyclonic anomaly and convergence over the Gulf of Mexico. In Central Africa (Fig. 4u–x), the easterly moisture fluxes from the western Indian Ocean are strengthened for HR-InCPEs, while the easterly fluxes converge with strong anomalous westerly fluxes from the South Atlantic for HR-WeCPEs. In general, HR-producing WeCPEs correspond to the most intense water vapor channels across all regions, facilitating substantial moisture convergence from adjacent oceans. Overall, the variations in water vapor channels associated with HR create a more humid environment and thicker warm-cloud layers, resulting in higher hydrometeor concentration and enhanced warm-rain coalescence, especially in WeCPEs that demonstrate a more maritime-like microphysical characteristic.

Conclusions

Our study elucidates the microphysical and environmental distinctions between precipitating CPEs producing relatively heavier and lighter rain rates, respectively (Fig. 5). This global view offers new insights into the mechanisms driving different convective rainfall intensities from a microphysical perspective. The HR-producing InCPEs can produce much more extreme rainfall due to more vigorous mixed-phase processes combined with a rough balance between raindrop breakup and coalescence in the liquid-phase processes, compared with HR-producing WeCPEs that are dominated solely by coalescence-driven warm-rain processes. Compared with LR-producing InCPEs, HR-producing InCPEs tend to exhibit stronger convective intensity. Below the melting level (approximately 5 km altitude), the transition from breakup to coalescence plays a key role in increasing rain rates in both WeCPEs and InCPEs. This transition is primarily driven by higher hydrometeor concentration resulting from higher environmental humidity. The HR-producing InCPEs and WeCPEs exhibit continental- and maritime-like characteristics, respectively, but with higher concentrations. Compared to those LR-producing CPEs with similar convective intensity, the higher humidity for HR-producing CPEs is caused by enhanced water vapor channels originating over oceans and is associated with larger horizontal scales of the CPEs themselves and their parent systems within which they are embedded. Notably, HR-producing WeCPEs show the most intense water vapor transport from adjacent oceans and thus the thickest warm-cloud layer and the most humid middle troposphere, compensating for the lack of strong mixed-phase processes for the HR production.

Fig. 5: Schematic diagram of the microphysical features for the HR-InCPEs, LR-InCPEs, HR-WeCPEs and LR-WeCPEs over the global convective hotspots.
figure 5

Moisture convergence anomalies are determined with respect to the average over all CPEs.

Full size image

Overall, our findings highlight the critical, joint role of microphysical processes and larger-scale atmospheric circulation in determining rain rates produced by convective storms worldwide. Moreover, by illustrating favorable regional moisture transport for HR-producing CPEs in global convective hotspots and comparing with projections of large-scale circulation changes in future climate, this study promotes awareness of the potential risks of increasing heavy precipitation over southern Asia and the maritime continent islands posed by future situations. This knowledge is essential for developing more accurate satellite-based rainfall retrieval algorithms and weather prediction systems, and also for better preparing for climate change-related challenges globally.

As this study is based on idealized model simulations, the analysis primarily focuses on the role of moisture in modulating rainfall efficiency. However, other factors—such as boundary layer depth, the strength of convergence lines associated with local temperature gradients and potentially aerosol—also modulate how efficiently a convective cloud transforms its energy into rainfall47,48,49,50,51. These processes merit further investigation in future studies using more comprehensive modeling frameworks.

Methods

Data

The Tropical Rainfall Measuring Mission (TRMM) version 8 data31 from 1998 to 2014 are employed, including observations from the Ku band precipitation radar, the microwave imager and the lightning imaging sensor covering global tropics and subtropics (from 36°S to 36°N). Also employed are the dual-frequency precipitation radar data of Global Precipitation Measurement version 06 A (GPM)32 from 2014 to 2020. As the successor of TRMM, the GPM has a wider latitude range from 52°S to 66°N and provides estimates of parameters of the DSD. While TRMM and dual-frequency GPM retrievals are subject to known uncertainties, particularly in heavy convective rain or snowfall52,53, we mitigate these issues by employing both datasets for cross-validation and focusing on global warm-season events to reduce snowfall-related and seasonal biases.

Only the maximum reflectivity and rain rate profiles of CPEs are used for both TRMM and GPM, with the DSD data selected from the corresponding maximum pixels for consistency. Moreover, the Z–M relationship27,54 is used to calculate the vertical profiles of liquid water content (LWC) and ice water content (IWC) using the reflectivity:

$${{\rm{LWC}}}=3.44\times {10}^{-3}{{{\rm{MAXREF}}}}^{4/7}{{\rm{g}}}{{{\rm{m}}}}^{-3}$$
(1)
$${{\rm{IWC}}}=1000\pi {\rho }_{i}{N}_{0}^{3/7}{\left(\frac{5.28\times {10}^{-18}}{720}{{\rm{MAXREF}}}\right)}^{4/7}{{\rm{g}}}{{{\rm{m}}}}^{-3}$$
(2)

Here \({\rho }_{i}\) is the ice density (917 kg m−3) and \({N}_{0}\) is the intercept index (4 × 106 m−4). It should be noted that the retrievals of LWC and IWC do not explicitly separate ice and rain contributions, potentially introducing bias in the mixed-phase layer (between about 5–7 km), where IWC may be overestimated and LWC may be underestimated55. While differences arise from warm-rain processes below 5 km, this limitation has few impacts on our conclusions.

