Trends in aerosol optical depth over Indian region: Potential causes and impact indicators

Authors


Abstract

[1] The first regional synthesis of long-term (back to ~ 25 years at some stations) primary data (from direct measurement) on aerosol optical depth from the ARFINET (network of aerosol observatories established under the Aerosol Radiative Forcing over India (ARFI) project of Indian Space Research Organization over Indian subcontinent) have revealed a statistically significant increasing trend with a significant seasonal variability. Examining the current values of turbidity coefficients with those reported ~ 50 years ago reveals the phenomenal nature of the increase in aerosol loading. Seasonally, the rate of increase is consistently high during the dry months (December to March) over the entire region whereas the trends are rather inconsistent and weak during the premonsoon (April to May) and summer monsoon period (June to September). The trends in the spectral variation of aerosol optical depth (AOD) reveal the significance of anthropogenic activities on the increasing trend in AOD. Examining these with climate variables such as seasonal and regional rainfall, it is seen that the dry season depicts a decreasing trend in the total number of rainy days over the Indian region. The insignificant trend in AOD observed over the Indo-Gangetic Plain, a regional hot spot of aerosols, during the premonsoon and summer monsoon season is mainly attributed to the competing effects of dust transport and wet removal of aerosols by the monsoon rain. Contributions of different aerosol chemical species to the total dust, simulated using Goddard Chemistry Aerosol Radiation and Transport model over the ARFINET stations, showed an increasing trend for all the anthropogenic components and a decreasing trend for dust, consistent with the inference deduced from trend in Angstrom exponent.

1 Introduction

[2] The anthropogenic activities associated with modernization, industrialization, and urbanization have been leading to a sustained increase in the amounts of aerosols (suspended particles in the atmosphere) and trace gasses emitted into the atmosphere. This is believed to cause changes to Earth's climate that are irreversible, at least at regional levels, if not globally [IPCC, 2007]. In this context, the need for quantitative information on the long-term trends in aerosol loadings and the spectral aerosol optical depth (AOD) is being increasingly felt important by the climate and atmospheric science community, besides concerns on implications to human health and reduction in crop yield. This information is also sought by policy makers, who look forward to mitigation strategies and negotiations. While it is imperative that the aerosol loading at a given region might show long-term changes associated with changes in anthropogenic activities essential for economic and social developments and the ever increasing needs in the energy, transport, and agricultural sectors, the indirect modifications to weather and climate would result in modifications in the natural aerosol systems such as sea spray and dust. There have been several recent studies on long-term trend in aerosols over distinct geographical regions based on satellite data [Massie et al., 2004; Mishchenko et al., 2007; Zhao et al., 2008; Zhang and Reid, 2010; Hsu et al., 2012] due to the advantage of the global coverage provided by satellites. However, such studies are more confined to oceanic environments, primarily due to the higher accuracy of satellite-derived AODs over oceans and the better spatiotemporal homogeneity of aerosol properties over oceans, in contrast to that over the landmasses. Over the land, the large heterogeneity in time and space of the aerosol loading due to heterogeneous source distribution, and the large and heterogeneous variation in the surface reflectance make the satellite-derived AODs to be less accurate [Jethva et al., 2009]. The “snapshot” view of the satellite also is a constraint while looking for long-term features. In addition, aerosol retrievals from satellites have serious issues such as uncertainties due to the assumptions within the algorithms, bias in aerosol sampling caused by frequent cloud disturbance, and uncertainties associated with instrument calibration. Thus, studies on long-term trend in aerosols over the land call for carefully obtained data from dense networks of aerosol observatories. Regional and global networks such as the Aerosol Robotic Network (AERONET) [Holben et al., 1998], Sky radiometer network (SKYNET) [Kim et al., 2004], European Aerosol Research Lidar Network (EARLINET) [Matthias et al., 2004], Micro Pulse Lidar Network (MPLNET) [Welton et al., 2001], ARFINET [Moorthy et al., 2013a] are designed and maintained essentially with the above objective.

[3] Global-scale analysis of long-term trend in aerosol optical depth (AOD) by several investigators based on satellite data has shown negligible and contrasting trends [Mishchenko et al., 2007; Hsu et al., 2012]. Nevertheless, they have brought out regional differences in trends, with increasing trend in AOD over South and Southeast Asian regions. In addition, there are reports on the continued decrease in surface-reaching solar radiation over these regions, indicating a global dimming due to aerosols [Wild et al., 2005]. There are a few similar studies over Indian region based on the satellite data [Massie et al., 2004; Porch et al., 2007; Dey and Girolamo, 2011]. However, none of these studies employed long-term ground-based AOD measurements from the networks.

[4] With a view to address this issue, over the south Asian region, a carefully planned activity has been taken up by the Indian Space Research Organization (ISRO) by setting up a national network of aerosol observatories, now known as ARFINET, being established in a phased manner over a period of years. This activity, with the long-term objective of generating an aerosol database over the Indian subcontinent, initiated in mid-1980s [Moorthy et al., 2013a] has now matured well with the network consisting of 35 observatories spread over the mainland and adjoining oceans as shown in Figure 1, where each circle represents the ARFINET observatories, superposed on the digital elevation map of India and is identified by its short name. A list of these observatories, their geographical locations, broad classification, and the period from when (for which) data are available is provided in Table 1. As this network evolved over a period of time, the data span varies between stations. Two of the stations have more than two decades of database. The data, obtained from these observatories, are used to examine the long-term trends in aerosol optical depths over the Indian region and delineate its subregional distinctiveness and seasonality. The results are analyzed to infer, at least qualitatively, the possible causes and implications to regional climate.

Figure 1.

Locations of ARFINET sites marked on a digital elevation map of India, where each circle represents the ARFINET observatories, identified by its short name. The triangle symbols indicate AERONET stations. A list of these observatories, their geographical locations, broad classification, and the period from when (for which) data are available are provided in Table 1.

Table 1. Details of ARFINET Stations
Serial NumberStation Name (Broad Classification)Station IDLatitude (°N)Longitude (°E)Altitude (m)Starting year
  1. a

    Presently not operational.

