Multi-Year Analysis of Aerosol Properties Retrieved from the Ångström Parameters for Different Spectral Ranges over Pune

The present study evaluates the temporal variation of aerosol optical depth (AOD500 nm) and the Ångström parameters [viz., Ångström exponent (AE, α), Ångström turbidity coefficient (β) and second order Ångström exponent (α′)] at a tropical observing site, Pune (18°32′N; 73°49′E, 559 m AMSL) during 2008–15. Six-year means for winter and premonsoon seasons together are found to be 0.534 ± 0.13, 1.054 ± 0.27, 0.254 ± 0.08 and 0.167 ± 1.33 for AOD500 nm, AE, β and α′ respectively. Average month-to-month variability of AOD500 nm, AE, β and α′ during 2008–15 depicts seasonal cycle with strong departures with respect to long-term means. Frequency distributions for AOD, AE and β are positively skewed (skewness = 0.77, 0.32 and 1.14 respectively) while it is negatively skewed for α′ (skewness = –0.18). Analysis of AE difference, curvature parameter difference (α2–α1) and AOD500 nm–AE440-870 nm contour density map reveals that the aerosol ensemble at Pune consists of four aerosol types viz., UI (urban/industrial), CM (clear maritime), DD (desert dust) and MT (mixed type). Their relative magnitudes, however, differ during winter and pre-monsoon seasons. Thus, the contour density map shows dominance of UI and relatively less occurrence of MT type aerosols during winter. In pre-monsoon, however, the aerosol scenario is driven by MT type aerosol although UI and DD type aerosols show their remarkable existence.


INTRODUCTION
The Earth-atmosphere energy balance acts as the driving factor in climate dynamics.This demands understanding of the processes perturbing the ensuing energy balance critical in discerning the climate system.Globally, numerous studies have revealed that the aerosol particles are one of the key players in these processes (Krishnamoorthy et al., 1993, Holben et al., 2001;Kinne et al., 2013;Mao et al., 2014).They affect the global climate system in different ways, for example, through the extinction of ground reaching solar irradiance, redistribution of energy across the Earth's atmosphere, influence on cloud microphysics and thereby the hydrological cycle.Aerosols produce cooling/warming of the atmosphere, determined by their intrinsic properties, like single scattering albedo, composition, etc., in opposition to the greenhouse gases which only produce warming.Although, aerosol optical, radiative and physical properties are now reasonably documented, the "level of scientific understanding" concerning the various climatic effects of atmospheric aerosols is still very inadequate.Large uncertainties, however, still exist in aerosol climate implications due to diversity in aerosol types, modifications in radiative, optical, physical and chemical properties, influence of dynamic and/or synoptic meteorology and prevalent mixing mechanisms in the atmosphere (IPCC, 2013).
Over India, the regional monsoon system, atmospheric dynamics, seasonally-changing air-mass patterns and spatiotemporal distribution of emission sources/sinks, mainly drive the strong seasonal and inter-annual variability in aerosols (Ramachandran and Cherian, 2008).The past studies have indicated gradual increase in aerosol load due to upward growth in population, urbanization, industrialization over Indian sub-continent exhibiting large spatio-temporal variability in aerosol trends (Dani et al., 2012;Ramachandran et al., 2012a, b;Moorthy et al., 2013).Aerosols can be a mixture of various types, like anthropogenic aerosols produced due to urbanization/industrialization, crop residue burning and seasonal forest fires (Jain et al., 2014;Pawar et al., 2015), dust generated in the Thar Desert or transported from Arabian Desert, Saudi Arabia (SA), and the United Arab Emirates (UAE) (Aher et al., 2014), and marine aerosols from adjoining oceans (Sinha et al., 2012).
AOD, an integrated extinction coefficient over a vertical column of a unit cross-section in the atmosphere, is an indicator of the degree to which aerosols prevent the transmission of light through absorption and scattering of light.Knowledge of the spectral dependence of AOD is an input parameter for requisite modeling of aerosol effects on the Earth-atmosphere radiation budget and for the reliable retrieval of aerosol optical properties from satellite sensors (Levy et al., 2007).AOD spectral dependence is proven to be different for different aerosol species as a result of a change in their chemical composition, physical and optical properties defined by parameters such as AOD, single scattering albedo (SSA), and aerosol size distribution.This spectral dependence is adequately expressed by the Ångström's empirical formula using which parameters AE and β can be retrieved (Ångström, 1964).
Estimation of AE in different spectral intervals is a useful technique for discrimination/characterization of different aerosol types (Kalapureddy et al., 2009;Kaskaoutis et al., 2010;Soni et al., 2011;Kanike et al., 2014).Identification of aerosol type and average particle size in the atmosphere is carried out based on the magnitude of AE.According to this, for AE ≤ 1, aerosol size distribution ensemble is dominated by coarse-mode aerosols at the effective radii greater than 0.5 µm while fine-mode aerosols having effective radii smaller than 0.5 µm prevail for AE ≥ 1.However, AE magnitude is found to be sensitive to the spectral wavelength (λ) used for AOD observation and its consequent estimation from AOD as a function of wavelength.This dependence leads to the curvature in AOD -λ variation which provides crucial input on the estimation of aerosol size distribution and is extremely sensitive to the aerosol type prevalent under given atmospheric condition (King et al., 1978).Curvature, further predicts a transformation from fine-to coarsemode aerosol fraction which is an established fact through many studies (Eck et al., 2001;Holben et al., 2001;Schuster et al., 2006;Kaskaoutis et al., 2007a, b;Kalapureddy and Devara, 2008;Kaskaoutis et al., 2009Kaskaoutis et al., , 2010)).In a theoretical investigation, Schuster et al. (2006), with the help of multiwavelength Mie computations analyzed the relationship between AOD spectral dependence and the associated aerosol size distribution by considering Ångström exponent's sensitivity to mono-modal and bi-modal aerosol size spectrum.
The present study investigates the temporal variability of aerosol optical properties, i.e., AOD, AE (retrieved by different methods), β, the second order Ångström exponent, α′ and correlation amongst curvature parameters (α 1 , α 2 ) of spectral AOD over Pune in the long-term observations .Attempt is also made to discriminate aerosols on the basis of AOD 500 nm -AE 440-870 nm contour density maps.

