Climatology of columnar aerosol properties and the influence of synoptic conditions: First-time results from the northeastern region of India



[1] Six years of spectral aerosol optical depths (AODs), from the northeastern part of India (Dibrugarh), are used to evolve a climatology for this region. The results indicate that the seasonal mean AODs at 500 nm go as high as 0.45 ± 0.05 during premonsoon season (March to May), decrease gradually through the monsoon (June to September) to reach the lowest value of 0.19 ± 0.06 during the retreating-monsoon season (October and November), and increase to 0.31 ± 0.04 in winter (December to February). The AOD spectra are generally flatter than those seen typically over continental sites of India (and elsewhere in the neighboring regions) with Ångström exponent α remaining below 1.0 during February through August, indicating a relatively low abundance of fine and accumulation mode aerosols. The columnar size distributions (CSD) retrieved from spectral AODs are, in general, bimodal with primary mode at ∼ 0.1 μm and secondary mode at ∼ 1.0 μm. High mass loading (∼309.5 ± 65.9 mg m−2) and effective radius (∼0.40 ± 0.09 μm) occur during premonsoon and are attributed to significant abundance of coarse (natural) aerosols. Cluster analysis of air mass back trajectories indicate significant transport of mineral dust from the arid regions of west Asia and northwest India across the Indo-Gangetic plains and marine aerosols advected from the Bay of Bengal contributing largely to the coarse mode aerosols during this season. On the other hand, the peculiar topography combined with the local conditions and the widespread rainfall lead to a more pristine environment during retreating-monsoon season with quite low AODs and columnar loading.

1. Introduction

[2] The Earth-atmosphere system is in a balance between the incoming and the outgoing energy; a perturbation to which is likely to have significant implications to global climate. Of the various factors that are responsible for climate changes, the effect of atmospheric aerosols, both natural and anthropogenic, remains most uncertain [Intergovernmental Panel on Climate Change, 2007]. Well-focused and integrated field campaigns have also been formulated, such as ACE-1 [Bates et al., 1998], ACE-2 [Raes et al., 2000], SCAR-B [Kaufman et al., 1998], TARFOX [Russell et al., 1999], INDOEX [Ramanathan et al., 2001], and ICARB [Moorthy et al., 2008]. Nevertheless, significant data gaps exist. Viewed in the backdrop of the above, the regional characterization of aerosol properties leading to a climatological feature is of great importance and relevance.

[3] In the regional characterizations of atmospheric aerosols over Southeastern Asian region, the northeastern parts of India is unique owing to its characteristic dense vegetation, vast water bodies and the unique topography, with mountains in the north, east and south and the highly developed Indo-Gangetic Plains toward the west. Recent studies have shown that presence of substantial amounts of absorbing aerosols in the Himalayan region can significantly influence the Asian summer monsoon [Lau et al., 2008]. Several regional experiments are being planned to evolve regional aerosol characteristics pertinent to this region such as JAMEX [Lau et al., 2008], EAST-AIRE [Li et al., 2007], SHARE-Asia [Lau et al., 2008]. Despite these, extensive observations on aerosol characteristics leading to a climatological picture of this region of India have not yet been carried out. With a view to bridge this gap, spectral AOD measurements were initiated at Dibrugarh (27.3°N, 94.6°E, 111 m a.s.l, in extreme eastern end of India) and regular measurements have been carried out since October 2001 as a part of the aerosol radiative forcing over India (ARFI) project of ISRO-Geosphere Biosphere Program (I-GBP). The data collected for 6 years since then have been used in the present analysis to infer climatological features of aerosol over this region, where there exists no other measurement of aerosols. The aerosol microphysical and optical properties thus obtained are examined for region specific processes and also compared with those reported at other parts of India and south and east Asia.

2. Northeast India: Topography and Climate

[4] Northeast India is the region bound between 22°N to 29.5°N and 88°E to 97.5°E, internationally bordered by China to its north and east, Myanmar to the southeast and south and Bangladesh to the southwest. Topographically, the great Himalayan ranges and Tibetan plateau to the north, the hills and mountain of Yanan to the east and the Meghalaya Plateau in the middle strongly influence the general tropical warm climate of the region. Many of these hills and mountains are quite high (between 1 km to 5 km a.s.l) and prevent the rain bearing monsoon winds from escaping this region on the one hand, while they do not allow the dry and cold winds of central Asia to enter the northeast region. The Meghalaya plateau orographically lifts the southwest monsoon winds, causing the world's heaviest rainfall in its southern margin. The experimental site, Dibrugarh (Figure 1), is located in the northeastern corner of upper Brahmaputra valley. It is enriched with forests scattered all over and large number of rivers and their tributaries and streams flow through the area, ultimately draining into the enormous river Brahmaputra, and eventually to the Bay of Bengal.

Figure 1.

Geographical location of the measurement site Dibrugarh (DBR) over the Indian subcontinent and adjoining south Asian regions.

[5] Dibrugarh experiences subtropical monsoon climate with mild winter, warm, and humid summer. The average annual temperature is 23.9°C and the average annual rainfall is 2760 mm with a total number of 193 rainy days. The climatological mean rainfall at Dibrugarh peaks in July in the annuals (∼520 mm), and the number of rainy days (22 days). The surface winds are generally low (<4 m s−1) throughout with very little seasonal changes, unlike at the southern and coastal regions of India. On the basis of the distribution of temperature, rainfall, rainy days, humidity, presence of fogs and thunderstorms, the climate of the area is classified into four seasons; winter (December through February), premonsoon (March through May), monsoon (June through September) and retreating monsoon (October and November). During winter, fair weather prevails with occasional fog and haze. Rainfall is scarce, with a seasonal total of 103 mm (∼4% of annual). The minimum temperature ranges between 8°C and 10°C and the maximum between 27°C and 29°C. During premonsoon season, the ambient temperature increases (with maximum in the range 28°C to 32°C) and rainfall increases to ∼24% of annual. During the monsoon season, rainfall is intense and about 65% of the annual rainfall occurs during this period. The maximum temperature ranges between 33°C and 37°C. The monsoon withdraws by the end of September or beginning of October, rainfall decreases abruptly (∼7% of annual) and the sky becomes progressively clear.

