In order to simulate trace gases and particulates interacting with the meteorological fields, the Weather Research and Forecasting model coupled with a chemistry module (WRF-Chem; see Grell et al., 2005) version 3.6.1 is used in this study. Within WRF-Chem, the Regional Acid Deposition Model, version 2 (RADM2), photochemical mechanism (Stockwell et al., 1997) coupled with the Modal Aerosol Dynamics Model for Europe (MADE) and the Secondary Organic Aerosol Model (SORGAM) (Ackermann et al., 1998; Schell et al., 2001) are included to simulate atmospheric chemistry and anthropogenic aerosol evolutions. MADE/SORGAM uses a modal approach to represent the aerosol size distribution and predicts mass and number concentrations of three aerosol modes (Aiken, accumulation, and coarse).

To resolve the convective system over the Maritime Continent in our simulations, two model domains with two-way nesting are designed. Here, Domain 1 (431×141 grid cells) has a resolution of 25 km, while Domain 2 (561×591 grid cells) has a resolution of 5 km (Fig. 1). Specifically, Domain 1 is positioned to include the tropical Indian Ocean in its western half in order to capture the path of the Madden–Julian Oscillation (MJO) and, at the same time, have a northern boundary constrained at 23∘ N in latitude to avoid potential numerical instability from the terrain of the Tibetan Plateau. Domain 2, which has a finer resolution, is positioned to cover mainland Southeast Asia as well as the islands of Sumatra and Borneo.

National Center for Environment Prediction FiNaL (NCEP-FNL) reanalysis data (National Centers for Environmental Prediction, 2000) are used to provide initial and boundary meteorological conditions and to perform four-dimensional data assimilation (FDDA) to nudge model temperature, water vapor, and zonal and meridional wind speeds above the planetary boundary layer (PBL) for Domain 1. The time frequency of nudging is every 6 h. The Mellor–Yamada–Nakanishi–Niino level 2.5 (MYNN) scheme (Nakanishi and Niino, 2009) is chosen to represent the planetary boundary layer in this study. Other physics schemes adopted in the simulations include the Morrison two-moment microphysics scheme (Morrison et al., 2009), the Rapid Radiative Transfer Model for GCMs (RRTMG) longwave and shortwave radiation schemes (Mlawer et al., 1997; Iacono et al., 2008), the Unified Noah land-surface scheme (Tewari et al., 2004), and the Grell–Freitas ensemble cumulus scheme (Grell and Freitas, 2014) (for Domain 1 only). As the main purpose of this study is to reveal fire aerosol–convection interaction through modeling a large quantity of convective systems continually over a relatively long period and taking into account the computational resources available to us, we have adopted a 5 km horizontal resolution that excludes the cumulus parameterization scheme. Previous studies have shown that a WRF model with a similar resolution without convection parameterization can still capture many critical characteristics of deep convection (Wagner et al., 2018). Our model evaluation, especially through the comparison of modeled results with sounding profiles, has demonstrated the same.

WRF-Chem needs emissions for gaseous and particulate precursors to drive its simulations. For this purpose, we have used the Regional Emission inventory in ASia (REAS) version 2.1 (Kurokawa et al., 2013). REAS includes emissions of most primary air pollutants and greenhouse gases, covering each month from 2000 to 2008. In addition, the Fire INventory from the U.S. National Center for Atmospheric Research (NCAR) version 1.5 (FINNv1.5) (Wiedinmyer et al., 2011) is also used in the study to provide biomass burning emissions. FINNv1.5 separately classifies burning of extratropical forest, tropical forest (including peatland), savanna, and grassland. Fire heat fluxes for four different types of fire are prescribed in WRF-Chem to calculate plume height (cf. Table 1 in Freitas et al., 2007). For peatland fire, we have set its heat flux as 4.4 kW m-2, which is the same as that of savanna burning and differs from that of the tropical forest burning by 30 kW m-2.  The plume rise algorithm in WRF-Chem, specifically modified to improve the representation of tropical peat fire, was described in Lee et al. (2017). It is worth indicating that the heat flux from biomass burning is not incorporated in the thermodynamic equation of the current WRF-Chem model. Note that the current fire emission inventories could underestimate near-surface fire aerosol concentration by ignoring some of the characteristics of smoldering burning as well (Shi et al., 2019).

