Simulation of South Asian aerosols for regional climate studies

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 117,, doi: /2011jd016711, 2012 Simulation of South Asian aerosols for regional climate studies Vijayakumar S. Nair, 1,2 Fabien Solmon, 1 Filippo Giorgi, 1 Laura Mariotti, 1 S. Suresh Babu, 2 and K. Krishna Moorthy 2 Received 11 August 2011; revised 19 December 2011; accepted 20 December 2011; published 24 February [1] Extensive intercomparison of columnar and near-surface aerosols, simulated over the South Asian domain using the aerosol module included in the regional climate model (RegCM4) of the Abdus Salam International Centre for Theoretical Physics (ICTP) have been carried out using ground-based network of Sun/sky Aerosol Robotic Network (AERONET) radiometers, satellite sensors such as Moderate Resolution Imaging Spectroradiometer (MODIS) and Multiangle Imaging Spectroradiometer (MISR), and ground-based black carbon (BC) measurements made at Aerosol Radiative Forcing over India (ARFI) network stations. In general, RegCM4 simulations reproduced the spatial and seasonal characteristics of aerosol optical depth over South Asia reasonably well, particularly over west Asia, where mineral dust is a major contributor to the total aerosol loading. In contrast, RegCM4 simulations drastically underestimated the BC mass concentrations over most of the stations, by a factor of 2 to 5, with a large spatial variability. Seasonally, the discrepancy between the measured and simulated BC tended to be higher during winter and periods when the atmospheric boundary layer is convectively stable (such as nighttime and early mornings), while during summer season and during periods when the boundary layer is convectively unstable (daytime) the discrepancies were much lower, with the noontime values agreeing very closely with the observations. A detailed analysis revealed that the model does not reproduce the nocturnal high in BC, observed at most of the Indian sites especially during winter, because of the excessive vertical transport of aerosols under stable boundary layer conditions. As far as the vertical distribution was concerned, the simulated vertical profiles of BC agreed well with airborne measurements during daytime. This comprehensive validation exercise reveals the strengths and weaknesses of the model in simulating the spatial and temporal heterogeneities of the aerosol fields over South Asia. Citation: Nair, V. S., F. Solmon, F. Giorgi, L. Mariotti, S. S. Babu, and K. K. Moorthy (2012), Simulation of South Asian aerosols for regional climate studies, J. Geophys. Res., 117,, doi: /2011jd Introduction [2] Extensive studies have been carried out to investigate the regional and global distribution of aerosols and their direct and indirect radiative forcings using a wide variety of climate models having different levels of complexity along with state-of-the-art satellites and ground-based measurements for the last two decades [Kaufman et al.,2002;bellouin et al., 2005; Forster et al., 2007; Myhre, 2009]. The assessments of aerosol climatic impacts are highly uncertain and this contributes significantly to the uncertainty in the total climate forcing estimates [Forster et al., 2007]. Forster et al. [2007] reported aerosol direct radiative forcing of 0.5 W m 2 with an uncertainty range of 0.9 to 0.1 W m 2 at the 90% 1 Abdus Salam International Centre for Theoretical Physics, Trieste, Italy. 2 Space Physics Laboratory, Vikram Sarabhai Space Centre, Thiruvananthapuram, India. Copyright 2012 by the American Geophysical Union /12/2011JD confidence level. This large uncertainty is at least partially attributed to the difference between the measured and simulated radiative forcings [Forster et al., 2007; Myhre, 2009], which questions the accuracy of regional distribution of aerosol properties simulated by climate models. [3] The role of aerosols over the Asian region has been a topic of high interest for the last decade [Lawrence and Lelieveld, 2010], as the anthropogenic emissions from China and India contribute substantially to the global aerosol loading in association with the rapid industrialization and increased energy demands in these countries. Anthropogenic aerosol emissions over Asia are also expected to continue to increase for the next few decades [Ohara et al., 2007]. The incomplete combustion of cheaply available energy sources, such as coal and wood, emits large amounts of black carbon (BC) and organic carbon (OC) into the atmosphere. These carbonaceous aerosols trap radiation and warm the atmosphere, in contrast to sulfate aerosols, which mostly reflect solar radiation and induce cooling at the top of the atmosphere [Forster et al., 2007]. Several studies have shown that large aerosol 1of17

2 loadings over the South Asian region could lead to changes in regional climate, via surface solar dimming and tropospheric warming, inducing, for example, change in the precipitation patterns and early onset of monsoon [Ramanathan et al., 2001; Lau et al., 2006; Meehl et al., 2008; Badarinath et al., 2010]. Even though several impact assessment studies have been carried out over the South Asian region, validation of the simulated aerosol fields over this region is still not exhaustive [Reddy et al., 2004; Ganguly et al., 2009; Menon et al., 2010; Lawrence and Lelieveld, 2010]. [4] The discrepancies between simulated and measured aerosol properties over the Indian region have not been thoroughly investigated [Dickerson et al., 2002; Menon et al., 2010; Lawrence and Lelieveld, 2010]. Therefore, this paper presents a comprehensive validation analysis including the spatial, seasonal, diurnal, and vertical profiles of aerosol BC, along with columnar aerosol optical depth (AOD) over South Asia using data collected from both ground-based networks of aerosol observatories and satellite measurements. The analysis is carried out on a multiyear simulation of aerosol amounts over the South Asia region with a regional climate model (the recently released RegCM4) [Giorgi et al., 2012] including an aerosol module. It serves to identify major systematic errors of the model and their origin, along with a comprehensive assessment of the capability of the RegCM4 to describe aerosol processes over this region. In section 2, we first describe the model, observation, databases, and experiment design. Section 3 then presents a validation analysis for different aerosol characteristics, while section 4 reports the main conclusions. 