  2. Urban: population > 2 million; semiurban: 2 million > population >0.5 million; rural: population < 0.5 million.

1Agartala (rural)AGR23.5091.2500432010
2Anantapur (semiarid, rural)ATP14.4677.6703502001
3Bhubaneswar (semiurban)BBR20.2085.8000782009
4Bangalore (urban)BLR12.9777.5909602003
5Chennai (urban)CHN12.7079.9200502011
6Dibrugarh (rural)DBR27.3094.6001112002
7Dehradun (Himalayan Foothills, semiurban)DDN30.3478.0407002007
8Delhi (urban)DEL28.6077.2002392012
9Goa (coastal, semiurban)GOA15.4673.8300702007
10Hyderabad (urban)HYD17.7578.7305572003
11Hyderabad (suburban)HDT17.4778.5805572007
12Hanle (Himalayan high altitude, pristine rural)HNL32.7878.9545202009
13Imphal (rural)IPH24.7593.9207652010
14Jaisalmer (desert, rural)JSL26.9270.9502252010
15Jodhpur (arid, semiurban)JDR26.2672.9902361987a
16Kadappa (semiarid, rural)KDP14.4678.8101382011
17Kharagpur (rural)KGP22.5087.5000282004
18Kolkata (urban)KKT22.5788.3700122012
19Kullu (Himalayan Foothills, high altitude, rural)KLU31.9077.1011542009
20Minicoy (island)MCY08.2073.0000021995a
21Mysore (semiurban)MYS12.3076.5007721989a
22Naliya (rural)NAL22.2368.8900502007
23Nagpur (urban)NGP21.1579.1503002008
24Nainital (Himalayan high altitude, rural)NTL29.2079.3019602002
25Ooty (High altitude, rural)OTY11.4076.7025202012
26Port Blair (island)PBR11.6492.7100602002
27Patiala (semiurban)PTL30.3376.4602492006
28Pune (urban)PUN18.5473.8504572005
29Rajkot (semiurban)RJK22.3070.7301422012
30Ranchi (semiurban)RNC23.2385.2306542010
31Shillong (high altitude, rural)SHN25.6091.9110332008
32Trivandrum (coastal, semiurban)TVM08.5077.0000021986
33Udaipur (semiurban)UDP24.6073.9005772010
34Varanasi (Indo-Gangetic Plain, semiurban)VNS25.3082.9600782012
35Visakhapatnam (coastal, industrialized, semiurban)VSK17.7083.1000201988

2 Database and Analysis

[5] Each of the ARFINET observatories is equipped with instruments for measurements of spectral AOD, either the Multiwavelength solar Radiometer (MWR) [Moorthy et al., 1989] or the handheld Microtops Sun photometer, both these instruments having been extensively intercompared and consistency established [Kompalli et al., 2010]. Examining Table 1, it emerges that AOD databases, long enough for trend analysis, are available from eight stations: Trivandrum (TVM), Visakhapatanam (VSK), Mysore (MYS), Anantapur (ATP), Hyderabad (HYD), Patiala (PTL), Dibrugarh (DBR), and Port Blair (PBR). Among these stations, TVM and VSK have the longest database, going well back to 25 years or more. For MYS, though the data started almost along with that of TVM and VSK, the station became nonfunctional after 2000. For the rest of the stations, the data are more recent and pertain to this (21st) century.

[6] Even though it could be argued that the database at some of these stations could have been longer for a statistically significant trend analysis, it should also be borne in mind that this is the only source of primary data over Indian region, having such a long temporal and spatial coverage where the AOD data have been obtained using intercompared instruments and following a common laid out protocol. With a view to improving the regional coverage, we have also considered two AERONET stations (Kanpur (data since 2001) and Gandhi College (GC) over the mainland) and a marine station at Hanimadhoo (data since 2005) in the Maldives (all shown in Figure 1 by the triangle symbol).

[7] The columnar spectral AOD estimated using the 10 channel MWR followed the principle of filter wheel radiometers [Shaw et al., 1973; Moorthy et al., 1989, 2007b]. The instrument details, AOD retrieval method, and error budget have been discussed in several earlier papers [Moorthy et al., 2007a, 2007b; Gogoi et al., 2009; Kompalli et al., 2010 being the recent ones]. Continuous measurements of direct solar flux at 10 narrow wavelength bands centered at 380, 400, 450, 500, 600, 650, 750, 850, 935, and 1025 nm are analyzed following Langley plot technique [Shaw et al., 1973; Moorthy et al., 2007a, 2007b] and AODs are estimated by subtracting the contribution due to molecular scattering and absorption from the total optical depth (obtained as the slope of the Langley plot).

[8] The narrowband interference filters with full width at half maximum bandwidth of 5 nm are used for the selection of spectral bands. A three cavity configuration ensured a near-uniform transmittance within the passband and a sharp reduction in the transmission beyond. The band-selected radiation is passed through a field-limiting optics that limits the total field of view of the MWR close 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 detector is proportional to flux incident at the entrance window, over several orders of intensity variations. The output is digitized using a 12 bit ADC and recorded along with GPS time and coordinates. The interference filters were periodically replaced either when deterioration in transmittance is noticed in the data or once in 3 years, whichever occurred earlier.

[9] Spectral AODs are estimated on all clear or partly clear days when unobstructed solar visibility is available for more than 3 h. In cases when the data length is confined to one half of the day, the entire measurements are considered as a single set and the mean AOD is deduced for the day. However, on days when the Langley plots showed distinctly different slopes for the forenoon (FN) and afternoon (AN) parts of the same day or on days where data spanned over the entire day, AODs are deduced separately for the FN and AN parts. 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. 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). The long-term stability of the instrument has been ascertained from the Langley intercepts, and any data showing variabilities > 5% have not been considered for the analysis.