Study Location
The data used in this work were collected at the Nowrosjee Wadia College (NWC), Pune (18°32'N, 73°51'E, and 559 m AMSL), India.NWC, Pune is an urban site, situated nearly 100 m from a state highway in Pune which carries heavy traffic towards the northeast and is surrounded by commercial and residential areas.It is situated at 1500 m away from the confluence of two rivers Mula and Mutha.The geographical terrain of this location makes it an ideal site for studying mixtures of aerosols transported from deserts and oceans in addition to the local emissions.As per the Indian monsoon system, climatologically, this observing site is marked with different air circulation patterns which follow four seasons, namely southwest monsoon (June-September), post monsoon (October-November), winter (December-February) and pre-monsoon (March-May).Seasonally, the air-masses affecting the observational area change their directions and are responsible for long-range transport of aerosols of different origin and characteristics (Fig. 1).The well-mixed anthropogenic aerosols are the prominent aerosol type during winter and pre-monsoon seasons due to low wind speed.In pre-monsoon, however, both dust and biomassburning plumes affect the observing site more or less to the same extent on certain occasions, thus producing a rich mixture of both coarse-and fine-mode aerosols.During monsoon, the low-altitude air-masses are a mixture of maritime and the desert air masses.

Air Mass Back Trajectory Analysis
Air mass back trajectory analysis is a useful tool to recognize the possible aerosol source regions and to trace back their pathways along which they reach the receptor site (Pawar et al., 2012;2015) For this, a 5-day back trajectory analysis has been carried out at 0600 UTC for individual days for a period of 2008-15 over NWC, Pune using Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) meteorological model (Draxler and Hess, 1998).Air-mass trajectories are computed at three different heights viz., 500 m (mixed layer height), 1500 m (boundary layer height) and 4000 m (free troposphere) for the present location.Seasonal mean trajectories are calculated by estimating the mean of all individual day back trajectories over 120 hours for above altitudes with respect to latitude and longitude corresponding to winter and pre-monsoon seasons during the study period.Figs.1(a) and 1(b) shows the mean back trajectories with ± 1 standard deviation for winter and premonsoon seasons respectively over NWC, Pune.During winter, air-masses are found to be of local origin at 500 m and 1500 m heights whereas at 4000 m there is an occurrence of strong long-range advection.For the pre-monsoon season, however, trajectories at all heights seem to originate from the westward regions in relation to the observing site.
Back trajectory analysis suggests that aerosols can be of mixed type during both winter and pre-monsoon.More specifically, the HYSPLIT air-mass back trajectory simulation predicts that the long range transport of dust aerosols can be from the continent of North Africa, Arabian Sea, Gulf countries etc. and Thar Desert (Alam et al., 2012;Aher et al., 2014).Also, local anthropogenic activities, advection from agriculture field waste burning is expected to contribute to the mixed type columnar aerosol content (Badarinath et al., 2009).

