Interactive comment on “ Evaluation of black carbon emission inventories using a Lagrangian dispersion model – a case study over Southern India ”

The manuscript by Gadhavi et al. presents a comparison of observed and model simulated equivalent black carbon (BC) concentrations. The observations were obtained with an aethalometer at a rural site in Southern India (Gadanki) during 2008 to 2012. The model simulations are based on a Lagrangian dispersion model (FLEXPART with NCEP Global Forecast Systems Final meteorological analysis data). For each day, a potential emission sensitivity (PES) field is obtained by a 10-day backward model run initialized from the receptor point. Model BC concentrations at the observation site are then calculated based on the PES using three different emission inventories.


Introduction
Black carbon (BC) is a component of soot, which is responsible for the absorption of visible light (Yasa et al., 1979).It is emitted into the atmosphere as a consequence of incomplete combustion processes like biofuel burning, running of inefficient diesel engines, forest fires, etc.Unlike other aerosols, BC aerosols are responsible for positive radiative forcing which is comparable to forcing by major greenhouse gases (Haywood and Ramaswamy, 1998;Jacobson, 2001;Bond et al., 2013).Presence of BC in the atmosphere also affects the hydrological cycle of Earth and regional climate (Ackerman et al., 2000).
Understanding the sources of BC, their geographical distribution and future changes is therefore important to improve climate modelling and would support development of policies exploring climate co-benefits of air pollution regulation controlling sources of BC.However, global BC emissions estimates are highly uncertain.Dickerson et al. (2002) estimated BC emissions of South Asia at between 2 and 3 Tg in year 1999 using the BC / CO ratio -a factor of 2 to 3 higher than bottom-up BC inventories, suggesting significant underestimation of BC sources in South Asia.The range of global BC emissions has been reported as 4 to 13 Tg yr −1 (Bond et al., 2013).Emissions from India contribute 7 to 14 % of global BC emissions (Bond et al., 2004;Schultz and Rast, 2007;Klimont et al., 2009Klimont et al., , 2015a, b) , b) and observed BC concentrations over India are significantly higher than in other regions (Suresh Babu and Moorthy, 2002;Suresh Babu et al., 2002;Ganguly et al., 2005Ganguly et al., , 2006a, b; Jayara- man et al., 2006;Ramachandran and Rajesh, 2007;Beegum et al., 2009;Gadhavi and Jayaraman, 2010;Ramachandran and Kedia, 2010;Vinoj et al., 2010;Raghavendra Kumar et al., 2011).Model-predicted BC concentrations over India are generally found to be factor of 2 to 6 lower than those observed (Ganguly et al., 2009;Nair et al., 2012;Bond et al., 2013;Moorthy et al., 2013).This raises the question whether the observed high BC concentrations over India are the result of transport from other places, relatively inefficient removal of BC compared to elsewhere, or underestimation of emissions from India.

Site description
Observations of BC have been carried out at the climate observatory of the National Atmospheric Research Laboratory in Gadanki.Gadanki (13.48 • N and 79.18 • E, 365 m a.s.l.) is a typical rural site in southern India, with no major industrial activities in the near vicinity.Gadanki has tropical wet climate and experiences a prolonged rainy season from both southwest and northeast monsoons unlike the northern and western parts of India.Monthly rainfall patterns over Gadanki for the years 2009 and 2011 are shown in Fig. 1.February to May is mostly dry.The rainy season starts in June and goes on until December with short lulls in between.The maximum rainfall over Gadanki in the year 2009 was observed during November whereas in the year 2011 it occurred during August with a comparable rain amount in November.The year 2009 was officially declared as a drought year for the state of Andhra Pradesh (in which Gadanki is located), whereas 2011 was a normal year.
Open biomass burning has a well characterized seasonal cycle over India (Joseph et al., 2009).Fire radiative power (FRP) is a measure of radiative energy emitted per unit time in a fire event.Its value is proportional to amount of material being burnt in the fire event.Fires detected using MODIS satellite sensor are characterized by FRP values using an empirical formula based on difference in brightness temperature at 4 µm with respect to non-fire pixels in the vicinity (Giglio et al., 2003;Justice et al., 2006;Davies et al., 2009).In Fig. 2, long-term (2000Fig. 2, long-term ( -2013) ) monthly median FRP values over the southern part of India (south of 18 • N latitude; henceforth referred as Peninsular India) and over the whole of India are shown.FRP is high during February to May and low during June to September.The largest differences in the seasonal variation of FRP between Peninsular India and whole India occur during October to November.As mentioned before, Peninsular India where the observations are carried out experiences two rainy seasons whereas North, West and Central India experiences only one rainy season.The northeast monsoon brings rain over Peninsular India during winter and reduces the number of fire events and hence FRP whereas absence of rain results in high FRP over other parts of India.