The default chemical profiles of several species in the lateral boundary condition are higher than their background concentrations in our study region and are thus equivalent for providing additional aerosol sources from boundaries. To prevent this, we have set NO, NO2, SO2, and all primary aerosol levels to zero at the lateral boundaries of Domain 1. We have also adjusted the ozone profile used for lateral boundary condition based on the World Meteorological Organization (WMO) Global Atmosphere Watch (GAW) station in Bukit Kototabang, Indonesia (Lee et al., 2019).

Two numerical simulations, both including anthropogenic emissions (mainly fossil fuel emissions) either with and without the biomass burning emissions (hereafter referred to as FFBB and FF, respectively), have been conducted to investigate the impacts of biomass burning aerosols on convective systems over the Maritime Continent through both direct and indirect effects. Our study focuses on the fire season from June to September 2008. Therefore, the simulations start from 1 May 2008 and last for 5 months. The first month is used as a spin-up period. Among the years with available emission data, both emission amount from biomass burning and total precipitation in 2008 approximate their ensemble mean or represent an average condition (Fig. S1 in the Supplement). Nevertheless, interannual variation in biomass burning emissions alongside precipitation in the studied regions do exist (Lee et al., 2017, 2018), and the influence of such variation on the effects of fire aerosol on convection should be addressed in future studies.

The primary target of this study is the convective systems associated with Sumatra squall lines and diurnal rainfall over Borneo. Thus, our analyses mainly focus on the convections over two specific regions: the Sumatra region (r1 in Fig. 1) and the Borneo region (r2 in Fig. 1). The area coverage of the Sumatra region (r1) is from 0 to 6∘ N and 97 to 103∘ E, while the area coverage of the Borneo region (r2) is from 1∘ S to 5∘ N and 109 to 115∘ E.

To examine the impacts of fire aerosols on cloud formation and rainfall intensity and amount, we have selected three convective systems each for the two focus regions to perform an in-depth case study. We first trace the path of individual convections and focus the analyses on the specific area of each of these convective systems to identify the impacts of fire aerosols. Table 1 shows the selected cases in the Sumatra region (r1) and the Borneo region (r2). The selected cases are chosen randomly from different fire periods of the two study regions. We did not set any criteria initially when we chose these cases. After we analyzed all cases, 3 mm 3 h-1 was set as the threshold to distinguish weak and strong convections.

The consequent analyses are then focused on the fire-season-wise statistics of convections for each study region. Table 2 and Fig. S2 show the fire periods in the two study regions. There is a total of 54 convective systems simulated during the fire periods in the Sumatra region (r1) and 35 convective systems in the Borneo region (r2).

The statistical quantities used in this study follow Wang (2005) to estimate the mean value over a specific region (e.g., r1 or r2). The cloud area mean quantities are defined as a function of output time step (t) by the following equation: 1c‾areat=1N(t)∑q>qmin⁡n>nmin⁡cx,y,z,t. Here c is a given quantity (e.g., cloud water mass). Equation (1) only applies to the grid points where both the mass concentration q and number concentration n of a hydrometeor exceed their given minima. The total number of these grid points at a given output time step t is represented by N(t). The cloud area mean quantities are used to present the average quantities of a given variable at a given output time step. Note that the cloud area mean quantities only apply to hydrometeors. For rainfall, the analyzed quantities are spatial averages over a specific area of the convective system for the case study or over the entire study region for longer-term statistical estimates.

Statistics covering the entire simulated fire season (∼4 months) for each study region have been derived to provide trend and tendency information regarding several aspects of the impact of fire aerosols on convections. In our simulations, PM2.5 concentration in FF during the fire periods, which can be regarded as the background value for FFBB simulation before adding fire aerosols, is 1.36±0.19 µg m-3 in r1 and 0.56±0.09 µg m-3 in r2. In comparison, PM2.5 concentration in FFBB is 11.37±10.41 µg m-3 in r1 and 10.07±7.73 µg m-3 in r2. Note that unlike in some other studies where the control simulations use constant aerosol concentrations, fire aerosol concentrations in our simulations can vary in response to changes in fire emissions or aerosol removal by rain scavenging due to precipitation change caused by fire aerosols themselves. Hence, the processes included in our simulations are closer to reality, and the results could better reflect the nature of fire aerosol–convection interaction over the Maritime Continent.