3. Aerosol Database

[6] Spectral aerosol optical depth (AOD) were estimated regularly at Dibrugarh using a 10-channel MultiWavelength solar Radiometer (MWR) [Moorthy et al., 2007] designed and developed following the principle of filter wheel radiometers. The details of the instrument, method of analysis and error budget are discussed in detail by several investigators [e.g., Moorthy et al., 2007; Gogoi et al., 2008]. The MWR makes continuous measurements of direct solar flux at 10 wavelength bands centered at 380, 400, 450, 500, 600, 650, 750, 850, 935 and 1025 nm. The bands are selected using narrowband interference filters with full width at half maximum (FWHM) bandwidth of 5 to 6 nm and a band shape factor 3, which ensures a near uniform transmittance within the passband and a sharp reduction in the transmission beyond. The filters are blocked beyond the passband, from far UV to far IR, and the transmission in the blocking range is <10−4 of that in the passband. The band-selected radiation is passed through a field-limiting optics that limits the total field of view of the MWR to <2°. The radiation is detected using a photodetector amplifier hybrid (UDT 455 UV of United detector technology) operating in photovoltaic mode. The output voltage of the UDT is proportional to flux incident at the entrance window, over several orders of intensity variations. This voltage is digitized using a 12-bit ADC and the data is recorded on to an IBM compatible PC along with the information of the wavelength and time of observation (accurate to seconds). The instrument operates in a fully automatic mode employing a passive equatorial mount and the data are collected at regular intervals of 2 min over the entire wavelength range. Spectral AODs were estimated from the MWR following Langley plot technique at each of the wavelengths on all clear and partly clear days when unobstructed solar visibility (for the portion of the sky with angular diameter of ∼10° with the sun at the center) was available for >3 h a day. The data collected during any day are considered as a single set if they span only 3−4 h and the AOD is considered as mean for the day. However, if the data span over more than 5 h of the day, then the forenoon (FN) and afternoon (AN) parts of the data are considered as two separate sets and AODs are deduced for each set separately and these are considered as independent data. The short-term temporal variations in the AOD (within the duration of a data set) are not considered in this study, which deals with the longer temporal (over a month) averages for climatology. This is generally valid, as the station is remote in nature, with no major local sources of aerosols within a radius of ∼100 km about the site. In the present study, a total of 629 sets of MWR data, spread over 6 years (October 2001 to December 2007), were analyzed to obtain the climatological features of aerosols properties at Dibrugarh.

[7] In the estimation of AOD by Langley technique, the stability of the instrument is important. This is ascertained by examining the temporal invariability of the Langley intercept, corrected for the daily variation in Sun-Earth distance. The long-term stability of the instrument was fairly good, with the Langley intercepts lying within 5% of the mean typically and 10% for the worst cases. The fluctuations are relatively higher at the shorter wavelengths compared to those at the longer wavelengths. Estimates have shown that typical error in the retrieved AOD is ∼0.01 excluding the variance of the Langley fit. The variance of the Langley intercept (typically 5%) along with the other uncertainties puts the uncertainty in AOD in the range of 0.02–0.03 at different wavelengths, the values tending toward the upper levels at shorter wavelengths (<500 nm) and during periods of high AODs (>0.5), which are rather less frequent and confined only to the premonsoon season. Thus these uncertainties are for the worst case, and also account for the effects of averaging and statistical (regression) analysis. As the typical AOD values are ∼0.3 to 0.4 at short wavelengths (380 nm to 500 nm range), this means that the worst uncertainties are ≤10%. Moreover, these errors are statistical in nature and are uncorrelated among different channels, and not systematic (like dark current or detector offset, and owing to model errors in accounting for the scattering and absorption by air molecules, which are all less than 0.1%). The channel 935 nm has strong absorption due to water vapor and this channel is mainly used for estimating columnar water vapor along with the window channel 1025 (details are given by Nair and Moorthy [1998]) and this information is used to correct for water vapor absorption at 850 nm (which is very small, ∼0.005 typically). As far as the absorption due to O3 is concerned, the maximum contribution to AOD works out to be 0.03 at 600 nm (close to the peak of the Chappius band). The optical depth due to O3 absorption as a function of wavelength is estimated for the wavelength range 450 to 700 nm using spectral absorption cross section from LOWTRAN (at 5 cm−1 resolution) and model altitude profile of O3 for Indian regions derived from rocket and balloon measurements (as explained by Moorthy et al. [1989]). The uncertainties in these are very small (<0.003).

4. Results and Discussion

4.1. Spectral Aerosol Optical Depth

[8] Spectral AOD (denoted by τpλ), estimated from the MWR are examined for temporal features; the temporal variation of the monthly mean AOD at 500 nm being shown in Figure 2, where the vertical bars through the mean are the standard errors. It is observed that during any year the AOD increases to attain peak value during March−April. This peak has a mean value (over the 6-year period) of ∼0.61 ± 0.07, even though highest value of monthly mean AOD at 500 nm (0.98 ± 0.31) occurred in March 2007. Subsequently, the AODs fall off rapidly toward the monsoon months. The annual pattern repeats nearly consistently, but with different magnitudes of mean AOD.

Figure 2.

Temporal variations of the monthly mean AOD at 500 nm for the period 2001–2006. The vertical lines through the bars shows the upper bound of the standard deviation of the mean.

