• Open Access

Inter-annual variations in natural and anthropogenic aerosol loadings over the seas adjoining India using a hybrid approach

Authors


Abstract

We report the inter-annual variations of natural and anthropogenic aerosol loading over the Arabian Sea (AS) and Bay of Bengal (BoB) for the period 2001–2010 using a hybrid approach. Mean (±1 standard deviation) decadal contributions of maritime, dust and anthropogenic aerosols to aerosol optical depth, τ are 16.6 ± 4.2 and 17.5 ± 4.6%, 44.3 ± 22.6 and 37.4 ± 20% and 39 ± 23.4 and 45 ± 20.5%, respectively, over the AS and BoB. Significant increase in τ (>0.007 or 2.3% per year) is mainly driven by anthropogenic aerosols. The results can be used for improved estimates of aerosol radiative forcing in these regions.

1. Introduction

Aerosols perturb the Earth's radiation budget directly by scattering and absorbing solar radiation and indirectly by modifying cloud properties. Numerous studies have been carried out to quantify these effects at various spatial and temporal scales (e.g. Yu et al. 2006 have summarized some of these studies), but considerable uncertainty exists at regional scale. Large uncertainties in emission inventories of anthropogenic and natural aerosols, their mixing and redistribution by wind make it difficult for the climate models to reproduce the observed space-time variability of aerosols in the Indian subcontinent. Limited ground-based measurements of aerosol chemical properties are inadequate to address this problem. Measurement of aerosol optical depth (τ) cannot be directly used to separate out the anthropogenic and natural components. To tackle this issue, two major approaches were adopted. Combinations of aerosol properties were used to qualitatively estimate the anthropogenic and natural fraction. For example, Bellouin et al. (2005) have separated the anthropogenic and natural aerosols based on the accumulation mode fraction threshold values. A combination of accumulation mode fraction and single scattering albedo (SSA) was utilized by Srivastava et al. (2012) to infer the dominant aerosol types in the Indo-Gangetic Basin. Another popular way to infer about dominant aerosol types includes analysis of τ and Angstrom Exponent (e.g. Kaskaoutis et al., 2010, 2011a). In another approach, representative aerosol composition has been derived by constraining the model-simulated spectral optical properties with in-situ observations (e.g. Dey and Tripathi, 2008). Both these approaches rely on aerosol properties, primarily spectral τ, and hence have limitations for applicability at regional scale because of unavailability of a dense network of long-term ground-based measurements. Moreover, such observations are mostly confined to the Indian landmass, while data over the oceans are available only for short periods typical of campaign mode mostly during the winter (e.g. Moorthy et al., 2010) and pre-monsoon (e.g. Kaskaoutis et al., 2010) seasons. In-situ measurements over the oceans are difficult to carry out during the monsoon season due to weather condition, thus limiting the applicability of the above two approaches.

τ shows strong seasonal cycle over the Arabian Sea (AS) and Bay of Bengal (BoB), adjacent to the Indian subcontinent (Dey and Di Girolamo, 2010; Kaskaoutis et al., 2010; Satheesh et al. (2006a, 2006b)). Every year, during the pre-monsoon (March–May) and monsoon (June–September) seasons, mineral dust is transported to the oceans from the Great Indian Desert, West Asia and Northeast African dust sources (Kaskaoutis et al., 2011b). Anthropogenic particles are transported to these regions from the post-monsoon to winter seasons (Moorthy et al., 2010). Thus, it is important to quantify the inter-annual variations of relative contributions of natural (dust and maritime) and anthropogenic aerosols to τ for further improving the estimates of aerosol radiative forcing. Here, we report a hybrid approach (by combining satellite aerosol products and regional climate model simulations) to differentiate optical depths of dust (τd) and maritime (τma) particles from anthropogenic (τan) component over the seas adjoining India. The variability of τma, τd and τan for the period 2001–2010 is examined and the climatic implications are discussed.