Instrumentation and data
Equivalent BC (EBC) concentrations are measured using an aethalometer (Model AE31; Magee-Scientific, USA), which has seven wavelength channels centred at 370, 470, 520, 590, 660, 880 and 950 nm.In this study, we report EBC values based on 880 nm channel data as it has minimum interference from other species and is considered to be the standard channel for BC measurement with this technology (Hansen, 2005).Details of the instrument and the typical set-up used at Gadanki are reported in an earlier study (Gadhavi and Jayaraman, 2010).The ambient air is drawn with a typical flow rate of 2.9 L min −1 for five minutes and passes through a quartz fibre filter fitted in an optical chamber.Changes in transmission of light through filter paper is monitored which is affected by accumulated deposition of light-absorbing particles on the filter paper.The changes in absorption coefficient of filter paper are converted to equivalent BC mass by dividing it with mass absorption cross-section 0.166 cm 2 µg −1 (at 880 nm).Assuming that most of light absorption is due to BC at 880 nm, for the convenience of comparisons with the model simulations, we refer to these measurements as BC.The error in estimating BC concentration is expected to be less than 10 % (Hansen, 2005; Gadhavi and Jayaraman, 2010 and references therein).

Emission inventory data
We have considered three emission inventories namely ECLIPSE, RETRO and SAFAR-India.The ECLIPSE (Evaluating the CLimate and air quality ImPacts of Short-livEd pollutants) global emission inventory has been developed using the GAINS Model (Greenhouse gas -Air pollution Interactions and Synergies Model; Amann et al., 2011).The sources considered range from wick lamps to thermal power stations, including residential combustion, transport, shipping, large combustion installations, industrial processes, waste and open burning of agricultural residues.This inventory does not include emissions from open biomass burning other than agricultural waste burning.Hence forest fire emissions are included from GFEDv3 (Global Fire Emissions Database; van der Werf et al., 2010).The ECLIPSE emission data set has been developed for the period from 1990 to 2050; the inventory extends to 2010 while the baseline projection until 2050 assumes implementation of existing environmental legislation and draws on the energy projection of the IEA -International Energy Agency's Energy Technology Perspective 2012 (ETP2012) (Klimont et al., 2015a, b).In this work, emission values for the year 2010 from version 5 of the inventory are used.Version 5 was recently released and has about 44 % higher emissions than version 4a inventory over India, mainly due to the addition of sources which were not previously considered (e.g.wick lamps).The original data set is available at 0.5 • × 0.5 • resolution including monthly resolution for several key source sectors; however in this study, the grid resolution has been reduced to 1 • × 1 • .Emission fluxes from the ECLIPSE + GFED inventory are shown in Fig. 3a.Hereafter, if not specifically mentioned, reference to ECLIPSE inventory implies ECLIPSE + GFED.The total BC emissions of India in 2010 are estimated at 1233 Gg yr −1 of which 52 Gg yr −1 are from forest fire emissions based on GFED.The major contribution originates from the Indo-Gangetic Basin (IGB) in the north and in a few pockets on the western coast of India.In contrast, in South and Central India BC emissions are relatively low.Within IGB, emissions are higher in Bihar, West Bengal and Haryana states of India and Bangladesh (a map of India with state names is provided in the Supplement).
The RETRO emission inventory is the outcome of the project REanalysis of the TROpospheric (RETRO) chemical composition over the past 40 years.The emission inventory for BC has two parts -one for anthropogenic emissions which includes biofuel burning, industrial combustion and H. S. Gadhavi et al.: Black carbon emissions in South Asia agricultural residue burning.BC emissions from forest fires over India are accounted for separately based on the Reg-FIRM model (Schultz et al., 2008).Schultz et al. (2008) had to reduce the literature values for carbon emissions per unit area over India to achieve consistency with reported emissions from the subcontinent which highlights inherent problems in the bottom-up inventory approach for emissions from biomass burning.The emission fluxes are monthly averages of BC in kg m −2 s −1 for each grid-box.Annual total BC emissions of India based on this inventory for the year 2010 are 697 Gg yr −1 out of which 31 Gg yr −1 are from forest fires.In Fig. 3b, differences between ECLIPSE and RETRO (ECLIPSE -RETRO) are shown, i.e.RETRO emission fluxes are lower than ECLIPSE emissions in all of South Asia.The difference is particularly high over the Bihar, West Bengal states of India, and Bangladesh and Myanmar.
Finally, we have considered the regional emission inventory SAFAR (System of Air quality Forecasting and Research)-India (Sahu et al., 2008).The SAFAR-India includes only anthropogenic emissions from fossil fuel, fuel wood, dung combustion and agricultural waste burning using district-level statistics on activities, population, farming, etc.In preparation of the inventory, Sahu et al. (2008) have used emission factors for biofuel combustion from Venkataraman et al. (2005), and emission factors for fossil fuel combustion are based on Cooke et al. (1999) for their "under-developedcountries" category.The inventory was updated after publication of Sahu et al. (2008).The latest inventory contains annual emissions for the years 1991, 2001 and 2011 at a spatial resolution of 1 • × 1 • .In this work, we have used emission values for 2011.Total BC emissions of India based on this inventory are 1119 Gg yr −1 .Although ECLIPSE and SAFAR inventory have comparable total emissions for India, their spatial and source distribution are significantly different.In Fig. 3c, spatial allocation differences between ECLIPSE and SAFAR (ECLIPSE -SAFAR) are shown.SAFAR inventory has comparable or marginally higher emissions in central, southern and western parts of India.Regions close to big cities like Mumbai, Delhi, Ahmedabad and Kolkata have significantly higher emissions in SAFAR compared to ECLIPSE.The opposite is true over Bihar, West Bengal and northeastern parts of India, where the ECLIPSE inventory is significantly higher than SAFAR.With respect to source distribution, the key difference is between large combustion plants (power plants and industrial boilers) and the residential sector.SAFAR estimates large BC emissions from power plant boilers while this source is very small in ECLIPSE.This is linked to the emission factors used, i.e.SAFAR uses values from Cooke et al. (1999) who suggested high emission factors for large industrial boilers but Bond et al. (2004) concluded that there is no evidence for so high values.ECLIPSE relies on smaller values as discussed in Bond et al. (2004) and Kupiainen and Klimont (2007).For the residential sector, ECLIPSE includes specific calculation of emissions from diesel generators and wick lamps; particularly the inclusion of the latter source resulted in additional BC emissions leading to higher estimates in version 5 of ECLIPSE.