[9] In view of the above, the AODs of individual months of different years are grouped together and the climatological means are estimated. The annual distribution of the climatological mean AOD is shown in Figure 3, at four representative wavelengths (380, 500, 750 and 1025 nm) spanning over the spectral range covered by the MWR. A rather systematic variation is seen with AOD peaking in the month of March, falling off gradually to the lowest value in July and increasing slowly superposed with some modulations, which show a weak secondary peak in September. Though the annual variations are quite similar over the entire wavelength range (of the MWR), these are sharper at the shorter (visible) wavelength regime (380 to 500 nm) compared to the NIR. Figure 3 also indicates a change in the wavelength dependence of AOD with seasons. The highest value of the monthly mean AOD at 500 nm is 0.54 ± 0.08 (in March) and the least is 0.09 ± 0.07 (in July). Variations in τpλ within a month are generally higher during the months of April to September, as indicated by the length of the error bars. Considering the entire data period, the climatological mean AOD at 500 nm for Dibrugarh is ∼0.29 ± 0.23 [annual mean values varying from 0.23 ± 0.17 in 2006 (lowest) to 0.37 ± 0.13 (highest) in 2007]. Here it is to be noted that this large annual variation in AOD with a peak in the premonsoon season is very important, especially in the context of the “Elevated Heat Pump Effect” [Lau et al., 2008], which demonstrates the role of elevated layers of absorbing aerosols in the Himalayan region in significantly modifying the monsoons.

Figure 3.

Annual variations of climatological mean AOD at four representative wavelengths: 380, 500, 750, and 1025 nm. Vertical lines through the points are the standard deviations of the mean. The peak in March and the low in July are clearly depicted at all of the wavelengths.

4.2. Ångström Parameters

[10] The spectral dependence of AOD contains information about the physical characteristics of aerosols that can be inferred from the Ångström relationship as [Ångström, 1961]

equation image

where the wavelength exponent α is a good indicator of the fraction of accumulation mode particles (radii < 1 μm) to coarse mode particles (r > 1 μm) and the turbidity coefficient β (equals τpλ at λ = 1 μm) provides a measure of columnar aerosol loading. The values of α and β were calculated by evolving a linear least squares fit between τpλ and λ (in μm) in log-log scale over the entire wavelength range of the MWR. The slope and intercept of the regression line give respectively the values of α and lnβ. Following this, α and β are estimated for each AOD spectrum and climatological monthly mean values are calculated. The annual variations of these, shown in Figure 4, shows transformation of the aerosols over Dibrugarh from a high accumulation mode domination (α > 1.0) during September through January to increased coarse mode domination (α < 1.0) during February through August. Turbidity coefficient β remains low (<0.2) for most part of the year (July through January). The highest aerosol loading occurs during March (β ∼ 0.36 ± 0.08). The climatological mean values of α (∼1.08 ± 0.29) and β (∼0.19 ± 0.11) indicate a relatively low abundance of fine or accumulation mode aerosols as well as low columnar aerosol loading at Dibrugarh compared to other continental locations in India as will be seen later.

Figure 4.

Annual variation of the climatological mean values of the Ångström parameters α (open squares joined by a thin line) and β (solid circles joined by a thick line). Vertical lines through the points are the standard deviations of the means.

4.3. Frequency Distribution of AOD and α

[11] In order to examine the distribution of AOD and α over the years and to find the weightings of high and low values to the climatological mean, the frequency of occurrence of these parameters are estimated and shown in Figures 5a and 5b, respectively. This shows a highly skewed distribution for AOD (at 500 nm), with >66% of values lying below 0.25; with a median of 0.15, indicating, in general, a low aerosol loading. Frequency of occurrence of higher AODs decrease progressively, and only in 1% of the cases AOD was ∼0.8. However, these higher AODs are responsible for the large standard deviations seen for the climatological mean. The frequency distribution of α is more symmetric with ∼43% of values are <1.0. However, α shows large variations in the months of September through January, with occasional very high values. In all ∼20% of AOD spectra showed α > 1.5 and this leads to the high standard deviation of 0.29 to the climatological mean value of 1.08. With a view to examining these from the perspective of distinct aerosol types that dominate in different seasons, we have shown the frequency of occurrence of AOD and α as a function of distinct seasons in Figures 6a and 6b. While the distribution of AOD is quite broad, indicating occurrence of AODs over a wide range (0.1 to 0.9) during the premonsoon and monsoon seasons, it becomes narrowly peaked in the retreating monsoon (ret-monsoon) and winter seasons, when AODs < 0.3 occur for >81% and 63%, respectively. The frequency distribution of Ångström exponent too exhibits clear seasonal variation in the distribution, with the mode changing from a low value of 0.7 during premonsoon, to 1.1 during winter through 0.9 in ret-monsoon season. During monsoon season, the distributions are quite broad for both AOD and Ångström exponent. This further indicates the dominant influence of a coarse mode aerosol type during premonsoon and monsoon seasons, which becomes rather weak during ret-monsoon and winter.

Figure 5.

Frequency of occurrence of (a) AOD (500 nm) and (b) Ångström exponent α for all data irrespective of seasons.

Figure 6a.

Frequency of occurrence of AOD (500 nm) examined separately for the distinct seasons, showing the broad nature of the distributions during premonsoon and monsoon seasons in contrast to the more sharp and skewed distributions for the other seasons.

Figure 6b.

Same as Figure 6a but for the Ångström wavelength exponent α.