[10] Extensive intercomparisons of AODs estimated using MWR with other commercial Sun photometers such as Multifilter Rotating Shadow Band Radiometer (Yankee Environmental Systems, USA), calibrated Microtops Sun photometer (Solar light Company, USA), and EKO Sun photometer (Model MS-120 of EKO Instruments trading Company Ltd, Japan) have been made at the ARFINET locations TVM and PBR as well as onboard ship cruises during the Indian Ocean Experiment. These have shown very good agreement with correlation coefficient of ~ 0.99, 0.88, and 0.92 and root-mean-square difference of 0.03, 0.05, and 0.04, respectively [Kompalli et al., 2010]. As such, AOD from such calibrated Microtops has also been used at some stations.

2.1 Statistical Technique Used for the Trend Analysis

[11] Several statistical methods are available to quantify trends in environmental and atmospheric state variables (e.g., total ozone, stratospheric temperature, UV radiation, or AOD) starting from simple linear to multiorder polynomials [Tiao et al., 1990; Weatherhead et al., 1998; Reinsel et al., 1999; Newchurch et al., 2000; Zhang and Reid, 2010]. It is the common practice to model these variables or trends as being smooth changes. The linear trend analyses have the practical advantage that it allows a simple approximation of the direction and magnitude of changes in the data. After considering several methods for trend analysis, the Intergovernmental Panel on Climate Change (IPCC) has also adopted the linear trend analysis in its fifth assessment report. Accordingly, as a first step, the monthly mean time series of AOD has been subjected to linear regression analysis to estimate the linear trend (year−1) in AOD. However, this procedure gives equal weightage to all the monthly means and thus could be significantly biased by the statistically ill-proportioned averages calculated from a smaller number of cloud-free observations (for example, rainy and cloudy seasons) [Yoon et al., 2011]. In the present study, we have estimated the statistical significance of the estimated trends following the method discussed by Weatherhead et al. [1998] as detailed below.

[12] In general, statistical trend analyses of time series data of geophysical or environmental variables over the years use the following form (equation (1)) of time series trend model [e.g., Tiao et al., 1990; Weatherhead et al., 1998; Reinsel et al., 1999; Newchurch et al., 2000; Zhang and Reid, 2010; Yoon et al., 2011; de Meij et al., 2012; Hsu et al., 2012]. Let Yt denote the monthly mean values of the geophysical variable (in our case AOD at the selected wavelength; 500 nm) at a time t (t = 1,2,…..N; where N is the total number of monthly mean data points). Then following the simple trend model of Weatherhead et al. [1998], Yt can be expressed as

display math(1)

where μ is an offset term representing the base value (AOD at the beginning of the time series), ω is the trend (year−1), Xt is the independent variable representing time in year, and εt is the noise on the time series (residuals), which is represented by a first-order autoregressive model

display math(2)

where φ is the autocorrelation coefficient (−1 < φ < 1) and ct is the white noise which is totally random. The trend estimated by using the least square fit of Yt versus t would yield ω, and its standard deviation, σω calculated as [Weatherhead et al., 1998]

display math(3)

where σε is the standard deviation of the noise (ε) and N is the total number of years. The significance of the trend is assessed using the ratio |ω/σω| (i.e., the absolute trend relative to its uncertainty). The trend is considered significant at a 5% significance level or 95% confidence level when this ratio is greater than 2 [Tiao et al., 1990; Hsu et al., 2012]. Unless mentioned otherwise, the term “significant trend” in this study refers to such trends.

3 Results and Discussion

3.1 Annual Trends in AOD and Their Subregional Distribution

[13] The spatial variation of annual trend in AOD at 500 nm, estimated as discussed earlier, is shown in Figure 2 and the detailed statistics are given in Table 2. Over most of the stations, AOD shows an increasing trend, except at ATP (semiarid) and GC (rural location in the IGP) where the trends are weakly negative. Among the stations considered here, TVM and VSK had the longest database; spanning more than 20 years and the temporal trend in AOD at these stations is shown in Figure 3 for the 500 nm wavelength. The slope of this line gives the trends as 0.009 year−1 with ω/σω of 5.56 over TVM (over the base value of 0.26 at 1986) and 0.0104 year−1 with ω/σω of 4.52 for VSK (over the base value of 0.30 in 1988, MWR data from 1988 to 2008, and MTOPS data from 2009 to 2011). The high values of ω/σω in these two cases indicate high-significance levels of the respective trends. Obviously, both the trends in AOD500 as well as the base value are higher at VSK than those over TVM, implying that the aerosol loading over VSK remains always higher than that over TVM. This is mainly attributed to the industrialized nature of VSK even far back in 1980s. Over MYS (a continental station where the AOD measurements were initiated in 1989 and discontinued after 1999), the trend in AOD during the period from 1989 to 1999 was 0.0057 year−1 over the base value of 0.32 in 1989.

Figure 2.

Spatial variation of annual trend in AOD (year−1) over Indian region. Each station is represented in the figure by a short three letter symbol inscribed in the colored circle where the color indicates the annual trend in AOD expressed in year−1following the color scheme at the right. Shades of green, orange, red, maroon, and black indicate the increasing trends in the order, while shades of blue decreasing trend.

Table 2. Detailed Statistics of Trend in AOD at 500 nm Over Indian Region From ARFINET Stations
StationStarting Year Base AOD (or y Intercept) at Starting YearRegression Coefficient (R)Slope (ω, AOD year−1)Standard Deviation of ω (σω)ω/σωTrend (% year−1)
  1. The trend estimated for the AERONET stations available over the Indian region is also included for comparison. Bold text highlights the annual trend in AOD shown in Figure 2. Asterisk (*): discontinued after 1999.