Model description
We have used the Lagrangian particle dispersion model (LPDM) FLEXPART v9.0 (Stohl et al., 1998(Stohl et al., , 2005)).The LPDM computes the trajectories of a large number of particles (infinitesimally small air parcels).Unlike ordinary air back-trajectory models, FLEXPART includes several processes important for aerosol dispersion and removal like diffusion by turbulence in the boundary layer and aloft, deep convective mixing, dry deposition and wet deposition.The representation of narrow plumes is not possible in Eulerian models whereas in LPDM one can track the particles correctly at subgrid scale.Furthermore, FLEXPART can be run in both forward-and backwardin-time modes.The output of the forward modelling from emission sources are simulated concentration fields, whereas a backward run of the model initialized from a receptor point (typically, a measurement location) provides source-receptor (S-R) relationships or potential emission sensitivity (PES) fields.A detailed description of the FLEXPART-based S-R relationship can be found in Seibert and Frank (2004).It is related to the residence time of particles in output grid cells.The S-R relationship describes the sensitivity of receptor y to source x.In the present case, the receptor y is a vector of 24 h average BC concentrations at Gadanki for different days, and source x is vector of area-averaged BC emissions in different grid-boxes at different time intervals.In case of FLEXPART based S-R relationship, the S-R relationship is a matrix M whose elements m il are defined by m il = y l x i (Seibert and Frank, 2004).Once the matrix M is known for a given source vector (emission inventory), receptor values (BC concentrations at measurement site) can be obtained by a simple matrix-vector multiplication.The backward (also known as retroplume) runs are particularly useful to understand the regional distribution of sources contributing to pollution at the observation site and the corresponding transport pathways and for evaluating emission inventories using point observations.
We have used the NCEP Global Forecast Systems Final -GFS-FNL; NCEP (2000); hereinafter referred to as FNL data -meteorological analysis data to drive FLEXPART.GFS-FNL data are available at 1 • × 1 • spatial resolution and at 6hourly temporal resolution.Vertically, the data are available at 26 pressure levels extending from the surface to 10 hPa.
We have used backward runs of FLEXPART to simulate the BC concentrations at Gadanki to understand the relative merit of various inventories for the comparison of modelled values with observations.Various settings for the model runs are summarized in Table 1.In the backward runs, BC particles were traced backward in time from the receptor site (Gadanki) for 10 days.The simulations were carried out for every day of the years 2009 and 2011.Since the  FNL data do not include precipitation values for the year 2009, the model particles were subjected to only dry deposition in the year 2009 whereas the particles were subjected to both dry and wet deposition in the year 2011.To calculate dry deposition, the particle density, aerodynamic diameter and standard deviation of a log-normal distribution were assumed to be 1400 kg m −3 , 0.25 µm and 1.25, respectively following Stohl et al. (2013).Below-cloud scavenging is modelled using a wet scavenging coefficient defined as λ = AI B , where A is the wet scavenging coefficient at precipitation rate (I ) equal to 1 mm h −1 , and B is the factor dependency (McMahon and Denison, 1979).We have set values of A equal to 2 × 10 −7 s −1 and B equal to 0.62 following Stohl et al. (2013).The in-cloud scavenging is simulated using a scavenging coefficient defined as λ = (1.25I 0.64 ) H −1 , where H is cloud thickness in metres (Hertel et al., 1995).The PES values in the bottom-most layer (so-called footprint layer; 0-100 m a.g.l.) were multiplied by the emission fluxes to calculate the BC concentration at the receptor.ber).Although the data period is not sufficient to do a thorough trend analysis, for the available data, no trend is observed.Also, there are no major differences in seasonal peak and low concentrations from 2008 to 2012.Hence, keeping computational time constraints in mind, the numerical simulations were carried out only for the relatively dry year 2009 and the normally wet year 2011.

Potential emission sensitivity
The model output is PES values on a three-dimensional grid.
Since BC is mainly emitted near surface, we focus here only on the PES of the bottom-most layer from 0-100 m a.g.l.(the so-called footprint layer) and refer to this simply as PES.PES maps for 5 different days representing different mete- During the transition period, the air travels over western and central India before reaching Gadanki (e.g.Fig. 5d).It is rare that significant PES values occur over Southeast Asian coun-tries or China, though in a few instances trajectories came from Myanmar, Southeast Asian countries and South China (e.g.Fig. 5b).The advantage of a dispersion model vis-à-vis a simple air trajectory model can be seen in Fig. 5c.The median trajectory shown with grey dots is found to pass over the Arabian Peninsula, though surface PES values are not significant over the Arabian Peninsula, but are substantial over the northern Indian Ocean.In such circumstances, simple air back-trajectory analysis may ascribe observed concentration to emissions over Arabia whereas in reality it is not being influenced by surface emissions over that region.To demonstrate the effect of wet deposition on PES, PES maps for 14 October 2011 are shown with and without wet deposition in Fig. 5e and f respectively.The week preceding 14 October 2011 had large rainfall over southern India and the Bay of Bengal.Precipitation maps for 6 days from TRMM satellite are provided in the Supplement.Wet deposition is generally the most important removal process for aerosol and its effect on PES can be seen by the reduction of the high PES area especially over the ocean.However, the highest PES values over India close to the observation site remain almost unaffected by precipitation.Simulated BC concentrations for this case with and without precipitation are 1.0 and 1.4 µg m −3 , respectively.