[12] In view of the strong influence of prevailing synoptic conditions in modulating the aerosol properties and to aid incorporation into regional models, we examined the climatological seasonal mean values of AOD and α. These are presented in Table 1, along with similar reports from earlier measurements by different investigators at other parts of India and countries adjoining Dibrugarh. It is apparent that the Ångström exponent α at Dibrugarh during winter (∼1.14) is comparable to those reported over dusty urban (Beijing and Gwangju) and coastal (Trivandrum, Gosan and Anmyon) locations, but higher than the oceanic (Arabian Sea) and pristine high-altitude (Nainital, Dunhuang) locations; indicating the strong presence of coarse mode aerosols associated with natural sources (dust or sea spray). On the other hand, industrialized urban and coastal locations such as Kanpur (urban, continental), Shirahama (urban), Pune (urban, industrial) and Goa (harbor, industrial) show comparatively higher values of α (ranging from 1.25 to 1.48) due to the strong accumulation mode sources associated with anthropogenic activities. It is also interesting to note the high values of α at Dalanzadgad, a high-altitude rural site in Mongolia, which is due to the extremely low dominance of coarse mode aerosols. During premonsoon and monsoon seasons, the aerosol characteristics, depicting lower value of α (∼0.85) at Dibrugarh is similar to those seen at the IGP (Kanpur) and coastal (Trivandrum) India. Even though Beijing, Dalanzadgad and Gosan also show similar characteristics during premonsoon season, the α values at these stations remain fairly high (∼1.4, 1.3 and 1.2, respectively) owing to the strong industrial component. During ret-monsoon season, Dibrugarh (α = 1.12) bears a resemblance only to Kanpur while the other continental and coastal locations show comparatively higher values of α (ranging from 1.2 to 1.5). The examination of Ångström exponent α thus indicate that the values of α at Dibrugarh are lower than those obtained in many continental sites, indicating only a moderate accumulation mode loading and a fairly strong presence of coarse mode aerosols associated with natural sources. The reduced accumulation mode abundance could be attributed to remote and fairly nonindustrialized nature of this region. Extensive measurements over different geographical environments have shown that the values of α < 1.0 indicate size distributions dominated by coarse mode aerosols (radii ≥ 0.5 μm) that are typically associated with dust or sea salt [Eck et al., 1999; Schuster et al., 2006]. The high values of α (>1.3) are typically associated with the continental aerosols [Behnert et al., 2007]. The influence of continental air mass, to the increased value of α over oceans adjacent to continents, is shown in earlier literature [Hoppel et al., 1990; Satheesh and Moorthy, 1997; Moorthy et al., 2005, 2007]. Examining our results and discussions that have foregone along with Table 1, it is clear that columnar aerosol property over Dibrugarh closely resemble that of a typical remote coastal or arid environments during premonsoon and monsoon seasons; while it behaves almost like a polluted and anthropogenically influenced continental site during winter.

Table 1. AOD and Ångström Exponent α Over Various Indian Locations as Well as Marine Locations Adjacent to India and East Asian Locations Along With That Observed During the Present Study
LocationLocation CharacteristicsInstrumentPeriodAODÅngström ExponentReference
Dibrugarh (27.3°N, 94.6°E, 111m)Rural continental site, northeast IndiaMWR2001–20070.310.450. study
Kanpur (26.43°N, 80.36°E, 142 m)Urban, industrial site, Indo-Gangetic plainCIMEL2001–20030.570.540.660.631.260.600.661.12Singh et al. [2004]
Beijing (34.0°N, 116.40°E, 92 m)Urban site, ChinaCIMEL2001–20050.500.801.000.511.100.901.251.15Kim et al. [2007]
Gwangju (35.1°N, 126.50°E, 60 m)Urban site, KoreaCIMEL2004–20050.290.460.610. et al. [2007]
Shirahama (33.7°N, 135.4°E, 10 m)Downwind site of city, JapanCIMEL2001–20050.190.340.870. et al. [2007]
Dunhuang (40.04°N, 94.79°E, 1300 m)Dust activity center, ChinaAureolemeter1999–20000. et al. [2004]
Nainital (29.37°N, 79.45°E, 2000 m)High-altitude station, central HimalayasMWR2002–20020.030.16-0.080.720.46--Sagar et al. [2004]
Dalanzadgad (43.6°N, 104.4°E, 1470 m)High-altitude rural site, MongoliaCIMEL1997–20050. et al. [2007]
Pune (18.53°N, 73.85°E, 570 m)Urban, industrial site, west coast of IndiaPrede Sun-sky radiometer2000–20040.380.42--1.40 --Pandithurai et al. [2007]
Trivandrum (8.55°N, 76.97°E, 3 m)Tropical coastal site, southern IndiaMWR2000–20030.430.400.290.381.000.850.321.20Moorthy et al. [2007]
Dona Paula, Goa (15.47°N, 73.81°E, 25 m)Coastal site, west coast of IndiaMicrotop-II2000–20020.410.48--1.481.14--Suresh and Desa [2005]
Gosan (33.3°N, 126.2°E, 50 m)Coastal background site, KoreaCIMEL2001–20050.290.400.400.291.000.801.201.40Kim et al. [2007]
Anmyon (36.5°N, 126.3°E, 47 m)Coastal background site, KoreaCIMEL2000–20050.310.550.500.311.101.001.401.30Kim et al. [2007]
Arabian Sea (4°N–20°N to 50°E–78°E)Oceanic regionMWR1995–20020.290.47--0.700.30--Satheesh et al. [2006a]
Bay of Bengal (10°N–25°N, 77°E–100°E)Oceanic regionMWR2000–2004-0.48-0.19-0.50--Satheesh et al. [2006b]

[13] The presence of dust is indicated by the fact that the largest τp(500) values correspond to the lowest α values [Holben et al., 2001]. Dey et al. [2004] have reported that maximum τp(500) > 0.8 was observed during the premonsoon season over Kanpur corresponding to the maximum decrease in α. Table 1 also clearly indicates that the highest seasonal mean AOD, but with the minimum values of α for six principal sites of East Asia occurred during March through April (or May). Over the Indian mainland, coastal and the adjoining oceanic regions, the seasonal mean values of AOD are highest with lowest α values in the premonsoon season. Examination of our results in the light of the above clearly indicates that the high AOD and lower value of α during premonsoon season at Dibrugarh is due the loading of dust into the atmosphere.

4.4. Retrieved Columnar Size Distribution

[14] The spectral AOD (τpλ) values contain information of height integrated (columnar) size distribution (CSD) functions [nc(r)] of aerosols, as both are connected through the Mie integral equation,

equation image

where Qext is the Mie extinction efficiency, which depends on the aerosol complex refractive index (m), radius (r) and wavelength of the incident radiation (λ); nc(r) is the columnar number density of aerosols (in a vertical column of unit cross section) in an infinitesimal radius range dr centered at r. In defining nc(r) this way, it is implicitly assumed that the number size distribution is height invariant or averaged over the vertical column. The radii limits raand rb to the integral are respectively the lower and upper cutoff radii of the particles, such that only those particles having sizes within the range ra to rb contribute significantly to Qext.

