The VR grid, which refines to 0.125∘ horizontal grid spacing (∼14 km) across south-central Asia, transitions to 0.25∘, followed by 0.5∘ and eventually 1∘ (Fig. 1a). The region of the analysis grid that is characterized by horizontal grid spacings less than 1∘ is located approximately between 60 and 120∘ E and 5 and 55∘ N. This encompasses all of India, the Bay of Bengal, and the Arabian Sea. Most of India and the Bay of Bengal are characterized by grid resolutions greater than 0.125∘. As in Zarzycki et al. (2013), the spectral element dynamic core is used in the Community Atmosphere Model, the atmospheric component of CESM, to solve the primitive hydrostatic equations on a fully unstructured quadrilateral mesh (Dennis et al., 2011). In addition to atmospheric fields, land surface fields are also treated on the same VR grid.

VR and uniform (UN) 1∘ resolution experiments are run to explore the sensitivity of BCD effects to model grid spacing. The number of horizontal computational grid cells increases from 48 602 in the UN experiment to 114 860 in the VR experiment. CESM experiments are conducted on 30 vertical levels; however the physics time step in VR simulations is 2 times smaller than that used in UN experiments (15 min compared to 30 min) to avoid numerical instability. Additionally, the dynamics time step is 9 s in VR experiments and 90 s in UN experiments. The grid setup between the VR and UN experiments is identical to that used in Rahimi et al. (2019).

Both VR and UN simulations are conducted using identical physics. CESM version 1.2 is used with CAM version 5.3 (Neale et al., 2010) and is coupled with the Community Land Model, version 4.0 (CLM4). CLM4 default dust erodibility data, defined on a 1.9∘×2.5∘ mesh, is used. Coupled into CLM4 is the SNow ICe and Aerosol Radiative (SNICAR) model, which prognostically treats albedo reductions associated with snow aging and snow grain size changes, as well as BC and dust deposition on snowpack absorption (Flanner et al., 2007). Parameterized cloud microphysics from Morrison and Gettelman (2008), shallow convection from Park and Bretherton (2009), and deep convection from Zhang and McFarlane (1995) and Richter and Rasch (2008) are used. Radiation is treated using the rapid radiative transfer model from Iacono et al. (2008), and aerosol impacts are simulated using the three-mode version of the Modal Aerosol Module (MAM3), described in Liu et al. (2012). A wavelength-independent refractive index of 1.95–0.79i is used for BC in shortwave bands. The dust refractive index used in this study varies in 16 simulated longwave bands (not listed here), but its shortwave refractive index varies between {(1.51 to 1.8) and (0.01 to 0.1)i}. The real component of dust refractive index varies only slightly in near-IR wavelengths (∼1.53), while its imaginary component varies between -0.004i and -0.03i.

Eight VR and four UN experiments are run to estimate the impacts of BCD-induced SDE and ARI on premonsoonal and monsoon climate fields (Table 1). Aside from the control experiments, defined to be CONT-vr and CONT-un for the VR and UN control experiments, respectively, seven VR and three UN perturbation experiments are run to quantify various BCD effects.

Each individual experiment is run for 11 years, and the first year in each simulation is neglected in the analysis to allow for spin-up. The simulations are run with prescribed climatological sea surface temperature and sea ice cover averaged from 1982 to 2001 (Hurrell et al., 2008). The greenhouse gas concentrations and anthropogenic aerosol and precursor gas emissions are prescribed at the level for the year 2000 from the Intergovernmental Panel on Climate Change's Fifth Assessment Report. After comparing the simulations to both gridded and point source reference data (locations shown in Fig. 1b), the means of various climate variables from the last 10 years of simulations are computed to evaluate the impacts of BCD-induced SDE and ARI across southern Asia. Furthermore, all simulated data (both VR and UN) across the region are interpolated to an identical 0.125∘ rectilinear grid for direct comparison.

For the VR perturbation experiments, one is run with BCD SDE turned off (noSDE-vr). This is achieved by setting the on-snow BCD deposition fluxes to zero in CLM4. Second, an experiment is run in which BCD ARI is turned off (noARI-vr). This is achieved by excluding the BCD volume from the calculation of bulk aerosol extinction, asymmetry parameter, and single-scatter albedo in CAM. Third, an experiment is run in which both BCD SDE and ARI are removed (noBCDrad-vr). The fourth and fifth perturbation experiments are run, identical to noSDE-vr and noARI-vr, but only the BC-induced SDE and ARI are removed, respectively (noBCSDE-vr and noBCARI-vr, respectively). The sixth and seventh VR perturbation experiments are identical to noBCSDE-vr and noBCARI-vr, except dust SDE and ARI are removed, respectively (noDSDE-vr and noDARI-vr, respectively). Three UN perturbation experiments are also run to assess BCD effects on a 1∘ mesh: noSDE-un, noARI-un, and noBCDrad-un.

The BCD effects on meteorological and hydrological fields can be found by subtracting the perturbation experiments from the control simulation (CONT-vr). For some variable x, the effect induced by a specific species (BC or dust) at a grid cell can be computed.

Change in x induced by SDE is 1SDEx=CONT-vrx-noSDE-vrx.

Change in x induced by ARI is 2ARIx=CONT-vrx-noARI-vrx.

Change in x induced by SDE + ARI is 3TOTALx=CONT-vrx-noBCDrad-vrx.

Change in x induced by BC SDE is 4BCSDEx=CONT-vrx-noBCSDE-vrx.

Change in x induced by BC ARI is 5BCARIx=CONT-vrx-noBCARI-vrx.

Change in x induced by dust SDE is 6DSDEx=CONT-vrx--noDSDE-vrx.

Change in x induced by dust ARI is 7DARIx=CONT-vrx--noDARI-vrx.