Modelled BC concentrations
BC concentrations are determined by multiplying the footprint PES values with emission fluxes from the various inventories for every grid point and then integrating over the whole globe.The method implies that BC emissions are uniformly distributed in a grid cell of height 100 m (height of footprint PES layer).For a surface source, the footprint PES layer should be as small as possible.However, a very shallow footprint layer is not ideal from statistical point of view, as the PES is calculated based on the mass (and, thus, approximately the number) of particles in the footprint layer.With a very shallow height (say, 10 m), one would need to release 10 times more particles (number of trajectories) than with a 100 m height of the footprint layer to arrive at the same statistical error for the footprint PES, whereas increasing the height of the PES layer will not introduce significant error as long as the boundary layer height is higher than the footprint PES layer.BC concentrations are calculated with the three emission inventories ECLIPSE, RETRO and SAFAR-India.
The SAFAR-India emission inventory is available only for the Indian region, hence inventory values outside India are set to zero.In the case of the ECLIPSE inventory, emissions outside India including shipping are found to contribute on average 6 % of the total modelled BC concentrations over Gadanki.There were only 36 days in the year 2009 that had more than 15 % of the BC originating from emissions outside India.Note that for the year 2009, wet deposition was not simulated.In the case of 2011, for which wet deposition was simulated, emissions outside India contributed 5 % on average and there were only 24 days when their contribution was more than 15 %.In Fig. 6, seasonal and annual averages of source contribution maps are shown.During winter the emissions from IGB region (North India) dominate the BC concentrations at Gadanki, whereas during spring, emissions from southern India dominate.During summer, the source region is very small resulting in low concentration of BC as shown later.Autumn is a transition period from southwest monsoon to northeast monsoon and hence BC concentrations at Gadanki are due to both northern and southern India emissions.On average, India north of 18 • N latitude contributes 43 % of simulated BC mass and the part north of 22 • N latitude contributes 33 % at Gadanki.The contribution increases to 67 and 57 % during winter from the two regions, respectively.
A comparison of observed and model-estimated BC concentration for the year 2009 is shown in Fig. 7a.There are no big differences between the three emission inventories.BC estimates based on RETRO are a little lower than for the other inventories, as expected, since total BC emissions of India (697 Gg yr −1 ) in RETRO are significantly lower than in the other two inventories.Overall, the seasonal pattern is well reproduced in the model runs.Several submonthly-scale variations of observed BC concentrations are also well reproduced by the model, confirming its ability to simulate the influence of short-term changes in the meteorological conditions.During autumn and winter, the observed values are reproduced by the model within around 30 % but the model underestimates the observed BC concentrations during spring and summer quite substantially.In Table 2, values of annual and seasonal averages, observation to model ratio, mean biases, root mean square differences (RMSD) and correlation coefficients (R) between observation and model for different inventories are shown.Overall, SAFAR has the smallest bias (0.8 µg m −3 and least RMSD (1.4 µg m −3 in 2009 and 1.1 µg m −3 in 2011) with comparable values for ECLIPSE.The bias is small during autumn in general.In fact, with the SAFAR inventory, the model overestimates the observed concentrations during autumn of 2009 by a small amount (0.104 µg m −3 ).The largest bias and RMSD are found during spring.Note that seasonal variations in model values are purely due to meteorology and transport as emission fluxes are constant within a month in the ECLIPSE and RETRO inventories and throughout the year for SAFAR inventory.Although the SAFAR inventory has seasonally fixed emission fluxes, the model's performance using the SAFAR inventory is not very different compared to using the ECLIPSE inventory.This is because BC emissions in ECLIPSE inventory has very small seasonal variation.Monthly BC emissions of India in ECLIPSE (excluding GFED) inventory vary from 93.7 Gg in September to a maximum 104.2 Gg in July mainly due to seasonal variation of agricultural waste burning, which varies from 1.2 to 10.5 Gg.Together with GFED, there is less than 4.1 % monthly variation of total BC emissions in a year in India.