[15] Under such conditions, nc(r) can be retrieved from spectral AOD estimates by numerical inversion of equation (2). Of the several methods described in the literature, we have adapted the constrained linear inversion method [King et al., 1978, King, 1982]. Since the MWR measures only the directly transmitted flux (unlike the AERONET, where almucantar measurements are also measured and used for retrieving size distributions), we used the spectral AODs and equation (2) to retrieve the columnar size distribution. The details of application of this method to the MWR data have been discussed by Moorthy et al. [1991] and Saha and Moorthy [2004]. The spectral AODs from the measurements and the corresponding errors formed the inputs. In the present case, we set ra = 0.05 μm and rb = 3.0 μm as optimal after performing a sensitivity analysis of the size range on Qext over the entire wavelength range of measurements. This range was divided into 10 coarse bins of unequal width in the numerical approach. We used wavelength-dependent complex refractive indices (1.48, −.0464), (1.48, −.0452), (1.48, −.0437), (1.48, −.043), (1.48, −0.0412), (1.48, −.0415), (1.48, −.0409), (1.47, −.0418), (1.47, −.0441), (1.47, −.0464), adapted from Lubin et al. [2002], considering spherical aerosols, externally mixed with 10% soot during winter. Each parenthesis represents one complex value with the first one being its real part and the next the imaginary part. The spectral AODs are reestimated using the direct Mie equation, after each iteration, and are compared with the input AOD spectrum, and the solutions are accepted only when the reestimated values agree with those from the measurements within the measurement errors. As the measurement errors are also part of the input, the solutions are weighted by these errors, with better accuracy around size ranges sensitive to more accurate (AOD) measurements [King, 1982]. In general, stable solutions were obtained after 6 to 7 iterations, for fairly low value (<0.1) of the smoothing factor, γrel [King et al., 1978]. Though it is likely that the aerosol refractive index would vary with season with the change in aerosol types, this is not considered in the inversion, as there exist no data on the seasonal refractive indices for aerosols over the Indian region. However, our inversions are from the spectral extinctions of direct solar radiation, rather than from the scattered sky radiances. Under such conditions, following King et al. [1978], the nature of the retrieved size distributions and the positions of the modes are insensitive or only very weakly dependent on changes in the aerosol refractive index (within the normally expected range for aerosols); only the absolute magnitude of the number concentration will be affected. Furthermore, in computing the Mie efficiency parameters (Q in equation (2)) we assumed spherical aerosols. It is known that the aerosol shape would deviate from spherical nature, particularly for coarse mineral dust, and under such conditions the phase function of the nonspherical coarse mode particles would be more similar to that of spherical fine mode aerosols, which would lead to an underestimation of the mean and effective radii. This effect will be more important in inversions involving scattered intensity as function of scattering angle (as is the case with the AERONET inversion [Dubovik and King, 2000]).

4.4.1. Annual Variations of CSDs

[16] Representative CSDs for each month of the study period have been obtained by inverting the monthly mean τpλ values. A total of 53 such CSDs were obtained and these were examined for general features and their seasonal distinctiveness. In general, the CSDs exhibited bimodal (BM) characteristics with a secondary (less prominent) large particle (coarse) mode preceded by a primary (predominant) small particle (accumulation) mode, the aerosol number density at which was more than 100 times higher than its value at the coarse mode. In most of the cases the accumulation mode was explicit; nevertheless, in a few cases it was only indicated by a slanting nature of the CSDs toward smaller values of r. In such cases the mode might be falling below ra [Moorthy and Satheesh, 2000] the lower radii limit used in the inversion and would thus represent a monomodal with mode < ra. A few of the monthly mean AODs yielded a unimodal (UM) distribution characterized by the occurrence of a single mode, while in two cases the CSDs appeared to follow a power law (PL) dependence.

[17] Typical examples of the above four types of CSDs are shown in Figure 7 (as representative examples) with nc(r) plotted against r on a log-log scale. The respective AOD spectra are shown in Figure 8, where the points with the vertical bars represent the monthly mean τpλ values (and standard deviations), estimated using the MWR, while the continuous lines represent the τpλ values reestimated from the corresponding retrieved size distributions.

Figure 7.

Plot of size distributions (nc(r) against r in a log-log scale) retrieved from the spectral AODs, showing typical examples of unimodal (UM), power law (PL), bimodal (BM), and a combination of power law and unimodal (PL+UM) distributions. Details are provided in the text.

Figure 8.

Plots showing the spectral AODs reestimated from the retrieved size distributions (solid lines) of Figure 7, along with the corresponding measured AOD spectra (points) that formed the inputs to the inversion. The measurement errors (statistically independent) are shown by the lines through the points.

[18] It is readily seen from Figure 8 that the wavelength dependence of AOD resulting in the unimodal or power law distribution on inversion is distinctly different from that resulting in bimodal distributions. The τpλλ plots depict a hyperbolic (negative) curvature in the case of unimodal distribution, whereas the curvature is parabolic (positive) toward the longer wavelength in the case of bimodal distributions. There is almost linear (in log-log scale) decrease of τpλ with λ in case of power law distribution. It has been reported in the recent literatures that the curvature of the AOD spectra in λ domain is an indicator of the nature of the aerosols and their size distributions [e.g., Kaufman, 1993; Eck et al., 1999, 2001; Reid et al., 1999]. Accordingly a negative curvature indicates aerosol size distributions dominated by the fine mode aerosols representing a dominance of anthropogenic or biomass burning aerosols, while a positive curvature indicates size distributions with significant coarse mode contribution (such as dust or sea salt). The negative curvature associated with the UM size distribution is typical of the accumulation mode sizes, while the positive curvatures for the bimodal distribution is caused mainly by the increased contribution of the coarse mode particles to the AODs at NIR wavelengths than that by fine mode aerosols at visible wavelengths [Schuster et al., 2006]. The values of Ångström exponent (derived for the entire wavelength range) for the four representative AOD spectra in Figure 8 were 0.43, 1.03, 1.39 and 1.32, respectively. However, it is important to note that the different plausible size distributions may have the same Ångström exponent, or in other words, Ångström exponent does not characterize the uniqueness of a particular type of distributions [O'Neill et al., 2001, 2003].