The anomalies computed in Eqs. (4) through (7) for variable x are linear, but may add together such that their combined effect is nonlinear. In other words, there are nonlinear interactions between BC and dust SDE and ARI that may be important when considering the perturbations to SAM climate. We emphasize the VR results, but we do briefly discuss their differences with UN results in Sect. 5.

For our analysis, we break up southern Asia into five distinctive subregions (Fig. 1c). We consider most of India separately from the Indo-Gangetic Plain (IGP) to gain a sense of the impacts exerted on the SAM by the TP regional aerosol effects. Across India and the IGP, we only consider land grid cells with elevations lower than 1200 and 600 m, respectively. We also divide the TP into the western TP (WTP) and eastern TP (ETP) since other studies (Lau et al., 2010; Qian et al., 2011) have found there to be noticeable differences in simulated aerosol effects between these two regions. For the TP analyses areas, we only consider grid cells with elevations greater than 3700 m. A fifth region, the TP foothills (TPF), is also considered to explore how BCD effects may impact orographic precipitation. For the TPF subregion, we only consider grid cells with elevations between 400 and 3700 m.

A validation of the simulated meteorology was performed in Rahimi et al. (2019), in which CONT-vr and CONT-un were compared to surface- and satellite-based datasets. They found that there were marked improvements in the simulated temperature, precipitation, and snow coverage across the TP and its southern mountain ranges when using a VR grid. This is important when simulating the SDE, which is fundamentally dependent on the spatial distribution of snow coverage. In this study, we further evaluate CESM's performance in simulating aerosols, as the SDE and ARI are fundamentally dependent on the spatial variability and magnitude of aerosol loading.

To accurately capture the impacts of SDE and ARI on South Asian climate, it is important for the simulations to adequately represent atmospheric aerosol characteristics, e.g., aerosol optical depth (AOD). We make use of satellite and ground-based AOD measurements, as well as point source BC measurements to evaluate the model performance. While in-atmosphere and in-snow BC measurements are available for model performance evaluation, in-snow and in-atmosphere dust measurements are lacking due to uncertainties in how dust is classified and the lack of data across southern Asia. These data and their respective uncertainties will now be presented.

Control simulations are compared to 1∘ and 0.5∘ Level-3 550 and 555 nm AOD data from the Moderate Resolution Imaging Spectroradiometer (MODIS; Platnick et al., 2015) and the Multi-angle Imaging SpectroRadiometer (MISR, MISR Science Team, 2015), respectively. MODIS data are averaged monthly from 2001 to 2014, while MISR data are similarly averaged from 2002 to 2014. Weaknesses in MODIS and MISR retrievals are associated with algorithm inadequacies, as well as retrieval difficulties over bright or cloud-filled pixels. More information on MODIS and MISR can be found in Sect. S1.1 and S1.2 of the Supplement, respectively.

AOD spectral radiometer measurements from 57 AErosol RObotic NETwork (AERONET) sites across central and southern Asia are used to evaluate simulated total, coarse-mode, and fine-mode AOD at 500 nm (Holben et al., 1998). Model results are linearly interpolated to AERONET site locations and averaged spatially (see Fig. 1b) between 60 and 110∘ E and between 5 and 40∘ N. AERONET data for South Asian sites were only available from 1998 onwards, and only data before 2016 are used. The simulated fine-mode AOD is the summation of the accumulation- and Aitken-mode AOD variables in CESM. AERONET site locations are denoted by the blue stars in Fig. 1b.

AOD information is available via the Max Planck Institute's Aerosol Climatology, version 2 (MACv2), providing present-day (year 2005) climatological estimates of monthly mean AOD at 550 nm on a 1∘ rectilinear grid (Stevens et al., 2017). Surface observations from AERONET are used in conjunction with anthropogenic and fire emission plume strengths from CMIP5 simulations to estimate preindustrial AOD based on present-day AOD. More information on the MACv2 product can be found in Sect. S1.3.

We compare our results against AOD data from the Modern-Era Retrospective analysis for Research and Applications version 2 (MERRA-2; see https://esgf.nccs.nasa.gov/projects/create-ip/, last access: 1 June 2018). Monthly 550 nm MERRA-2 data are available on a 0.5∘ by 0.625∘ grid and analysis is performed using data from 1980 to 2017. More information on the MERRA-2 product can be found in Sect. S1.4.

Finally, simulations are compared to point atmospheric BC measurements from 13 sites discussed in He et al. (2014) and 11 sites discussed in Yang et al. (2018). The site locations are shown in Fig. 1b (hollow black circles). Additionally, 26 measurements of in-snow BC, as discussed in He et al. (2014) are used to evaluate our simulations. Locations of in-snow BC measurements are also given in Fig. 1b (red triangles). Site metadata for in-snow and in-atmosphere BC measurements, as well as simulated and observed BC concentrations are given in Table 2.

Focusing on annual AOD averaged between 0 and 60∘ N and 60 and 140∘ E, in which several major BCD emissions sources lie, CONT-vr and CONT-un simulate AOD values of 0.147 and 0.143, respectively. On the other hand, MODIS, MISR, MACv2, and MERRA-2 depict regionally higher annual AOD values of 0.285, 0.202, 0.235, and 0.216, respectively. While both CESM simulations reasonably capture the global annually averaged AOD compared to satellite observations (see Sect. S1.5), they do not capture the generally larger annual AOD values across south-central Asia. Because the buildup of aerosols across southern Asia has been identified to affect the premonsoonal and monsoonal properties (Lau et al., 2011, 2017), the CESM simulations' underprediction of annual AOD by almost a factor of 2 across South Asia must be kept in mind.

