Four types of datasets from the years 2002 to 2012 (11 years) are used in this study, which include (1) precipitation, (2) aerosols, (3) clouds, and (4) other meteorological fields.

We used both station data of gauge-based precipitation and gridded data of aerosols, clouds and other meteorological variables. Gridded datasets in this study were downloaded with the horizontal resolution of 1∘×1∘, which is consistent with the resolution of MODIS Level-3 products. To unify the datasets, we interpolated all the gridded datasets onto the selected 176 rainfall stations using the average value in a 1∘×1∘ grid as the background condition of each rainfall station; i.e., the stations in the same 1∘×1∘ grid have the same aerosol, cloud and meteorological conditions.

Our previous study (Zhou et al., 2018) reported the distinct peak shift of heavy rainfall diurnal variation between clean and polluted days using the indicator of AOD over the BTH region during early summer. Similarly to our previous study, the PDF of the heavy rainfall peak time shows that the maximum of the rainfall peak is about 2 h earlier on the polluted days (20:00 LST) than that on the clean days (22:00 LST) (Fig. 3a). To comprehensively recognize the changes in rainfall diurnal variation associated with air qualities, here we examined the PDF of the start time, the duration and the intensity besides the peak time of heavy rainfall.

As shown in Fig. 3a, the start time of heavy rainfall exhibits a significant advance on the polluted days. The secondary peak on the early morning is ignored here because the early-morning rainfall is usually associated with the mountain winds (Wolyn et al., 1994; Li et al., 2016) and the nighttime low-level jet (Higgins et al., 1997; Liu et al., 2012) that is beyond the scope of this study. The time for the maximum frequency of heavy rainfall initiation is around 6 h earlier on the polluted days, shifting from around 00:00 LST on the clean days to 18:00 LST (Fig. 3a). Regarding the rainfall durations, the average persistence of heavy rainfall on polluted days is 0.8 h longer than that on clean days (Table 2). According to the PDF shown as in Fig. 3a, the occurrence of short-term precipitation (≤ 6 h, Yuan et al., 2010) decreases, while that of long-term precipitation (> 6 h, Yuan et al., 2010) increases. The intensity of hourly rainfall exhibits a insignificant increase on the polluted days.

The distinct behaviors of heavy rainfall diurnal variation between clean and polluted days have been well demonstrated using the indicator of AOD. Using CDNC as the indicator of CCN, the above-mentioned results are also significant, as shown in Fig. 3b. The start time and peak time of heavy rainfall on the polluted condition also show significant advances compared with that on the clean condition, with the average advances of 2.2 and 2.6 h, respectively (Table 2). The duration of heavy rainfall on the polluted condition is also prolonged, which is 0.5 h longer on average (Table 2). Similarly to the results based on AOD, the difference of rainfall intensity between clean and polluted conditions using CDNC does not pass the 95 % statistical confidence level either.

Hence, the results using either AOD or CDNC show that the start and peak times of heavy rainfall occur earlier and that the duration becomes longer under pollution. We found that the AOD and CDNC only have an insignificant positive correlation, which denotes that the selected cases could be different between using AOD and CDNC. The differences between the two indicators might be attributed to the non-linear relationship between CCN and aerosol pollution (e.g., Jiang et al., 2016), the misdetection of AOD when the humidity is high (Boucher and Quaas, 2012), the calculation uncertainty of CDNC, and the sampling differences between AOD and CDNC. Since the two indicators represent aerosols from the different perspectives, we cannot identify which one is more reliable. Because the change in rainfall intensity is not significant based on either AOD or CDNC, the following analysis only focuses on studying the changes in start time, peak time and duration of heavy rainfall along with aerosol pollution.

Using the indicator of AI, we further investigated the distinct behaviors of heavy rainfall diurnal variation related to absorbing aerosols and scattering aerosols, respectively. The PDFs of start time, peak time and duration of heavy rainfall under the extreme circumstances of absorbing aerosols and scattering aerosols are compared in Fig. 4. Here, we briefly named the days with extremely large amounts of absorbing aerosols as absorbing aerosol days and with more scattering aerosols as scattering aerosol days. The start time of heavy rainfall on absorbing aerosol days shows a significance earlier compared with that on scattering aerosol days (Fig. 4a), with a 0.7 h advance on average (Table 3). Similarly, the rainfall peak time also shows earlier on absorbing aerosol days (Fig. 4b), with an average advance of 1.6 h (Table 3). The rainfall duration on scattering aerosol days is longer than that on absorbing aerosol days, which are 6.0 and 5.0 h, respectively, on average (Table 3). All the above-mentioned differences between the two groups have passed the 95 % statistical confidence level. The results indicate that the absorbing aerosols and scattering aerosols may have different or inverse effects on the heavy rainfall that absorbing aerosols may generate the heavy rainfall in advance, while the scattering aerosols may delay and prolong the heavy rainfall.