4.4.2. Physical Parameters of CSDs

[19] From the retrieved CSDs, the following parameters representing the physical state of the aerosols were estimated following the expressions

[20] Effective radius

equation image

For a given aerosol size distribution, Reff is a measure of the total volume to area of aerosols and gives the radius of an equivalent monodispersion that would exhibit the same scattering properties [Hansen and Travis, 1974]. Columnar mass loading

equation image

where d is the mean density of aerosols assumed as 2.2 g cm−3 [Pruppacher and Klett, 1978]. Columnar number density

equation image

The columnar content of accumulation (Na) and coarse mode (Nc) aerosols were also estimated, considering particles smaller than 1 μm (diameter) or 0.5 μm in radius to represent the accumulation regime.

[21] In addition to the above, the physical parameters of the retrieved CSDs were determined by evolving analytical fit, with minimum RMS error, to a bimodal log normal distribution function of the form

equation image

where Noi are scaling parameters, which depend on the total aerosol concentration, rmi and σi are respectively the mode radii and standard deviations with i = 1 representing the primary (small particle) mode and i = 2, the secondary (large particle) mode. These parameters representing the physical state of aerosols were found to fall under three broad categories, namely unimodal (UM), bimodal (BM), and power law (PL), as defined earlier. The UM size distribution was characterized by a single mode (peak) in the columnar number density-size spectrum with the number density dropping off on either side of the peak. The BM distribution essentially depicted a secondary mode having a peak number density about 2 orders less than that at the primary mode, which at times was only indicated. Occasionally, the columnar number density also showed a monotonic decrease with increase in the particle radius depicting inverse power law dependence with particle radius and exhibiting a PL or Junge distribution. In cases where the CSDs were the combination of a power law and a secondary unimodal distribution (PL+UM) with the primary mode only being indicate, the parameterization was carried out using the equation

equation image

where ν is the power law index. Following this rm1, rm2, σ1, σ2, ν, N01 and No2 are estimated for the CSDs and when the distribution depicted only a single mode, i in equation (6) was set = 1. An example of the analytical fit of a retrieved BM (points joined by the continuous line) CSD to equation (6) is shown in Figure 9, with the dotted line representing the analytical best fit, for the winter of 2005.

Figure 9.

Typical example showing the analytical parameterization with minimum RMS error of a retrieved size distribution (solid line joining the points) using a bimodal lognormal distribution (equation (6)) shown by the dotted line. The parameters (mode radii and standard deviations) of the analytical distribution and the RMS error of the fit are also given.

[22] To explore the relationship between the spectral dependence of AOD and the size distribution of aerosols, we examined the variation of the effective radius (Reff) of size distribution with Ångström exponent (α) in Figure 10a. The steady decrease in Reff is indicated with the increase in α with a correlation coefficient of 0.53. This feature becomes clearly explicable if we examine the variation of Reff with the ratio (No2/No1) of peak concentrations of coarse mode to accumulation mode (deduced from the best fit analytical expressions, equation (6)), shown in Figure 10b. A very good positive correlation with a high correlation coefficient of 0.89 is seen indicating a rather sharp increase in Reff as the amplitude of the coarse mode increases more relative to that of the accumulation mode. It is also important to note that there are several occasions when the accumulation mode amplitude is only less than 100 time the coarse mode amplitude; indicating the relative abundance of the coarse mode; which readily explains the frequently observed flatter AOD spectra and lower values of α (discussed in section 4.3).

Figure 10.

Scatterplots of the effective radius (Reff), estimated from the retrieved CSDs against (a) Ångström wavelength exponent (α) and (b) ratio of peak concentration at the coarse mode (N02) to that at the accumulation mode (No1), estimated from the analytical fit.

[23] In Figure 11, we examine the variation of Reff with mL (Figure 11a) and Ångström exponent β (Figure 11b). Even though the variation in top panel is a natural consequence of higher aerosol volume being contributed by the large particles, the observation in Figure 11b that the increase in β being associated with increase in Reff indicates that the increase in the columnar abundance is associated with a larger increase in the coarse mode concentration.

Figure 11.

Scatterplot of Reff against (a) mass loading mL and (b) Ångström coefficient β.

[24] With a view to delineating the seasonal distinctiveness, and evolving seasonal aerosol models, we examined the climatological values of rm1, rm2, σ1, σ2, Reff, mL, Nt and No2/No1, by averaging each of the parameter with respect to individual months, irrespective of years. The annual variations of the mode radii (rmi) and its standard deviations (σi) of the mode are shown in Figure 12. The most striking feature is the nearly steady radii of rm1 and rm2 over the years. They do not show any remarkable seasonal changes, but just remain around mean value 0.11 ± 0.07 and 0.99 ± 0.10; that is, mainly there are two types of aerosols, whose size distributions does not change significant, only the relative contribution of this to columnar abundance with season lead to changes in α, β or Reff. In other words, the strengths of the sources those contribute to the secondary mode change seasonally. The mean values of σ1 and σ2 are generally ∼0.37 ± 0.03 and 0.21 ± 0.04, respectively; that is, the primary mode is, in general, broader than the secondary mode. The above observations indicate the existences of two modes in the columnar size distribution of aerosols, attributable to distinct source mechanisms. Since the basic characteristics of aerosol size distributions (mode radii and standard deviations) do not show any clear seasonal change, implying a steady source characteristics.

Figure 12.

Annual variations of (a) mode radii, rm1 and rm2, and (b) standard deviations σ1 and σ2 of the mode. The dashed lines represent the mean values. The striking feature is the nearly steady value of rm1 and rm2 over the years.

[25] Now, examining the temporal features of other derived parameters in Figure 13, we note that both Reff and mL increase from January to reach the peak values by March−April (Reff = 0.43 ± 0.07 μm in April, Figure 13a); and whereas mL = 335.2 ± 28.5 mg m−2 (Figure 13b). Comparable higher value of Reff (∼0.37 ± 0.05) is seen in the month of October, too. Except during the premonsoon months (March to May), the values of mL remains below 200 mg m−2, while Nt is relatively featureless (Figure 13c). Figure 13d shows that during March to August (excluding a drastic drop in July), the ratio of No2/No1 increases by more than 2 orders of magnitude from its value during the December or January, indicating increased dominance of large particles in the size spectrum.