As mentioned before, simulations for the year 2009 were carried out without including the wet deposition process.When including wet deposition for the year 2009, the underestimation may even be larger than that reported here.However, in Fig. 7b and c, we show a comparison for the year 2011 without and with wet deposition, respectively.It can be seen that the wet deposition has very little effect and hardly produces perceptible differences between Fig. 7b and c.Overall, wet deposition reduces modelled BC values by only 8 % when using the ECLIPSE inventory.Seasonally, the wet deposition is found to be reducing modelled BC values by 5 % in winter, 6 % in spring, 14 % in summer and 15 % in autumn.Such seasonal influence is expected as the maximum rain over Gadanki is received during summer and autumn (see Fig. 1).There were about 76 days in the year 2011 when wet deposition reduced the BC concentrations by more than 15 %.In summary, wet deposition is not a major factor that causes underestimation of model BC values over Gadanki.This is a result of the relatively short rainy season over major parts of India and the short transport times during the rainy season for a major fraction of the BC between its emission and the arrival at Gadanki, rendering precipitation scavenging an ineffective process for this particular site.This result is site specific and does not imply that wet deposition is globally of minor importance.On the contrary, it is the main removal mechanism for BC in the model.
In Fig. 8, the average age spectra (measuring the time between BC emission and BC arrival at Gadanki) of modelled BC values estimated using the ECLIPSE inventory are shown for the full year as well as for the four seasons.One can see that on average about 30 to 40 % of BC mass is of age 4 days or more.During winter this value increases to 65 %.In other words, a large fraction of BC mass during winter is due to long-range transport of BC particles from northern India.If the dry deposition process is the reason for underestimation then one may expect larger model biases in winter.Instead, during winter, the comparison between model and observations is better.Hence, dry deposition may also not be an important factor for causing the underestimation.
The differences between observation and model are in fact correlated to biomass burning activity (see Fig. 2; also see seasonal maps of fire hotspots overlaid on PES in the Supplement).Since, SAFAR-India inventory considers only anthropogenic emissions and do not include forest fire emissions, such underestimation during spring (open biomass burning season) is expected for it.The RETRO inventory has BC emissions from forest fires and in the case of ECLIPSE inventory, forest fire emissions are included from GFEDv3.In spite of this, ECLIPSE and RETRO inventories have similar underestimation like SAFAR-India during spring.The modelled BC values using ECLIPSE and RETRO are about a factor 2.1 and 3.5 lower than the observed BC concentration (2.8 µg m −3 ) respectively.The variation of monthly BC emissions from forest fires in GFED and RETRO are similar to the monthly variations of FRP shown in Fig. 2 with max-imum emissions of 45 Gg in the case of GFED and 24 Gg in the case of RETRO during March.However, BC emissions from forest fires over India in GFED and RETRO are lower than anthropogenic emissions by a factor of 23 annually and by a factor of 2.2 to 2.8 during March (peak biomass burning season).The year 2009 was a drought year and had a higher number of forest fire events.This is reflected in higher observed BC concentrations during spring of 2009 compared to the spring of 2011 (see Figs. 4 and 7).In spite of the low BC values during spring of 2011 (being a normal year from drought or forest fire events perspective), all the three inventories still significantly underestimate the observations (see Table 2).The fraction of BC particles of age less than 4 days is 61 % in the year 2009 and 70 % in the year 2011 (see Fig. 8).In other words, freshly emitted particles over southern India form a major part of the total BC load during summer and spring (see Fig. 6).Hence, our analysis suggests that underestimation is due to underestimation of emissions over southern India; however, it is difficult to pinpoint sectors that are being underestimated for BC emissions.Contextual information such as underestimation being correlated to FRP suggests that BC emissions from open biomass burning may be the main sector responsible for underestimation of BC concentration at Gadanki.Gustafsson et al. (2009) and Sheesley et al. (2012) apportioned carbonaceous aerosols using radiocarbon technique over two locations influenced by air masses from India and found biomass burning contributing to the extent of 70 % of total mass of carbonaceous aerosols.Pavuluri et al. (2011) studied the correlation of BC with levoglucosan and nonsea-salt K + at Chennai (a major city in southern India) and found that biomass burning is the major source of them during winter and summer.Lelieveld et al. (2001) estimated the contribution of biomass burning in CO in the range of 60 to During summer when wind speeds and direction are conducive for dust aerosol, atmosphere may have a high level of dust amount.Dust is a weakly absorbing type of aerosol.The mass absorption cross-section of dust is 9 times lower than the BC mass absorption cross-section (Zhang et al., 2008).However, during summer, when BC concentrations are low,