TVM (8.5°N, 77.0°E, 2 m amsl) (Rural coastal location on the west coast of the southern tip of India)1986Annual0.260.4210.00890.00165.563.47
DJFM0.270.6410.00920.0019.193.39
AM0.270.2090.00370.00162.311.36
JJAS0.270.4610.00730.00116.632.69
ON0.270.5870.00870.00165.433.2
2000Annual0.370.2310.0110.00472.342.97
DJFM0.380.2360.00640.00183.551.69
AM0.380.3040.01230.00641.923.25
JJAS0.380.2280.00950.00432.22.51
ON0.380.2930.00930.00611.522.46
VSK (17.7°N, 83.1°E, 20 m amsl) Industrialized coastal location on the east coast of India1988Annual0.300.4520.01040.00234.523.47
DJFM0.420.5850.01020.00176.002.42
AM0.420.4500.00930.00234.042.20
JJAS0.420.3780.00510.00143.641.21
ON0.420.5120.00730.00193.841.73
2001Annual0.340.5210.02660.00743.597.91
DJFM0.490.3330.0130.0052.592.65
AM0.490.7740.03110.00853.656.34
JJAS0.490.9480.04750.01892.519.78
ON0.490.8690.03550.01033.447.31
MYS* (12.3°N, 76.5°E, 772 m amsl) Industrialized urban location in central peninsula1989Annual0.320.1820.00570.00232.481.8
DJFM0.250.0270.00050.00140.350.19
AM0.330.5550.01010.00175.943.05
JJAS0.330.362−0.01350.00265.19−4.09
ON0.320.470.0290.00833.498.92
ATP (14.62°N, 77.65°E, 331 m amsl) Rural semiarid location in central peninsula2001Annual0.440.0680.00320.00510.630.72
DJFM0.370.124−0.0030.00221.36−0.8
AM0.380.1760.00760.003822.01
JJAS0.360.401−0.01730.00672.58−4.85
ON0.380.260−0.00710.00242.95−1.88
HYD (17.75°N, 78.73°E, 557 m amsl) Industrialized urban location close to central India2003Annual0.510.220.01270.00512.492.46
DJFM0.490.4310.01380.00255.522.8
AM0.480.960.02920.01661.756.07
JJAS0.480.5150.01870.00424.453.87
ON0.480.598−0.01960.00474.17−4.07
PTL (30.33°N, 76.46°E, 249 m amsl) Urban location on the western part of northern India2006Annual0.480.1080.00880.00791.111.84
DJFM0.430.7310.04050.01582.569.42
AM0.430.761−0.02130.01531.39−4.99
JJAS0.430.121−0.00420.00271.55−0.97
ON0.430.013−0.00040.00470.08−0.09
DBR (27.3°N, 94.6°E, 111 m amsl) Rural location in the far eastern part of northern India2002Annual0.30.210.01350.00512.654.49
DJFM0.290.4970.01970.00484.16.79
AM0.290.4430.01550.00315.05.34
JJAS0.270.2070.00720.00252.882.62
ON0.430.013−0.00040.00470.08−0.09
PBR (11.64°N, 92.71°E, 60 m amsl) Island location in Bay of Bengal2002Annual0.290.2290.00970.00442.23.38
DJFM0.260.4310.00980.00244.083.82
AM0.290.8200.03550.01712.0712.41
JJAS0.340.740.01510.00652.324.49
ON0.180.503−0.02710.00634.3−14.78
Kanpur (26.4°N, 80.3°E, 50 m amsl) Industrialized urban location in IGP (AERONET Station)2001Annual0.570.1560.00890.00461.9341.56
DJFM0.540.4430.0140.00383.682.6
AM0.540.067−0.00150.0020.75−0.27
JJAS0.540.0220.00040.00190.210.07
ON0.530.5140.01860.00394.763.53
MCOH (6.8°N, 73.2°E) Island location in northern Indian ocean (AERONET station)2005Annual0.220.3840.02080.00623.3549.38
DJFM0.20.6270.01540.0053.087.6
AM0.20.8520.02620.00858−6.98
JJAS0.200.1650.00650.00125.413.2
ON0.200.5140.01860.00394.763.53
Figure 3.

Long-term temporal variation of AOD at 500 nm over TVM and VSK. Each circle represents the monthly mean AOD at 500 nm and the solid line through the points represents the linear least squares fit indicating the long-term trend in AOD, its slope yielding the trend (year-1). The gap in the data from July 1991 to December 1993 is due to eliminating the AOD values for this period because of the significant modification by the Pinatubo volcanic eruption, the only strongest natural impact during the period under study.

[14] The remaining stations, where AOD measurements were initiated on or after 2000, also depicted increasing trends by varying extent as given in Table 2. With a view to examine the rate of increase at TVM and VSK during the period when the other stations also had data, the database since 2000 was analyzed separately and the trends are estimated. It is interesting and important to note that at both these stations, the rate of increase has significantly increased in the last decade from 0.009 year−1 to 0.011 year−1 at TVM and from 0.0104 year−1 to 0.0266 year−1 at VSK. This indicates that the AOD (and hence aerosol columnar loading) over the peninsula is not only increasing steadily, even the rate of increase also has increased, the situation being more alarming over the industrialized eastern coastal location (VSK).

[15] During the same period (2002–2011), statistically significant (with | ω/σω| > 2) values of increasing trend in AODs at 500 nm are seen over HYD (0.0127 year−1 from the base value of 0.51 in 2003), DBR (0.0135 year−1 from the base value of 0.3 at 2002), and PBR (0.01 year−1 from the base value of 0.29 at 2002), and increasing trend, but with low-statistical significance (|ω/σω|< 2), over PTL (0.0088 year−1 from the base value of 0.48 at 2006) and KNP (0.0089 year−1 from the base value of 0.57 at 2001).

