1 Introduction

The Northern India and Himalayan foothills (NIHF) region is an essential component of the Indian monsoon (Gadgil et al. 2003). Summer monsoon rainfall over this region feeds the Ganges, and the Indus rivers, providing fresh water essential for agriculture, industrial and livelihood over hundreds of millions people in the region. The Himalayan foothills (HF), an escarpment that rises to over 5 km above the Indo-Gangetic Plain (IGP) over Northern India (NI), provides a major barrier to the northward advance of the Indian monsoon rainfall, as well as contributing to the thermal contrast between the Tibetan Plateau and the Indian Ocean (Boos and Kuang 2010; Wu et al. 2012). It also provides strong orographic forcing to the prevailing southwesterly monsoon flow, giving rise to strong vertical ascent and development of vigorous thunderstorm cells with intense precipitation during the monsoon season (Houze et al. 2007; Rasmussen and Houze 2012; Das et al. 2014). The importance of the NIHF orography in forcing strong convection and extreme precipitations, in relationship to the large-scale Indian monsoon has also been reported (Barros and Lang 2003; Barros and Lattenmaier 1994).

Recently, an increasing number of studies have suggested that absorbing aerosols (mainly desert dusts and carbonaceous aerosols) can affect the interannual and intraseasonal variability of the Indian monsoon rainfall (Lau et al. 2006; Lau and Kim 2006; Lau 2014; Manoj et al. 2011; Hazra et al. 2013; Vinoj et al. 2014; Kim et al. 2015; D’Errico et al. 2015; Sanap and Pandithurai 2015, and many others). Based on these studies, Lau (2016) has proposed a new paradigm arguing that natural aerosol, particularly dust and carbonaceous aerosols from natural sources should be considered an essential component of an intrinsic aerosol-monsoon weather and climate system. Specifically for the NIHF region, the Elevated Heat Pump (EHP) hypothesis (Lau et al. 2006) posited that through atmospheric diabatic heating and circulation feedbacks, atmospheric heating due to absorbing aerosols from both anthropogenic and natural sources accumulated over the Indo-Gangetic Plain during the pre-monsoon period can, lead to increased rainfall and advance of the monsoon rainy season over the NIHF. At present, effects of absorbing aerosols in increasing monsoon rainfall and circulation consistent with EHP, but with significantly different regional spatial and temporal details have been found in many global climate models (Lau et al. 2006; Meehl et al. 2008; Randles and Ramaswamy 2008; Wang et al. 2009; Collier and Zhang 2009; Henriksson et al. 2014; D’Errico et al. 2015; Jin et al. 2015). One of the main reasons for the regional differences is that state-of-the-art climate models, due to their coarse resolution generally, are still unable to resolve the complex topography and interactions of monsoon winds, precipitation and aerosol transport, sources and sink processes over the NIHF. For more realistic simulations, the use of high-resolution regional atmospheric model with realistic representation of aerosol and monsoon processes is required. In this paper, using the NASA Unified Physics Weather Research Forecast (NU-WRF) model, we have carried out numerical experiments aimed at providing a better understanding of the physical processes involved in aerosol-monsoon interactions, including aerosol radiative and microphysical effects, as well as induced dynamical feedbacks over the complex topography of the NIHF region. The experiments will be conducted based on a case study of the 2008 Indian monsoon.

1.1 The 2008 Indian monsoon

The all-Indian monsoon rainfall for the full monsoon season (June–September) during 2008 was near normal. However, rainfall was excessive in the north, but deficient in the central and southern India for most of the season (Rao et al. 2011). Abnormally heavy rain started in early June, and lasted through early August. The heavy rainfall eventually led to the breaching of the Kosi river in early August, with disastrous flooding over Nepal and many states in northern India including, Assam, Andhra Pradesh, Bihar, Maharashtra, Uttar Pradesh, West Bengal and Orissa. The flood affected over 2.3 million people in the northern part of the state of Bihar, prompting the Indian government to declare a national disaster. The heavy rain in June (Fig. 1a) was found mainly over NIHF, extending southeastward toward the Bay of Bengal. At the same time, rainfall was suppressed over southwestern, central and southern, and extreme northeastern India. The summer of 2008 also witnessed an early onset of the monsoon over northern India (Tyagi et al. 2009). Most interesting, it was also a year with unusually large loading of aerosols over the Arabian Sea, northwestern India, and across central Indian during the early monsoon season (Fig. 1b). In this study, we will examine the possible roles of aerosols in influencing this rather unusual Indian monsoon.