2. Experiment, Data Analysis and Modeling 2.1. RegCM4 and Experiment [5] In this paper, we use the fourth generation of the regional climate model (RegCM4) [Giorgi et al., 2012] of the Abdus Salam International Centre for Theoretical Physics (ICTP) to simulate the aerosol fields over the South Asian region. Details about RegCM4 and the parameterization schemes available in the model can be found in other studies [Giorgi et al., 1993; Pal et al., 2007; Giorgi et al., 2012], and hence are not repeated here. Briefly, RegCM4 is a limitedarea, hydrostatic model with sigma-p vertical coordinates. The main parameterization schemes used in our experiments are (1) the biosphere-atmosphere transfer scheme [Dickinson et al., 1993] for land process parameterization, (2) the radiative transfer scheme from the National Center for Atmospheric Research Community Climate Model, Version 3 [Kiehl et al., 1996], (3) nonlocal boundary layer scheme from the work of Holtslag et al. [1990], (4) moist convection from the study by Grell [1993] and (5) the subgrid explicit moisture scheme (SUBEX) for nonconvective precipitation [Pal et al., 2000]. Some of these schemes have received modifications as detailed in the study by Giorgi et al. [2012]. The model code is available in the public domain ( eforge.escience-lab.org/gf/project/regcm/), and it is supported through a large user community (RegCNEt) [Giorgi et al., 2006]. [6] The aerosol module of RegCM4 includes sulfate, BC (hydrophilic and hydrophobic), OC (hydrophilic and hydrophobic), dust and sea salt [Qian and Giorgi, 1999; Solmon et al., 2006; Zakey et al., 2006, 2008]. The model uses an online emission scheme for dust aerosols at four size bins (0.01 1, 1 2.5, 2.5 5, 5 20 mm) based on wind speed, soil surface characteristics, and particle size [Zakey et al., 2006]. Following the modified parameterization of Monahan et al. [1986], wind-dependent emission of sea-salt aerosols is calculated for two size bins (accumulation and coarse mode) [Zakey et al., 2008]. The emission fluxes (ng m 2 s 1 )of SO 2, BC, and OC are directly fed into the model from global emission inventory data. The current version of RegCM4 uses the emission inventory of SO 2 from the EDGAR database [Solmon et al., 2006], and BC and OC emissions from the updated inventory of Junker and Liousse [2008] (hereafter Liousse inventory) for the year Monthly mean data of biomass burning emissions are described by Liousse et al. [1996]. The mixing ratio of sulfate aerosols is estimated based on the emission of SO 2 and its conversion to SO 4 as described by Qian et al. [2001]; it uses an oxidant climatology provided by the model for ozone and related chemical tracers (MOZART) chemistry transport model. Carbon chemistry is not explicitly calculated, but the transformation of BC aerosols from hydrophobic to hydrophilic is represented as described in the studies by Cooke et al. [1999] and Solmon et al. [2006]. [7] The mass mixing ratio of the resulting 12 prognostic aerosol variables are calculated by considering advection, vertical and horizontal turbulent diffusion, convective transport, surface emission, wet removal by large-scale and convective precipitation, dry deposition and production, and loss because of physicochemical transformations [Solmon et al., 2006]. The model assumes the following constant dry deposition velocities for all tracers except dust and sea salt over land and ocean, respectively: hydrophilic BC (0.025 and 0.02 cm s 1, hydrophobic BC (0.025 and cm s 1 ), hydrophilic OC (0.025 and 0.02 cm s 1, hydrophobic OC (0.025 and cm s 1 ), SO 2 (0.3 and 0.8 cm s 1, and SO 4 (0.2 and 0.2 cm s 1 )[Solmon et al., 2006]. The dry deposition scheme of size-segregated dust and sea salt aerosols include gravitational settling, Brownian diffusion, impaction, and interception [Giorgi, 1986; Zakey et al., 2006]. Wet deposition of aerosol is estimated using the parameterization of Giorgi [1989] and Giorgi and Chameides [1986] for resolvable scale precipitation and convective precipitation, respectively. [8] The optical properties of aerosols used in this study are taken from Mie calculations for externally mixed aerosols at 12 spectral bands, including a visible broadband ranging from 350 to 640 nm. The mass-specific extinction efficiencies (a) at the visible broadband (at nm wavelength) for the dust aerosols are 2.45, 0.86, 0.38, and 0.17 m 2 g 1 for the four size bins [Zakey et al., 2006], respectively, for hydrophilic and hydrophobic BC are 6.6 and 9.4 m 2 g 1 and for OC 4.3 and 3.4 m 2 g 1, respectively. The hygroscopic growth of the optical properties of sulfate, hydrophilic BC and OC, and sea salt is taken into account following the parameterization of Kiehl et al. [2000] and Kasten [1969]. [9] The domain of simulation is shown in Figure 1. This is essentially the same as the South Asian domain of the coordinated regional climate downscaling experiment (CORDEX) program [Giorgi et al., 2009]. Following the CORDEX specifications, the model horizontal grid spacing is 50 km. Out of the 11 Aerosol Robotic Network (a Sun photometer network) (AERONET) stations considered for this study, 2of17

3 Figure 1. South Asian domain of model simulation (dotted yellow boundary). Intercomparison sites of Sun/sky radiometers (AERONET stations, solid circles) and in situ measurement sites of BC (ARFINET, open stars) measurements. Solar Village, Kanpur, Trivandrum, and Kharagpur are identified. Kanpur and Solar Village are identified in Figure 1, as the AOD data from these sites are investigated in greater detail. Similarly, long-term BC data from Kharagpur and Trivandrum are also examined in detail. The model simulation covers the 3 year period , with six-hourly lateral boundary conditions provided from ERA-Interim reanalysis of meteorological observations (pressure, temperature, humidity, wind speed, and wind direction) [Uppala et al., 2008] Observation Database [10] In general, aerosol modules in climate models and chemical transport models are validated using quality-assured satellite data and/or ground-based in situ measurements of aerosol parameters, mainly AOD, made at individual or networks of stations [Holben et al., 1998; Kinne et al., 2006]. In this study, both