[16] Despite presenting a gloomy picture of increasing atmospheric opacity over several stations across India, the rate of increase that has emerged from the study remains comparable to (and sometimes higher than) the values reported for several regions in several recent investigations. The global mean annual trend in AOD is reported to be weakly positive (0.00078 ± 0.00019 year−1) [Hsu et al., 2012] which is based on SeaWiFS data from 1998 to 2010, in contrast to the decreasing trend (−0.0014 year−1) reported by Mishchenko et al. [2007] using the advanced very high resolution radiometer data for the period 1991–2005. The increasing trend reported by Hsu et al. [2012] is statically significant (|ω/σω|= 4.21), with a value of 0.00080 ± 0.00019 year−1 over global oceans while over the land, the reported trend of 0.00058 ± 0.00041 year−1, interestingly lower than that over oceans but with a lower statistical significance (with |ω/σω| of 1.39). They have reported an increasing trend in AOD over northern Indian region (0.0063 ± 0.0020 year−1 with |ω/σω| of 3.2), southern China (0.0049 ± 0.0018 year−1 with |ω/σω| of 2.69), Arabian Peninsula (0.0092 ± 0.0013 year−1 with |ω/σω| of 7.8), and Africa (Sahel) (0.0049 ± 0.0031 year−1 with |ω/σω| of 1.55), all are several times higher than the global mean values. Examining regional and global AOD trends over oceans using 10 years (2002–2009) of satellite-derived AOD (at 550 nm) products, Zhang and Reid [2010] have reported a systematic difference in the decadal trends. While MODIS data showed a consistent increase in AOD at the rate of 0.01 per decade, MISR did not show any significant trend. They have reported a negligible global trend of ± 0.0003 year−1 over global oceans. However, regionally, they found statistically significant increasing trend in AOD over Bay of Bengal (0.0069 year−1, comparable to the value 0.01 year−1 over PBR an island station in BoB), Arabian Sea (0.0065 year−1), and coastal China (0.062 year−1). Analyzing the long-term global AERONET data, for the period from 1998 to 2009, Yoon et al. [2012] have reported clear increasing trends (in AOD at 440 nm) over Beijing (0.00717, + 1.06%), Mauna Loa (located at 3397 m amsl, representative of free tropospheric aerosols, 0.00049 year−1 (+ 3.16%)), Mongu (0.01580 year−1, 5.7%), and Solar Village (+0.01881 year−1, + 6.12% ).

3.2 Seasonality

[17] The seasonally changing synoptic-scale meteorology is known to modulate the AOD periodically (intraseasonal/seasonal/annual and interannual time scales) through advection and precipitation [Moorthy et al., 2007a; Gogoi et al., 2009]. Over Indian region, the most important characteristic feature of synoptic meteorology comprises of seasonally changing air mass type and precipitation associated with the Asian summer monsoon from June to September. During the period from December to March, the prevailing winds are generally northeasterlies, from the inland continental regions toward the Indian Ocean, constituting a continental, dry air mass. The precipitation is rather insignificant, RH is low, and temperatures are in general low to moderate, except over the southern peninsula. During June to September, the steady onshore wind brings in moist marine air from the Indian Ocean over the continent through the Arabian Sea and Bay of Bengal, constituting the marine air mass. October–November and April–May months mark the transition from the moist marine air mass to dry continental air mass and vice versa. During the dry continental air mass period (DJFM), there is little rainfall or deep convection over Indian region and the sky is generally cloud free whereas the wet marine air mass period, also known as the summer monsoon period (JJAS), is characterized by widespread rainfall over most of the Indian region. Around 80% of the average annual rainfall (~ 1182 mm) over Indian region occurs during this period whereas the rest is contributed, almost equally, by premonsoon and post monsoon rainfalls. In addition, the generally cloudy/overcast skies during the monsoon seasons restrict the AOD measurements and the number of datasets during this period is fewer than that during DJFM period. Though there are significant regional variations in the quantum of rainfall received in each season, the seasonality in air mass and precipitation pattern is more or less maintained, except perhaps in the south and southeastern part of the peninsula where the significant rainfall occurs during the return monsoon from October to December also.

[18] In view of the strong modulations on the regional aerosol loading and aerosol types by the synoptic Asian monsoon system, we examined the seasonality in the observed trends over India. For this, we considered two contrasting seasons: the dry continental air mass period from December to March (DJFM) and the humid marine air mass period from June to September (JJAS). The results are given in Table 2 and in Figure 4 along with that for whole year for comparison. Interestingly, while at TVM, a near-equatorial coastal station, the AOD trends did not show a significant seasonal shift from DJFM (0.0092 year−1) to JJAS (0.0073 year−1), over VSK, the magnitude of the trend was significantly higher during DJFM (0.0081 year−1) compared to JJAS (0.0014 year−1) despite both being coastal locations. However, during the AM period, the AOD trend over TVM reduces to 0.0037 year−1, nearly half of its value during DJFM and JJAS periods. A small reduction in AOD trend during AM (April–May) period compared to that during DJFM period is also observed over VSK.

Figure 4.

Seasonal changes in the spatial variation of AOD trend over Indian region.

[19] Moving over to the semiarid, semirural continental location, ATP, the AOD trend has been consistently negative (except during AM periods), more negative (−0.017 year−1) during the monsoon season (JJAS period) compared to that during winter (DJFM) period (−0.003 year−1). In contrast to this, a strong positive trend is seen over the urban location, HYD, in central peninsula, about 300 km north of ATP, especially during the JJAS period (0.0187 year−1).

[20] Over the Indo-Gangetic Plains in the north, AOD shows consistent increasing trend during DJFM period from Patiala to Dibrugarh. The trends are very weak or inconspicuous in other seasons. Over PTL, the AOD trend is found to be positive and significantly high during DJFM period (0.0405 year−1) and negative during rest of the seasons. Over KNP, the increasing AOD trend during DJFM period (0.014 year−1) changes to a weak negative (AM period, −0.0015 year−1) or negligible trend (0.0004 year−1) during JJAS period. Over DBR, the AOD trend is found to be consistently positive (except during ON period) with high value during the DJFM period (~ 0.02 year−1) and low during JJAS period (0.0072 year−1). Surprisingly, the AOD trend over the AERONET station located at Gandhi College, Ballia in IGP, showed consistently negative trend in all the seasons.

[21] At this juncture, it would be interesting to examine the trends over Indian mainland with that over the oceanic regions adjoining it. Examining the data from MCOH in the northern Indian Ocean, which is under the influence of continental flow mainly from Indian landmass during DJFM, while under the influence of cleaner marine air mass during JJAS, the analysis shows that there is a steady increase in AOD (0.0154 year−1) in winter and a surprisingly higher rate of increase, 0.0262 year−1, during the premonsoon months compared to the very low trends during JJAS period (0.0065 year−1). During the JJAS period, the prevailing wind over the Indian Ocean is strong southwesterly and is directed from the ocean to land. Hence, the major source of aerosols over the island station, MCOH, during this period is sea-salt production by wind. The positive sign of the trend during JJAS is interesting as the prevailing marine air mass would imply an increasing trend in natural (sea-salt) aerosols.