Fig. 1
figure 1

Spatial distribution of observed a rainfall anomalis (mm day−1) from TRMM and b AOD from MODIS, for the Indian monsoon during June 2008. Rainfall anomalies are computed relative to 1998–2013 climatology

2 Modeling strategy and methodology

We use the NASA Unified physics Weather Research Forecast (NUWRF) regional atmospheric model to carry out ensemble forecast experiments to investigate the role of aerosol-monsoon dynamic interaction in the NIHF region. NUWRF has been developed at GSFC to unify physics in WRF with NASA’s current physics package for aerosol and cloud microphysics, as well as assimilation systems, common to GEOS-5, the Goddard Cumulus Ensemble cloud model (GCE) (Tao et al. 2009), the Goddard Chemistry and Radiation Transport (GOCART) model (Chin et al. 2004, 2009, 2014; Ginoux et al. 2001), and the Land data Information System (LIS) (Peters-Lidard et al. 2007, 2015; Tao et al. 2013; Lang et al. 2014; Shi et al. 2014). Briefly, GOCART uses the following modules in aerosol simulation: emission, which includes dust, sulfur, black carbon, organic carbon, and sea-salt emissions; chemistry, which includes in-air and in-cloud oxidations of sulfate precursors (SO2 and DMS); turbulent dry deposition, which is calculated by aerodynamic resistance; gravitational settling, which is a function of air viscosity and particle size; and wet deposition, which accounts for the scavenging of soluble species in convective updrafts and rainout/washout in large-scale precipitation.

Dust size distribution is calculated by solving the continuity equation for eight size bins ranging from 0.1 to 10 μm in radii, representing mineral dust types from fine clay to coarse sand particles. The continuity equation includes macroscopic advection by winds, parameterized eddy diffusion and moist convection. The removal mechanisms include dry deposition at the surface by impaction, wet deposition in and below clouds, and gravitational settling. Dust emission is computed as a function of the surface wind at 10 m, a threshold frictional velocity, and a source function which is depended on geographic locations and dust sizes (Ginoux et al. 2001; Gillette and Passi 1988).

For aerosol-cloud microphysics, NUWRF computes cloud droplet concentration using aerosol mass directly predicted by GOCART/WRF-Chem at each time step. For a given air temperature and supersaturation, cloud condensation nuclei (CCN) is calculated from the aerosol species predicted by GOCART based on the Koehler curve (Koehler et al. 2006; Andreae and Rosenfeld 2008), while the concentration of ice nuclei (IN) is obtained following the approach of Demott et al. (2010). Both CCN and IN are diagnostic parameters calculated from the WRF-Chem/GOCART-predicted aerosol mass concentrations in the Goddard one-moment microphysics scheme. CCN is used to calculate the auto-conversion of cloud liquid water (Qc) to rain water (Qr) following Liu and Daum (2004), while IN is used to calculate 1) the conversion of cloud ice (Qi) to snow (Qs) due to the Bergeron process and 2) the growth of cloud liquid water (Qc) to cloud ice (Qi) due to deposition.

The NUWRF team has also coupled GOCART with LIS which provides key information (e.g., soil moisture and soil porosity) that GOCART uses to calculate dust emissions. GOCART has also been coupled with the Goddard radiation and microphysics scheme to simulate the direct and indirect aerosol effects on energy budget, cloud, and precipitation. Parameterization schemes to estimate biogenic secondary organic aerosols have also been integrated from WRF/Chem into NUWRF. In addition, several model utilities have been developed and modified to facilitate NUWRF applications, including implementation of time varying GOCART aerosols types and distribution as initial and boundary conditions, and incorporation of pollutant emissions inventories for both anthropogenic and natural sources. For the present experiments, the single scattering albedo of dust aerosol at the visible band (0.4–0.7 μm) were specified in the range 0.95 to 0.79, representing 8 size categories of dust from fine clay to coarse dust particles. The much higher absorption (lower single scattering albedo) for larger size particles is consistent with recent observations indicating coating of fine black carbon particle on the larger size sand particles in Asian monsoon regions (Ganguly et al. 2005; Satheesh et al. 2008; Eck et al. 2010).