ground-based and satellite data are used to validate the aerosol properties simulated by RegCM Ground-Based Observations [11] Aerosol Radiative Forcing over India (ARFI) is a research project under the Geosphere Biosphere Programme (GBP) of Indian Space Research Organization (ISRO) aimed at understanding the radiative and climatic implications of atmospheric aerosols over the Indian region. One of the main objectives of the ARFI project is to establish a network (ARFINET) of aerosol observatories (currently 33 stations) located at distinctively different environments spanning the entire Indian territory. The network stations of ARFI are well equipped with instruments to measure physical, chemical, and optical properties of aerosols [Beegum et al., 2008, 2009]. In this study, near real time and continuous measurements of mass concentration of BC aerosols are made using a multiwavelength Aethalometer (Magee Scientific, USA) to validate the model simulations. The Aethalometer uses a filter-based optical attenuation technique to measure the BC mass concentration by assuming a mass-specific absorption cross section of 16.6 m 2 g 1 at 880 nm [Hansen et al., 1984]. Aethalometers aspirate the ambient air (above 10 m) at a flow rate of 3 LPM, and measurements were made at 5 min intervals [Nair et al., 2007]. Inherent uncertainties associated with the filter-based optical attenuation technique include shadowing and multiple-scattering effects, which can contribute up to 20% of uncertainty in the measurements [Arnott et al., 2005; Subramanian et al., 2007; Lack et al., 2008; Nair et al., 2008]. Variations in assumed massspecific absorption cross section of BC (16.6 m 2 g 1 ) can be another source of uncertainty in the estimation of BC mass concentration using Aethalometer. Multiyear measurements of BC mass concentration made at Kharagpur (22.5 N, 87.3 E) and Trivandrum (8.5 N, 77 E) are discussed in detail and BC values available in the literature are used for other stations. Trivandrum is a suburban location far away from local sources and industries, whereas Kharagpur lies at the mouth of the outflow region to the Bay of Bengal from Indo-Gangetic Plain (IGP). Aethalometers were operated at Trivandrum and Kharagpur for the period [12] AERONET is a worldwide network of CIMEL Sun/sky radiometers maintained by NASA to measure optical (spectral AOD, spectral single-scattering albedo), and microphysical properties of aerosols on the basis of direct Sun and sky radiance measurements [Holben et al., 1998]. These automatic radiometer data are widely used for validating the satellite retrievals as well as climate model simulations [Stier et al., 2005]. The uncertainty in AERONET measured AODs at wavelengths >440 nm are In this study, monthly mean values of level 2.0 (quality-assured) AOD at 500 nm made at 11 locations are used (see Figure 1) Satellite Observations [13] In this study, we use the validated level 3 gridded AOD data from the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Aqua satellite and the 3of17

4 Multiangle Imaging Spectroradiometer (MISR) onboard the Terra satellite. We have merged the deep blue AOD of MODIS over land and MODIS AOD over ocean to obtain a complete spatial coverage over the study domain. By considering the advantage of the multiangle measurements over bright land surfaces, monthly mean AODs at 558 nm retrieved from MISR are also used. MISR views the same scene at nine different angles, and this precludes the assumption of absolute reflectance of the land surface [Abdou et al., 2005]. 3. Results and Discussions 3.1. Simulated Monsoon Climatology [14] In this section, we first provide a brief analysis of the model performance in reproducing the monsoon climatology for the simulation period, as this is important for the aerosol transport and removal. Figure 2 compares the seasonal evolution of precipitation and low-level (850 hpa) winds between the RegCM4 and observations from the Tropical Rainfall Measuring Mission data set [Huffman et al., 2007] for precipitation and the ERA-Interim reanalysis for wind. The model generally captures the evolution of the monsoon circulation, with northeasterly monsoon flow in winter (December January February, DJF) and southwesterly flows in summer (June July August September, JJAS). The evolution of precipitation over land is also realistically reproduced, with a clear summer monsoon maximum over the Indian subcontinent. The main model deficiency is found over the Bay of Bengal and Arabian Sea in JJAS, where the model substantially overestimates precipitation. This problem is mostly tied to the use of prescribed sea surface temperatures (SSTs) over these regions, as demonstrated by an experiment in which an earlier version of RegCM was coupled to a regional ocean model by Ratnam et al. [2009]. The evaporation feedback associated with interactive SST is thus important for an improved simulation of precipitation over the Indian Ocean; however, Figure 2 shows that even without such coupling, the simulation of land precipitation, which is most important for aerosol modeling, is realistic. On the other hand, the model s systematic bias in summer ocean precipitation should be taken into proper consideration in the evaluation of the aerosol simulation Model Validation Against Satellite AOD [15] Aerosol properties vary spatially and temporally in association with the heterogeneities in the sources and sinks characteristics and because of the transport by large-scale circulation patterns (long-range transport). As satellites are the only means to spatially cover large regions or the entire globe with frequent observations, we first use the gridded monthly mean products of MODIS and MISR for validating the RegCM4 aerosol module. Figure 3 shows the AOD retrieved from MISR at 558 nm and MODIS at 550 nm along with the AOD simulated by RegCM4 at nm for the winter monsoon (DJF), premonsoon (March April May, MAM), summer monsoon (JJAS) and retreating monsoon (October November) phases. Several studies addressed the seasonal transformation of aerosols over the Indian region and more details are available in studies by Moorthy et al. [2007], Dey and Tripathi [2008], Ganguly et al. [2006], and Satheesh et al. [2006]. In general, RegCM4 simulates the broad features of spatial and seasonal variation of AOD: (1) high AOD over west Asia and Taklamakan