Figures 2 and S1 show the spatial distribution of Asian May and June (MJ) averaged and annually averaged AOD, respectively, from simulations, satellite measurements, and aerosol reanalysis. Data reveal higher MJ AOD values across southern Asia, the Tarim Basin eastward into northern China, and eastern China. The lowest AOD values are located across the TP, Russia, and northwestern Micronesia. CONT-vr and CONT-un overpredict AOD values over Asian dust sources (the Taklamakan and Gobi deserts) and the northern Arabian Sea, and simulations underpredict AOD values across India, eastern China, and oceanic regions compared to MISR, MODIS, MACv2, and MERRA-2 data.

Across deserts, the overestimation of AOD may be due to the fact that CLM4 uses a default erodibility dataset originally designed for use at a 1.9∘×2.5∘ grid. The fact that many areas of our domain are refined to 0.125∘ grid spacing may lead to an overestimation of dust emissions across the region, correctable by tuning the dust emission factor. Over heavily polluted regions (e.g., east China), CONT-un and CONT-vr underprediction of AOD compared to observations may be due to the underestimation of anthropogenic aerosol emissions and the missing treatment of secondary aerosol production in the models (Fan et al., 2018). Across oceanic regions, the undersimulated AOD by models is most likely the result of inadequate sea salt emissions, which are not a focus of this study.

Figure 3a–c show simulated versus observed annually averaged AOD from 57 AERONET sites (shown in Fig. 1b). CESM experiments underpredict AOD compared to AERONET measurements. Fine-mode aerosols contribute to most of the AOD underprediction. CONT-vr simulates a fine-mode and coarse-mode mean AOD bias (ΔAOD‾) of -0.189 and -0.062, respectively, while CONT-un simulates similar biases. Furthermore, CONT-vr slope values for the total, fine-mode, and coarse-mode best-fit lines of 0.157, 0.135, and 0.402, respectively, are simulated.

AOD underprediction by CONT-vr and CONT-un is also evident in a time series depicting the monthly variability of AOD averaged over the AERONET sites (Fig. 3d). Simulations and observations generally show a similar pattern, with larger values in the spring. CESM simulations underpredict AERONET AOD by a factor of ∼1.5. MISR, MACv2, and MERRA-2 appear to best agree with AERONET observations, while MODIS generally shows monthly AOD values that are 10 %–30 % higher than AERONET observations. Additionally, CONT-vr, CONT-un, MERRA-2, and MISR correlate better with AERONET than MODIS and MACv2, with CONT-vr having the highest r value of 0.874.

Figure 3e shows a time series of annually averaged monthly AOD interpolated to 18 AERONET sites across the IGP and TPF between 25 and 30∘ N and 70 and 90∘ E. Similar to the 57-site average, both CESM simulations underpredict monthly AOD by a factor of 1.5 to 2 across the IGP and TPF. However, in contrast to the 57-site average, the 18-site r values of the CESM simulations with AERONET are notably lower (see Sect. S1.5).

The CESM-simulated wet bias described in Rahimi et al. (2019) could have the effect of overpredicting wet scavenging of aerosols. This idea is reinforced when looking at point BC measurements across India, where simulated BC wet deposition dominates over dry deposition (Fig. S2), and simulated atmospheric BC concentrations are much lower than observations (Fig. 4a); simulations may be washing out too many aerosols. In addition, incorrect emissions may further contribute to the simulated bias in aerosol amounts (Zhao et al., 2011; Fan et al., 2018). Non-simultaneity between simulations and observation data may also play a role in skewing the interpretation of simulated aerosol features. Both anthropogenic emissions in Asia and dust emission in the Middle East have experienced significant decadal increasing trends during the first decade of the 21st century (e.g., Hsu et al., 2012; Jin et al., 2018). These trends may partially explain why the CESM experiments conducted with the year 2000 emissions underpredict AOD compared to observations. It is important to keep in mind these considerations when interpreting the relatively poor model performance in simulating AOD.

Figure 4a depicts an almost unanimous underprediction of atmospheric BC concentrations across the 24 measurement sites by the CESM experiments, with average aerosol biases of -2.77 and -2.76 µg m-3 for CONT-vr and CONT-un, respectively. The largest underpredictions occur over urban sites such as Delhi (observed 13.5 µg m-3), Dibrugarh (observed 8.9 µg m-3), and Lhasa (observed 3.7 µg m-3; see Table 2). Averages of observations depict a mean concentration across all sites of 3.24 µg m-3, while CONT-vr and CONT-un underpredict this value by a factor of 6.9 and 6.8, respectively. Additionally, several sites see a simulated underprediction of BC concentrations by more than a factor of 10.

The widespread underprediction of atmospheric BC does not necessarily translate to an underprediction of in-snow BC mixing ratio as seen in Fig. 4b. CONT-vr and CONT-un simulate a bias of -10.4 and +22.7 µg kg-1, respectively, when comparing to the station-averaged BC mixing ratio of 54.6 µg kg-1. This indicates that CONT-vr is more comparable to observations magnitude-wise. It is also noteworthy that several sites showcase a CONT-vr-simulated in-snow BC mixing ratio that is an order of magnitude different from that simulated in CONT-un. These large differences in simulated in-snow BC between the VR and UN experiments can be attributed to large meteorological and terrain differences between the two experiments, especially for He et al. (2014) sites across the Himalayas. For instance, Fig. S2 shows that at East Rongbuk (28.02∘ N, 86.96∘ E), CONT-vr simulates the terrain height to be more than 2.5 km higher than CONT-un, culminating in lower monthly temperatures and snow water equivalent (SWE) increases of more than 300 mm compared to CONT-un. Despite the increased SWE at Rongbuk, CONT-vr simulates tens of millimeters less precipitation than CONT-un owing to the smaller south-to-north upslope zone simulated in the VR experiment (Rahimi et al., 2019). The smaller VR-simulated precipitation correlates with 100–300 µg m-2 d-1 less wet-scavenged BC compared to the CONT-un experiment. The decreased wet deposition coupled with the larger SWE amounts in CONT-vr thus favor lower in-snow BC mixing ratios than CONT-un.