INSTRUMENTATION AND OBSERVATIONAL DETAILS
AOD measurements were carried out from the campus of the Nowrosjee Wadia College, Pune during winter (December-February) and pre-monsoon (March-May) seasons from 2008 to 2015 by operating Microtops II Sun photometer (MT-II).Each optical filter has ~10 nm full width at half maximum (FWHM) and the long-term stability of interference filter in filter-detector combination in the optical block is approximately 0.1 nm per year.In a 2.5° field of view (FOV), a sun target and spectrometer pointing assembly are permanently attached to the optical block of MT-II, which are further laser-aligned to ensure accurate alignment of the optical channels of the spectrometer to better than 0.1°.Measurements were normally performed at 15 minutes' interval from 08:00 hrs in the morning till 17:30 hrs in the evening, when the weather conditions are free of clouds and also when there are no clouds near the Sun's field of view (FOV).
MT-II measures the direct-beam solar irradiance at 5 wavelengths viz., 440, 500, 675, 870 and 1020 nm.These measurements with the instrument's internal calibration constant viz., ln I 0 , the extraterrestrial constant at each of these wavelengths, is used to retrieve spectral AOD by employing Lambert-Beer-Bouger attenuation law.Typical errors in MT-II measured AOD retrievals is in the range of 0.002-0.021(Devara et al., 2001;Morys et al., 2001;Pawar et al., 2012).High accuracy in AOD estimation is needed in this kind of study as any bias in these estimations may significantly affect retrieval of the Ångström parameters, namely, α, β, and α′ which are derived from AODs at different wavelengths.The required high accuracy in AOD measurements was ensured by excluding perturbed irradiance measurements due to invisible clouds as they can contribute up to two standard deviations to the daily mean AODs and AE values.The manual errors in Sun pointing were further minimized by carrying out three sets of subsequent observations in each 15-min time interval.Out of the three data sets, an observation possessing standard deviation less than 5% was retained for further analysis enhancing the reliability in the data processing.This data selection procdure was further augmented by considering AOD data sets which produced best second-order polynomial fit to the ln AOD versus ln λ plot judged by the highest coefficient of determination (R 2 = 0.99) (Kaskaoutis et al., 2007a;Kalapureddy et al., 2009).After eliminating the perturbed data by implementing the above listed criteria, the available number of observing days, producing statistically significant AOD data sets at Pune during winter and pre-monsoon seasons were about 300 with on an average 10,000 AOD spectral curves for the observing period 2008-15.

METHODOLOGY AND ERROR ANALYSIS
For wavelength (λ) expressed in µm, the simplest method to quantify the changes in spectral AOD is to estimate Ångström parameters (α and β) using Eq. ( 1), As stated above, AE(α) provides first-hand information on the aerosol size distribution while the Ångström turbidity coefficient (β), corresponding to AOD at unit λ (1 µm), is linked to the columnar mass loading of coarse-mode aerosols (Aher and Agashe, 1998;Moorthy et al., 2007).In this work, both α and β are retrieved by using the linear least-squares method (LSM, use of four wavelengths) to Eq. ( 1) after discarding the water vapour and mixed gases absorption bands i.e., for the total spectral band in the range 440-870 nm.Additionally, α is also derived by using the Volz method (VM, use of two wavelengths) in the shorter wavelength band (440-500 nm) and longer wavelength band (675-870 nm) (Kaskaoutis and Kambezidis, 2008) using Eq. ( 2), here, τ 1 and τ 2 are AODs at wavelengths λ 1 and λ 2 .
As is well known, aerosols at a given observing site may be produced from different sources (such as those from biomass burning, fossil fuels, urban activity, sea-spray, mineral dust, etc.).The aerosol ensemble at a given site may therefore be a probable mixture of many of these types with real aerosol size spectrum spanning over wider size ranges.In this case, a more precise empirical relationship between aerosol extinction (AOD) and the wavelength of observation is obtained with a second -degree polynomial approximation (Pedros et al., 2003;Schuster et al., 2006) as in Eq. ( 3), where α 0 , α 1, and α 2 are coefficients of the second degree polynomial with α 2 providing information on the spectral curvature of the fitted function.As a special case, when α 2 = 0, (i.e., for insignificant curvature in the chosen spectral region), coefficients α 0 and α 1 become ln β and α (Eq. 1) respectively (Beegum et al., 2009).The deviation of the spectral AOD curve from its linear trend of variation is also quantified by estimating the second order Anstrom exponent (which is the first order derivative of the AE and second order derivative of AOD), α′ given in Eq. ( 4).
where λ i-1 , λ i and λ i+1 are the discrete wavelengths at 440, 675 and 1020 nm and τ i-1 , τ i and τ i+1 are the corresponding AODs.The magnitude of the variables α 2 and α′ just like AE enables discrimination of aerosols types in fine-and coarse-mode aerosols generated from variety of sources.Thus, for α 2 ˂ 0 and α′ ˃ 0, the curvature in the lnAOD -lnλ plot may be negative (convex type) indicating dominance of fine -mode aerosols in aerosol size distribution.Alternately, for α 2 ˃ 0 and α′ ˂ 0, the curvature in the lnAOD -lnλ plot could be positive (concave type) suggesting presence of coarse -mode aerosols in their size distribution (Eck et al., 1999l;Reid et al., 1999;Eck et al., 2001;Schuster et al., 2006;Kedia andRamchandran, 2009, 2011;Soni et al., 2011).