For our model experiments, we followed the ensemble forecast methodology commonly used in monsoon numerical weather prediction up to a week to investigate the impacts of aerosol-monsoon dynamic interaction on short time scales. We used nested domains with 9 km resolution in an inner domain (60°–100°E, 10°–40°N) covering the Indian subcontinent and eastern Indo-China, and an outer domain with 27 km resolution, spanning the large-scale monsoon domain (40–120°E, 5°S–40°N), with 61 layers in the vertical (Fig. S1). We selected the early monsoon period June 1–July 15 2008 for our study. This period covered the monsoon onset transition phase (June 11–17), and several subsequent heavy rain events over NIHF (See discussion in Sect. 3.2). We carried out 45, 7-day forecast experiments forced by prescribed observed sea surface temperature, and with atmospheric and aerosol boundary conditions initialized at intervals of one day starting from June 1, first with no aerosol (NA) effects, which is still the common practice for most numerical weather forecasts in monsoon regions (Krishnamurti et al. 1991). A second identical set of experiments was carried out but with the inclusion of aerosol radiative (RAD) effect. A third set of experiments same as the control, except with both aerosol radiative and microphysics interactive effects (RADM) were carried out for the onset period June 11–17. Initial and boundary conditions for meteorology and aerosols were obtained respectively from NCEP GFS 1° × 1° global analysis, and from GOCART. By comparing the RAD and RADM to NA forecasts over the NIHF region in a realistic setting with a high-resolution regional model and sophisticated aerosol physics, we can gain a better understanding of intrinsic, aerosol-monsoon interaction for the Indian monsoon, involving no change in anthropogenic aerosol emissions. In following discussions, all model anomaly quantities are defined with respect to NA. For brevity, we shall refer to RAD anomalies as due to Aerosol-Radiative Interaction (ARI) effects. Effects of Aerosol-Cloud-microphysics-Interaction (ACI) will be diagnosed from comparison between RADM and RAD anomalies. Aerosol impacts will be examined using two different kinds of averaging. First, ensemble forecast averaging for day-1 through day-7 for a specific date, will be compared to observation to assess aerosol impacts on forecast skills. Second, ensemble averages of all forecasts as a function of forecast days, will be used to assess the aerosol impact as a function of forecast lead time. For comparing model results to observations, we use rainfall data from the Tropical Precipitation Measuring Mission (TRMM), and winds and temperature data from the NASA MERRA2, Modern Era Retrospective analysis for Research and Applications -version 2 (Bosilovich et al. 2015).

3 Results

To facilitate comparison of model results to observations, and set the stage for the model output analysis, a brief description of the observed large-scale features of the Indian monsoon of 2008 is presented first.

3.1 Observed large-scale monsoon features during June 2008

The excessive rainfall over northern India in June 2008 (see Fig. 1) was accompanied by an upper tropospheric (500–200 hPa) anomalous warming (deviation from climatological mean) northwest of the Indian subcontinent, and an anomalous large-scale anticyclone (Fig. 2a), with strong easterlies over northern Indian and the Arabian Sea (20–35N), consistent with thermal wind balance in an environment with positive temperature gradient (warmer-north and colder-south). The importance of the tropospheric temperature gradient in controlling the timing of the onset of the Indian monsoon is well known (Li and Yanai 1996; Goswami and Xavier 2005; Wu et al. 2012). In the lower troposphere, an anomalous low-level cyclonic circulation cell was found over the northern Arabian Sea, coupled with an anticyclonic cell over southeastern India and the southern Bay of Bengal (Fig. 2b). Associated with these circulation cells were stronger low-level northeasterlies over the Arabian Peninsular and strengthened low-level southwesterlies over the Arabian Sea and westerlies over central India. The former would increase dust emission over the desert surface, and the latter would increase dust and moisture transport from the Arabian Sea to northern India. Strong anomalous cooling of the lower troposphere was found over the Indo-Gangetic Plain (20°–30°N), and anomalous warming above with maximum upper troposphere warming anchored over the Tibetan Plateau (Fig. 2c). These features were accompanied by anomalous ascending motions and increased water vapor over the NIHF with sinking motion and increased dryness over southern India, and northern Indian Ocean (Fig. 2c, d). Increased monsoon westerlies were found in the lower troposphere over central and Northern India, and enhanced easterlies, from the upper troposphere to near surface over the NIHF (Fig. 2d). The aforementioned anomalies signal an enhancement of the early part of the Indian monsoon consistent with the EHP effect due to dust aerosols (Lau et al. 2006, 2008; Lau and Kim 2006). It should be noted that spring and summer of 2008 coincided with the decay phase of the 2007–2008 La Nina, with significant anomalous cooling over the equatorial central and eastern Pacific and warming of the southern tropical Indian Ocean during the boreal spring and summer of 2008. These SST anomalies would have significantly influenced the aforementioned anomalies of the Indian monsoon (Rao et al. 2011; Kim et al. 2015). However, given the prescribed observed SST in the control experiment (NA), SST effects including possible impacts of aerosols on SST are already included in the prescribed observed SST and lateral boundary conditions in NA. Hence model differences (with respect to NA) will only be attributed to aerosol effects associated with aerosol-monsoon interaction on fast times scales, up to 7 days - the longest lead time chosen for the model forecast experiments.