during premonsoon and summer monsoon seasons, (2) high AOD over the Arabian Sea during summer, and (3) low AOD values over the Indian Ocean. The simulated AOD values are close to the satellite observations during the premonsoon, whereas an underestimation is observed during winter over the IGP. The discrepancy dominates over the regions where anthropogenic activities are more compared to the natural aerosol regimes. The apparent discrepancies between the AODs from simulation and retrieved from MODIS data over west Asia and northern Arabian Sea during summer season are attributable to (1) larger uncertainties in the MODIS retrieval over bright surfaces, (2) effect of clouds, and (3) inherent uncertainties in model simulations. Several observations have shown the presence of large abundance of dust over these regions, advected by the strong winds from African, Arabian, and Indian deserts [Moorthy et al., 2005; Zhu et al., 2007]. The consequent brightening of the background reduces the contrast leading to larger uncertainty in the AOD retrievals. Furthermore, these regions exhibit frequent and widespread cloud cover during this season associated with the Asian summer monsoons, rendering fewer cloud-free days and increased cloud contamination in the retrieved AOD values. [16] It is noteworthy that, in general, the MODIS AOD is higher than the MISR AOD over land and vice versa over the ocean. It should, however, be stressed that the MODIS retrievals of aerosol properties are not highly accurate over the South Asian region, especially over the Indian continent because of the inaccurate representation of surface albedo in MODIS retrieval algorithm [Jethva et al., 2009]. Abdou et al. [2005] reported that MISR AOD values are close to the AERONET measurements over land. Prasad and Singh [2007] noticed that the correlation between MISR and MODIS AOD with AERONET data have a seasonal dependence, and overall, MISR performs better over the IGP because of its multiangle, multispectral, and narrowband measurements. Conversely, MODIS aerosol observations over the ocean are highly correlated with shipborne and island measurements [Vinoj et al., 2004] Model Validation Against AERONET Measurements [17] Figure 4 shows the comparison between AODs simulated using RegCM4 at nm with mean monthly AOD at 500 nm measured using CIMEL radiometers at seven AERONET sites distributed across the South Asian domain. Note that most of these stations are located in west Asia, and only a few stations are available in the Indian region. It can be seen that the model performance varies over different regions depending on the dominant aerosol species. For example, RegCM4 reproduces fairly well the AOD over the dust-dominated locations (Solar Village, Bahrain, Muscat, and Hamim), whereas significant underestimations are observed over areas such as Kanpur, Lahore, and Karachi, where anthropogenic aerosols dominate in the total aerosol loading. In general, most of the simulated AODs lie just close to the 1:1 line drawn in Figure 4 (right). To obtain a quantitative picture, we have done a linear regression analysis over western Asia; the intercept, slope, and correlation coefficient (R) of the best fit to the AOD at each site are 4of17

5 Figure 2. (a d) Simulated and (e h) observed precipitation (mm/day) and 850 hpa wind (m s 1 ) in the four seasons: DJF (Figures 2a and 2e), MAM (Figures 2b and 2f), JJAS (Figures 2c and 2g), October November (ON, Figures 2d and 2h). Precipitation observations are from the Tropical Rainfall Measuring Mission data set while observed winds are from the ERA-Interim reanalysis. 5of17

6 Figure 3. Spatial distribution of AOD measured using MODIS and MISR and simulated by RegCM4 for the winter (DJF), premonsoon (MAM), summer (JJAS), and retreating monsoon (ON) seasons of year as follows: (1) Solar Village: AOD AERONET (R = 0.62), (2) Bahrain: AOD AERONET (R = 0.65), and (3) Hamim: AOD AERONET (R = 0.85). Over polluted locations the best fit values are as follows: (1) Kanpur: AOD AERONET (R = 0.35), (2) Lahore: AOD AERONET (R = 0.21), and (3) Karachi: AOD AERONET (R = 0.7). It should be noted that the model simulates the AOD in broadband frequencies (at nm) and radiometer-derived AODs are at narrowband centered at 500 nm. In general, it is found that simulated AOD values are in good agreement with AERONET AOD over dust-dominated areas, whereas model tends to underestimate AOD over anthropogenically impacted regions. [18] It is well known that aerosols over the South Asian region have a strong seasonal dependence (Figure 3), and many models have difficulties in reproducing this seasonal variability over the region. To investigate the model s ability 6of17

7 Figure 4. Scatterplot of monthly mean AODs simulated using RegCM4 and measured using AERONET at (left) west Asia (dust dominated) and (right) subcontinental urban locations for the period Dotted lines indicate the 1:1 line drawn between the simulated and measured AODs. Linear regression fit to the entire data on the each plot is shown as solid line and the regression coefficients (slope, intercept, and correlation coefficients) are given in the respective plots. to capture the seasonal variation of AOD, we considered two AERONET stations, Solar Village (24.9 N, 46.4 E) and Kanpur (26.5 N, 80.2 E). The former represents a west Asian dust-dominated region, whereas the latter experiences high dust loading during the premonsoon and summer monsoon seasons and high anthropogenic and biomass burning aerosol loading during the winter season. The comparison of AERONET AODs at 500 nm with simultaneous RegCM4 AOD values at the model band nm over Solar Village and Kanpur are shown in the first and second panels of Figure 5, respectively. In general, the model captures the magnitude and seasonal variation of AOD at the Solar Figure 5. Temporal variation of monthly mean AOD simulated and measured at Solar Village and Kanpur for the period 2005 to of17