RESULTS AND DISCUSSION
In the following sub-sections, we present the temporal features of various aerosol optical properties, i.e., AOD, AE (retrieved by two different methods), β, α′ and also investigate correlation between curvature parameters α 1 , α 2 (determined from the polynomial fit to a plot of ln AOD against ln λ) and AOD 500 nm over Pune during 2008-15.Also, the results on the discrimination of aerosol types based on AOD 500 nm -AE 440-870 nm contour density maps are enumerated.
On seasonal basis, AE is high (≥ 1.05) during winter months (December-February) which are associated with either high or low AOD 500 nm values with respect to long-term mean .However, during pre-monsoon months (March-May) as far as the occurrence of AE is concerned, it is low (< 1.05) and is found to be associated with either high or low AOD 500 nm values with reference to the longterm mean.Since AE is ratio of number concentration of fine-to coarse-mode aerosols, the occurrence of high and low AE in winter and pre-monsoon seasons respectively indicate the dominance of fine-and coarse-mode aerosols during these seasons.Soni et al. (2011) have observed the similar variation of AE/AOD 500 nm at the observational site in Delhi for the period 2007-08.Over South India, a systematic seasonal variation of AOD 1020 nm has been observed with high values ~0.4 during monsoon and low during post-monsoon and winter seasons with values ~0.1.AOD 400 nm also has seasonal variation but not as systematic as has been observed for AOD 1020 nm .The difference between the AOD 400 nm and AOD 1020 nm become low during June, July and August and high during November, December, March and April suggesting changes in columnar aerosol size distribution (Suman et al., 2014).
These observations, further, reveal that the high AODs observed in the winter/pre-monsoon seasons are associated with their systematic diurnal trend.This is analyzed by following the sampling procedure used by many researchers (Smirnov et al., 2002;Wang et al., 2004).This aspect is clear in Fig. 4, which displays the calculated percentage departures of AOD 500 nm from the respective daily mean values averaged for each half-an-hour for the period of observations.It reveals that during winter, AOD shows maximum positive percentage departure (30%) in the forenoon (FN).During afternoon (AN), however, the departure becomes -34%.Thus, on an average, at NWC, for the winter season, AOD depicts positive departure during FN and negative departure during AN in relation to mean AOD.These departures are presumably caused by the winter haze (light brown in colour) produced mainly from places of human habitation, generally settles during the night and rises up in the FN and gets dissipated by thermal turbulence and mixing with upper winds by noon (Mani et al., 1969).Further, the dynamics of the meteorological processes in the local atmospheric boundary layer (ABL) along with an increase in combustion activities influence the winter diurnal variability (Pandithurai et al., 2007;Pawar et al., 2012).This indicates that aerosol loading is high during FN and low during AN producing high observed AOD in winter.
For pre-monsoon season, in the morning, AOD departure is negative (-27%).As the day progresses, the departure becomes positive in the time interval from 11:30 to 14:30 hrs and afterwards reverses sign.The occurrence of maximum AOD in the hot dry pre-monsoon months at NWC, Pune could be attributed to the presence of tropical continental air-mass pattern over western, central and north India during this season with its source region in southwest Asia.It is the driest and hottest air over the Indian sub-continent, with marked instability, intense insolation and turbulence which leads to the development of dust raising winds and dust storms.This indicates that the source of the aerosols is of local origin as well as influx caused by the long-range transport which depends on conditions in the first few km of the atmosphere and the dispersion of dust from the ground to higher levels (Mani et al., 1969).877.7 mm, with -1.7% departure).This behavior of the Indian monsoon can be termed as 'contrasting monsoon' and is responsible for thermal structure of the atmosphere.As such the rapid industrialization and urbanization are found to be the causal factors producing long-term  rising trend in the surface air temperature over major Indian cities including Pune (Kothawale et al., 2016).The rising temperatures cause dust generation from strong surface heating and high winds producing aerosol influx into the atmosphere thereby increasing columnar aerosol loading (Smirnov et al., 2002).For example, the surface air temperatures recorded during the pre-monsoon of 2002 lead to an increase of AOD as compared to 2001 (Bhawar and Devara, 2010).The occurrence of high AODs during 2008-10 could be assigned to contrasting monsoon and long-term temperature trend cited above.Also, Bhawar and Rahul (2013) have observed a 30-40% increase in the aerosol occurrence frequency (AOF) derived from the observed CALIPSO aerosol extinction profiles at lower altitudes (below 6 km) during 2009 and a 5-8% enhancement in AOF at higher altitudes in the year 2008.