[22] Similar seasonality in trends has been reported by a few others recently. Based on MISR observation for the period from 2000 to 2010, Dey and Di Girolamo [2011] have reported similar seasonal shift in the AOD trends over Indian region while based on MODIS data for the period (2000–2006) over Mediterranean basin; Papadimas et al. [2008] have reported the seasonal changes in AOD trend with a decreasing trend in summer (−0.0014 year−1) changing to an increasing trend (+0.0012 year−1) during winter.

3.3 Discussion

[23] This is the first attempt to examine the long-term trends in AOD by analyzing long-term AOD data from ground-based Sun photometer data from a network representing diverse geographical and environmental regions over the Indian mainland and adjoining oceans. Our study has revealed a generally increasing trend in the AOD (in the range 0.008 to 0.02 year−1) except at two locations. The trends are significant and high over peninsular region where the data go back to 1980s and the rate of increase also appears to increase in the recent decades. Seasonally, the rate of increase is consistently high during the dry DJFM period (winter season) over the entire domain. This is also the season when the effect of natural aerosols (marine aerosols and mineral dust) is least abundant over this region and aerosols have longer lifetime in the atmosphere due to weaker wet removal processes, lower dispersion by thermal convections, and increased confinement by the shallow winter atmospheric boundary layer capping. The stronger winter conditions along with the low-level anticyclone prevailing over central India during this season is most conducive for inhibiting vertical dispersion and keeping the ventilation coefficient low. Winds are also low during this season. The effect would be especially strong over the northern Indian plains, where the minimum temperatures dip to very low values during winter and foggy conditions occur frequently. The orography of the region with Himalayas on the north and the Vindhya-Satpura range and Chota Nagpur plateau to the south providing a channel over the IGP which narrows from west to east adds to the confinement. Additionally, fuel use for domicile purposes including keeping the environment warm also contributes to high-aerosol loading in this season. As such, anthropogenic aerosol species are likely to dominate during this season. Other studies [Moorthy et al., 2005; Nair et al., 2007; Moorthy et al., 2013b] have also shown that the wintertime aerosols over Indian region are dominated by accumulation mode aerosols and the contribution of BC to the total aerosols is ~ 10% or more (the annual highs). Thus, the increase in aerosol load over Indian region during DJFM period could be logically attributed to increased anthropogenic emissions (which continues to grow), favorably synoptic and mesoscale meteorological conditions, and the orography.

[24] In contrast to this, the weak and inconsistent/inconspicuous trends in AOD during AM and JJAS (premonsoon and monsoon seasons) appear to be more controlled by natural sources such as transported aeoline dust over the IGP and marine aerosols over peninsular Indian region, offset by the extensive wet removal by the monsoon rain fall system. As both these processes exhibit large interannual variations at regional and subregional scales, which are not consistent (in space/time), the net impact on aerosol loading (AOD) would be inconsistent over the years and this large variation marks any trend, if at all present. Similar results for the dust-dominated season have been reported by Hsu et al. [2012] for the dust-dominated season over eastern China and this has attributed to the large interannual variability of the dust load.

[25] It is important to note that significant transport of Arabian dust to the Indian region leading to an increase in the AOD during AM period has been reported by several investigators [Moorthy et al., 2005; Niranjan et al., 2007; Beegum et al., 2008]. It is especially interesting to note that the AOD trends over the IGP stations (PTL and KNP) are either negative or negligible during these seasons, when the columnar AOD over this region is significantly high and influenced by intense dust activity. The net aerosol loading over this region is the resultant of local production and strong advection; competing with the wet removal, the trends become insignificant. To better delineate these aspects, we analyze the spectral variation of AOD and examined the trends.

3.1.1 Natural Versus Anthropogenic

[26] Spectral variation of AOD contains information on columnar size distribution of aerosols. In general, over most of the ARFINET stations, the AOD depicted a significant spectral variation with higher values at shorter wavelengths and falling off toward longer wavelengths. A steeper spectral dependence implies relative dominance accumulation mode aerosols (generally associated with anthropogenic activities) as compared to cases with more flat spectral dependence where coarse mode dominance increases. The smaller particles have comparatively longer lifetime in the atmosphere compared to the coarser-sized aerosols which are mostly associated with natural sources. With this in view, we have examined the annual trends in AOD at 400 nm, 500 nm, 750 nm, and 1025 nm (four wavelengths, two in the visible and two in the near IR) for different stations and the results are shown in Figure 5. In general, the rate of increase (per year) in AOD is found to be higher at the shorter wavelengths, even though exceptions are also noticed. Interestingly, the AOD trend is negative at two industrialized urban locations (VSK and HYD) at the longest wavelengths.

Figure 5.

Spectral changes in AOD trend over ARFINET stations.

[27] With a view to quantifying this, the Angstrom wavelength exponent α has been estimated from the individual AOD spectra by least square fitting the relation [Angstrom, 1964]

display math

where τp(λ) is the AOD at wavelength λ, α is the Angstrom wavelength exponent, and β is the turbidity coefficient (numerically equal to the AOD at 1 µm), a measure of the columnar aerosol loading. The long-term variation of α estimated from the spectral AOD for TVM and VSK (stations with longest database and contrasting human activities) are shown in Figure 6, respectively, in the top and bottom panels. While the positive trends in α implies steepening of AOD spectra due to increasing abundance of accumulation mode particles, linked to anthropogenic sources, negative trends suggest increase in the relative abundance of natural coarse mode aerosols or reduction in human activities. A consistently increasing trend (implying the increasing anthropogenic impact) is clearly discernable in the bottom panel, at VSK, with α increasing at the rate of ~ 0.038 ± 0.004 year−1. This trend is nearly 4 times higher than the corresponding trend over TVM (~0.011 ± 0.003 year−1) where there has been no increase in industrial activities over the last four decades; the only discernible change is the increasing urbanization and number of automobiles. The trend in α (estimated as above) across the ARFINET stations is shown in Figure 7. Except for the semiarid stations, HYD and ATP, and the northern Indian station, PTL (where significant dust intrusion occurs) and DBR (a station in the northeastern region surrounded by tall mountains on all the three sides with an opening to the outflow from the IGP) showed a positive trend in α. On the other hand, the stations prone to significant dust impact (either advected or local) showed a decreasing trend in α. The decreasing (negative) trend is in-line with a similar trend reported by Hsu et al. [2012] and Yoon et al. [2012], based on the analysis of AERONET data from Solar village (a station in the Arabian Desert).