To further verify the different behaviors of heavy rainfall diurnal variation associated with two different types of aerosols, we purposely re-examine the above-mentioned phenomena using BC/sulfate that can represent typical absorbing/scattering aerosols over the BTH region. BC has its maximum center over the BTH region (Fig. 5a) and our previous study has indicated that the radiative effect of BC low-level warming may facilitate the convective rainfall generation (Zhou et al., 2018). The percentage of sulfate is also large over the BTH region (Fig. 5b) and sulfate is one of the most effective CCN that influence the precipitation in this region (Gunthe et al., 2011). Accordingly, we selected the cases with different AOD of BC and sulfate to compare their roles in the diurnal variation of heavy rainfall. The methods have been described in Sect. 2.2.2. The PDFs of the start time, peak time and duration of heavy rainfall in the cases with more/less BC are shown in Fig. 6a, respectively. The most striking result is that the maximum frequency of rainfall start time in the more BC cases evidently shifts earlier (Fig. 6a). Meanwhile, the mean peak time in the more BC cases shows 1.1 h earlier than that in the less BC cases (Table 3). And the duration of heavy rainfall is slightly shortened by the averaged 0.2 h in the more BC cases. The features in the more BC cases are consistent with the above results of absorbing aerosols. In contrast, when the sulfate has a larger amount, the mean start time of rainfall is delayed by 0.5 h, while the duration shows a significant increase by 1.5 h on average (Table 3). The behaviors in the more sulfate cases are also similar to the above results of scattering aerosols, except for the peak time that shows later in the scattering aerosol cases but a little earlier in the more sulfate cases (Table 3).

Moisture supply is an indispensable factor for the precipitation formation, and it also has an important impact on AOD (Boucher and Quaas, 2012). Since the southwesterly circulation can transport not only pollutants, but also plenty of moisture to the BTH region (Wu et al., 2017), more pollution usually corresponds to more moisture for the BTH region (Sun et al., 2015), so that it is hard to completely remove the moisture effect on the above results in a pure observational study. Here we attempt to recognize the moisture effect on the heavy rainfall to further understand the above aerosol-associated changes. Because the moisture supply for BTH is mainly transported via low-level southwesterly circulation, we purposely used the SH at 850 hPa as the indicator of moisture condition.

Using the similar percentile method, we compared the heavy rainfall characteristics in the relatively wet (SH more than the 75th percentile) and relatively dry (SH less than the 25th percentile) environments regardless of the aerosol condition, as shown in Fig. 7a. The results show that the start time of heavy rainfall is delayed by 0.9 h, the peak time is 0.6 h earlier and the duration is prolonged by 2.0 h on average in the wet environment, which is similar to the results of the more sulfate cases. Besides, the same results are obtained using a different moisture indicator, e.g., the 850 hPa absolute humidity. These results indicate that the advance of a heavy rainfall start time on the polluted days is not caused by more moisture supply, while the longer duration and earlier peak in the more sulfate cases might be related to the increased moisture. To further identify the role of sulfate, we examined the sensitivities of the results associated with sulfate under a different moisture condition. In the dry (SH less than the 25th percentile) and intermediate cases (SH between the 25th and 75th percentiles), the heavy rainfall still shows a later start time, earlier peak and longer duration with the increase in sulfate, while the change in peak time is not significant in the dry cases; in the high-moisture cases (SH more than the 75th percentile), it shows an earlier peak and shorter duration in the more sulfate cases. Therefore, we suppose that the impact of sulfate aerosols on the heavy rainfall is sensitive to moisture, and notably the sulfate could contribute to the longer duration in the polluted cases when it is relatively dry.