Frequency Distribution of AE, β and α′
Analysis of frequency distribution of the overall daily AOD, AE, β and α′ data portrayed in Fig. 5 reveals a great dispersion of the values thus denoting variability in both columnar aerosol loading and aerosol size distribution.The frequency distributions for AOD, AE and β are positively skewed with skewness indices 0.77, 0.32 and 1.14 respectively while the frequency distributions for α′ is negatively skewed (skewness = -0.18).The cumulative frequency counts (CFCs) for AOD (in the range 0.32 to 0.87), AE (in the range 0.58 to 1.56) and β (in the range 0.075 to 0.525) are found to be 88%, 94% and 96% correspondingly.For α′, CFCs come to 48% and 52% with respective ranges having magnitudes -3.0 to 0 and 0 to 3.0.The observed large spread in the occurrence frequency of both AOD and Ångström turbidity coefficient (β) confirms that there exists a moderate to very high columnar aerosol loading over Pune owing to seasonally dominant/changing anthropogenic and natural aerosol sources (Iqbal, 1983;Aher and Agashe, 1998).The occurrence of large values of α′ is presumably due to the presence of the fine-mode dominated aerosol size distributions and close to zero or negative values of α′ are characteristic of size distributions with a dominant coarsemode (Eck et al., 2001;Kedia andRamchandran, 2009, 2011).Further, from the season wise splitting (figure not shown) of frequency distribution shown in Fig. 5, it is seen that during winter AE > 1 on ~87% occasions (concurrently, α′ < 0 on ~36% occasions) depicting predominance of finemode aerosols.During pre-monsoon, however, the occurrence frequency reverses with AE < 1 on 62% occasions and α′ < 0 on ~65% occasions highlighting the presence of coarsemode aerosols in the atmosphere.The present study also corroborates well with similar observations carried at the downtown location in Delhi (Soni et al., 2011).According to their study, α′ values are positive with a higher magnitude in October-December months, less than 1 in February and March, and nearing zero in April-August.

Monthly Variation of Ångström Exponents
Fig. 6 depicts a systematic box -whisker patterns along with the associated binned data of the daily mean Ångström exponents derived by using Volz method (Eq.( 2)) at two non-overlapping, narrow spectral intervals used in the observations, viz., shorter (440-500 nm), longer (675-870 nm) wavelengths and by employing LSM method to spectral AODs at four wavelengths in the spectral range 440-870 nm during 2008-15 over Pune.The solid circles in each boxwhisker plots are the mean of the sampled data while the whiskers show standard deviations and the boxes represent 25 percentile (lower line), median (middle line) and 75 percentile (top line) of AEs.The minimum/maximum values of AEs for each month are represented by lower and upper solid stars respectively.As can be seen in Fig. 6, the mean, median, 25% and 75% percentiles of AEs are found to be significantly different for all three spectral intervals during December-May (2008-15).This indicates a significant change in the monthly aerosol type and columnar aerosol size distribution.Interestingly, for all the three spectral intervals, AE ≥ 1 during December-March with large standard deviations while AE < 1 for April-May months with comparatively less standard deviations indicating the nature of the prevalent relationship between spectral dependence of AE and aerosol type (Kaskaoutis et al., 2007b;Kumar et al., 2013;Pawar et al., 2015).Similar kind of Ångström exponent variation as a function of different wavelength ranges used for its estimation had been the subject of many aerosol measurement studies (Eck et al., 1999(Eck et al., , 2005;;Sinha et al., 2012).