Figure 13.

Annual variations of (a) effective radii, reff, (b) mass loading, mL, (c) integrated content of aerosols, Nt, and (d) ratio of coarse to accumulation mode aerosol concentration, No2/No1. Highest aerosol loading mL and Reff occur in the premonsoon season showing large abundance of coarse mode aerosols.

[26] Despite the above characteristics on the annual variations, the seasonal mean CSDs were always bimodal except for the premonsoon 2005. Highest aerosol loading Nc/Na and Reff occur in the premonsoon season indicating dominance of advected coarse mode aerosols. Subsequently all these parameters decrease through monsoon season and reach the low, in general, during the ret-monsoon season. From the interdependency of mL and Reff it is evident that the occurrences of the lower values of Reff and mL during monsoon season are associated with the decrease in the total concentration and the relative abundance of coarse mode aerosols, washed out by rain. However, the lower value of mL occurring, in general, during ret-monsoon season owing to lack of replenishment of coarse particles, which have been washed out by the extensive rain.

[27] Earlier studies at several continental locations in Asia have shown the coarse (secondary) mode in the size distribution to be associated with continental dust or sea salt aerosols. Investigations at East Asian sites (Anmyon, Gosan, Beijing and Shirahama; in Table 1), have shown the coarse mode to occur between 2 to 4 μm during the Asian dust storms in premonsoon (MAM) seasons [Kim et al., 2007]. On the basis of Kanpur AERONET measurements, Dey et al. [2004] have shown steady bimodal size distributions (volume) during the period of strong dust activity (April–July) with a coarse mode centered around 5 μm with an average effective radius Reff > 0.4 μm. However, these measurements were made somewhat within the source regions of the dust storms. The site Dibrugarh is quite far away from active dust source regions of west Asia and central India, and surrounded by forests, dense vegetation and water bodies, which do not favor local dust production. Under such conditions, what we encounter at DBR would be advected (transported) dust, which is much finer in size than those observed near the source regions of the arids. Hess et al. [1998] have described the transported dust with a lognormal distribution having mode radii of 0.5 μm. This might account for the differences in the mode radii seen at DBR from those reported for Kanpur. The differences in the inversion schemes also might be contributing. Notwithstanding all these, the bimodal nature is persistent at all these stations. The coarse mode concentration is found to significantly decrease during monsoon associated with wash out. On the basis of multiyear measurements from the coastal location Trivandrum in south India, Moorthy et al. [1991] have shown that the CSDs transform from unimodal during winter season to bimodal with a coarse mode (at ∼0.8 μm) during monsoon season, and attributed it to the advection of sea-spray aerosols from the adjoining ocean. Similarly, analyzing the CSDs over a tiny island location Minicoy (in the Arabian Sea), Moorthy and Satheesh [2000] have reported the presence of a strong coarse particle mode rm2 ∼ 0.86 μm and attributed it to sea-spray aerosols. Examining the results at Dibrugarh in the light of the above and its particular topography, allowing advection only from the Indo-Gangetic plains or Bay of Bengal, it appears that the strong presence of the coarse mode aerosols and the low value of α during most part of the year is associated with either mineral dust or marine aerosol component or both.

5. Role of Transport

[28] In order to establish the links between the synoptic air masses and the optical and microphysical properties of aerosols over Dibrugarh, the 5-day isentropic back trajectories are analyzed on all days when the MWR observations were made using the Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model. The trajectories are considered at three different height levels, namely, 500 m, 1800 m and 3600 m above ground following the considerations of Moorthy et al. [2003]. Given the regional peculiarity of Dibrugarh (section 2), only the higher trajectories would be capable of long-range transport and as such, we focus on them. For this, we separated the trajectories into different clusters, to ascertain the primary pathways that favor advection of aerosol particles originated elsewhere. The main criterion of the trajectory clustering is to minimize the variability among the trajectories within a cluster and maximize the variability between the clusters. Initially each trajectory is defined to a cluster, in other words if there are N trajectories there are N clusters. Now, for every combination of trajectory pairs, the cluster spatial variance (SPVAR) is calculated. SPVAR is the sum of the squared distances between the endpoints of the cluster's component trajectories and the mean of the trajectories in that cluster. Then the total spatial variance (TSV), the sum of all the SPVAR, is calculated. The pairs of clusters combined are the ones with the lowest increase in TSV (which is initially zero). After the first iteration, the number of clusters is N - 1. Clusters paired always stay together. For the subsequent iterations, the clusters are either individual trajectory or the cluster that were initially paired. Again, every combination is tried, and the SPVAR and TSV for each are calculated. The iterations are continued until the last two clusters are combined.

[29] Initially during clustering iterations, the TSV increases faster, subsequently slowly and at a nearly constant rate. At some point in the iterations, it again increases rapidly, indicating that the clusters being combined are not very similar. This latter increase suggests where to stop the clustering and is clearly seen in a plot of percentage change in TSV versus number of clusters (Figure 14). The iterative step just before the large increase in the change of TSV gives the final number of clusters. It is readily seen from Figure 14 that the TSV increases abruptly when the number of cluster decreases to 2. The most acceptable value of cluster can thus be considered here as 4 or 3. Following the above procedure, the cluster mean trajectories and their percent contributions are evaluated for each season at each different height. These are shown in Figure 15 (for premonsoon and monsoon) and Figure 16 (ret-monsoon and winter) respectively. It is readily noticed from Figures 15 and 16 that the cluster-mean trajectories distinctly change their orientation, traversing distinct geographic locations at different seasons, at different heights. Seven distinct cluster mean trajectories are found, which are assigned to distinct groups in accordance with their location of origin as mentioned in Table 2. The percentage contribution of each group to the total is also indicated in Figures 15 and 16. Subsequently, we grouped individual days with trajectories possessing a common cluster at all the three levels, along with the days on which the trajectories possess the same cluster at least in two levels. The values of AOD, α, and β for individual days were separated into each group and averaged and the mean values are given in Table 3. The values appearing after the “±” symbols represent the standard errors. The following emerge from Table 3.