Fig. 2
figure 2

Horizontal distribution of anomalies of a tropospheric temperature (°C) and winds (ms−1) averaged between 500–200 hPa, b 850 hPa winds and specific humidity (gKg−1) in the lower troposphere (1000–600 hPa; and latitude-height cross-sections of c meridional overturning circulation (streamline) and temperature, and d zonal winds (contour) and specific humidity

3.2 Impact on model forecasts

Figure 3 compares the distribution of AOD, rainfall and 850 hPa winds in NA with observed rainfall from TRMM, AOD from MODIS, and winds from MERRA2 for the onset phase of the 2008 Indian monsoon (June 11–17). The NA experiment simulates the mean spatial distribution of aerosol AOD reasonably well compared to MODIS, featuring relatively high aerosol loading over the northern Arabian Sea, and northwestern Indo-Gangetic Plain, and low aerosol loading over northeastern India and the Bay of Bengal (Fig. 3). Overall, the AOD magnitude is underestimated, in some regions with only half the magnitude of MODIS AOD. An examination of the aerosol composition in GOCART has determined that a large-fraction of the AOD over the Arabian Sea and northwestern India and Pakistan is contributed mainly by desert dust (65–70%) transported from the Middle East deserts and the Thar desert of northwestern Indian subcontinent.

Fig. 3
figure 3

Horizontal distribution of AOD a in control model simulation (NA), and b from MODIS observation. Rainfall and winds in c NA experiment, and d from TRMM, and MERRA2. Unit of rainfall in mm day−1, and winds in ms−1

The model mean rainfall and wind patterns show broad similarities to observations (Fig. 3c, d), but with notable differences. The model low-level westerlies are too strong. As a result, precipitation upwind over the mountain ridges of the Western Ghats, and Indo-China are overestimated. The model also missed the heavy rainfall band extending from the east coast of northern India into the Bay of Bengal. These are mostly due to the excessive monsoon low-level westerlies across central and southern India. As shown in following discussion, these biases are not related to model aerosol physics, because they are essential unchanged in the RAD and RADM simulations. Another notable feature is the excessive model rainfall over the southern Indo-Gangetic Plain, and deficient rainfall over the HF compared to TRMM. In the following, we will demonstrate how this bias is mitigated in the RAD and RADM experiments and how we can use the incremental improvement to explore the physical processes underlying aerosol-monsoon interactions over NIHF region. Despite the aforementioned discrepancies, the overall AOD and rainfall distributions in the model are consistent with dust aerosol emissions from the arid region of Southwest Asia, the Thar desert, and Middle East deserts and transport across the Arabia Sea, with strong wet removal in rainy regions.

Comparing the rainfall spatial distribution of RAD, and difference between RAD and NA (Fig. 4a, b) depicts a shift of the monsoon rain band from NI to HF during the onset phase, but little differences elsewhere. This means that given the large-scale aerosol and meteorological initial condition, and on time scale of less than 7 days, aerosol-monsoon interaction effects are mostly found over the NIHF region. The latitudinal rainfall profile averaged over the 80°–90°E sector shows clearly that the excessive NA model rainfall over the NI (23°–25°N), and deficient rainfall over the HF (25°–30°N) compared to TRMM (Fig. 4c, d). ARI effects (Fig. 4e) clearly shift the onset maximum rainfall northward towards the HF, mitigating but not completely removing the model rainfall bias.

Fig. 4
figure 4

Distribution of a total rainfall in RAD, b rainfall anomalies in RAD. Latitudinal profile of rainfall averaged between 80–90E in c TRMM observations, d NA, and e RAD, with positive (negative) RAD-minus-NA differences indicated by red (blue) shading. The boundaries of Northern and Southern Domain defined based on local tropograhy used for later analysis are also shown

To further assess the impacts of aerosol-monsoon interactions on the model forecasts, Fig. 5 shows the time series of the 1–7 days forecast rainfall, for NA and RAD, and TRMM, for the extended integration period through 1 June to July 15 over the northern domain (ND), and the southern domain (SD), defined along the curvature of the Himalaya mountain range (see domain marker in Fig. 4b) to better capture the rainfall anomaly along the HF, arising from aerosol-monsoon interaction. For the ND (Fig. 5a) the model forecasts show reasonable skills in capturing the major episodes of heavy rainfall events. However, after the onset, the model rainfall events tend to be overly active, with only much shallower breaks compared to TRMM observations. It is also noted that clear differences between RAD and NA forecasts tend to emerge after forecast day 3–4, when the control by the initial conditions on the model forecasts are beginning to diminish. Beyond forecast day-4, the number of days with predicted increased rainfall is much higher than that with reduced rainfall, in RAD compared to NA, with a frequency ratio of approximately 2-to-1 indicating a strong impact of absorbing aerosols in enhancing rainfall. The time series of the Day 5–7 mean forecasts (Fig. 5b) show clear enhanced rainfall in RAD relative to NA, during the on-set phase of rainfall events, indicating an advance (~1–5 days) of the onset of monsoon rainfall over the ND (Fig. 5c) due to ARI. The rainfall enhancement and advance of rain events in ND can be seen to last through the entire month of June and to early July. Conversely over SD (Fig. 5c), rainfall is reduced in RAD compared to NA. Beyond forecast day-4, the number of days with reduced rainfall compared that with increased rainfall is approximately 9 to 1, as a result of the northward shift of convection and rainfall away from the SD to the ND. Note that the SD experiences reduction in rainfall mostly during the peak and more often during the dissipation phase of the rainfall events (Fig. 5d).