2. Hybrid approach

We used Terra-Moderate Resolution Imaging Spectroradiometer (MODIS) retrieved τ (C005, Level 3) at 550 nm wavelength. MODIS is a large swath (∼2330 km) multi-spectral sensor, routinely monitoring τ at 10 × 10 km resolution across the globe. Details of MODIS aerosol retrieval are well documented in the literature (Remer et al., 2008) and hence are not repeated here. The quality of MODIS-τ was globally validated and utilized by numerous researchers to study the variability of aerosol characteristics at global as well as regional scales including India (e.g. Ramachandran and Cherian, 2008). Here, analysis was carried out over the AS (6°–19°N and 65°–73°E) and BoB (12°–21°N and 87°–92°E). MODIS-τ can be represented as sum of three components:

display math(1)

We chose southern Indian Ocean (bounded within 62°–100°E and 55°–10°S) to derive an empirical relation for wind-dependent production of τma following Satheesh et al. (2006a) and further partitioning into fine and coarse mode. This region is far away from the landmass and is assumed to contain negligible dust and anthropogenic aerosols (Babu et al., 2010). MODIS-τ showed exponential relationship with mean wind speed (U) at 1000 hPa from NCEP reanalysis data set (Figure 1) with a high degree of correlation (R = 0.74, statistically significant at 95% confidence level following t-test) and a root mean square error of 0.04:

display math(2)
Figure 1.

Wind-dependence of MODIS-τ with wind speed U at 1000 hPa.

The empirical constant ‘0.062’ represents the non-maritime background τ in absence of wind. Natural sulfate particles from marine planktons may contribute to this background τ and the relative contribution of sulfate to τ decreases relative to that of maritime particles with an increase in wind speed (Satheesh et al., 2006a, 2006b). Then τma can be estimated as function of U by:

display math(3)

As, no other source of maritime particle production exists, this empirical relation is assumed to be valid for the AS and BoB as well and utilized to estimate τma during the entire study period from corresponding NCEP-derived U at 1000 hPa. The empirical regression constant ‘0.10’ in Equation (1) is very close to the value (0.09) reported in Satheesh et al. (2006a), whereas the regression constant ‘0.062’ is slightly smaller and may be attributed to a longer period of data compared to limited period and smaller area considered in the previous study.

Now, fine mode optical depth can be expressed as:

display math(4)

where ff is the total fine mode fraction and fma,f, fd,f and fan,f are the fractions of τma, τd and τan in the fine mode, respectively. Substituting for τan from Equation (1) to Equation (4), τd can be estimated as:

display math(5)

Assuming entire anthropogenic contribution in the fine mode, Equation (5) can be rewritten as:

display math(6)

provided fma,f, ff and fd,f are known for our study regions. fma,f and ff are derived from MODIS aerosol product. Mean (±1σ) fma,f calculated over the southern Indian ocean (where fma,f = ff) is found to be 0.38 ± 0.08 and close to the value 0.31 ± 0.1 derived from AERONET data in Amsterdam Island (77.57°E, 37.81°S).

We used ICTP-Regional Climate Model, RegCM4.1 to calculate fd,f in absence of any in-situ data from these regions. RegCM4.1 is a hydrostatic model with sigma-p vertical coordinates that includes an inbuilt aerosol module for mineral dust (Zakey et al., 2006). Dust emission results from saltation causing horizontal flux and sand blasting, which creates a vertical flux of dust into the atmosphere depending on the wind strength. Dust loading is calculated at four size bins ranging from 0.1 to 1.0 µm, 1.0 to 2.5 µm, 2.5 to 5.0 µm and 5.0 to 20 µm (Das et al., 2013). The soil textures are taken from United States Department of Agriculture data set. RegCM4.1 is driven by initial and 6-hourly updating lateral boundary conditions from National Centers for Environmental Prediction (NCEP)-National Center for Atmospheric Research (NCAR) reanalysis data. Simulations are carried out at 30-km horizontal resolution and 18 vertical layers with the model top set at 50 hPa considering Holtslag planetary boundary layer scheme and Grell cumulus parameterization scheme (Giorgi et al., 2012). The model captures the spatial variability of wind field and precipitation reasonably well, but with a high bias in 850 hPa wind over the AS and high bias in monsoon precipitation over the BoB. This may lead to a high bias in fd,f over the AS and BoB during the pre-monsoon and monsoon seasons, respectively, that is estimated by the ratio of τd simulated using the first two size bins and all four bins (Das et al., 2013).