Figure 14.

Total spatial variance (TSV) versus clusters in premonsoon season for trajectories at 1800 m above ground.

Figure 15.

Cluster mean trajectories along with their spread during premonsoon and monsoon seasons arriving at DBR at two different height levels from ground, but above the atmospheric boundary layer. The percentage contribution of each cluster to the total is also indicated.

Figure 16.

Same as Figure 15 but for ret-monsoon and winter seasons.

Table 2. Cluster Groups and Region in Which They Originated
GroupRegion in Which They Originated
1Local (L)
2Bay of Bengal (BoB)
3Mainland of India (MI)
4West Asia (WA)
5Southern China (SC)
6East Asia (EA)
7Mixed (M)
Table 3. Mean Values of AOD, α, and β With Respect to Each Cluster of Trajectories at Four Seasons of Premonsoon, Monsoon, Ret-Monsoon, and Winter
SeasonGroupaAOD (500 nm)αβ
PremonsoonL0.31 ± 0.050.77 ± 0.070.21 ± 0.02
 WA0.45 ± 0.030.85 ± 0.050.28 ± 0.02
MonsoonL0.29 ± 0.081.07 ± 0.120.16 ± 0.02
 MI0.37 ± 0.080.91 ± 0.150.19 ± 0.03
 EA0.33 ± 0.050.98 ± 0.090.16 ± 0.02
Ret-MonsoonWA0.18 ± 0.021.05 ± 0.090.09 ± 0.02
 L, BoB0.18 ± 0.100.95 ± 0.090.10 ± 0.02
 MI0.18 ± 0.101.0 ± 0.050.10 ± 0.05
WinterMI0.26 ± 0.031.03 ± 0.060.12 ± 0.02
 WA0.31 ± 0.021.14 ± 0.030.14 ± 0.01

[30] 1. During premonsoon season, the highest value of AOD (mean = 0.45 ± 0.03) and β (mean = 0.28 ± 0.02) were contributed by the upper air westerlies from west Asia, traversing across the IGP toward northeastern India, while the AODs remained lower associated with trajectories confined to local regions.

[31] 2. During monsoon season, the highest contributions to high AOD (mean = 0.37 ± 0.08) were associated with the trajectories from central India while the advection from the east contributed 0.33 ± 0.05. Low AODs and steeper spectral dependence were encountered associated with trajectories confined to the local regions and from Bay of Bengal. It might be recalled that on the basis of AOD measurements from Port Blair in the BoB, Moorthy et al. [2003] have shown increased abundance of fine aerosols associated with advection from east Asia, which led to steeper AOD spectra.

[32] 3. The influence of long-range transport was not seen conspicuously in the ret-monsoon season. During this period, large numbers of trajectories (∼54%) were confined locally and ∼24% observations exhibited mixed characteristics. This leads to lower α and annually lowest AODs, signifying the rather pristine nature of the locale.

[33] 4. During winter season, the trajectories at all levels were purely westerly, originating either at the Indian mainland or west Asian locations. However, the low surface temperatures and extreme winter conditions are not conducive for generation and advection of dust and consequently the AODs remain moderate and α high (∼1.14 ± 0.03).

[34] In the light of the above, it may be concluded that during premonsoon season the increase in AOD over Dibrugarh is related, in general, to the higher-level transport of the desert and arid region of west Asia, polluted air from the industrialized and urban regions of India. Coming to monsoon season, the long-range transport of aerosols from south and east Asia, Bay of Bengal, and the mainland of India modulate the aerosol properties. However, there is also a rapid decrease in the aerosol concentration during this season due to the extensive washout. For aerosols in the size range 0.1 to 1.0 μm, wet removal is the chief removal mechanism [Flossmann et al., 1985; Moorthy et al., 2007]. During ret-monsoon season, the low rainfall coupled with low relative humidity and large diurnal variation in the ambient temperature leads to progressively drier land surfaces. The local sandy area (including the banks of the vast river Brahmaputra) also contributes to dust input along with those form the widespread vegetations and the dense tea plantations around Dibrugarh. In addition, Badarinath et al. [2007] have also reported significant correlation between forest fire occurrences and the variations in the aerosol concentrations over the northeastern India. However, these are seasonal and the abundance varies from year to year.

6. Conclusions

[35] The long-term investigations of AODs and columnar size distributions over the northeastern location (Dibrugarh) of India revealed the following.

[36] 1. The AOD shows clear seasonal variations with a high seasonal mean value of ∼0.45 ± 0.05 at 500 nm during premonsoon season, decreasing gradually through the monsoon season to reach the lowest value of 0.19 ± 0.06 during ret-monsoon season and increase to 0.31 ± 0.04 in winter.

[37] 2. The AOD spectra are generally flatter than those seen typically over continental sites of India and the neighboring regions. The climatological mean value of α and β are 1.08 ± 0.29 and 0.19 ± 0.11, respectively, indicating a relatively low abundance of fine or accumulation mode aerosols as well as low columnar aerosol loading.

[38] 3. The retrieved columnar size distributions are, in general, bimodal with primary mode at ∼0.1 μm and secondary mode at ∼1.0 μm. The occurrence of high mass loading and effective radius during premonsoon season is attributed to significant abundance of coarse (natural) aerosols during this season.

[39] 4. Cluster analysis of air mass back trajectories indicate that the increase in AOD over Dibrugarh during premonsoon season is related, in general, to significant transport of mineral dust from the arid regions of west Asia and northwest India across the Indo-Gangetic plains. On the other hand, the peculiar topography combined with local conditions and widespread rainfall lead to low AODs and columnar loading during ret-monsoon season.

[40] 5. These inferences are basically derived from the columnar AOD measurements. Other complementary measurements of aerosol microphysics and chemistry are needed for a comprehensive characterization of aerosols in this region, and this is planned in the coming years.


[41] This study was carried out under the Aerosol Radiative Forcing over India (ARFI) project of ISRO-Geosphere Biosphere Program (ISRO-GBP). We acknowledge NOAA Air Resources Laboratory for the provision of the HYSPLIT transport and dispersion model and READY website ( used in this publication.