Correlation between Ångström Exponents at Short and Long Wavelength Intervals
As stated earlier (section 'methodology'), the curvature in lnAOD-ln λ plot could be neagative (α 2 ˂ 0 and α′ ˃ 0, convex type) or positive (α 2 ˃ 0 and α′˂ 0, concave type) implying dominance of fine-or coarse-mode aerosols respectively.Fig. 7 portrays the comparison of AE values retrieved at two narrow, short and long spectral intervals for the observation days in winter/pre-monsoon together, based on the nature of curvature indicator parameters, α 2 /α′.From the figure, it can be seen that a straight line with unity slope is a line where two AE values viz., AE 440-500 nm and AE 675-870 nm are expected to be identical.This is also a line corresponding to zero curvature (α 2 = 0) represented by Eq. ( 3).For the period under study, Fig. 7 depicts that at Pune, ~74% of the AOD spectra considered for analysis are convex-type spectral curves (α 2 < 0, i.e., negative curvature) and ~26% fall into a group having concave-type curves (α 2 > 0, i.e., positive curvature).From the plot, it is seen that AE 675-870 nm > AE 440-500 nm leads to negative curvature in spectral curve while AE 440-500 nm > AE 675-870 nm gives rise to positive curvature effect (Kaskaoutis et al., 2007b;Soni et al., 2011).This is on account of the fact that the rate of variation of AE for negative curvature is noteworthy at shorter wavelengths, while in the case of the positive curvature, it is remarkable at longer wavelengths.
On a seasonal basis, ~17% and ~36% AE values fall in α 2 > 0 group during winter and pre-monsoon seasons respectively (Fig. 7).The relatively high percentage of AE values during pre-monsoon for α 2 > 0, indicates that the coarse-mode aerosols probably of desert dust origin out number fine-mode aerosols (Aher et al., 2014) as depicted in Fig. 1.Alternately, AEs get grouped into α 2 < 0 on ~83% and ~64% occasions respectively, during winter and pre-monsoon seasons.This observation, on the other hand, corresponds to the proportionately higher number concentration of the finemode aerosols in the size distribution in the atmospheric column over Pune during winter than during pre-monsoon.This finding corroborates the results reported by Eck et al. (1999Eck et al. ( , 2010)); Schuster et al. (2006); and Kaskaoutis et al.

Ångström Exponent Difference versus AOD
The investigation of the Ångström exponent difference, defined as AED = (AE 440-500 nm -AE 675-870 nm ) as a function AOD 500 nm , gives information of AOD influence on the spectral dependence of AE.This is readily obtained by constructing a scatter diagram of AED against AOD 500 nm .The results obtained are displayed in Fig. 8 for winter/premonsoon together during the period of observations reported.For high turbid conditions, i.e., for AOD 500 nm in the range 0.5 to 1.0 for 34.32% of the observing days, AED is found to be positive and large whereas it is near zero and low for 65.68% of the observing days (Fig. 8).Positive and negative AED values indicate the positive and negative curvature respectively in AOD spectra and AED close to zero show absence of spectral variability in AE.The occurrence of the higher average fine-mode fraction obtained over Pune is supported by the frequency distribution analysis of AED ploted in Fig. 8 (Kedia and Ramachandran, 2009).This aspect  is clearly seen in Fig. 8, which shows that about 92.70% of AEDs lie in the bin ends -2 to + 2 out of which 77.74% lie in the bin ends -1 to + 1 out of all the observing days.
AED data in winter and pre-monsoon seasons shown in the same figure, further, reveals key features about the spectral dependence of AE.During winter, for AOD in the range 0.3 to 1.0, AED is found to be negative on 80.29% of days while it is positive for remaining 19.71% days during the period 2008-15.From these observations, it is clear that over Pune, fine-mode aerosols dominate over coarse-mode (desert dust) aerosols in the winter season, although a small signature of coarse-mode aerosols is seen on account of positive AED values (positive curvature; α 2 > 0).The small signature of coarse-mode aerosols is due to long transport of dust aerosols (Fig. 1(a)).Soni et al. (2011) over Delhi and Kaskaoutis et al. (2007b) at Alta Floresta (Brazil) and Ispra (Italy) have reported similar results indicating enhanced presence of anthropogenic/fine-mode aerosols under turbid atmospheres.For the pre-monsoon season, however, the scenario is quite different with AED < 0 on 59.05% observing days in AOD interval 0.3 to 1.0 and AED > 0 on 40.95% observing days in the AOD interval 0.3 to 1.0 (Fig. 8).Thus, during pre-monsoon, at Pune contributions from both fineand coarse-mode more or less equally dominate (Fig. 1(b)).Synoptically, results indicate contrasting seasonal aerosol characteristics at the observing site.