Discussion is warranted about the magnitude of uncertainty in the hybrid approach. Uncertainty may arise due to errors in the regression constants of Equation (3) and absolute value of U. τma is less sensitive to an increase in U until U exceeds 6 m s−1 (Figure 1). Even beyond 6 m s−1, τma varies by ˜0.005 for a change of U by ˜6 m s−1. Hence a large error in U is required to induce significant uncertainty in estimated τma and τd. An error of ±10% in the empirical constant ‘0.062’ in the Equation (3) translates to an error of ±7.6% (±11%) in τd for U lower (greater) than 6 m s−1. The uncertainty in estimated τd is <2% for an error as large as 50% in the constant ‘0.10’ of the Equation (3). These factors may be important only in the monsoon season (when U is large), but given the large mean seasonal fd values (Table 1), the error will not invalidate the overall conclusions. Error due to absolute errors in MODIS-τ and ff retrievals over oceans due to cloud contamination and uncertainties regarding aerosol microphysics is another factor to consider. As τma is estimated using Equation (3), any error in MODIS-τ and ff will lead to error in estimated τd only. Overestimations of MODIS-τ and ff tend to overestimate and underestimate τd, respectively. Global validation of MODIS C005 aerosol product suggests better retrieval quality over ocean than over land. For example, MODIS-τ at 550 nm shows strong correlation (R = 0.91) with AERONET-τ with a slope of 0.94 and very low (0.005) intercept for the best-fit regression line (Remer et al., 2008). Standard error in MODIS-τ is <0.01 for τ < 0.5 (Shi et al., 2011), a condition observed in our study region. This translates into a maximum possible underestimation of 0.01 in estimated τd. ff agrees to within 25% of the AERONET-retrieved values over ocean for τ > 0.15 (Remer et al., 2002). For a ±25% error in MODIS-ff (since τ in our study area always exceeds 0.2, Table 1), estimated τd has an uncertainty envelope of ˜30%. It is more likely that MODIS-ff is underestimated as the dust particles are mostly considered in the coarse mode in the retrieval algorithm. We note that overestimation of τd due to the underestimation of ff may partially be compensated by the overestimation due to an error in MODIS-τ. This may lead to an underestimation of τan values shown in Figure 1. The third factor is the error in simulated fd,f. Sensitivity study reveals that 10% error in fd,f translates into an error of ˜0.01 in estimated τd. In general, τd will be overestimated for an overestimation in fd,f, which is the case since the model does not consider any local sources (Das et al., 2013). Finally, error may creep in due to the assumption of entire anthropogenic fraction in fine mode. The underestimation in estimated mean seasonal τd over both the regions increases (Figure 2) with an increase in fraction of τan in coarse mode with larger sensitivity observed during the seasons dominated by anthropogenic particles. Size-segregated data of anthropogenic aerosols are unavailable from these regions. Considering fan,f = 0.83 ± 0.05 (following Bellouin et al., 2005, where ff > 0.83 was considered to be of mainly anthropogenic sources and hence ff may be considered equal to fan,f), the error in mean seasonal τd is maximum (>0.04, which translates into >50%) during these seasons; whereas it is minimum (˜0.03 translating to ˜15%) during the monsoon season. This also leads to an overestimation of τd. To summarize, all these factors lead to an underestimation of τan and hence the mean seasonal estimates of fan (summarized in Table 1) are on the lower side. Size-segregated chemical measurements are required to estimate regional mean seasonal values of fd,f and fan,f. Until then, we proceed to examine the seasonal and inter-annual variations of τma, τd and τan in view of these uncertainties and conservative estimates of τan (Figure 3).