Figure 6.

Long-term temporal variation of α over TVM and VSK. The solid spheres represent the monthly mean values of α and the dashed lines through the points represent the liner least square fit, the slope of which provides the trend in α.

Figure 7.

Spatial variation of trend in Angstrom exponent (α).

3.1.2 Anthropogenic Impact

[28] There are several possible reasons for the observed increase in anthropogenic impact. The population in India increased 185% during the last 50 years from 434.9 million in 1961 to about 1241.5 million in 2011 (http://censusindia.gov.in). During this period, the Indian economy also witnessed significant improvement, with the gross domestic product growth jumping from 4% in 1960s to 8.4% in 2006–2011, most of this increase happening in the last three decades. The total number of vehicles in India (automobiles) grew enormously from 0.4 million in 1960 to a whooping 142 million in 2011, which directly leads to increased vehicular emissions (despite improved control measures) and indirectly to increased need for power generation to support these manufacturing industries and the other supplementary needs. The large increase in population sets a proportionate increase in the demand on energy for domicile needs, fertilizer use to improve crop production, and change in land use pattern. All these should invariably lead to increased emissions, including particulate, which would reflect in the AOD trends.

[29] At this juncture, it is very interesting to examine the present-day atmospheric turbidity levels with the historical records available from measurements made in India more than four decades ago. Mani et al. [1969] have made extensive study of Angstrom turbidity coefficients (very close to AOD at 1025 nm) over several locations in India based on the data collected during 1966–1967 using Volz Sun photometers. A comparison of the annual mean AOD at 1025 nm retrieved from the post 2006 spectral AOD data over the ARFINET locations (except over JDR where the data were collected during 1999–2001) coinciding with the locations for which Angstrom turbidity coefficients (β) have been reported by Mani et al. [1969] is shown in Figure 8. It not only puts the change in the last nearly 50 years in perspective but also reveals the alarming situation; with phenomenal increase in the turbidity coefficient (β). Assuming the increase to be linear over the four decades, the rate of increase in β over TVM and VSK from 1966 to 2006 is estimated as 0.006 year−1 and 0.003 year−1, respectively. The recent (1986–2011) values of long-term trend in AOD1025 (~AOD at 1 µm) over TVM and VSK are 0.005 year−1 and −0.009 year−1, respectively. While the value is very much comparable to the long-term average over TVM, it appears to have changed largely over VSK; even the sign of the trend has changed from increasing to decreasing. In this context, it is also important to note that over the last 50 years, there have been no increase in industrial activities in and around TVM, the development being limited to increase in the transportation sector, urbanization, and population whereas there has been consistent increase in the industrial activities (steel smelting plants, thermal power plants, and several small-scale industries) at VSK.

Figure 8.

Intercomparison of the annual mean values of turbidity coefficient estimated using the spectral AOD obtained over ARFINET stations after 2006 with that reported over same location by Mani et al. [1969] based on the data collected during 1966–1967 period.

[30] In this context, it is important to keep in mind that during the period 2000 to 2009, the stratospheric AOD also showed a significant increasing trend within the latitude bands of 50°S to 50°N. The long-term trend in stratospheric AOD (between 20 and 30 km), deduced from Stratospheric Aerosol and Gas measurements II, Global Ozone Monitoring by Occultation of Stars, and Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation, for the latitude bands of 50°N–20°N, 20°N–20°S, and 20°S–50°S, respectively, are 0.0001 year−1, 0.0002 year−1, and 0.00007 year−1, respectively [Vernier et al., 2011]. They attributed this increasing trend to an increase of SO2 entering the stratosphere associated with coal burning in Southeast Asia.

3.1.3 Goddard Chemistry Aerosol Radiation and Transport (GOCART) Simulations

[31] At this point, it would be interesting to examine how the chemical transport models simulate the long-term trend scenarios. For this we have used the Goddard Chemistry Aerosol Radiation and Transport (GOCART) model data. The GOCART model is a global chemical transport model based on the compilation of emission inventories of black carbon (BC), primary organic carbon (OC), and SO2 emissions from land-based anthropogenic sources, marine traffic, air traffic, biomass burning, and volcanoes for the period 1980–2000 by Diehl et al. [2012]. It simulates the global distribution of AOD from various aerosol types, such as sulfate, dust, sea-salt, and carbonaceous (OC and BC) aerosols, from which the composite AOD is estimated as the sum of the AOD due to each component. The details of GOCART model and the validation of GOCART model against observation are given in Chin et al. [2002]. Over the Indian region, extensive evaluation of BC aerosols simulated using GOCART was carried out using ARFINET data by Moorthy et al. [2013b]. The monthly mean AODs at 500 nm simulated using the GOCART and available at http://gdata1.sci.gsfc.nasa.gov/daac-bin/G3/gui.cgi?instance_id=aerosol_monthly for the period January 2000 to December 2007 have been used to estimate trend in AOD. Using these data, the annual trends of the AODs of the individual aerosol species as well as the composite AOD at each of the stations considered in this study are estimated. The results are shown in Figure 9. In general, the composite AOD simulated by the GOCART showed increasing trends though with varying magnitudes.

Figure 9.

Trends in AOD of different aerosol species at the ARFINET stations, as simulated by GOCART. The trends in composite AOD as simulated by model and obtained from observation are shown, respectively, with star and filled circle symbols.