8 Figure 6. Ratio of measured to simulated BC over the Indian stations during the (top) premonsoon and (bottom) winter seasons. The color scheme and size of the circles indicate the magnitude of the ratio of measured to simulated BC. The numbers on the labels indicate the source of data; 1: Land Campaign II data [Nair et al., 2007], 2: ICARB [Beegum et al., 2009], 3: Kumar et al. [2011], 4: Pathak et al. [2010], 5: Ganguly et al. [2006], 6: Sreekanth et al. [2007], 7: Ram et al. [2010b], and 8: Rengarajan et al. [2007]. Village, where mineral dust aerosols dominate throughout the year. Compared to the Solar Village, RegCM4 tends to underestimate the AOD at Kanpur, especially during winter. One of the possible reasons would be the uncertainty in representing the seasonal and regional anthropogenic emission sources (fossil fuel and biomass burning) accurately [Ganguly et al., 2009; Menon et al., 2010]. On the basis of the Indian Ocean Experiment (INDOEX) experiment, several authors have reported that the aerosol fields simulated by climate models underestimate AOD over the Indian region especially during the winter season [Dickerson et al., 2002; Reddy et al., 2004] Black Carbon Over South Asia [19] Even if the model-simulated AODs are close to observations, there can be large discrepancies in the chemical composition and vertical distribution of aerosols. In this regard, BC is one of the most uncertain components of the aerosol system, especially over South Asia [Kinne et al., 2006, Koch et al., 2009] Validation Against Measurements [20] The spatial distributions of the ratio of measured to simulated BC over the Indian region during the premonsoon (MAM) and winter (DJF) seasons are shown in Figure 6. The majority of data used in Figure 6 were taken during the Land Campaign II (December 2004) and Integrated Campaign for Aerosols, gases and Radiation Budget (ICARB) experiment (MAM, 2006), while the data from the other stations were collected at different times. The measurements at Kanpur during premonsoon and at Hissar (semiarid region in the northwestern India) during winter were made using a thermal technique [Ram et al., 2010a, 2010b], whereas all other measurements were made using an optical attenuation technique [Nair et al., 2007]. Most of these stations are part of the ARFINET network and more details can be found in other studies [Nair et al., 2007; Beegum et al., 2009; Pathak et al., 2010]. The main features are as follows: 1. Model simulations are closer to the measurements during the premonsoon than the winter season. 2. Simulated BC values are 5 to 7 times lower than the 8of17

9 Figure 7. Monthly mean variation of measured (open circles) and simulated (solid circles) BC mass concentration at (top) Kharagpur and (bottom) Trivandrum for the period measurements at Hyderabad, Dibrugarh, and Port Blair during the winter and premonsoon seasons, mainly because of the influence of severe biomass burning events [Kharol and Badarinath, 2006; Kaskaoutis et al., 2011]. The measured BC at Port Blair is 7 times higher than the simulated BC during winter and 8 times higher during premonsoon. Moorthy and Babu [2006] have shown that, being in the downwind region of the IGP and East Asia, the BC mass concentration at Port Blair is significantly influenced by the air masses traveling from these aerosol-laden regions. 3. Significant underestimation of simulated BC is observed during both seasons over Trivandrum and Visakhapatnam. 4. The model performs well over Minicoy (ratio of model to measurement is close to 1), an island in the south Arabian Sea, which is influenced by the advection of air masses from the southern peninsular India (during winter) and west Asia (during premonsoon). 5. In contrast with other areas, the model overestimates BC at Nainital (station representing the Himalayan foothill region) during the winter. [21] Generally, seasonal variation of BC mass concentration shows winter high and summer low over the Indian region in association with the reversal of large-scale synoptic circulation pattern of the Indian monsoon [Moorthy et al., 2007]. Figure 7 illustrates the intercomparison of measured and simulated BC mass concentrations at Kharagpur and Trivandrum for the entire simulation period ( ). The model underestimates the BC concentration at both stations; these underestimations are much higher during winter at Kharagpur and during the monsoon season at Trivandrum. The BC mass concentration observed at Kharagpur is much higher than at Trivandrum, which indicates that industrial and residual emissions are much stronger over the IGP compared to those in the southern peninsula [Nair et al., 2007]. It is well known that the anthropogenic aerosol loading over northern India increases during the winter season because of increased emissions from local and residual sources and limited dispersion of aerosols (reduced ventilation) because of the decrease in boundary layer height and wind velocity [Nair et al., 2007]. In addition, local orography of the IGP also supports the accumulation of aerosols during winter [Nair et al., 2007]. This confinement of aerosols during the winter season leads to severe fog events, health issues, and visibility reduction [Gautam et al., 2007]. [22] The annual mean values of measured near-surface BC at Kharagpur and Trivandrum are and mg m 3, respectively. The corresponding values of simulated BC are and mg m 3, respectively. The model, therefore, substantially underestimates near-surface BC concentrations at both sites. A similar underestimation is also found at Minicoy (an island in the Arabian Sea, 6.26 N, 73 E), where model values are closer to the observation during winter (measurements: mg m 3 and model: mg m 3 ) and premonsoon season (measurements: mg m 3 and model: mgm 3 ) but significantly underestimated during summer (measurements: mgm 3, model: mg m 3 ) (values were taken from Vinoj et al. [2009]). As seen in Figure 2, this underestimation in the simulated BC over Minicoy during summer is mainly associated with the wet removal of BC because of the excess rainfall over the southeastern Arabian Sea. In general, the large difference between the model and measurements can be attributed to uncertainties in the inventory data, measurement uncertainties, or systematic errors in the simulation of atmospheric 9of17