We also investigate the distributions of moisture and rainfall behaviors in the clean and polluted cases, respectively, using AOD and CDNC (Fig. 7b, c). The results show that the relationship between moisture and rainfall start time/peak time/duration is not linear. The distribution of SH exhibits a slight increase with pollution in the AOD cases, indicating that the polluted cases selected by AOD are accompanied by more moisture than the clean cases. However, when fixing the moisture at a certain range, especially at the relatively dry condition (for example, the SH between 8 and 12 g kg-1), we can detect the similar phenomena of earlier start/peak time and longer duration in the polluted cases based on either AOD or CDNC. To further clarify the characteristics of heavy rainfall associated with pollution, we removed the samples with high SH (SH more than the 75th percentile) and found that the results in Sect. 3.1 remain; that is, the start/peak time of heavy rainfall is in advance and the duration is prolonged with the increase in AOD / CDNC when SH is less than 12.95 g kg-1 (75th percentile) (Fig. 8).

The above results indicate that the advance of heavy rainfall start in the polluted cases is independent of moisture condition, while the advance of peak time and longer duration could be influenced by the moisture effect. For the earlier peak time of heavy rainfall, we suppose the role of BC (absorbing aerosols) might be dominant because the change in peak time in the former analysis is more significant (Table 3), although the sulfate and moisture also have a positive contribution. The increased sulfate (scattering aerosols) contributes to the longer duration of heavy rainfall (Fig. 6b), but the role of sulfate is somewhat sensitive to the moisture condition. With the increase in sulfate, the duration is longer when the moisture condition is relatively dry, while it becomes shorter when it is extremely wet. Overall, when removing the extremely high moisture cases, the earlier start/peak time and longer duration of heavy rainfall associated with aerosol pollution are significant.

Using AOD as the macro aerosol indicator, as shown in Fig. 9, the PDF distribution shows that the CF on the polluted condition is evidently larger than that on the clean condition. The average CF is 62.8 % on the clean condition and 89.3 % on the polluted condition (Table 4). The average CTP on the polluted condition is 487.3 hPa, which is larger than 442.3 hPa on the clean condition, indicating that the cloud top height is lower on the polluted days. The COT, CWP and CER were further analyzed for the liquid and ice portions of clouds as shown in Fig. 9. Both liquid and ice COT on the polluted condition exhibit significant increases compared with that on the clean condition. The mean amount of liquid COT is increased by 3.1, and ice COT increases by 6.2 (Table 4). Similarly to COT, the amounts of liquid and ice CWP also increase under pollution, which increase by 33.6 and 88.2 g m-2, respectively. In addition, the liquid CER is increased by 0.8 µm and the ice CER is decreased by 2.8 µm on the polluted days. The differences of the above-cloud properties between clean and polluted cases have all passed the 95 % statistical confidence level.

Using CDNC as the micro aerosol indicator, the above-mentioned changes in cloud properties are consistent with that using AOD, except for liquid CER (Fig. 9). Since the calculation method of CDNC is not independent of the liquid COT and liquid CER, we would not directly compare the results of liquid COT and CER based on CDNC with those based on AOD here. But according to other variables that are independent of the CDNC calculation, we found that the cases with more CDNC are accompanied by the increase in CTP, ice COT and liquid and ice CWP, which increase by 90.2 hPa, 24.4, 112.4 and 224.1 g m-2, respectively (Table 4), all of which are consistent with the results based on AOD. The CER of ice clouds also shows a consistent decrease by 9.5 µm on the polluted condition based on CDNC. We noticed that the changes in COT/CWP/CER for both liquid and ice based on CDNC are much larger than that based on AOD, which indicates that these cloud properties might be more sensitive to the indicator of CDNC rather than AOD.

According to the above comparison, the concurrent changes in cloud properties along with heavy rainfall show consistent results using the two aerosol indicators (AOD and CDNC). The pollution corresponds to the increase in CF, ice COT, liquid and ice CWP but the decrease in cloud top height (the increase in CTP corresponds to the decrease in cloud top height) and ice CER. The liquid COT and liquid CER are also increased with the enhanced pollution in the AOD analysis. Besides, the above-mentioned results exhibit significance when we limited the moisture to the drier condition (SH less than the 25th percentile) or intermediate condition (SH between the 25th and 75th percentiles). When the moisture is higher (SH more than the 75th percentile), the change in CTP becomes insignificant based on CDNC.