Fig. 5
figure 5

Time series of a individual 7-day rainfall forecasts June 10–July 21 and b ensemble mean of Day 5–Day 7 forecasts for the Northern Domain, and same for c and d except for the Southern Domain (See domain definition in Fig. 4b). Histogram in a and c indicates observed TRMM rainfall in the Northern Domain and the Southern domain respectively. Red (blue) dots and shading indicate enhanced (reduced) rainfall in RAD relative to NA. Unit of rainfall in mmday−1

3.3 Large-scale interactions

In this section, we explore aspects of aerosol-monsoon large-scale interaction, which governs the northward shift of rainfall from the SD to the ND. The variations of anomalous RAD rainfall and AOD are coherent (Fig. 6a), generally with enhanced rainfall in the ND region (north of 27°N) where AOD is low, and suppressed rainfall in the regions of high AOD in the SD (south of 27°N). This is consistent with increased wash out of aerosols in wet regions, and longer aerosol life times in more stable dry regions. Wind anomalies (Fig. 6b) show a rapid strengthening and northward displacement of westerlies and southerlies in the SD accompanying the monsoon onset over the ND, starting around June 16. This is followed by a major event during June 17–21 with increase in AOD, strengthening southwesteries in the SD coupled to enhanced rainfall in the ND. A similar event involving coherent variations of rainfall, AOD and winds took place during July 3–6. Note that anomalous low-level southerlies after the monsoon onset (~June 16) remain unchanged in sign, extending across approximately 27°N, from the SD to the ND. This means that the ambient moisture in the SD is being advected northward by the strengthened southerlies, thus increasing (decreasing) atmospheric moisture in the ND (SD). The resulting changes in convective stability in terms of the vertical profiles of moist static energy in the two regions will be discuss in Sect. 3.4. The aforementioned features indicate strong aerosol-monsoon interactions associated with advection of aerosols and increased precipitation downwind of the strengthening monsoon southwesterlies, consistent with EHP. The fluctuations in AOD, winds, and rainfall in the ND and SD are associated with development of westward propagating monsoon depression along the Himalayan foothills (See Sect. 3.4, for more detailed discussion). The overall aerosol radiation effects on the large-scale monsoon environment are examined next.

Fig. 6
figure 6

Latitude-time cross-section of ensemble mean forecasts during June 1–17, 2008 for a AOD (contour) and rainfall anomalies (in % change), and b zonal wind (shaded) and meridional wind (contour). Unit of wind in ms−1

The RAD anomalies of upper tropospheric temperature and winds (Fig. 7a) shows clearly a warmer upper troposphere over northern India and regions further north with increased upper level easterlies are found near 18–25°N induced by ARI. The upper level anomalies are coupled to a strengthened low-level monsoon southwesterlies and a pronounced increase in moisture over northwest and northern India (Fig. 7b). Reduced moisture associated with the dry air masses and desert dust transport is found over the central Arabian Sea, and northeastern India. Except for the magnitude, the anomaly patterns are quite similar to those shown in Fig. 2a, b. As stated previously, SST effects are already included in the prescribed observed SST. The present experiments are focused only on the fast responses of aerosol-monsoon dynamic interaction. The results here suggest that ARI reinforces, and adds regional details to, the SST forcing.

Fig. 7
figure 7

Horizontal distribution of RAD anomalies in a temperature averaged between 500–200 hPa and 200 hPa winds, and b lower tropospheric moisture and winds during June 10–17, 2008

Additional regional features induced by aerosol radiative effects are depicted in meridional-height anomaly cross-sections (Fig. 8). Over the Arabia Sea (60–70E), the dust aerosol layer is thick, extending from the surface to 600 hPa, with maximum concentration in the 800–700 hPa layer (Fig. 8a). Pronounced atmospheric warming is found within the dust layer, and cooling near the top of the dust layer mainly due to longwave radiation effects (Zhang and Christopher 2003; Zhu et al. 2007; Lau et al. 2009). Substantial anomalous warming is found over the upper troposphere 300–200 hPA layer, in conjunction with the enhancement of the South Asian High and associated easterly warm advection (Fig. 7a). Surface cooling is found over land surface north of 25°N, reflecting solar attenuation effects due to aerosols (Ramanathan et al. 2005). The differential warming/cooling in the vertical increases stability in the lower troposphere, but enhances instability above, giving rise to weak rising motion in the mid- troposphere (Fig. 8c). As a result, except near the center of dust layer, the atmosphere over the Arabian Sea is generally moister (Fig. 8b). Given that the same sea surface temperature is prescribed in all experiments, the source of moisture is likely due to enhanced surface evaporation from the strengthened surface westerlies over the Arabian Sea, as well as from increased mixing of the moist marine cloud water with environmental air. Wind anomalies show increased easterlies (westerlies) in the upper (lower) troposphere indicating a strengthening of the large-scale monsoon flow. The colocation of low-level westerlies with regions of maximum aerosol loading and dry air masses signals strong westward advection of dust aerosols. As a result of the aforementioned stability changes induced by ARI, an anomalous meridional circulation with northward onshore flow towards southern Pakistan, and largely mid-tropospheric weak rising motion is found over the region. This anomalous rising motion is likely to enhance convective mixing and further moistening of the atmosphere over the ocean. However, the region remains dry in the absolute sense, with no precipitation (Fig. 8d). Over the western Indian region (70°–80°E), the aerosol-monsoon characteristics are similar to those over the Arabian Sea except that the signals are stronger, including a clear northward shift of the center of moderate precipitation, accompanying a strengthening of the local monsoonal meridional circulation (Fig. S2).