Averaged throughout the entire modeled fire period, cloud water mass (Qc), cloud droplet number concentration (Qnc), and raindrop number concentration (Qnr) in FFBB differ substantially from those in FF, demonstrating the influence of fire aerosols. Figure 8 shows that adding fire aerosols in FFBB would increase Qc by 14 % and Qnc by 226 % in r1 and Qc by 18 % and Qnc by 349 % in r2. Another pronounced change in response to adding fire aerosols is a decrease in Qnr by 44 % in r1 and 47 % in r2. Although this shows an increase in snow mass (Qs) and graupel mass (Qg) and a decrease in rainwater mass (Qr) after adding fire aerosols, the uncertainty of these hydrometeor changes is large.

In Sect. 3.2, we discussed the significant rainfall increase that occurred in weak convective systems after adding fire aerosols due to the aerosol invigoration effect. On one hand, regardless of the strength of convection, the mean 3-hourly rainfall during the fire periods is 1.06±0.85 mm in FF and 1.09±0.86 mm in FFBB over the Sumatra region (r1) and statistically does not change significantly when responding to fire aerosols. The rainfall difference in the Borneo region (r2) between FF and FFBB is also insignificant (1.32±1.20 mm 3 h-1 in FF versus 1.35±1.14 mm 3 h-1 in FFBB). On the other hand, we have found that the impacts of fire aerosols appear in several other rainfall patterns. For instance, the daily maximum and minimum rainfalls display clear differences between the FFBB and FF simulations, specifically in r2 rather than in r1 (Fig. 9). For r1, the impacts of fire aerosol are reflected in event-wise statistics, e.g., higher event-wise maximum and minimum rainfall intensity in FFBB than in FF, identified in 30 out of 54 convective events in total. These are mostly weak convective events in r1. Interestingly, somewhat opposite to the rainfall statistics in r1, the intensity of event-wise maximum and minimum rainfall in r2 is higher in FF than in FFBB. The daily rainfall peak of 3 h rainfall in r1 is mostly less than 3 mm; in comparison, one-third of convective events in r2 have daily maximum 3 h rainfall exceeding 3 mm (Fig. 9c), suggesting that the convective systems in r2 tend to develop stronger than in r1 and the fire aerosols significantly suppress the maximum rainfall intensity of strong convections in r1. We roughly used 1.25 mm 3 h-1 of the domain-averaged rainfall to classify weak and strong convective systems. We find that the conclusions regarding differences between hydrometers and rainfall in weak systems between the FF and FFBB experiments stay the same and that such differences are still not that significant in both regions (Table S1 and Fig. S8).

We have categorized the maximum rainfall based on its values in the afternoon and at midnight. We find that the heavy maximum rainfalls in r2 that tend to occur at midnight (Fig. 9c) are associated with the anticyclonic circulation formed in western Borneo induced by southeasterly winds from southern latitudes and turn northeastward along the west coast of Borneo, owing to the terrain of Borneo and sea breeze from the South China Sea. The vortex produced by such a circulation leads to strong updraft and then strong convection. Note that this anticyclonic circulation is different from the Borneo vortex, the latter appears as a persistent feature of the boreal winter climatology and is related to the northeasterly from the South China Sea and cold surge events (Chang et al., 1983, 2005).

The low-level wind pattern of Borneo convections is similar to the westerly regime, especially the weak westerly (WW) regime identified by Ichikawa and Yasunari (2006). According to their analysis, the WW regime tends to occur in boreal summer. Its composites include an anticyclonic feature with the weak wind field over Borneo. The deep convective storms developed in the WW regime tend to stay close to the west coast, which is associated with the lower-level convergence enhanced by the prevailing wind and local circulations around that area, resulting in localized rainfall over the offshore region of the west coast. Based on our simulations, the onset of convection occurs in the afternoon over the western mountain range of Borneo. These storms would consequently evolve into widespread shallow storms in the evening over the western part of the island. The maximum rainfall appears on the west coast because of a local westward propagating rainfall system that develops around midnight or in the early morning.

The comparison of the maximum rainfall between FF and FFBB in Fig. 9 shows that fire aerosols tend to reduce the maximum rainfall, especially for high-intensity rainfall events. In other words, fire aerosols have substantial impacts on the nocturnal convections, which are associated with the local anticyclonic circulation in western Borneo. This effect on nocturnal convections in western Borneo by fire aerosols will be discussed further in the next section.