Similar to what was noted in Sect. 3.2, the temporal inconsistency between point source BC measurements and the CESM experiments must be kept in mind. BC measurements were conducted between 1999 and 2013, while simulations are run with year 2000 anthropogenic emissions. Our results could therefore be biased depending on the trends in BC emissions after the year 2000.

All-sky direct radiative effect (DRE) and in-snow radiative effect (ISRE) diagnostics are computed online in CONT-vr. The all-sky DRE is computed by subtracting a diagnostically computed top-of-atmosphere (TOA) energy balance without aerosols from that computed with aerosols present following Ghan et al. (2012). ISRE is computed in the SNICAR code via a similar method.

While both aerosols contribute to positive ISRE for all months across the TP region, their DREs are more complicated. BCD combines to incite a generally positive DRE across South Asia during MJ and JA, and the pattern of the DRE is similar during these time periods, respectively. The spatial distribution of BCD DRE during MJ is shown in Fig. 5, while that for JA is shown in Fig. S3. Positive dust-induced DRE values of +4 to +9 W m-2 are simulated across the Tarim Basin and the Gobi desert.

BC is simulated to exert unanimously positive DREs across southern and central Asia during MJ and JA. BC induces a DRE of between +1 and +4 W m-2 across India during MJ (Fig. 5a), with the magnitude weakening by JA (Fig. S3a). The highest BC DRE values of nearly +2 W m-2 are found across the IGP in northern India, while DRE values of +0.5 W m-2 or less are found across the TP during MJ into JA.

Dust is simulated to exert both positive and negative DREs across southern and central Asia during MJ and JA (Fig. 5b). Dust effectuates a DRE of nearly +10 W m-2 (+2 W m-2) during MJ south of the Gobi desert across the Ghar desert (north of the TP across the Tarim Basin). Dust-induced DRE values of between +1.5 and +4 W m-2 are also simulated across the central and northern Ghat Mountains of India during MJ. Meanwhile, dust induces a DRE of around +1 and -1 W m-2 across the TP and east-central India, respectively, during MJ. Areas with negative dust DRE values are typically characterized by low surface albedo; dust brightens the planetary albedo and thus cools the TOA. During JA, dust incites a DRE that is small (less than +1 W m-2) across east-central India (Fig. S3b). Also during JA, dust-induced DRE values in excess of +4 W m-2 are simulated across the northern Arabian Sea as dust is transported eastwards from Saudi Arabia by lower-tropospheric westerlies.

Despite the most prominent SWE reductions occurring due to BC SDE (discussed later), diagnostically computed ISRE values across the WTP indicate that BC and dust contribute to similar regionally averaged seasonal values of between +1 and +3 W m-2 from March through June (Fig. 5c–e), Together, BC and dust contribute to an annual maximum ISRE of +7 W m-2 across the WTP in May (Fig. 5c), while this maximum occurs in March across the ETP (Fig. 5d; +2.8 W m-2) and TPF (Fig. 5e; +2.2 W m-2). The ISRE maxima occur in boreal spring for all three TP regions as the solar elevation angle and South Asia BC and dust burdens increase during this time.

The largest ISRE values occur in MJ compared to JA, with the mountains on the southern and western TP periphery being characterized by ISRE values greater than +10 W m-2, locally (see Fig. S4). This heterogeneity in the ISRE spatial pattern mimics the heterogeneity in the SWE and 2 m temperature anomaly patterns across the TP (shown later), especially for the BCD SDE anomaly patterns. These features, obviously attributable to the VR mesh, are not captured in the UN experiments.

Together, BCD contribute to statistically significant (SS; t values in excess of 0.9) ARI- and SDE-induced 2 m temperature (T2) changes during MJ and July and August (JA) across Tibet and South Asia. Warming in excess of 3 ∘C across the WTP and 1.3 ∘C across the ETP is shown in Fig. 6a. Meanwhile, the collective impacts (ARI + SDE) of BCD contribute to cooling across most of India due to cloud coverage increases (to be discussed in Sect. 4.4), with SS values of -0.7 to -1.2 ∘C across western India and the IGP region. By JA, BCD effectuates T2 patterns similar to those in MJ across southern Asia, as shown in Fig. S5a. However, the areas of cooling characterizing much of India are shifted north and west to include most of Pakistan, and the areas of warming characterizing a majority of the TP are much reduced compared to MJ, especially across the southern TP. We note that T2 changes of 0.1–0.2 ∘C are simulated across portions of the Arabian Sea. These values should not be interpreted as significant due to the fact that sea surface temperatures (SSTs) are prescribed.

The sign and magnitude of T2 changes vary as a function of effect type (SDE or ARI) and by species (BC or dust). BCD SDE-induced T2 warming can exceed ARI T2 warming over complex terrain, as indicated by our results. Values of +2 ∘C are simulated across the WTP mountain chains (Fig. 6b), such as the western Himalayas, Kunlun, Karakoram, and Hindu Kush during MJ. Within the WTP, SDE-induced warming reaches 0.5 to 1.3 ∘C. Additionally, BC generally contributes to more simulated SDE warming than dust across the WTP during MJ and JA (Figures 6d and S5d, respectively), while dust contributes to a majority of the SDE-induced warming across the ETP during MJ.