10 Table 1. Comparison of Seasonal Variation of BC Mass Concentration Over Kharagpur Simulated by RegCM4 Using Liousse [Junker and Liousse, 2008] and REAS [Ohara et al., 2007] Inventory Data With in Situ Measurements a BC Concentration (mg m 3 ) Inventories Used Junker and Liousse Seasons Measurements [2008] REAS Winter monsoon (DJF) Pre monsoon (MAM) Summer monsoon (JJAS) Retreating monsoon (ON) a DJF, December January February; MAM, March April May; JJAS, June July August September; ON, October November. processes. Additional discussion of the relative role of these sources of uncertainty is presented in the following sections. [23] BC is one of the main aerosol components over South Asia, contributing 50% to the surface PM 2.5 over the IGP [Ram and Sarin, 2011]. Koch et al. [2007] noted that South Asia contributes 7 to 8% to the global BC aerosol loading, and almost 80% of the anthropogenic non-biomass-burning BC is transported to the rest of the world via large-scale circulations. Even though several studies have highlighted the climatic potential of light-absorbing carbonaceous aerosols over this region [Meehl et al., 2008; Lau et al., 2006], large uncertainty persists in the magnitude of the simulated BC mass concentration and its major sources [Novakov et al., 2000; Menon et al., 2010; Lawrence and Lelieveld, 2010]. In fact, it is extremely challenging to simulate aerosol fields, and in particular BC, over the South Asian region because of the lack of accurate information on the source characteristics and microphysical properties of the particulate material [Gustafsson et al., 2009; Ganguly et al., 2009] Role of Inventories [24] Large differences in measured and simulated BC mass concentrations call for the need to re-examine the emission inventory data as well as inherent uncertainties in the measurement techniques used over this region. In order to explore the uncertainties related to the inventory data, we ran an additional RegCM4 simulation in which the Liousse inventory [Junker and Liousse, 2008] was replaced by the Regional Emission Inventory in Asia (REAS) [Ohara et al., 2007]. The comparison of seasonal variation of BC mass concentration simulated at Kharagpur using the Liousse and REAS inventories is given in Table 1. Kharagpur is selected because of its unique location, at the mouth of the outflow region to the Bay of Bengal, and because of the availability of continuous and long-term data. Even though in both the cases BC mass concentration at Kharagpur is underestimated, the simulation using the REAS inventory values are closer to the measurements; in fact, during summer monsoon, the REAS values are higher than the measurements. Recently, Menon et al. [2010] also reported large differences between the global climate model (GCM)-simulated (GISS ModelE) and measured BC mass concentration over India. These authors used the Bond et al. [2004] and Sahu et al. [2008] inventories for their simulations and found that the differences between the measurements and simulated BC were much higher than the differences between simulations with the two emission inventories. Later, Ganguly et al. [2009] and Menon et al. [2010] also found large differences between the simulated and measured BC mass concentrations over the Indian region. Some of the earlier studies also showed that AODs simulated by climate models are substantially lower than the measured AODs over the Indian region [Reddy et al., 2004; Ganguly et al., 2009], especially during winter. This has been commonly attributed to the inadequate representation of the aerosol inventories [Ganguly et al., 2009]. [25] Indeed, inventories of aerosol emissions over India are quite uncertain because of the lack of spatially resolved basic activity data [Sahu et al., 2008]. On the basis of carbon monoxide (CO) and BC measurements during INDOEX, Dickerson et al. [2002] reported large discrepancies between the simulated and measured BC concentrations over the Indian region, attributing them to missing sources of BC or analytical errors in the measurements. These authors estimated a BC emission flux of 1.7 to 2.8 Tg yr 1 (using shipborne data) over the Indian region, whereas the emission rates of BC reported from regional and global inventories (REAS Tg yr 1, Beig 1.3 Tg yr 1, and Bond 0.48 Tg yr 1 ) are much lower than the top to bottom approach [Bond et al., 2004; Ohara et al., 2007; Sahu et al., 2008]. In addition, inventory-based estimates show that biomass burning contributes 77% to the total BC [Ohara et al., 2007] emitted from South Asia, whereas measurements show fossil fuel emissions (50 90%) as the dominant BC source [Novakov et al., 2000]. The wide range of values reported by different BC inventories and contrasting results from the bottom-up and top-down approaches reveal the complexity of the problem [Gustafsson et al., 2009; Lawrence and Lelieveld, 2010] Measurement Uncertainties [26] The measurements of BC mass concentration over the Indian region were carried out mostly using the Aethalometer working on the principle of optical attenuation technique [Nair et al., 2007, 2008; Beegum et al., 2008]. This technique suffers from inherent uncertainties associated with filter loading and multiple scattering effects [Arnott et al., 2005; Nair et al., 2008]. Moreover, the Aethalometer uses a manufacturer-specific mass absorption cross section of 16.6 m 2 g 1 at 880 nm to calculate the BC mass concentration from the measured change in the attenuation because of the aerosols deposited on the filter. Several studies have shown that the mass absorption efficiency has a large spatiotemporal heterogeneity depending on the source characteristics [Ram et al., 2010b]. Ram et al. [2010b] showed that the mass-specific absorption cross section of BC aerosols at Kanpur is higher (20.7 m 2 g 1 ) than the value used in the Aethalometer measurements (16.6 m 2 g 1 ). To further investigate the dependence of the measurement uncertainties on the discrepancy between measured and simulated BC concentration (Figures 6 and 7), the mass concentration of elemental carbon (EC) measured using the thermal technique is compared to corresponding simulated values (Figure 8). [27] Ram et al. [2010a] have reported a year-round measurement of EC mass concentrations at Kanpur, a polluted station located in the IGP using the thermal technique (Sunset Laboratory, USA); more details are available in the work of Ram et al. [2010a, 2010b]. An intercomparison of EC mass concentration measured using the thermal technique [Ram et al., 2010a, Table 4] and simultaneous values simulated using RegCM4 are shown in Figure 8. It is clear 10 of 17