[32] Over TVM and PBR, the trends in composite AOD from GOCART simulation are comparable to those in the observed AOD. Over VSK, the trend in the GOCART-simulated AOD is 1.32% and that in the observed AOD during the periods from 1988 to 2011 and 2001 to 2011 are 1.73% and 6.94%, respectively. At ATP, in contrast to the negative trend in the observed AOD, trend estimated from the GOCART-simulated composite AOD showed a positive trend (2.83%) with significant value of |ω/σω|. Over HYD, DBR, and KNP, the magnitude of the trend in the GOCART-simulated AODs (1.01%, 0.8%, and 0.58%, respectively) is significantly less than that in the observed AODs (2.46%, 4.49%, and 1.56%, respectively).

[33] Examining the species-specific AODs simulated by GOCART, the sulfate and BC AODs depicted strong increasing trend over all the stations. Modeling study by Streets et al. [2006] has indicated that the trends in surface radiation are correlated with the trends in the emissions of sulfur dioxide, BC, and AODs. Based on the analysis of emission inventories (EMEP, REAS, and IPCC inventories) and AODs derived from satellites (MODIS and MISR) and AERONET, de Meij et al. [2012] reported that the AOD trends are generally dominated by changing anthropogenic SO2 emissions, especially over Europe and Asia. Lu et al. [2011] also reported upward trends in fine anthropogenic aerosols over China and India using emission inventories for sulfate and carbonaceous aerosols. However, a decreasing trend in dust AOD is observed over all the stations in the GOCART simulation. Here it is to be noted that the AOD trend over the stations in the IGP also showed either negative or negligible trend during the period, AM and JJAS months, when the dust sources are active. Similarly, the semiarid station, ATP, also showed a decreasing trend in AOD. Thus, the negative trend in AOD over these stations during certain period (AM and JJAS months) appears to be strongly associated with the significant contribution of dust AOD to the total.

3.1.4 Indicators of Possible Climate Impact

[34] Climate impact simulations have shown that increased aerosol abundance may lead to change in the hydrological cycle and precipitation [Ramanathan et al., 2005; Lau et al., 2006]. With a view to exploring any such indications over the Indian domain, we analyzed the long-term daily gridded rainfall data at 1° × 1° resolution (based on the measurements at 2140 stations in India for the period 1951–2004) from the India Meteorological Department (IMD) [Rajeevan et al., 2006]. The results, shown in Figure 10, revealed some interesting results. A decreasing trend in the number of rainy days over most of the study regions is noticed and the possible role of positive feedback effect of this leading to the significant increasing trend in AOD during the DJFM period. Recently, several investigators have reported an enhancement in premonsoon rainfall over northern India and the Himalayan foot hills regions based on IMD database [Lau and Kim, 2010; Gautam et al., 2009]. However, the amount of summer monsoon rainfall did not show any significant long-term trend; rather, it showed interannual variations. Nevertheless, Goswami et al. [2006] have reported that the moderate rain (from 5 mm/day to < 100 mm/day) events during the summer monsoon showed a decreasing trend which is partly offset by the increasing trend in heavy rain events (≥ 100 mm/day), making the trend in seasonal total rainfall insignificant. Over the peninsular India, significant increase in the advected coarse mode aerosols of marine origin associated with the strong southwesterly winds during the June to September period is observed [Moorthy et al., 2007a]. Thus, the aerosol trend over peninsular India during the JJAS period is a combined effect of local production, marine advection and wet removal.

Figure 10.

Time series of the number of rainy days over ARFINET station during the DJFM period.

3.1.4.1 Reduction in Direct Solar Flux

[35] Based on the spectral AOD data made over the wavelength range from 380 nm to 1025 nm, the broadband AOD380–1025 values are estimated. The long-term variation of the regional average of broad band AOD380–1025 during 1995 to 2012 is shown in Figure 11. The slope of the linear fit yields the long-term trend in broadband AOD380–1025 as 0.004 year−1 for a base value of 0.28 in 1995. This translates to 0.4% year−1 decrease in the direct short wave (380 nm–1020 nm) radiation flux over Indian region.

Figure 11.

The long-term variation of all India average of broadband AOD380–1025 during 1995 to 2012.

4 Summary and Conclusion

[36] Long-term spectral AODs measured from a network (ARFINET) of observatories spread across India are analyzed to quantify the long-term trends with statistical significance. The major outcomes from this study are the following:

  1. [37] Statistically significant and consistent increasing trends in AOD are seen at most of the locations, especially where the data go back to 1980s. However, two stations showed a weakly deceasing trend also.

  2. [38] Examining the recent values of Angstrom turbidity coefficients (β) with those reported from Sun photometric measurements ~ 50 years ago revealed that the trend has been consistent over the decades and a phenomenal increase in aerosol loading has taken place.

  3. [39] Seasonally, the rate of increase is consistent and high during the dry winter months (December to March) over the entire region whereas the trends are rather inconsistent during the premonsoon (April to May) and summer monsoon period (June to September).

  4. [40] A significant increasing trend in the Angstrom wavelength exponent (α) is observed associated with the increasing trend in AOD during DJFM period implying the significant increasing trend in anthropogenic contribution in the building up of aerosols.

  5. [41] The insignificant trend in AOD observed, over the Indo-Gangetic Plain (IGP), a regional hotspot of aerosols, during the premonsoon and summer monsoon seasons is mainly attributed to the combined effects of dust transport and wet removal of aerosols associated with the monsoon rainfall; both being highly varying interannually.

  6. [42] Contributions of different aerosol chemical species to the total AOD simulated using GOCART over the ARFINET stations showed an increasing trend for all the anthropogenic components and a decreasing trend for dust.

Acknowledgments

[43] This work is carried out under the ARFI project of ISRO-Geosphere Biosphere Programme (ISRO-GBP). We thank all the ARFINET investigators for the sustained efforts and support rendered over the years in maintaining the network and collecting data. Special mention is to B. V. Krishna Murthy, who had sown the seeds for this activity in the early 1980s. We thank the AERONET (data available at http://aeronet/gsfc. nasa.gov) PIs and their staff for establishing and maintaining the sites used in this investigation. GOCART model outputs were obtained from the Giovanni web portal. We also thank the three anonymous reviewers for their useful comments and suggestions.

Ancillary