BCD ARI drive a majority of the simulated T2 changes across southern Asia during MJ (Fig. 6c) and JA, with the largest T2 changes occurring during MJ. While MJ changes in T2 associated with ARI are brought forth by both BCD collectively, the TP area has more expansive T2 changes as a result of dust-induced ARI (Fig. 6g); a much larger swath of 1.3 to 2 ∘C warming occurs across the WTP compared to the simulated BC-induced ARI (Fig. 6e). Across the IGP, dust ARI brings forth cooling of more than 0.7 ∘C during MJ, while BCD ARI cool portions of central and southern India by 0.5 to 0.7 ∘C during MJ. By JA, dust ARI (BC ARI) contributes to most of the cooling (warming) across the IGP and Pakistan (northern TP), with simulated T2 changes from -0.2 to -0.7 ∘C (+0.2 to +1.3 ∘C). With these effects in mind, it should be noted that BC is underestimated compared to surface observations,

While understanding the BCD-induced changes in T2 is important from an anthropogenic perspective, these changes are inadequate when examining the influence of BCD on SAM dynamics. This is because BCD-induced SAM changes depend on the thermal characteristics of the tropospheric column. For this reason, Sect. 5 will make use of the 300–700 hPa mean column temperature differences instead of T2 differences when examining the circulation changes brought about by BCD effects on the SAM.

BCD effects contribute to large reductions in SWE across the TP and TPF from April through June. Peak BCD-induced SDE plus ARI reductions in SWE of 75 mm (49 %) are simulated in May across the WTP (Fig. 7a), while peak reductions of 25 mm occur in April across the ETP and TPF (61 % and 49 %, respectively; Fig. 7b, c). Even though the largest regionally averaged T2 warming results from BCD-induced ARI, the largest reductions in SWE are due to simulated BCD-induced SDE. By June, BCD SDE contributes to SWE reductions across the WTP of greater than 50 mm (62 %), while BCD ARI contributes to reductions in SWE of 10 mm (< 20 %) or less for all months. It is noted that the largest percent changes in SWE associated with BCD effects occur in the summer months, when SWE is minimized across the TP.

BC SDE drives a majority of SWE changes across the TP with reductions of greater than 30 mm (17 % to 44 %) or more across the WTP from March through June, but other effects are important too. BC ARI, dust ARI, and dust SDE all contribute to reduced SWE in excess of 10 mm from March through June across the WTP. Compared to the WTP, BCD effects across the ETP and TPF bring forth smaller reductions in SWE, but BC SDE still contributes to the largest effects on SWE. Additionally, the largest reductions in SWE are typically found along the western and southern TP periphery (Fig. 8).

BCD effects that lead to changes in TP area SWE can directly impact runoff, which replenishes main waterways across the region. It is found that BCD SDE drive runoff increases (decreases) from February through June (June through September) across the WTP and ETP, with peak runoff increases (decreases) of 1.2 mm d-1 (1 mm d-1) across the WTP occurring in May (July) (see Fig. S6), constituting a 77 % (36 %) increase (decrease) in runoff. The peak runoff increases across the WTP (ETP) correlate with maximum BCD SDE reductions in SWE of 75 mm (25 mm), which occur in May (April).

BCD effects bring forth responses in the mass fields, which impact simulated SAM cloud fraction (CF) and specific humidity (q). Driven primarily by BC and dust ARI, q increases in excess of 1 g kg-1 are simulated from mid-March through June across India, the IGP, and TPF; moisture changes peak during June across the IGP (+3.5 g kg-1) and TPF (+2.5 g kg-1) (Fig. 9). BCD-induced SDE contributes to smaller q changes from April through July across the IGP and TPF of +0.7 and +0.8 g kg-1, respectively. Interestingly, the q changes from dust-induced SDE and BC-induced SDE across the TPF and IGP do not add linearly during June (see Fig. 9c, e). Focusing specifically on the IGP, as the q changes are similar between the IGP and TPF, dust (BC) SDE contributes to a q change of +0.3 g kg-1 (+1.2 g kg-1), but we see a total (i.e., BC SDE + dust SDE) q change of +0.7 g kg-1. This result could be due to an increase in precipitation due to the combined effects of BCD SDE, which would act to deplete the available water vapor, locally. The positive q changes just discussed are dramatically reduced from July through September. Across the WTP and ETP, q changes of +1 g kg-1 are simulated in June, while smaller q changes are simulated for the spring and summer months across both regions.

Specific humidity increases due to BCD effects correlate reasonably well with increases in CF, especially across South Asia from April through September (see Fig. S7). An increase in CF, primarily driven by BCD ARI, is simulated across India, the IGP, and TPF from April through June. Net peak CF increases occur in May across India (12 %) and June across the IGP (15 %) and TPF (16 %) as seen in Fig. S7. CF increases of 10 % or more due to aerosol effects across southern Asia during MJ have been noted previously (Lau et al., 2010). By September, the CF increases vanish across India, the IGP, and TPF. Across the ETP, simulations indicate that CF increases of 3 % are due to BCD SDE in June, with slightly larger increases (as much as 7 %) in June and July across the WTP (see Fig. S7) also due to BCD SDE.

Compared to SDE, BCD ARI generally brings forth the largest changes in CF across South Asia from May through August. The spatial distributions of MJ (JA) CF changes are shown in Fig. 10 (Fig. S8). CF increases of 7 % or more across India and CF decreases of 5 %–10 % or more across the central and northern TP are simulated (Fig. 10a). The Arabian Sea, western Ghat Mountains, and TPF are characterized by the largest positive CF changes, which can exceed 15 %. Furthermore, while BC ARI patterns are similar to those of dust ARI during MJ, the magnitudes of dust-induced ARI changes are generally larger, especially across the Arabian Sea and the ETP (Fig. 10e, g).