Fig. 8
figure 8

Height-latitude cross sections of a temperature (color shaded) and AOT (contoured), b specific humidity (color shaded) and zonal wind (contoured), c meridional streamlines with vertical upward (downward) motions shaded red (blue), and d precipitation anomalies for the Arabian Sea-Pakistan-Afghanistan region (60–70E). Positive (negative) anomalies are indicated by solid (dashed) contours, or red (blue) color. Same for e, f, g and h, except for the northern and northeastern India sector 80–90°E

The ARI effects on precipitation and circulation are strongest over the domain covering eastern India and the western Bay of Bengal (80°–90°E). As shown in Fig. 8e, the ARI-induced upper tropospheric warming over the Tibetan Plateau and regions to the south are very pronounced. This occurs in conjunction with a pronounced elevated thick aerosol layer, pushing against the southern slopes of the Tibetan Plateau by the increasing monsoon southerlies (See Fig. 6b). Here the dust loading is lower compared to that over the Arabian Sea and western India, due to removal by both dry and wet depositions during the long-distance transport from the Middle East deserts to eastern India. However, still noticeable are the cooling near the top of the dust layer, warming within and below the dust layer and near the surface, characteristic of light-absorbing dust layers (Zhu et al. 2007; Lau et al. 2009). A signal of the aerosol semi-direct effect reducing convective potential, in the form of strong surface cooling due to aerosol solar attenuation and warming induced by aerosol heating can be found over the HF (25–35°N, Fig. 8e). However, this effect is overpowered by the increased low-level moisture advection into this region by the strengthened southwesterlies, as evident in the increased moist static energy in the lower troposphere (700 hPa) relative to the regions above (See discussion in Sect. 3.4, Fig. 11). Except within the dust layer, specific humidity is substantially increased over the entire atmospheric column. The increase is most pronounced and coincides with a strong low-level westerly jet near 25–30N, (Fig. 8f). Coupled to the wind, temperature, moisture and aerosol anomalies, is an enhanced local meridional circulation, with strong rising motion in the HF, and sinking motion to the south (Fig. 8g), and the northward displacement of the maximum rainfall over the HF (Fig. 8h).

3.4 Mesoscale Interactions

In this section, we present results on the mesoscale aspects of aerosol-monsoon interactions including both ARI and ACI effects, based on comparisons of the RADM, RAD and NA experiments. Here, we focus on the evolution of mesoscale development of monsoon rainfall systems during the onset phase (June 11–17) of the 2008 Indian monsoon. As shown in Fig. 9 (panels in left column), in the absence of aerosol effects (NA), the onset of the Indian monsoon over northern India was associated with increased transport of moisture to the NIHF regions by the increasing monsoon southwesterlies, resulting in the spontaneous development of a monsoon depression (cyclonic circulation cell) over the northern Bay of Bengal as shown in the NA Day-4 forecast. At Day-5, a secondary cyclone was spurred over the northern-central IGP. Subsequently, both monsoon cyclones amplified and propagated northwestward, over the IGP. Such evolution of monsoon depressions is well known during the monsoon onset, and active phases of the monsoon over northern India (Krishnamurti and Bhalme 1976; Saha et al. 1981). In RAD (panels in middle column), the effect of ARI noticeable starting around Day-5, in a weakening of the primary monsoon depression compared to NA, and in the development of intense small-scale convection cells further northwestward up the HF. The suppression of the primary monsoon depression, and the development of vigorous convective cells along the HF continue, with the most pronounced signal at Day-7. In RADM (panels in right column), the inclusion of ACI further enhances the ARI effects (See discussions related to Figs. 10, 11). The suppression of the primary monsoon onset cyclone is already noticeable at Day-4, and the subsequent transformation (Day-5 to Day-7) of the monsoon onset cyclones (in NA) is similar to ARI effects, except with faster speed, increased intensity and with more numerous convective cells. The metamorphosis of the monsoon depression in NA to invigorated line precipitation cells along the HF, due to combined effects of ARI and ACI is most striking at forecast Day-7.