According to these results, we made the following speculation: first, the CF, liquid and ice COT and CWP increase with pollution, because the aerosols serving as CCN can nucleate a larger number of cloud droplets which in a moisture-sufficient environment can hold more liquid water in the cloud. Second, the CTP increases (the cloud top height decreases) under pollution, because the earlier start of the precipitation process (Fig. 3) inhibits the vertical growth of clouds. Third, the ice CER decreases under pollution using either AOD or CDNC, because the increased cloud droplet number leads to more cloud droplets transforming into ice crystals and causes the decrease in ice CER (Chylek et al., 2006; Zhao et al., 2018; Gryspeerdt et al., 2018). However, the results of liquid CER might have uncertainties. The liquid CER is increased when AOD increases (Fig. 9), which might be related to the aerosol humidification effect, the misdetection of AOD and cloud water, and the earlier formation of the clouds and precipitation on the polluted days. Since we cannot distinguish the liquid part of mixed-phase clouds from liquid (warm) clouds in the observation, the above-mentioned change in liquid cloud properties might come from that of both the liquid (warm) clouds and the liquid part of mixed-phase clouds. Likewise, the above-mentioned change in ice cloud properties might come from that of both ice (cold) clouds and the ice part of mixed-phase clouds. Currently the physical processes of cold clouds and mixed-phase clouds have not been clarified yet, including the diffusional growth, accretion, riming and melting process of ice precipitation (Cheng et al., 2007, 2010), which needs numerical model simulations to be further explored.

Section 3.3 has shown that the diurnal variation of heavy rainfall with more moisture supply is similar to the changes in heavy rainfall with more sulfate aerosols. We assume that the moisture under the cloud base and the sulfate serving as CCN both influence the cloud properties (Yuan et al., 2008; Jiang et al., 2008; Jung et al., 2013; Qiu et al., 2017). To identify the effect of CCN on clouds and its sensitivity to moisture, using CDNC to represent CCN, we purposely investigated the changes in above-cloud properties on the different conditions of the CDNC and the low-level moisture (SH at 850 hPa), respectively.

We categorized all cases of heavy rainfall into four groups, which are (1) clean and dry, (2) polluted and dry, (3) clean and wet, and (4) polluted and wet, and checked the changes in the above-cloud properties, as shown in Table 5. To retrieve the comparable samples, here “clean/polluted” refers to the CDNC on that day less/more than the 25th/75th percentile of the CDNC among the heavy rainfall days, and similarly, “dry/wet” refers to the SH on that day less/more than the 25th/75th percentile of itself among the heavy rainfall days. The average CDNC is 125.54 cm-3 on the dry condition and 120.71 cm-3 on the wet condition, and the average SH is 11.62 and 11.73 g kg-1 on the clean and polluted conditions, respectively; thus, we consider the CDNC or SH to remain almost the same when the other condition changes. We tested the significance of differences between groups 1 and 2, groups 1 and 3, groups 2 and 4, and groups 3 and 4. Because the CF is fixed above 80 % when calculating the CDNC (see in Sect. 2.1.3), here the selected groups all belong to the condition of higher CF.

Comparing the results of groups 1 and 2, which are both on the dry condition, we can identify the influence of CDNC on the cloud properties, which represents the effect of CCN. The changes in these cloud variables are the same as that in Sect. 4.1, that the CF, ice COT and liquid and ice CWP are increased on the polluted condition while the cloud top height and ice CER are decreased based on CDNC. Among these variables, the ice COT and liquid and ice CWP are especially larger on the polluted condition, which are 3–5 times larger than that on the clean condition (Table 5). On the wet condition, comparing groups 3 and 4, the changes are similar in that the CF, ice COT and liquid and ice CWP are increased and the ice CER is decreased, but the change in CTP becomes insignificant. However, the changes in these variables on the dry condition are evidently greater than that on the wet condition, which indicates that these cloud properties might be more sensitive to CDNC on the dry condition. The above comparisons indicate that with the increase in CDNC (CCN), the CF, ice COT and liquid and ice CWP are increased, while the ice CER is decreased regardless of the moisture amount.