While SDE magnitudes are generally smaller compared to those induced by BCD ARI, BC SDE drives MJ CF changes of +5 % to +15 % across the IGP and southwestern TP (Fig. 10d). Meanwhile, CF reductions of 2 % to 5 % are simulated due to BC SDE across northern TP during MJ, making the spatial pattern of CF changes induced by BC SDE (Fig. 10d) similar to those induced by BCD ARI (Fig. 10c). In addition, dust SDE contributes to increases in CF of 4 % to 7 % across the southern TP during MJ (see Fig. 10f).

BCD effects contribute to almost unanimously increased precipitation across southern Asia during premonsoonal months, as seen in Fig. 11a, with BCD collectively contributing to values in excess of +6 mm d-1 across the eastern Arabian Sea, the eastern Bay of Bengal, and the TPF. Here, we define “precipitation” to be sum of liquid precipitation plus ice precipitation, and we find that changes in total precipitation are driven by changes in liquid precipitation; simulated changes in snow precipitation are minimal (not shown). Elsewhere, BCD contributes to SS changes of between +1 and +4 mm d-1 across India and changes in MJ precipitation of around +1 mm d-1 over the southern and eastern TP. It seems as though the large-scale pattern in MJ precipitation changes is regulated by dust ARI (Fig. 11g). However, precipitation changes of between +1.5 and +5 mm d-1 are simulated across the Bay of Bengal and Arabian Sea associated with BC SDE (Fig. 11d).

From July through August, precipitation increases of 1 mm d-1 are simulated across India, driven primarily by dust ARI. Meanwhile, dust ARI-driven precipitation increases of 2 mm d-1 (16 %–35 %) are simulated through September across the IGP (see Fig. 12). Additionally, BCD SDE contributes to slight increases of 0.5 mm d-1 (2 %–17 %) or less across the IGP from May through August. Across India, however, with the exception of July, BCD-induced SDE contributes to decreased precipitation from March through October, with a peak reduction in June at 1.2 mm d-1 (10 %). These increases, primarily induced by dust ARI, are similar to those reported in Jin et al. (2015) and slightly larger than those found in Vinoj et al. (2014).

The TPF region is characterized by the largest precipitation increases relative to the other subregions due to an enhancement of BCD effects by the complex terrain, with BCD effects bringing forth increases in precipitation from April through August (see Fig. 12c). BCD contribute to unanimous SDE- and ARI-induced precipitation increases in June across the TPF, with dust and BC ARI (both approximately +4.5 mm d-1; +25 % and +23 %, respectively) and BC SDE (+4.3 mm d-1; +21 %) dominating the enhanced precipitation; the collective impacts of BCD enhance precipitation by more than 7 mm d-1 (47 %) in June. Dust SDE contributes to smaller MJ anomalies of less than +1 mm d-1 (3%–6 %) across the TPF.

BCD ARI-induced precipitation differences drive changes in runoff across India and the IGP, with runoff increases induced from April through October. Maximum precipitation increases of 2.5 mm d-1 (31 %) and 2 mm d-1 (27 %) occur a month ahead of runoff increases of 2 mm d-1 (41 %) and 1.6 mm d-1 (85 %) across India and the IGP, respectively (Fig. S6). The maximum precipitation (runoff) increase occurs in June (July) across India, while the maximum precipitation (runoff) increase occurs in July (August) across the IGP. In contrast to India and the IGP, the maximum precipitation increases in TPF driven by BCD ARI occur in the same month (June) as runoff, with precipitation (runoff) increases of 7 mm d-1 (6 mm d-1) representing increases of 47 % (58 %). The reasons for the differences in runoff–precipitation phase between India–IGP and the TPF may be the result of larger runoff effects associated with BCD SDE across the TPF compared to the IGP and India, which contributes to runoff increases of more than 3.5 mm d-1 (26 %) in June (Fig. S6c, e).

The changes induced by the combination of BCD effects on the premonsoonal and monsoonal meteorology can be examined by considering thermal vorticity, ζT (Bluestein, 1992). Analogous to thermal wind, ζT is defined to be the difference between the upper-level geostrophic vorticity (ζg,above) and the lower-level geostrophic vorticity (ζg,below) within a column: 8ζT≡ζg,above-ζg,below, where ζg=∂vg∂x-∂ug∂y, so 9ζg=1f0∇p2Φ, where f0 is the constant Coriolis parameter, ug and vg are the x and y components of the geostrophic velocity components, respectively, and Φ≡gz is the local geopotential height. Taking the derivative of Eq. (9) with respect to independent variable t, we find the local tendency in ζg to be 10∂ζg∂t=1f0∇p2∂Φ∂t.

Substituting Eq. (10) into Eq. (8), we get 11∂ζT∂t=1f0∇p2∂Φ∂tabove-∂Φ∂tbelow=1f0∇p2∂ΔΦ∂t, where ΔΦ≡Φabove-Φbelow is the average column thickness. Here, variable t can be time, but because we are applying Eq. (11) to independently run experiments, t is more accurately an independent variable denoting case. For simplicity, we define all variables that are subject to the operator ∂/∂t to be a tendency. Equation (11) states that the thermal vorticity tendency is proportional to the Laplacian of the layer thickness tendency, which is proportional to the mean layer temperature. This section will thus utilize the 300–700 hPa column-averaged temperature to evaluate circulation changes.

For synoptic-scale flows, it can be shown via scale analysis that the local vorticity tendency is dominated by the stretching of Earth's vorticity, or 12∂ζg∂t=-δf0, where δ≡∂u∂x+∂v∂y. Equation (11) can be written as 131f0∇p2∂ΔΦ∂t=-f0δabove-δbelow=-f0Δδ.

In the above series of equations, u and v are the zonal (x) and meridional (y) components of the 3-D wind field. The upper- and lower-tropospheric divergence of the horizontal wind field is given by δabove and δbelow, respectively, and Δδ≡δabove-δbelow.