Observed Curvatures in Aerosol Optical Depth Spectra
The presence of real aerosol size distributions may give rise to the curvature in lnAOD-lnλ variation as discussed earlier.The plausible reasons for the observed curvatures in aerosol optical depth spectra may be the prevalence of various aerosol types having wide-ranging formation mechanisms (Eck et al., 1999).As a result of this, the aerosol size spectrum spans over wider size ranges and also may be multimodal.The magnitude of curvature inherently associated aerosol optical depth spectra can be precisely determined by fitting a second degree polynomial to the measured spectral variation of AOD by employing Eq. ( 3).This procedure yields parameters α 1 and α 2 as coefficients of polynomial fit which can also be used to derive relevant information about aerosol types (Kaskaoutis et al., 2009).Schuster et al. (2006) have hypothesized that for aerosol size distributions exhibiting bimodal structure, a close approximation, α = (α 2 -α 1 ) can classify the dominant aerosol types.According to this hypothesis, for (α 2 -α 1 ) ≥ 2, AOD spectra are predominated by anthropogenic/fine-mode aerosols while (α 2 -α 1 ) ≤ 1 indicates a preponderance of bigger i.e., coarse-mode aerosols.However, a wide range of fine-mode fractions or a mixture of both fine-and coarse-modes may prevail in the atmospheric column for AOD spectra for (α 2 -α 1 ) values lying between 1 and 2. In the present work, to investigate above hypothesis and the dominant modes of aerosols over study area, a seasonal plot of the difference (α 2 -α 1 ) against AOD 500 nm has been constructed.Results shown in Fig. 9 reveal that during winter, the difference (α 2 -α 1 ) is found to lie between 1 and 2 [i.e., 1 < (α 2 -α 1 ) < 2] for 86.39% of AOD spectra while it is less than 1 for 13.61% of the AOD spectra.On the other hand, for pre -monsoon season, the difference (α 2 -α 1 ) is < 1 for 61.96% cases, while for 39.04% cases of AOD spectra it falls in the range 1 < (α 2 -α 1 ) < 2. These observations emphasize that at Pune, during winter, the particulate contribution to AOD spectra predominantly comes from a wide variety of anthropogenic fine-mode fractions or mixture of modes with a significantly less share of coarse-mode fractions to the aerosol size distribution.In contrast, however, the AOD spectra during pre-monsoon denoted a sizable signature of particles from coarse-mode fraction in the aerosol size distribution.This presumably be comprised of sea salt and desert dust particles with relatively small but significant contribution of a mixture of anthropogenic aerosols.These results agree well with the research findings reported elsewhere (Holben et al., 2001;Kaskaoutis et al., 2007a, b).