Fig. 9
figure 9

Time sequence showing the genesis and evolution of monsoon rainfall and mesoscale convective complex from forecast Day 4 to Day 7, for a NA, b RAD and c RADM. Extreme heavy precipitation (>120 mm day−1) is shaded red

Fig. 10
figure 10

Time-height distribution of different hydrometeor species for a the Northern Domain in NA, and b the Southern Domain in NA. Same for c and d, except for the RAD anomalies, and same for e and f, except for RADM anomalies. Purple contour, color shading, and black contour denotes respectively concentration of cloud-ice, snow and graupel, and liquid phase clouds and rain. Positive (negative) anomalies in c through f are indicated either by solid (dashed) contours or red (blue) shading. Units are in 10−5 kg m −3

Fig. 11
figure 11

Vertical profile of MSE (KJ Kg−1) showing a mean distribution for the Northern Domain and b the Southern Domain. Same for c and d except for RAD anomalies for forecast Day 6, and e and f except for RADM anomalies for forecast Day 6

The invigoration of deep convection by ARI and ACI in the HF region is further examined via the vertical distributions of hydrometeors associated with the deepening convection in the ND and SD. Without aerosol effects (NA), the distributions of liquid water (cloud and rain), cloud ice, snow and graupel indicate stronger development of deep convection in SD than in ND (Fig. 10a, b). Comparing RAD (Fig. 10c, d) with NA (Fig. 10a, b), ARI shifts the maximum cloud ice upwards from 250 to 200 hPa, enhances the concentration of snow and graupel in the 500–300 hPa layer, and increases liquid water content in the lower troposphere in the ND. In the SD, the effects are reversed, with substantial reduction in deep convection as evidenced in substantial loss of cloud ice, and snow and graupel in the mid to upper troposphere, as well as suppressed cloud and rain water in the lower troposphere. Comparing RADM (Fig. 10e, f) and RAD (Fig. 10c, d) anomalies indicates that deep convection is further enhanced by ACI in ND, as evident in the larger magnitude and extent of the increase in cloud ice, and snow/graupel in the mid- and upper troposphere. This appears to be at the expense of reduced rate of increase in cloud and rain water in the lower troposphere compared to RAD. In the SD, the stronger reduction hydrometeors of all kinds suggest a further suppression of deep convection by ACI. For the domain as a whole, the change in total rainfall due to inclusion of ACI is relatively small (<5%). Most of the changes by ACI appear to be in the nature of the deep convection and not in the total rain. These results are consistent with modeling and observations indicating enhanced glaciation, and formation of deeper clouds by aerosols in a moist environment [Rosenfeld et al. 2008; Fan et al. 2013].

The changes in the nature of the deep convection induced by ARI and ACI can be viewed in terms of changes in moist static energy (MSE) of the atmosphere over the ND and SD during the onset phase (Fig. 11). Judging from the stronger negative gradient (reduction with height) in mean MSE in the lower troposphere (Fig. 11a, b), it can be inferred that in the absence of aerosol influence (NA), the SD is more (compared to the ND) convectively unstable, and convection is likely to be deeper, as reflected in the larger loading of hydrometeors of all types (Fig. 10a, b). Under ARI forcing, the lower troposphere in ND (Fig. 11c, d) is further destabilized with increase in negative gradient MSE between 800 hPa and 600 hPa, while the mid- to upper troposphere between 600 and 300 hPa are stabilized, as evidence in the positive gradients in MSE between these two levels. Above 300 hPa, the MSE reflects again a convective de-stabilization of the atmosphere. Computations of separate contributions to the MSE (Fig. S3) shows that the destabilization in the lower troposphere is mainly due to increased moisture, while stabilization is mainly due to temperature change, i.e., aerosol semi-direct effect in heating the lower troposphere and cooling of the surface. Changes in MSE in the mid-to upper troposphere is associated with temperature change due to moist adiabatic ascent of heated air from below reaching maximum altitude. Essentially, in the ND, the low-level moisture effect overpowers the aerosol semi-effect, invigorating more intense convection and rainfall. Clearly, ACI enhances the MSE convective stability modulation due to ARI, favoring enhanced deep convection in ND (Fig. 11e). In contrast, for the SD under ARI (Fig. 11f), both the aerosol semi-direct effect and the reduction in moisture by advection (see discussion related to Fig. 7) stabilizes the lower troposphere, leading to the dissipation of the monsoon depression and suppressed precipitation. This effect is further amplified by ACI (Fig. 11f). As noted previously, the northward advection of the un-precipitated moisture from suppressed rainfall in SD by the strengthened southerlies can further increase in convective instability and rainfall in ND. This is the essence of the EHP effect. Overall, it can be argued that the MSE changes provide the physical underpinnings of anomalous deep convection, rainfall and moisture transport induced by ARI and ACI in the NIHF region.