11 Figure 8. Monthly mean variation of mass concentration of elemental carbon (EC) measured and simulated over Kanpur during Measurements were taken from Ram et al. [2010a]. that the model values are close to the EC measurements in contrast to the BC values shown in Figures 6 and 7. Even though several authors, including Ram et al. [2010b], reported a good agreement between BC measured using the Aethalometer and EC measurements using the thermal technique [Sharma et al., 2002], this apparent inconsistency is because of the fact that the EC values reported by Ram et al. [2010b] were measured only during daytime hours [Ram et al., 2010b, also personal communication, 2011], whereas BC values are daily averages of 288 measurements per day. It should be noted that aerosols over the Indian region exhibit large diurnal variations, with BC values being almost 4 to 6 times higher at night than at noon [Nair et al., 2007; Beegum et al., 2009]. Hence, in order to understand the model biases it is important to examine the diurnal variation of BC in detail BC Diurnal Variations [28] The diurnal variation of measured BC mass concentrations at Kharagpur and Trivandrum is shown in Figure 9 for December and June 2006, which represent the typical diurnal pattern of BC during the winter and summer monsoon seasons. The diurnal variation in BC mass concentration is significantly higher during the winter than the summer season at both stations. BC values measured using the Aethalometer at noontime are 2 4 times lower than at night. The model also shows a substantial diurnal variation of nearsurface BC; however, this is of much smaller amplitude than the observed. Simulated BC values are very close to the measurements at noontime but are underestimated during the night. Hence, it is clear that the better agreement between the model and the EC concentration shown in Figure 8 is because these measurements were taken during daytime. The large mismatch between the model BC and daily mean BC values shown in Figure 7 is attributed to the inability of the model to reproduce the extremely high nighttime BC values observed at Kharagpur and Trivandrum, especially during winter. The diurnal cycle of near-surface aerosol concentration is because of the fact that the vertical transport by the turbulent boundary layer and by deep convection is much more efficient during daytime than at night, when the boundary layer is shallow and stable conditions prevail (especially in winter). While the model appears to capture the daytime vertical transport, it evidently under-represents the very stable nighttime conditions and therefore overestimates the vertical aerosol transport and underestimates the near-surface aerosol concentrations. In fact, the nonlocal boundary layer scheme used in RegCM4 tends to overestimate vertical mixing in very stable conditions [Giorgi et al., 2012]. It is clear from Figure 9 that amplitude of the diurnal variation of BC is higher over Trivandrum compared to Kharagpur during summer. Being a coastal station, Trivandrum experiences land-sea breeze circulation, which plays a significant role in modulating the diurnal variation of aerosol properties, compared to Kharagpur, which is more inland. It is observed that residential biomass burning activities have diurnal as well as seasonal variations. Over the rural regions of India, cooking is mainly done using wood, leaves, and cow dung as fuel during the early morning and in the evening hours. Hence, the BC measurements may respond to these local emissions, whereas models use average and time-invariant emission fluxes of aerosols. Ram and Sarin [2011] found significant difference in the chemical composition of the aerosols samples collected during the daytime and nighttime. It is extremely difficult to model these numerous and spatially distributed emission sources because of their heterogeneous and time-varying nature. In addition, models use annual average emission inventories, which may not be valid over India because of the seasonality in some of the anthropogenic sources, such as bonfire used for heating. [29] The mean monthly variation of measured and simulated BC at noontime over Kharagpur and Trivandrum for the period (Figure 10) shows that the simulated values are more comparable to observations at noontime 11 of 17

12 Figure 9. Monthly mean diurnal variation of BC mass concentration measured using Aethalometer (open circles) and simulated using RegCM (solid circles) for the month of December (a) over Trivandrum and (c) over Kharagpur and for the month of June (b) over Trivandrum and (d) over Kharagpur. Shaded vertical bars indicate the rise and setting times for Sun. Figure 10. Monthly mean variation of BC mass concentration measured and simulated over Kharagpur and Trivandrum at noontime during Vertical bars represent the standard deviation of the data. 12 of 17

13 Figure 11. Vertical distribution of noontime and nighttime BC mass concentration simulated using RegCM4 over (left) Kharagpur and (middle) Trivandrum during December and June. (right) Measurements of BC were made onboard the aircraft over Kanpur on 4 January 2004 (adapted from Tripathi et al. [2005]). Kindly note that vertical scales of both the plots are different. rather than daily averages. Compared to Trivandrum, comparison was better over Kharagpur. [30] On the basis of simultaneous measurements of aerosols and boundary layer parameters, Nair et al. [2007] reported that diurnal variations of BC mass concentration over the Indian region is mainly associated with boundary layer dynamics, whereas vehicular and traffic emissions modulate the diurnal BC pattern as seen in most of the developed and midlatitude countries [Moorthy and Babu, 2006; Nair et al., 2007; Sreekanth et al., 2007; Beegum et al., 2009; Kumar et al., 2011]. Similar diurnal patterns have been observed over other South Asian countries such as Pakistan [Dutkiewicz et al., 2009] and western China [Engling et al., 2011]. Diurnal variations of BC observed at midlatitudes peak during the morning and evening hours associated with traffic density [Sharma et al., 2002; Venkatachari et al., 2006; Järvi et al., 2008; Saha and Despiau, 2009] and show significant differences on weekdays and weekends because of the decrease in vehicular density and shift of the peak traffic hour [Järvi et al., 2008]. In contrast, diurnal patterns over the South Asian region depend mostly on the boundary layer evolution, even over industrialized areas of the IGP [Nair et al., 2007]. Therefore, the