Seasonal Discrimination of the Aerosol Types
Fig. 10 shows the realistic characterization of the aerosol types using contour density maps for winter and pre-monsoon seasons during the study period.The contour density map makes use of the relationship between AOD 500 nm and AE 440-870 nm because of their strong wavelength dependence.These 2-dimensional binning of AOD 500 nm versus AE 440-870 nm patterns have been observed at several locations and for different aerosol types (e.g., biomass smoke, anthropogenic aerosols, desert dust) (Masmoudi et al., 2003;Kim et al., 2004;Ogunjobi et al., 2004;Eck et al., 2010).Further, the method of contour density maps is based on the sensitivity of these parameters to different, somewhat independent, microphysical aerosol properties (Kalapureddy et al., 2009;Kaskaoutis et al., 2011).Hence, AOD 500 nm -AE 440-870 nm contour maps qualitatively indicate the amount (viz., the columnar loading) and dimension (viz., size) of the atmospheric aerosols.In these maps, the rectangular areas denote urban/industrial (UI), clean maritime (CM), desert dust (DD) and mixed type (MT) aerosols respectively, and the boundaries of these areas correspond to the selected threshold values of AOD 500 nm and AE 440-870 nm .
During winter and pre-monsoon, AOD 500 nm values are found to range from 0.1-1.0 and 0.2-1.0 while AE 440-870 nm spans over 0.6-1.5 and 0.2-1.5 respectively.In order to cater for this wide range in both AOD 500 nm and AE 440-870 nm , the threshold values have been slightly altered from those used by Pace et al., (2006) for the observing site Lampeduca (Central Mediterranean).Therefore, AOD 500 nm < 0.2 with AE 440-870 nm < 1.3 represent CM aerosols while AOD 500 nm > 0.2 and AE 440-870 nm > 1.0 can be used to characterize UI aerosols.Next, AOD 500 nm values > 0.25 associated with AE 440-870 nm < 0.7 are indicative of DD particles transported over oceanic areas.Finally, over the remaining gaps, it is difficult to discriminate aerosols and as such they are considered as MT in conformity with the various aerosolmixing mechanisms in the atmosphere such as coagulation, condensation, humidification, and gas-to-particle conversion.
The results of this analysis are displayed in Fig. 10 for winter and pre-monsoon during 2008-15.In the winter, the AE 440-870 nm values are significantly higher (> 0.8) with AOD 500 nm (> 0.3) (Fig. 10(a)).The maximum density area is observed for the pair (AOD 500 nm , AE 440-870 nm ) = ~(0.45-0.8,1.15-1.4).The AE 440-870 nm values of this magnitude are indicative of fine-mode aerosol size distributions (Eck et al., 2005).This emphasizes UI/anthropogenic aerosol field in the vertical atmospheric column which considerably outnumber the MT, CM and DD types of aerosols.The MT aerosols show their presence in aerosol load over Pune although much less as compared to UI type.
In the pre-monsoon season (Fig. 10(b)), the winter maximum density area get converted to MT type with (AOD 500nm , AE 440-870nm ) = ~(0.55-0.65,0.8-1.0).In addition to MT type the UI and DD type aerosols also show their presence prominently.However, contribution from CM type appears to be negligible.

SUMMARY AND CONCLUSIONS
The month-to-month variability in the parameters describing optical features of aerosols i.e., AOD, AE, andα′ have been analyzed over NWC, Pune during 2008-15 to study long-term characteristics of aerosol optical properties.(iii) For the period under study, for AOD in the range 0.3 to 1.0, AED is found to be negative on 80.29% of days while it is positive for remaining 19.71% days during winter On the other hand, during pre-monsoon AED is negative on 59.05% and is positive on 40.95% observing days.The difference (α 2 -α 1 ) during winter is found to lie between 1 and 2 for 86.39% of AOD spectra while it is less than 1 for 13.61% of the AOD spectra.Alternately, for the pre-monsoon season, it is less than 1 for 61.96% cases, while for 39.04% cases of AOD spectra it falls in the range 1 < (α 2 -α 1 ) < 2. This statistical data indicates that aerosol ensemble at Pune consists both of fine-as well as coarse-mode aerosols although their relative magnitudes differ during winter and pre-monsoon seasons.(iv) Analysis of the contour density maps of AOD 500 nm against AE 440-870 nm also consolidate the above statistical information given for AED and (α 2 -α 1 ) difference.Thus, the contour density map shows dominance of UI and relatively less occurrence of MT type aerosols during winter.In pre-monsoon, however, the aerosol scenario is driven by MT type aerosols although UI and DD type aerosols show their remarkable existence.

Fig. 1 .
Fig. 1.Seasonal averages of 5-days back trajectories at 500 m, 1500 m and 4000 m height levels for the period 2008-15 over Pune during (a) winter and (b) pre-monsoon.

Fig. 10 .
Fig. 10.Contour density maps of the AOD 500 nm versus AE 440-870 nm in (a) winter (b) pre-monsoon seasons for the study period.