Comparing the results of groups 1 and 3, we can get the changes in cloud properties related only to moisture on the same clean condition. A common feature is that CF, CTP, COT and CWP both for liquid and ice exhibit increases along with the increase in moisture. Compared with the CTP on the clean and dry condition, it increases on both polluted and dry condition (group 2) and clean and wet condition (group 3), but on the former condition its increase is larger, which indicates that the influence of moisture on CTP might be secondary compared to the CDNC (CCN) effect. Similarly, comparing the COT/CWP in groups 2 and 3 to that in group 1, the increases in COT and CWP both for liquid and ice in group 2 are much larger than that in group 3, which indicates that the influences of moisture on COT and CWP may not overcome the influence of CCN. With the increase in moisture, the change in liquid CER is not significant on the same clean condition, but the ice CER is significantly decreased. On the polluted condition, comparing groups 2 and 4, we found the COT and CWP both for liquid and ice on the wet condition are evidently smaller than that on the dry condition, which indicates that increasing the moisture might partly compensate for the influence of CDNC (CCN) on COT/CWP. Besides, the liquid CER exhibits a slight increase with increased moisture in the same polluted environment, which may further support the idea that the increased CCN could nucleate more cloud water with increased moisture.

The results above indicate that both CDNC (CCN) and moisture have impacts on cloud properties. They both contribute to the increases in CF, CTP, COT and CWP, in which the influences of CDNC (CCN) on COT and CWP are significantly larger than moisture. The increases in both CDNC and moisture correspond to the significant decrease in ice CER, while only the increase in CDNC corresponds to the decrease in liquid CER, and that might be ascribed to the calculation method of CDNC. To reduce the uncertainties, we have tested the SH at different levels (e.g., 700 and 800 hPa) and different moisture indicators (e.g., absolute humidity) to verify these results and found that most cloud variables show similar changes, except for the CTP and the liquid CER, which indicates that the changes in CTP and liquid CER are more sensitive and have larger uncertainties. Since the behaviors of cloud changes are similar along with the increase in either CDNC (CCN) or moisture but more sensitive to the former, the results in Sect. 4.1 might actually reflect the combined effect of CCN and moisture, and the aerosol (CCN) effect on these cloud properties might be dominant on the polluted days.

Therefore, considering the results from this subsection and Sect. 3.3 that with the increase in aerosols, the changes in cloud features become smaller in the higher moisture environment than that in the drier environment and the duration of heavy rainfall is relatively shortened when it is extremely wet (Sect. 3.3), we speculate that the sulfate (CCN) effect might be suppressed in a relatively wet environment. Due to the limitations of observational study, we currently cannot figure out the respective roles of aerosols and moisture.

In this study we used two aerosol indicators, AOD and CDNC, which discriminate the pollution levels for different purposes. AOD is a good proxy for the large-scale pollution level, but it stands for the optical feature of aerosols and cannot well represent CCN when we focused on the aerosol–cloud interaction (Shinozuka et al., 2015). CDNC is a better proxy for CCN compared with AOD, which facilitates the study on the cloud changes associated with aerosol pollution. But the retrieved CDNC has larger uncertainties. First, the assumptions in the calculation of CDNC are idealized that CDNC is constant with height in a cloud and cloud liquid water increases monotonically in an adiabatic environment (Grosvenor et al., 2018), but the target of this study is the convective clouds with rainfall that may be not consistent with the adiabatic assumption. Second, as indicated by Grosvenor et al. (2018), the uncertainties in the pixel-level retrievals of CDNC from MODIS with 1∘×1∘ spatial resolution can be above 54 %, which come from the uncertainties of parameters and the original COT and CER data used in the calculation, and also the influence of the heterogeneity effect from thin clouds. To reduce the influence of the heterogeneity effect as much as possible, we have attempted to limit the conditions of CF, liquid COT and CER when calculating CDNC in the study. Besides, this study primarily focuses on the relative changes in CDNC, which may also be influenced by the potential systematic biases in the CDNC calculation, but actually reduced the uncertainties of absolute values. Another problem with CDNC in this study is that the CDNC could be influenced by updraft velocity because both increased CCN and updraft velocity could enhance aerosol activation and increase CDNC (Reutter et al., 2009). Since we cannot get any in-cloud long-term updraft data, we used the vertical velocity at 850 hPa obtained from ERA-Interim reanalysis data to roughly represent the cloud base updraft and investigated the possible relationship between CDNC and updraft. The results show that there is no significant correlation between CDNC and vertical velocity, although the updraft is relatively intensified in the polluted cases. We also examined the change in rainfall based on CDNC under three certain ranges of vertical velocity (less than the 25th percentile, between the 25th and 75th percentiles and more than the 75th percentile) and found that the primary results are similar.