correct representation of the near-surface mass concentration requires improvements in the model boundary layer transport processes, especially during nighttime stable conditions. In reality, emissions sources of BC are not constant in time (as assumed in the model), and this can also contribute to the observed discrepancy between measured and simulated diurnal patterns [Venkatachari et al., 2006] BC Vertical Distribution and Transport [31] Figure 11 shows the simulated diurnal evolution of the vertical distribution of BC mass concentration at Kharagpur and Trivandrum for December and June (representing the winter and summer season, respectively), along with the diurnal variation of vertical profiles of BC measured at Kanpur on 4 January It can be seen that BC aerosols residing within the boundary layer show different distributions during day and night, with higher nighttime values and differences across seasons. However, the simulated diurnal variations are lower than indicated by the measurements. Figure 11 (right) shows the measured vertical profiles of BC mass concentrations at Kanpur as part of the Land Campaign II during the morning (10:39 to 11:58 LT) and evening (16:06 to 17:35 LT) hours of 4 January 2004 (adapted from Tripathi et al. [2005]). These authors reported a large variation in the BC concentration during the evolution of the boundary layer. This comparison thus confirms that the model tends to overestimate turbulent vertical transport during stable nighttime conditions. [32] Vertical profiles of BC aerosols were made during the ICARB experiment over Bhubaneswar, Chennai, Trivandrum, and Goa during MAM 2006 period [Babu et al., 2008]. Figure 12 shows the vertical profiles of BC mass concentration measured using the Aethalometer onboard the aircraft during the ICARB 2006 campaign and simulated by RegCM4. It can be seen that the simulated and measured BC mass concentrations are in broad agreement at most of the Indian sites, except Trivandrum. As all the measurements were made within the convectively evolved daytime boundary layer, the agreement between the model and measurements is in line with the earlier discussion and support the conclusions from Figure 10. From Figures 10 and 12, it is clear that the model simulates the surface and vertical distribution of BC mass concentration close to the measurements during the daytime (unstable and convective boundary layer), in contrast to that measured during the nighttime (stable boundary layer conditions). These results again suggest that the nonlocal boundary layer scheme in RegCM4 tends to 13 of 17

14 Figure 12. Intercomparison of vertical profiles of BC mass concentration measured onboard aircraft during ICARB and simulated using RegCM4 over Bhubaneswar, Chennai, Trivandrum, and Goa. underrepresent the stagnant conditions in the nighttime stable boundary layers. This aspect of the model needs to be improved in future developments. [33] Several investigators have studied the vertical distribution of aerosols over the Indian region using lidar observations from different stations (Trivandrum, Visakhapatnam, Bangalore, Gadanki, Ahmedabad, Hyderabad, and Kharagpur) for the last two decades. Even though these experiments produced aggregated vertical profiles of composite aerosol extinction coefficients and their seasonal variations, not much information is available on the vertical distribution of aerosol composition. There are a few attempts to understand the vertical distribution of BC using airborne measurements over specific locations, such as Hyderabad, Kanpur, Bhubaneswar, Chennai, Trivandrum, Goa, and Port Blair [Moorthy et al., 2004; Tripathi et al., 2005; Babu et al., 2008, 2011]. These flights were made as a part of the Land Campaign II, ICARB, Winter-ICARB, and MT-continental tropical convergence zone field campaigns. Intercomparison of BC profiles measured onboard the aircraft and simulated using climate models does not exist over the Indian landmass. During the winter season, BC concentrations decrease drastically to very low values above the boundary layer, whereas significant amounts of BC are observed in the free troposphere during the premonsoon period [Moorthy et al., 2004; Babu et al., 2011]. As the aircrafts measurements in the unpressurized mode is limited only up to 3 km, Babu et al. [2011] have used high-altitude balloon measurements to obtain vertical profiles of BC up to 10 km. These authors reported an elevated aerosol layer at 4.5 km at Hyderabad in central India. These observations confirm the need for examining the vertical distribution of BC aerosols to evaluate the model outputs. For our experiment, the intercomparison of simulated and measured vertical profiles of BC during ICARB shows a good agreement at distinct locations, indicating that the underestimation of modeled BC is not consistent throughout the column. This agreement between the model and measurements is encouraging, as global models generally overpredict the BC concentration above the boundary layer over tropics and low-latitude regions [Koch et al., 2009]. It is thus clear from our comprehensive validation exercise that a number of areas need to be further investigated in order to improve our understanding of the effects of aerosols on climate and air quality over South Asia. 4. Conclusion [34] The spatial, temporal, and altitude distributions of aerosols over the Indian region, for the period , have been simulated using RegCM4. These estimates were examined using ground-based, airborne, and satellite measurements over the same region, during the same period. The highlights are as follows: 1. The RegCM4 simulations reproduced the observed seasonal and spatial variations in AOD over South Asia reasonably well. While over the west Asian regions, where mineral dust is a major contributor to the total aerosol loading, the model simulated the aerosol properties close to the in situ measurements, and it underestimated AODs over the regions where anthropogenic activities dominated. 2. Although RegCM4 simulated the observed seasonal trend in the (near-surface) BC mass concentration over the Indian region, the absolute magnitudes were underestimated by as much as 2 to 5 times, with a larger underestimate in winter. 3. Examination of the diurnal evolution revealed that the simulated concentrations were close to the measurements 14 of 17