Case studies
Since emission inventories for the years 2009 and 2011 are kept the same, the difference in model values between these two years is purely due to meteorology.In the rest of the paper, we focus on the year 2009.As noted in the previous section, the model underestimates BC values during spring and summer.The underestimation of BC concentrations may be related to underestimation of biomass burning activity during the dry season and subregionally incorrect anthropogenic emission fluxes.Here we discuss a few cases from the year 2009 which provide insight into these aspects.
In Fig. 9, a comparison of modelled and observed BC concentration over Gadanki for the year 2009 is shown.It is similar to Fig. 7a, but zoomed-in for three different periods.Note the sudden decrease in BC concentration for both observation and model on 8 January in Fig. 9a.From 1 to 6 January, high PES extended along the east coast of India and Indo-Gangetic Basin (IGB) region.However, from 7 January onward the PES region started shifting away from the coastline towards the central Bay of Bengal (BoB).On 8 January, the high PES region was a narrow region stretching eastward up to Andaman Nicobar Island and then turning northward up to Bangladesh.Oceanic regions do not have many BC sources except for exhaust from ships plying in the region.Simultaneous decrease in observed and modelled values in this case is indicative of the fact that a large part of the BC observed over Gadanki is transported rather than of local origin.Notice that in the model, BC variations are mainly due to changes in air mass transport (and precipitation in the case of year 2011) because the emissions are kept constant for at least 1 month.From 9 January onward, the high PES region moves again closer to the east coast of India but does not penetrate deep inland.However, on 13 January, the high PES region is found to be covering the whole of Bangladesh, and Bihar and West Bengal states of India.On 13 January, a BC peak is found in observations and in the model.Again on 16 January, the high PES region moved toward central BoB and away from the east coast of India (see Fig. 5b).However, unlike the 8 January event, in this case the high PES region is a little further north and east in the BoB and touches southern Myanmar.The decrease in the case of models is higher than was found for 8 January but the decrease in observed BC concentration is not as big as for 8 January.This is possible if emission fluxes over the high PES region are underestimated.
From 16 January onward, the PES region moves toward India systematically and deep inland over the IGB region.On 21 January the PES region covers the whole of West Bengal, Bihar and Delhi, and large parts of Orissa, Uttarpradesh and Haryana states of India in and around IGB. On 21 January both model and observation have high concentration close to 4 µg and there are relatively small differences between model and observations.This is indicative that the inventory values are realistic for this region, although perhaps still underestimate true emissions.During February and to the middle of March, observed BC concentrations are increasing whereas model values are systematically decreasing.During this period the high PES region has moved away from India towards the BoB, whereas the PES region in the immediate vicinity of Gadanki has moved southward over Tamil Nadu state of India.The large divergence between observation and model is an indication that the inventoried emission fluxes are significantly underestimated over southern India.This is also a period of high biomass burning activity in the region of the high PES (see Fig. 2).Hence, underestimation may be related specifically to underestimation of open biomass burning in southern India.
From 22 April, the high PES region moves to the west of the observation site, over Karnataka state of India and the Arabian Sea (see Fig. 5d), occasionally moving south of the observation site over entire Tamil Nadu.During later part of May, and June and July months, the high PES region is mostly over the Arabian Sea with a very small region over land due to strong winds (see Fig. 5c).Both model and observations have low values during this period.However, the model systematically underestimates the observations by a factor of 2 to 3 (see Fig. 9b).
From 5 September onward the PES region covers Andhra Pradesh, Madhya Pradesh and Gujarat states of India.On 22 September, the model significantly overestimates the observation and for some days it remains higher than observations (see Fig. 9c).The high PES region during this period lay over Andhra Pradesh and Tamil Nadu border and over southern Karnataka.From 16 October onwards the PES pattern moves entirely north of the observation site.At the beginning the pattern covers central and western India, but on 1 November the PES pattern is similar to that found during January-February and extends all the way up to the northwest border of India covering the entire Indo-Gangetic region (see Fig. 5a).On this day, the concentration is the highest in model with similar values in observations.During September-December, differences between observation and model estimates are small.Summarizing the above description, observed and modelled BC values are high when winds are from northern and western India, with relatively small differences between model and observations, indicative of relatively small errors in the emission inventories over this region.When winds are from South or South Central India, the observed values are high but the model values are substantially lower.Coincidentally, this is also the period of high biomass burning activity over southern India.The differences between the model and observation thus suggest that open biomass burning emissions over southern India have been underestimated in all the three inventories.