Table 1. Mean seasonal τ as retrieved by MODIS and fma, fd and fan as derived by the hybrid approach over the Arabian Sea (1st value) and Bay of Bengal (2nd value) during 2001–2010
YearSeasonτfmafdfan
2001Winter0.212, 0.2550.21, 0.180.16, 0.270.62, 0.55
Pre-monsoon0.269, 0.3210.15, 0.140.55, 0.500.30, 0.36
Monsoon0.414, 0.3410.23, 0.260.69, 0.550.08, 0.19
Post-monsoon0.255, 0.1640.16, 0.320.36, 0.130.47, 0.55
2002Winter0.224, 0.2650.23, 0.180.22, 0.250.56, 0.57
Pre-monsoon0.279, 0.3300.14, 0.140.62, 0.550.24, 0.30
Monsoon0.386, 0.3910.22, 0.230.67, 0.620.11, 0.15
Post-monsoon0.259, 0.1660.12, 0.300.34, 0.180.54, 0.52
2003Winter0.221, 0.2600.21, 0.170.26, 0.230.53, 0.60
Pre-monsoon0.291, 0.3670.13, 0.120.64, 0.630.23, 0.25
Monsoon0.408, 0.3210.23, 0.260.68, 0.570.09, 0.18
Post-monsoon0.253, 0.2100.15, 0.170.26, 0.210.59, 0.62
2004Winter0.226, 0.2700.20, 0.160.27, 0.230.53, 0.61
Pre-monsoon0.313, 0.3090.12, 0.130.68, 0.580.20, 0.28
Monsoon0.411, 0.3680.22, 0.220.72, 0.590.06, 0.19
Post-monsoon0.221, 0.2830.17, 0.140.29, 0.220.54, 0.64
2005Winter0.206, 0.2620.23, 0.180.25, 0.220.52, 0.60
Pre-monsoon0.251, 0.3340.13, 0.110.51, 0.500.36, 0.39
Monsoon0.421, 0.3450.23, 0.240.66, 0.540.11, 0.22
Post-monsoon0.236, 0.2150.18, 0.210.26, 0.190.56, 0.60
2006Winter0.243, 0.2480.20, 0.190.22, 0.250.58, 0.56
Pre-monsoon0.267, 0.3000.15, 0.120.54, 0.470.31, 0.41
Monsoon0.394, 0.4460.23, 0.210.69, 0.640.08, 0.16
Post-monsoon0.220, 0.2970.16, 0.150.18, 0.140.66, 0.71
2007Winter0.265, 0.3600.17, 0.160.26, 0.220.57, 0.62
Pre-monsoon0.326, 0.3200.09, 0.120.58, 0.450.33, 0.43
Monsoon0.423, 0.2670.16, 0.300.64, 0.390.20, 0.31
Post-monsoon0.286, 0.2090.13, 0.250.28, 0.100.59, 0.64
2008Winter0.263, 0.3000.19, 0.160.29, 0.210.52, 0.63
Pre-monsoon0.304, 0.4340.10, 0.090.54, 0.540.36, 0.37
Monsoon0.553, 0.4110.15, 0.210.73, 0.550.12, 0.24
Post-monsoon0.315, 0.2270.11, 0.170.23, 0.170.66, 0.67
2009Winter0.282, 0.3360.14, 0.130.20, 0.210.66, 0.66
Pre-monsoon0.375, 0.4270.09, 0.090.59, 0.540.31, 0.37
Monsoon0.417, 0.4070.21, 0.230.70, 0.560.09, 0.21
Post-monsoon0.283, 0.2760.13, 0.120.25, 0.220.62, 0.66
2010Winter0.264, 0.3270.15, 0.140.19, 0.220.65, 0.64
Pre-monsoon0.356, 0.3670.11, 0.120.53, 0.490.37, 0.38
Monsoon0.386, 0.3970.26, 0.170.70, 0.600.04, 0.23
Post-monsoon0.261, 0.3060.15, 0.200.27, 0.170.58, 0.62
Figure 2.

Error in mean seasonal dust optical depth (Δτd) due to deviation from the assumption of fan,f = 1 (solid and dotted lines are for the AS and BoB, respectively).

Figure 3.

Inter-annual variations of mean seasonal τ (top panel) and relative contributions of τma, τd and τan to total τ  over the Arabian Sea (middle panel) and the Bay of Bengal (bottom panel). ‘W’, ‘Pr’, ‘M’ and ‘Po’ in the x-axis represent ‘winter’, ‘pre-monsoon’, ‘monsoon’ and ‘post-monsoon’ seasons, respectively. The corresponding years are also mentioned in the x-axis. The ‘deficit’ monsoon seasons in the years 2002, 2004 and 2009 are indicated by arrow.