4 Conclusions

Using the NU-WRF model, we have conducted numerical forecast experiments to investigate the dynamics of aerosol-monsoon interaction over the complex topography of the NIHF region, under realistic initial and boundary conditions. Results show that aerosol-monsoon interactions have strong impacts on regional monsoon rainfall and circulation during the onset phase of 2008 Indian monsoon, consistent with the EHP hypothesis. Key findings include:

  • The complex and steep topography of the Himalaya foothills facilitate the build up of thick layers of dust aerosols transported by monsoon southwesterlies from the Arabian Peninsular deserts across the Arabian Sea, and accumulated over northwestern and central India during the pre- and early monsoon season of 2008.

  • Via aerosol induced dynamical feedback, ARI warms the upper troposphere over the Tibetan Plateau, increases the mid-to-upper tropospheric north-south temperature gradient, and thus strengthens the early Indian monsoon by shifting maximum monsoon rainfall from the Indo-Gangetic Plain to the Himalayan foothills regions, and advancing the timing of monsoon onset over HF by 1–5 days.

  • Aerosols have both local and non-local effects, which interacts with each other. Generally, regions of high aerosol loading are associated with drier air masses and suppressed rainfall, and heavy rain regions with low aerosol loading due to wash-out. The semi-direct (local) effect of dust aerosols leads to the stabilization and weakening of a developing monsoon depression over northeastern India and northern Bay of Bengal during the onset phase of the monsoon.

  • Non-local effects are exerted mainly through moisture and circulation feedback. Heating by thick dust layer over the Arabian Sea and northern India, enhances the large-scale southwesterly low-level monsoon flow and transport of moisture from the Arabian Sea to the Indian subcontinent. The enhanced southwesterly moisture transport increases low-level MSE, leading to the development of intense convective cells over the Himalaya Foothill regions.

  • While ARI dominates the large-scale aerosol-monsoon interaction, ACI further enhances the ARI effects, by suppressing development of monsoon depression over northeastern India, while intensifying the ice-phase precipitation and deep cloud processes in the HF, and thereby speeding up the transformation the monsoon depression into intense meso-scale cells with heavy rain in the forms line convection along the rugged terrains of the HF.

Our results are in general agreement with recent modeling studies using the WRF-CHEM. Krishnamurti et al. (2013) showed that the development of monsoon depression is suppressed by aerosol via increased cloud condensation nuclei (CNN), over northeastern India and Bay of Bengal. However, their results did not indicate development of mesoscale-scale deep convection over the Himalayan foothills, most likely because aerosol number concentration was prescribed in their experiments, which therefore did not include the full interactions of aerosol radiation, microphysics, and removal processes, with monsoon dynamics. Jin et al. (2015) using WRF-CHEM and satellite observations also found that the influx of Middle East dust, mainly through radiative forcing, could redistribute and increase Indian summer monsoon regional rainfall, consistent with our results. However the relative low resolution model (~35 km) they used would be unable to resolve the aerosol-monsoon dynamic interaction over the complex topography of Himalayas regions. Moreover, they focused on the mean response of the entire Indian monsoon season (June–August), while we emphasize the short-term (<7 days) interactions during the onset phase of the monsoon. We also note that, even with the 9-km resolution used our experiments, the resolution is still relative coarse, in terms of realistic representation of aerosol-cloud microphysics. Cloud resolving scales (<2–3 km) is highly desirable for better representation the interaction of aerosol with the complex topography of the Himalaya foothills. However, for such high resolution simulation, only short-term integrations, with single or few ensemble simulations are feasible, because of the prohibitive computational demands. Furthermore, as noted in the discussion in Sect. 3.2, while NUWRF simulated the AOD spatial distribution reasonably well, the magnitude of the aerosol loading was significantly underestimated, compared to MODIS observation. Assuming that a heavier aerosol loading will produce stronger impacts on monsoon, it is plausible that the present results may have underestimated the actual impacts of aerosols in reality.

As a caveat, we note that our results are based on a singular monsoon year of 2008. More studies need to be carried out to ascertain the robustness of the results. Nonetheless, the fundamental physical processes revealed by this study, together with many recent studies of aerosol impacts on monsoon weather and climate, lend support to the paradigm that aerosol from natural sources is an essential components of the Asian monsoon climate, affecting variability of the monsoon on diverse spatio-temporal scales (Lau 2016).

Finally, an important message from our study, consistent with many contemporary studies on aerosol-monsoon interactions (See Li et al. 2016 for a comprehensive review), is that aerosol–monsoon interaction is important not only for climate change but also for short-term weather (3–7 days) forecasts, and very likely for medium to long-term forecasts (>7 days) as well as seasonal-to-interannual predictions of the Asian monsoon.