In addition to AOD and CDNC, we also applied ultraviolet AI and AOD of BC/sulfate to identify different types of aerosols. We found that the AI has a weak positive correlation with AOD from MODIS, which indicates the results on absorbing aerosol days might represent the results on polluted days if identified by AOD. To avoid the uncertainty, we re-examined the results using AI when removing the polluted cases identified by AOD and found that the major results remain. The comparisons of BC/sulfate AOD cases also have uncertainties because they are retrieved from MACC reanalysis data. Although the above four indicators have their own uncertainties, currently we cannot find more reliable datasets in a long-term observational record. The major findings using these four indices could well identify the changes in rainfall and clouds accompanied by aerosols, but are insufficient to clarify the aerosol effect on clouds and precipitation.

This study has clearly identified the relationship of the aerosol pollution and the diurnal changes in heavy rainfall and associated clouds over the BTH region. However, although this work has attempted to exclude the impacts from the meteorological background, particularly circulation and moisture, the observation study still has its limitations on studying aerosol effects on rainfall and clouds: first, the observational datasets have their noise and uncertainty, including the misdetection of CF in the satellite product when AOD is large (Brennan et al., 2005; Levy et al., 2013) and the mutual interference between liquid and ice clouds (Holz et al., 2008; Platnick et al., 2017); second, the meteorological co-variations cannot be completely removed, thus bringing uncertainties of the results: e.g., the meteorology might affect the relationship between AOD and CF (Quaas et al., 2010; Grandey et al., 2013) and the relationship between AOD and CTP (Gryspeerdt et al., 2014a); third, the different types of aerosols cannot be completely well separated, although we used the AI index and AOD of BC/sulfate to identify the respective effects of absorbing aerosols and scattering aerosols. In addition, we selected the extreme ranges of AOD / CDNC to compare the characteristics of heavy rainfall and associated clouds, which could bring such uncertainties that these extreme conditions might be related to distinct microphysical processes or meteorological backgrounds. We further examined the results using the middle range of AOD and CDNC such as 25th–50th percentile versus 50th–75th percentile. The results are basically the same except that the peak time change is not significant based on AOD. Numerical model simulations are necessarily applied to further study on the specific impact of aerosols on the heavy rainfall. And the detailed processes of aerosol effect on the precipitation formation of mixed-phase and cold clouds also needs further exploration in our future study.

Using the gauge-based hourly rainfall records, aerosol and cloud satellite products and high temporal resolution reanalysis datasets during 2002–2012, this study investigated the different characteristics of heavy rainfall on the diurnal timescale on the clean and polluted conditions, respectively. Based on the macro and micro aerosol indicators, including AOD from the MODIS aerosol product and calculated CDNC from the MODIS cloud product, three significant features of heavy rainfall diurnal change associated with aerosols are found; that is, the rainfall start and peak times occur earlier and the duration becomes longer under pollution.

The different relationships of absorbing/scattering aerosols and the heavy rainfall diurnal changes were distinguishable using ultraviolet AI from OMI and reanalysis AOD of two aerosol types (BC and sulfate). The absorbing aerosols (BC) correspond to the earlier start and peak times of heavy rainfall, while the scattering aerosols (sulfate) correspond to the delayed start time and the longer duration. Considering the plausible effect of moisture, further analysis indicates that the duration of heavy rainfall in the presence of more sulfate is prolonged on the relatively dry condition, but is shortened on the extremely wet condition.

By comparing the characteristics of cloud macrophysics and microphysics variables, using both AOD and CDNC we found that the CF, ice COT, and liquid and ice CWP are increased on the polluted condition but that the cloud top height and the ice CER are reduced. Liquid COT and liquid CER are also increased in the AOD analysis. Comparing the influences of CDNC, which represents CCN, and SH at 850 hPa, which represents moisture, respectively, on these cloud variables, the cloud properties show consistent changes with the increase in CDNC and moisture, but are more sensitive to the CDNC (CCN).

According to these results, we speculate that both aerosol radiative effect and cloud effect have impacts on the diurnal variation of heavy rainfall over the BTH region. The heating effect of absorbing aerosols, especially BC, increases the instability of the lower to middle atmosphere so that it generates the heavy rainfall occurrence in advance. And with the sufficient moisture supply, the increased aerosols could nucleate more liquid water in the cloud, leading to the longer duration of heavy rainfall.