3. Results

The temporal variability of mean (±1 standard deviation, σ) seasonal relative fractions of dust (fd), maritime (fma) and anthropogenic (fan) aerosols over the AS and BoB during the 10-year period as derived by the hybrid approach is shown in Figure 1 along with total τ. τ shows similar seasonal variation over the two ocean basins with maxima in the monsoon season and minima in the post-monsoon season. Mean (±1σ) wintertime fma, fd and fan over the AS (0.192 ± 0.03, 0.233 ± 0.03 and 0.574 ± 0.05, respectively) are similar to the corresponding values (0.146 ± 0.02, 0.263 ± 0.05 and 0.589 ± 0.06, respectively) during the post-monsoon season. Over the BoB, mean (±1σ) fma, fd and fan for the winter season are 0.166 ± 0.02, 0.231 ± 0.02 and 0.602 ± 0.03, respectively, whereas the corresponding values of the post-monsoon season are 0.202 ± 0.06, 0.163 ± 0.05 and 0.633 ± 0.06 respectively. In the pre-monsoon season, mean (±1σ) fd increases to 0.577 ± 0.05 and 0.520 ± 0.05 over the AS and BoB, at the expense of anthropogenic particles for which fan reduces to 0.301 ± 0.06 and 0.355 ± 0.05, respectively. Detection of mixed type aerosols as the most frequent type in the analysis of ship-borne measurements of spectral τ and Angstrom Exponent in 340–1020 nm range (Kalapureddy et al., 2009) further attests the necessity of the estimates of the relative proportions of natural and anthropogenic components to quantify aerosol forcing. The AS is closer to the dust sources (i.e. West Asia, Africa and Great Indian Desert) than the BoB. However, dust from the Great Indian Desert is transported across the Indo-Gangetic Basin by predominantly north-westerly winds in this season to the northern BoB. On the other hand, the Western Ghats along the west coast of Indian peninsula channels the dust-laden air mass to the southern BoB (Dey and Di Girolamo, 2010). This combined influence of topography and meteorology leads to almost similar fd over the two ocean basins. Dust loading continues to be high in the monsoon season, but fd is larger in the AS (0.688 ± 0.03) than the BoB (0.560 ± 0.06) due to its proximity to the dust sources. Overall reduction in dust loading over BoB in the monsoon season is attributed to change in wind pattern that somewhat restricts the transport along the oceanic pathway and washout by precipitation. Previous estimates of fma= 0.2 over the northern AS (Satheesh et al., 2006a) during the monsoon months of the year 2001–2003 are very close to our estimates of fma = 0.21. However, fma in our estimates during the post-monsoon to winter season is lower by 10% than the previous estimates.

Next, we focus on the inter-annual variations of fma, fd and fan over these two regions (Figure 1 and Table 1). τan has increased at the rate of 0.0062 and 0.0087 (significant at 95% confidence level following t-test) per year in the winter and 0.007 and 0.006 per year in the pre-monsoon season over the AS and BoB, respectively. Larger fan in the last half of the last decade (Figure 1) suggests an increase in anthropogenic pollution load over the ocean. However, no statistically significant change in synoptic wind (Dey and Di Girolamo, 2011) means that transport strength remains more or less same. Hence, emission from anthropogenic sources over the landmass may have increased over the years. In the pre-monsoon season, the significant increase in τan is not reflected in fan due to larger variability of natural aerosols (Figure 1). τan increases by 0.004 per year over the BoB in the monsoon season and by 0.008 per year over the AS in the post-monsoon season. τd and τma do not show any trend and hence the significant increase in τan drives the increasing trend of τ over the last decade. Note that the signal of increasing τan is strong enough to be detected despite of the conservative estimates (as discussed in previous section) in our approach and is also supported by satellite-based estimates (Dey and Di Girolamo, 2011; Kaskaoutis et al., 2011b).