Using Eq. (13), we can link BCD-induced temperature changes to circulation changes during premonsoonal months. Figure 13a shows simulated MJ anomalies in 300–700 hPa column-averaged temperature. BCD-induced columnar warming of between 1 and 3 ∘C occurs in a belt between 20 and 40∘ N, within a zone of climatologically maximized west-southwesterly upper-tropospheric flow (see Fig. S9c). These strong upper-tropospheric winds are responsible for transporting dust downstream of major emission sources such as northern Africa and the Middle East. The warming in this belt is also due to changes in clouds through evaporation and regional circulation changes (Fig. 10a). Under this warming scenario, ∂ΔΦ∂t>0 (i.e., the layer expands). If it is assumed that the geopotential field is the linear combination of sinusoidal functions, the Laplacian of a positive quantity will contribute to a negative thermal vorticity tendency by Eq. (11). Hence, a warming–expanding column will bring about Δδ>0 via Eq. (13), and upper-tropospheric anticyclonic ζg by Eq. (12) (and Φ rises; Fig. 13a, d) will tend to be generated atop lower-tropospheric cyclonic ζg by Eq. (12) (and Φ falls). Via continuity, these changes in the vorticity field lead to changes in vertical motion under the assumption of incompressibility, a commonly used assumption on regional and global scales. As aerosols heat the atmosphere, ∂ΔΦ∂t>0, ∂ζT∂t<0, and Δδ>0. In an atmosphere where divergence increases with increasing height ∂δ∂z>0, the continuity equation subject to incompressibility is 14∂δ∂z=-∇x,y2w, where ∇x,y2 is the one-dimensional Laplacian and w is the vertical component to the 3-D velocity vector. Under such conditions, w must be positive if it is assumed that w(x,y,z,t) is a linear combination of sinusoidal functions. BCD warming brings forth an enhancement of the rising branch of the thermally direct monsoon circulation.

The application of Eqs. (13) and (14) to our results couples the total (BCD ARI and SDE) thermodynamic changes to the circulation changes during MJ. Maximal column-averaged temperature anomalies (Fig. 13a) are collocated with an upper-tropospheric anticyclonic anomaly in the mass field (Fig. 13d) across southwest into southern Asia during MJ, with 250 hPa pressure surface height rises of between 6 and 9 dm across south-central Asia (Fig. 13a).

The response in the mass field due to BCD ARI and SDE is not confined only to the upper troposphere, where the geostrophic approximation is most applicable. In fact, BCD-induced cyclonic changes in the 850 hPa flow are simulated (color fill, Fig. 13b), with an intensification in the WLLJ of as much as 5 m s-1, extending from eastern Africa, bifurcating the Arabian Sea, and protruding into southeast Asia during MJ (Fig. S13a). The magnitude of WLLJ intensification is smaller across India (2 to 4 m s-1), but its magnitude across the Bay of Bengal is more comparable to the intensified westerlies in the Arabian Sea. To this feature's north, there are simulated BCD-induced easterlies of -1 m s-1 across eastern Iran into Russia, but the most pronounced low-level flow changes lie across the open sea. This is to be expected, as the lower-tropospheric geostrophic assumption is more erroneous over land compared to oceans due to the far greater surface friction over land. More generally, friction may explain why the anomalous 850 hPa vorticity feature is more diffuse than the 250 hPa vorticity feature during MJ across south-central Asia.

The BCD-induced anomalies in the mass fields, discussed by invoking thermal vorticity arguments, lead to changes in the vertical motion pattern, water vapor budget, and precipitation patterns across South Asia during MJ. The 850 hPa q increases across South Asia of 1 to 2 g kg-1 are simulated (Fig. 13c) due to a stronger WLLJ (Fig. 13b). Meanwhile, BCD warming brings enhanced rising vertical motion by Eq. (14) across South Asia. These increases in w and q correlate with precipitation increases of +2 mm d-1 across India during MJ, while precipitation increases of greater than 6 mm d-1 are simulated across the western Ghats and TPF (see Fig. 13c). The latter increases may be due to increased orographic effects (not explicitly examined in this study), as low-level upslope flow over the Ghats is enhanced by a strengthened WLLJ, and upslope flow over the TPF is enhanced by stronger cyclonic flow across India (Fig. 13b). Positive precipitation anomalies increase in magnitude towards the east across the Arabian Sea and Bay of Bengal as eastward-moving precipitation systems gain sensible and latent heat from the open waters.

Into JA, the belt of maximum 300–700 hPa column heating shifts north (Fig. S11a) along with the subtropical jet. Meanwhile, the area of WLLJ intensification shifts north and shrinks significantly compared to MJ (Fig. S11b). A precipitation dipole is simulated across the western mountains of India during JA, with increases (decreases) in excess of 4 mm d-1 located further north (south) in the mountain chain. Precipitation increases greater than 4 mm d-1 are also simulated across the western IGP and the southern half of Pakistan (Fig. S11c), with the maximum BCD precipitation increases tending to occur on the eastern nose of the WLLJ anomaly (as in MJ).

It is noted in Sect. 4.5, when compared to other subregions, TPF precipitation is more sensitive to BC SDE, in addition to BC ARI and dust ARI (see Fig. 12). This could indicate the presence of the elevated heat pump effect (Lau et al., 2010), which develops in proximity to the TP as anomalous BCD-induced heating of the TP leads to enhanced anabatic upslope flow. This effect, coupled with enhanced southerly cyclonic flow incident on the TPF during MJ, may explain the enlarged precipitation enhancements across the TPF relative to other subregions considered in this study. Orographic enhancement of BCD effects over complex terrain was not explicitly examined in this study and remains the subject of future work.