Conclusions
Several field studies over India and in adjoining oceans have found high amount of absorbing aerosols.However, models are found to underpredict the observed high concentrations of absorbing aerosols.Using the Lagrangian particle dispersion model FLEXPART and three emission inventories, we compared the simulated BC concentrations with BC measurements at a rural site in southern India.As for the other models, FLEXPART underestimates the observed BC concentrations.We found that 93 to 95 % of the model BC concentration is the result of emissions from India.Northern India is a major source of anthropogenic BC particles, but southern India also has significant BC emissions.This study identifies a potentially significant underestimate of emissions in southern India, which is reflected in a large difference in the observed and modelled BC values in Gadanki during spring when the winds are predominantly from the south.We suggest that the key source for which the emission fluxes may be underestimated is open biomass burning.This is not to rule out the possibility that anthropogenic emissions may also be underestimated.
In the three emission inventories that we evaluated, the ECLIPSE inventory has the highest emissions (1.2 Tg yr −1 ), with similar emissions in the SAFAR-India inventory (1.1 Tg yr −1 ).This is also reflected in the comparison between the modelled and observed BC concentration over Gadanki.Modelled BC values based on ECLIPSE and SAFAR-India are higher than the values based on the RETRO inventory.However, they are not high enough to resolve underestimation in most of the seasons.The overall ratio of observation to model is found to be 1.5 for the SAFAR inventory, 1.7 for ECLIPSE inventory and 2.4 for RETRO inventory.Although the ECLIPSE inventory has the highest emissions over India, it is the SAFAR-India inventory that has the lowest ratio because of differences in spatial distribution in emission fluxes.The SAFAR-India inventory has higher emission fluxes over southern India compared to ECLIPSE.
The Supplement related to this article is available online at doi:10.5194/acp-15-1447-2015-supplement.