fd does not show any significant difference over the AS during the El Niño (2002–2003 and 2009–2010) and La Niña years (2007–2008). However, fd is much larger in the El Niño years (0.587) relative to the La Niña years (0.470) over the BoB. This implies an enhanced dust transport to the BoB during the El Niño years mostly through the Indo-Gangetic Basin route and not over the oceans along the Indian coastline, which is also observed in the satellite data of absorbing index (Abish and Mohanakumar, 2013). Larger fma over the AS during the monsoon season of El Niño years than in the La Niña years is attributed to a stronger zonal wind. However, over the BoB, a larger increase in τd outweighs the increase in τma during the El Niño years. Also note that fan is higher during the La Niña years relative to El Niño years. Thus the inter-annual variations of natural and anthropogenic aerosols over the oceans are highly influenced regional scale circulation. However, how much the zonal and meridional circulations are influenced by aerosols need further investigation. Also, fd is lower in the deficit years than in the normal monsoon years (all other years are normal years in the last decade except the three deficit years marked in Figure 1), probably due to larger transport of dust from East Africa by stronger Somali jet. Larger aerosol loading over the oceans during the deficit years was also observed by Rahul et al. (2009). This is in contrast to the observations over land (e.g. Kaskaoutis et al., 2012), where higher dust loading was observed during the deficit years (with respect to normal years) because of larger emission of local soil dust and dust transport from the Great Indian Desert. This further indicates the different dynamics of aerosol transport over the land and the oceans that must be understood in order to explain the observed inter-annual variations of natural and anthropogenic aerosols in the subcontinent.

4. Discussion and conclusions

In this study, a hybrid approach (by combining satellite data and climate model simulations) is adopted to examine inter-annual variations of natural and anthropogenic aerosols over the seas surrounding the Indian landmass. Our approach complements the lack of continuous observation of aerosol composition in these regions. The seasonal and inter-annual variability of the three components over the two ocean basins as discussed here are qualitatively supported by the satellite-based and ship-borne studies in the literature. The seasonal statistics (keeping in mind the uncertainties and the assumption of entire anthropogenic load in fine mode) may be used to evaluate simulations of aerosol distribution by chemical transport models. The estimates of radiative forcing may further be improved by considering the representative values of non-spherical and spherical dust. For example, optical properties of non-spherical dust differ from spherical dust of similar composition (Mishra et al., 2008). Recently, Dey and Di Girolamo (2010) have reported mean seasonal non-spherical fraction to τ (i.e. fnsp) to be 0.39 (0.27) and 0.51 (0.37) for the AS (BoB) in the pre-monsoon and monsoon seasons, respectively, based on Multiangle Imaging SpectroRadiometer (MISR) data. Both MISR climatology and this approach show an increase in relative abundance of dust in the monsoon season from the pre-monsoon season and larger values over the AS than BoB. In the MISR algorithm, fnsp is entirely contributed only by dust, whereas the spherical fraction is contributed by all types including dust. Considering mean fnsp from Dey and Di Girolamo (2010) and fd from this study, 67.6 and 74.1% (i.e. fnsp/fd) of the dust particles are estimated to be non-spherical in the pre-monsoon and monsoon seasons over the AS. The corresponding values over the BoB are 51.9 and 66.1%, respectively. These values may help in estimating the aerosol radiative forcing more accurately in view of the sensitivity study by Mishra et al. (2008). In the other two seasons, fd is low and hence perhaps the assumption of dusts as spherical particles may not lead to any significant error in the estimated radiative forcing.

The major conclusions are as follows:

  1. Annually, maritime, dust and anthropogenic particles contribute 16.6% (17.5%), 44.3% (37.4%) and 39% (44.9%) to τ over the AS (BoB) with a strong seasonal cycle.
  2. In the last decade, a significant rise (2.3% per year) in τ in these regions is driven by anthropogenic particles (the trend may be even higher given the conservative estimates in this hybrid approach).
  3. Combining previous estimates of non-spherical fraction from Dey and Di Girolamo (2010) and the present approach, non-spherical particles are estimated to contribute 67.6% (51.9%) to dust optical depth over the AS (BoB) in the pre-monsoon season, which further increases to 74.1% (66.1%) in the monsoon season.

Acknowledgements

The work is supported by research grant from Department of Science and Technology, Govt. of India under contract SR/FTP/ES-191/2010 (Fast Track Scheme) through a research project operational at IIT Delhi (IITD/IRD/RP2509). The authors acknowledge ICTP for providing the RegCM4.1 Model. The first author is thankful to CSIR for providing scholarship to carry out research work in IIT Delhi. NCEP reanalysis data are obtained from NOAA CIRES Climate Diagnostics Centre. MODIS data are downloaded from Atmospheric Science Data Centre. PI and staff of the Amsterdam Island AERONET site are acknowledged for establishing and maintenance of the site. We acknowledge the anonymous reviewers for their valuable comments that helped improving the manuscript.

Ancillary