To investigate the environmental conditions for CPEs, we use the hourly data on pressure levels from ERA533, including the specific humidity, horizontal winds, and temperature. The ERA5 data are interpolated onto the center positions of CPEs at the nearest time.

Convective precipitation event

In the study, the analyses are event-based using the TRMM/GPM precipitation feature dataset56. We extract the convective precipitation events (CPEs) consisting of more than four convective-type pixels from the precipitation features during warm seasons (April–September for the northern hemisphere; October–March for the southern hemisphere). MAXHT30 is used as a measure of the convective intensity14.

The TRMM CPEs are classified into three categories by convective intensity14: intense CPEs (InCPEs; MAXHT30 > 9 km), moderate CPEs (MoCPEs; 5.5 ≤ MAXHT30 ≤ km) and weak CPEs (WeCPEs; MAXHT30 < 5.5 km). The height of 9 km (5.5 km) corresponds to approximately −30 °C to −25 °C (−5 °C to 0 °C), thus indicating a deep (shallow) mixed-phase layer. Note that, compared with TRMM, GPM shows a lower frequency of CPEs with MAXHT30 below 4 km (Fig. S12 in Supplementary), possibly related to the sidelobe clutter in GPM and the beam-mismatch in TRMM57,58. According to the cumulative distribution functions of MAXHT30 (Fig. S12), the heights of 9 km and 5.5 km correspond to the 95th and 79th percentiles for TRMM, while for GPM, the corresponding heights for the same percentiles are 11 km and 7.5 km. Therefore, to match the percentiles between TRMM and GPM, we use 11 km and 7.5 km to define the three convective-intensity categories of CPEs in GPM. This use of MAXHT30 thresholds does not influence our major conclusions, as this study focuses on comparisons between HR- and LR-producing CPEs with comparable convective intensity, rather than discussions of absolute values of the associated metrics. Note that TRMM/GPM snapshots capture convective systems at various life cycle stages, resulting in MAXHT30 distributions (Fig. S12) that reflect a mix of shallow cumulus, congestus, and deep cumulonimbus clouds59. This may introduce some variability in the convective and microphysical statistics.

Due to the poor detection performance of the Precipitation Radar aboard TRMM for light rain rates60, there are significant inconsistencies between TRMM and GPM in the identification of precipitation events with maximal near-surface rain rate (MAXRR) below 10 mm/h61. Moreover, the sample size of GPM InCPEs with a MAXRR of <10 mm/h is limited, especially over southeastern China and the maritime continent islands (Fig. S13a). Therefore, only results from CPEs with a MAXRR exceeding 10 mm/h are presented in the main text. Note that the contrasts of vertical reflectivity profiles between the heavier and lighter convective rainfall remain consistent even with these much lighter rainfall events being included (Fig. S13b–e). Moreover, each convective-intensity category of CPE is further classified into three subgroups based on MAXRR intensity (Table S1): high rain (HR; MAXRR > 60 mm/h), medium rain (MR; 30 ≤ MAXRR ≤ 60 mm/h) and low rain (LR; 10 ≤ MAXRR < 30 mm/h), using the 65th and 85th percentiles of the global sample (i.e., 30 and 60 mm/h, respectively).

Design of numerical simulations

The microphysical discriminations among the different CPE types are also investigated using the advanced version of Cloud Model 1 (CM1; version cm1r21.0)62. The CM1 is a three-dimensional, time-dependent and nonhydrostatic model utilized extensively for idealized simulations, especially of precipitating convection. The model is configured with a domain of 100 km × 100 km and a grid spacing of 200 m in the horizontal, and the open radiative lateral boundary conditions are used. In the vertical, there are 117 levels with a spacing of 150 m. Four sets of ensemble simulations (CASE1, CASE2, CASE3, and CASE4) in a 60-min period are conducted, using the double-moment Morrison microphysics scheme63. The environmental conditions are initialized using modified soundings (Supplementary Discussion 2 and Fig. S10) in the inflow regions of two storms producing extreme hourly rainfall over a typical monsoon coast (South China) but with different convective intensities for CASE1 and CASE3, a supercell case over the central Oklahoma of USA for CASE2, and a shallow convection case from the Shaowu station (58725) in southeast China for CASE4.

All simulations are initialized with a warm bubble located 1.5 km above ground level with a vertical radius of 1.5 km and a horizontal radius of 10 or 15 km. Each radius configuration includes 11 ensemble members, by adding an initial perturbation of the warm bubble from −0.5 to 0.5 K to the ensemble mean of 2 K, resulting in a total of 22 ensemble members to offer more complete information concerning the possible uncertainty of the initial state than a single simulation. Other configurations applied in the simulations are provided in Table S2 of the Supplementary.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.