Figure 2 .
Figure 2. Monthly median fire radiative power values obtained from MODIS satellite for whole India and peninsular India (south of 18 • N latitude).

Figure 3 .
Figure 3. (a) ECLIPSE (ECLIPSE v5 + GFED v3) BC emission inventory over South Asia.(b) Difference between the ECLIPSE and RETRO emissions (ECLIPSE -RETRO).(c) Difference between ECLIPSE and SAFAR India emissions (ECLIPSE -SAFAR).Here and in rest of the paper political borders are shown for cursory region identification and may not be accurate.

Figure 4 .
Figure 4. Daily mean BC concentration observed at Gadanki (black dots) and their ±1σ standard deviation (orange vertical bars).

Figure 5 .
Figure 5. Selected examples of footprint potential emission sensitivity (PES) maps (also known as source-receptor relationships) using 10 days of backward runs (retroplumes) of FLEXPART from Gadanki.Panels (e) and (f) are PES maps of 14 October 2011 local time with and without wet deposition respectively.See Supplement for the PES maps for other days.

Figure 8 .
Figure 8. Fraction of simulated BC mass at Gadanki with particles of different age for year (a) 2009 and (b) 2011.Age 0-1 days represents contribution from day 1 for the backward simulation.Age 2-3 days represents day 2 and day 3 contribution, Age 4-6 days represents day 4 to day 6 and Age 7-10 represents day 7 to day 10.Note that 2009 simulations are without wet deposition whereas 2011 simulations are with wet deposition.

Figure 9 .
Figure 9.As Fig. 7a, but zoomed in for period (a) January to April, (b) May to August and (c) September to December.

Table 2 .
Average, ratio, bias, RMSD and correlation coefficient between modelled and observed BC concentrations when using different inventories for the years 2009 and 2011.
Winter: December to February; spring: March to May; summer: June to August; autumn: September to November.RMSD: root mean square deviation * Note: calculations for 2011 are with wet deposition; calculations for 2009 are without wet deposition.ECLIPSE inventory includes ECLIPSE v5, GFED v3 and shipping emissions.