Different species (BC or dust) and different radiative effects (SDE or ARI) regulate the BCD impact on the SAM and premonsoonal meteorology. Figure 14 depicts the vertical structure of BCD effect-wise changes in temperature, w, CF, and q across northern India during MJ, averaged horizontally between 25 and 30∘ N and 75 and 80∘ E (north-central India).

The combined effects of BCD SDE and ARI contribute to low-level cooling (1.2 ∘C) beneath middle- and upper-tropospheric warming (as large as 2.1 ∘C), which increases thermodynamic stability. Above 3 km, dust ARI drives atmospheric warming as large as 1.5 ∘C. BC ARI contributes to atmospheric warming as large as 1.1 ∘C in a pattern similar to that of dust ARI. The BC SDE contributes to middle-tropospheric temperature changes that are similar in sign to dust ARI but half the magnitude, while dust SDE contributes the least to atmospheric heating compared to other effects (Fig. 14a) during MJ. Finally, the lower-tropospheric cooling is the result of cloud increases through w, q, and CF increases.

Accompanying the BCD-induced warming of the 3–15 km layer is an increase in tropospheric w up to 17 km during MJ. As with temperature, BCD ARI drives w increases that peak around 12 km at 0.5 cm s-1. Dust ARI accounts for a large portion of the total w increase, with BC ARI-induced w increases that are ∼30 % less. Meanwhile, BC SDE contributes to slightly stronger upward vertical velocities than BC ARI below 11 km, while the reverse is true above this altitude. This result may indicate the presence of the elevated heat pump effect (Lau et al., 2010) associated with strong warming on the southern TP periphery. BC SDE-induced warming of the TP periphery may induce locally strong solenoidal circulations due to horizontal density variations between the warming TP and the adjacent free atmosphere. These circulations could manifest as lower-tropospheric anabatic branches of rising vertical motion across northern India, which may be stronger than the larger-scale thermally direct rising motions induced by BC ARI. These anabatic circulations were not explicitly studied in this work and are the subject of future work. In any case, BCD ARI would induce positive w anomalies throughout the tropospheric column across northern India, while BC SDE may generate rising vertical motion only in a layer of the atmosphere in which BC SDE-induced quasihorizontal temperature gradients could develop in proximity to the TP. As for temperature, dust SDE contributes to relatively small upward vertical velocity enhancements across northern India.

Interestingly, BCD ARI by itself corresponds with larger upward vertical velocity changes than those associated with the combined effects (SDE and ARI) of BCD, even though the combined effects lead to the strongest middle-tropospheric warming during MJ. This is because the combined effects of BCD greatly increase the thermodynamic stability of the atmosphere across northern India (Fig. 14a) such that w increases are depressed. Hydrostatically, warming of the mean atmospheric column tends to initiate a thermally direct rising bubble. However, if the heating is vertically nonuniform within the column, as is the case depicted in Fig. 14a, changes in atmospheric thermodynamic stability may actually reduce the buoyancy of the rising bubble. This increased atmospheric stability can be tied to negative CF anomalies as well because as the atmosphere becomes more stratified, increases in turbulent mixing and entrainment may incite the evaporation of clouds. Across northern Tibet, dust ARI and BC ARI actually stratify the atmosphere so much that CF anomalies approaching -10 % are simulated during MJ (see Fig. 10c). This leads to enhanced surface warming that persists into the monsoonal period (Figs. 6 and S5).

The BCD-induced effects on MJ upward vertical velocities, driven primarily by dust ARI and to a lesser extent BC ARI and BC SDE, correlate positively with CF anomalies. This is especially true for upper-tropospheric clouds (Fig. 14c). CF increases of 2 % are effectuated mainly by dust ARI below 4 km, while BC SDE primarily drives CF increases in excess of +6 % around 17 km. The increase in upper-tropospheric clouds is the result of increased convective precipitation, while the low-level CF increases result from increases in boundary layer moisture of nearly 3 g kg-1 (Fig. 14d) coupled with vertical velocity increases of 0.1 cm s-1 below 5 km and increased low-level thermodynamic stability.

Enhancements in w by BCD effects may not only be directly related to the synoptic response described in Sect. 5.1, but rather indirectly related via mesoscale features that form as a result of downscale energy cascade. While not explicitly examined in this study, low-level jet cores are characterized by mesoscale vertical circulations that may promote or inhibit upward vertical motion. Additionally, anabatic circulations associated with horizontal density and temperature gradients may do the same. These mesoscale features may enhance or depress the synoptic response induced by BCD effects and require more intense study.

VR experiments are comparable with UN experiments in simulating BCD effects (Fig. 15). These experiments simulate warming (cooling) across the TP (India) during MJ. However, VR experiments reveal a larger area of warming > 1.3 ∘C across the central and eastern TP (Fig. 15a) due to BCD effects compared to UN experiments (Fig. 15d). Additionally, UN simulated BCD effects bring about stronger cooling across India and Pakistan (Fig. 15d).

These differences between VR and UN experiments result from a stronger BCD SDE in the UN experiments. Aerosol-induced snow melting is much stronger in the UN experiments (not shown), leading to a much stronger warming of the WTP (Fig. 15e) than in the VR experiments. This brings about a stronger columnar warming in the UN experiments (not shown) that leads to stronger Arabian Gulf moisture flow into India (Fig. S12) through a stronger WLLJ (Fig. S13) and higher CF increases (Fig. S14) in the UN versus the VR experiments during MJ. Because Rahimi et al. (2019) showed that CONV-vr significantly outperformed CONT-un in simulating TP regional snow cover, it is reasonable to assume that CONT-vr more accurately simulates BCD SDE compared to CONT-un. Simulated BCD SDE should thus be quite different in terms of meteorological perturbations during MJ.