To further analyze the effects of fire aerosols on nocturnal convections, we have categorized convective events into nocturnal convections (NC) and non-nocturnal convections (non-NC), based on whether the maximum rainfall occurs from midnight to the early morning or from the late afternoon to the evening. Figure 10 shows the diurnal time series of precipitation averaged over the Borneo region (r2) in FF and FFBB. Again, 3 h mean rainfalls of nocturnal convections are higher than those of non-nocturnal convections in both simulations and fire aerosols weaken the maximum nocturnal rainfall intensity by about 9 %.

Nocturnal convections tend to stay close to the west coast, associated with a lower-level convergence enhanced by the prevailing wind and local circulations that are mainly related to the land breezes from inland western Borneo. The strong convergence near the surface over the offshore region of the west coast causes the weak westerly monsoon windflaws and local land breezes to merge during the nighttime. However, during the fire periods, the daytime absorption of fire aerosols (e.g., black carbon) can cause an atmospheric warming (even without fire-generated heating flux being incorporated in the model). This could increase near-surface air temperature and weaken land breezes and surface convergence. As a result, the nocturnal convections in FFBB cannot develop as strongly as those in FF. On the other hand, both nocturnal and non-nocturnal convections are initiated over the western mountain range under a prevailing wind of the sea breezes from the South China Sea. The increases in near-surface temperature owing to the fire aerosols can enhance this prevailing wind from the ocean and thus lead to a higher convective rainfall in FFBB during the onset stage of the nocturnal convections as well as non-nocturnal convections. Figure 11 illustrates the sea breeze increase in FFBB during the daytime (20:00 LT) and the land breeze decrease in FFBB during the nighttime (02:00 LT). The times of 20:00 LT and 02:00 LT are chosen here because of the peak of difference.

Diurnal evolution of vertical profiles clearly indicates that mass mixing ratio of total hydrometeors, temperature, and vertical velocity differ in both daytime and nighttime between FF and FFBB for those nocturnal convections (Fig. 12). The differences of near-surface temperature between FF and FFBB are more pronounced during the period after sunset (Fig. 12d). The differences of near-surface temperature mainly happen over land, and the higher near-surface temperature in FFBB weakens the land breezes and near-surface convergence along the coast. Starting from late afternoon (about 17:00 LT), vertical velocity increases with time until sunrise the next day in both simulations (Fig. 12e), due to the convergence of the monsoon windflaws and local land breezes during the nighttime, which matches very well with the of mass mixing ratio of total hydrometeors (Fig. 12a and e). Noticeably, the main differences in vertical velocity and hydrometeor mass mixing ratio between FFBB and FF also start to become evident in the evening. Because of the weaker convergence near the surface in FFBB, the differences in vertical velocity at the higher altitude between FFBB and FF peak at nighttime. The temperature increase from aerosol absorption seems small (please note that the direct heating from fire is not included in the WRF fire plume model), but we do see the change of vertical velocity owing to the aerosol heating effect. Based on our analysis, the temperature increase is mainly associated with the thermodynamic perturbation from the absorption of sunlight by fire aerosols. This also seems consistent with the analysis of Zhang et al. (2019). Indeed, should the heat flux generated by fires be incorporated in the model, the warming effects from biomass burning would be much stronger and also persist in a nocturnal time frame.

As a summary, the schematics shown in Fig. 13 illustrate the impact of biomass burning activities on nocturnal convections in the Borneo region. In the daytime, under the prevailing wind of sea breezes from the South China Sea, convections develop over the western mountain range. Because near-surface heating from the absorption of sunlight by fire aerosols could enhance the prevailing wind from the ocean, convective rainfall becomes higher at the onset stage of the nocturnal convections (still during daytime) due to biomass burning activities (Fig. 13b). At nighttime, convection moves to the offshore region of western Borneo. The strong convergences near the surface merge the weak westerly monsoon windflaws with local nighttime land breezes to form an anticyclonic circulation (Fig. 13c). During the fire periods, the daytime near-surface warming by fire aerosols could also further weaken land breezes and surface convergence. Hence, the nocturnal convections during fire events would not develop as strongly as in days without fires (Fig. 13d versus 13c).